Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

Journal of Power Sources 506 (2021) 230192

Contents lists available at ScienceDirect

Journal of Power Sources


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

A two-dimensional analytical unit cell model for redox flow battery


evaluation and optimization
Yunxiang Chen a ,∗, Jie Bao b , Zhijie Xu a , Peiyuan Gao a , Litao Yan b , Soowhan Kim b , Wei Wang b
a
Physical and Computational Sciences Directorate, Pacific Northwest National Laboratory, Richland, WA, 99354, USA
b
Energy and Environment Directorate, Pacific Northwest National Laboratory, Richland, WA, 99354, USA

HIGHLIGHTS GRAPHICAL ABSTRACT

• A two-dimensional analytical unit cell


model is proposed and widely validated.
• Activation, equilibrium, and pump en-
ergy losses are dominant sources of en-
ergy losses.
• A guideline for reducing activation, equi-
librium, and pump energy losses is pro-
posed.
• Electrode microstructure design depends
on an optimal porosity predicted by the
model.
• The model differs vastly from its 0-
dimension version in energy losses and
efficiency.

ARTICLE INFO ABSTRACT

Keywords: Cell performance optimization is important for improving the system efficiency of a redox flow battery. To gain
Redox flow battery better insights into key controlling factors of system efficiency, this work first proposed a theoretical model
Cell voltage model for a unit cell by extending a two-dimensional analytic model to a full cell. The model is then used for cell
Energy losses
performance optimization after validating it with experimental and numerical modeling data. With the results,
Validation and optimization
the activation, equilibrium, and pump energy losses are identified as the dominant battery energy losses. A
Mass transport
Electrode microstructure design
guideline for reducing these sources is also proposed. Following the guideline, the mass transport coefficient is
shown as a key control factor of equilibrium energy loss and Coulombic efficiency (CE). Approaches are then
proposed to improve CE and system efficiency. The mass transport also controls pump rate optimization by
reducing equilibrium energy loss. An optimal electrode porosity design can further improve a battery’s system
efficiency based on an optimal porosity predicted by our model. The model also demonstrates distinct behaviors
and overestimation in system efficiency when reduced to a zero-dimensional model. With the model, the
guideline, and new insights, this work provides a reliable and efficient tool for the evaluation and optimization
of redox flow batteries.

1. Introduction energies from solar and wind power to existing electric grids [1–
6]. Among existing RFBs [6,7], the vanadium redox flow battery
Redox flow batteries (RFBs) represent an economically viable en- (VRFB) is the most mature and has been commercially constructed
ergy storage technology that can integrate intermittent renewable at megawatt scale [8,9]. Despite such progress, massive deployment

∗ Corresponding author.
E-mail address: yunxiang.chen@pnnl.gov (Y. Chen).

https://doi.org/10.1016/j.jpowsour.2021.230192
Received 22 January 2021; Received in revised form 13 May 2021; Accepted 20 June 2021
Available online 25 June 2021
0378-7753/© 2021 Elsevier B.V. All rights reserved.
Y. Chen et al. Journal of Power Sources 506 (2021) 230192

of RFBs across the globe has not yet been achieved due to the high accuracy of the cell voltage, but significantly underestimates the pres-
costs of design, manufacture, and operation of battery systems [10– sure drop [35,36], and therefore overestimates the system efficiency
12]. Though organic redox electrolytes are believed to significantly when used for predicting cell performance. An improper selection of the
reduce the costs [13–15], their chemical instability and low solubility specific area also affects the local current density and thus changes the
render their large-scale applications in the near term [15]. Therefore, relative importance of equilibrium potential and activation potential to
for given redox materials, optimizing the battery performance in terms the cell voltage. Estimating the effects of mass transport coefficient on
of voltage and system efficiency is the most effective approach for cost cell voltage and performance is another source of arbitrariness. Though
reduction [16]. As the voltage of a single cell unit is the basis for the the mass transfer coefficient can generally be estimated by empirical
stack- and system-scale optimization [6,17,18], developing a reliable formulas linking a Sherwood number to Reynolds number and Schmidt
unit cell model is necessary for accurate prediction, evaluation, and number [16,29,37–42], the fitted coefficients and exponents vary a
optimization of the performance of RFBs at cell, stack, and system lot, which could result in large uncertainty in cell voltage prediction,
scales. especially at the end of discharge [41,43].
Performance at the cell scale is mainly determined by reaction To resolve the above issues and maintain the accuracy as 2D and
kinetics, mass transport, and pump consumption [6,16]. Specifically, 3D models and the computational efficiency as 0D models, this work
the reaction kinetics are affected by the reaction rate constant and integrates a 2D analytic solution to active species concentrations, a
the distributions of current density and active species concentrations new mass transport coefficient model, and a new porosity-specific
at electrode interfaces. The mass transport is usually determined by area relationship [34] to create a new unit cell voltage model (Sec-
the relative importance of electrolyte advection and diffusion at the tion 2). Such a model is then calibrated with cell voltage data from
electrode interface, which is further controlled by the local distribu- two experiments performed at Pacific Northwest National Laboratory
tion of flow velocity and active species concentrations. The velocity (PNNL), a 2D porous media model [26] and an experiment reported
distribution also determines the pressure drop between the cell inlet in the literature [31], to test its wide applicability under different
and outlet, which controls pump energy consumption. Due to the conditions (Section 3). The model’s calibration performance, evaluation
important roles of the local distribution of flow, concentrations, and of sources of potential losses and energy losses, and suggestions for cell
current density, two-dimensional (2D) and three-dimensional (3D) nu- performance optimization are included in Section 4.
merical models that solve the coupled conservation laws of mass,
charge, momentum, and energy with artificial or scanned electrode 2. Model equations
microstructure are the most suitable tools for understanding and im-
proving cell performance [6,17,19]. For instance, the first 2D transient Existing cell voltage models assume uniform concentrations within
model was introduced by [20] who studied the 2D distribution of electrodes, which results in less accurate predictions of equilibrium,
current density, concentrations, and electronic and ionic potentials activation, and concentration potentials that are affected by the con-
using a porous media assumption. Later, similar models were used centration gradient at the electrode–electrolyte interface [31–33]. By
to study the effects of heat transfer [21], applied current density, leveraging a new 2D analytical model for concentrations from our
porosity, mass transfer [22], hydrogen [23] and oxygen [24] evo- previous work [34], this paper aims to derive an analytical unit cell
lution, cross-over and water transport [25], and parameter selection model that enables better predictions of these potentials.
on cell performance [26]. By solving the flow in an X-ray computed
tomography scanned electrode using a lattice Boltzmann method and 2.1. Potentials
solving the concentration and charge transport using a finite volume
method, [27] introduced the first 3D pore-scale model for RFB and Previous studies show that the cell voltage (𝐸𝑐 ) of an aqueous RFB
studied the effects of flow rate, electrode microstructure, and inlet can be represented by six components (Eq. (1)), including the formal
state of charge (SoC) on cell performance [28]. Using a similar method potential of redox couples (𝐸 0 ), chemical potentials of active species
𝑎 ), chemical potentials from water and proton (𝐸 ), activation po-
(𝐸𝑒𝑞
with artificial electrodes, [29] further studied the oxygen evolution 𝑝𝑤
and the relationship between porosity and permeability, specific area, tentials due to charge and mass transfer resistance (𝐸𝑎𝑐𝑡 ), concentration
and effective diffusivity. The results demonstrate the significant impact potentials due to concentration gradients at electrode interface (𝐸𝑐𝑜𝑛 ),
of pore size and distribution on permeability, specific area, and gas and potential losses due to electron transport in current collectors and
evolution. electrodes and proton transport in the membrane (𝐸𝑜ℎ𝑚 ) [31,41,44–
Though 3D models are accurate for predicting flow and transport 46]. The summation of the first two terms is the equilibrium potential
processes, their scales are usually limited at millimeter scale due to attributed to vanadium ions, i.e., 𝐸𝑒𝑞 = 𝐸 0 + 𝐸𝑒𝑞
𝑎 , while the summation

high computational costs [27,29], which questions the efficacy of using of the first three terms is the open-circuit cell voltage, i.e., 𝐸𝑜𝑐𝑣 =
these models to predict and optimize the cell performance at cell and 𝐸𝑒𝑞 + 𝐸𝑝𝑤 [31].
larger scales [30]. The 2D numerical models are less expensive than
𝐸𝑐 = 𝐸 0 + 𝐸𝑒𝑞
𝑎
+ 𝐸𝑝𝑤 + 𝐸𝑎𝑐𝑡 + 𝐸𝑐𝑜𝑛 + 𝐸𝑜ℎ𝑚 (1)
the 3D models, however, their computational time still limits their ef-
fectiveness for cell-scale prediction and optimization [23]. Due to these For a vanadium flow battery, the formal potential is calculated by
issues, the most commonly used numerical models for cell performance the difference of formal potential of VO+ 2
/VO2+ (𝐸𝑃0 ) in the positive
prediction and optimization are zero-dimensional (0D) models that 3+ 2+ 0
electrode and that of V /V (𝐸𝑁 ) in the negative electrode, i.e., 𝐸 0 =
assume the distributions of flow, concentration, and current density are 𝐸𝑃0 − 𝐸𝑁0 .

uniform within the electrode [31–33]. This assumption is usually valid The chemical potentials of active species for a VRFB can be de-
at low current density, high active species concentration, and high flow termined by the chemical potential difference due to reactions of
rate [33]; however, it is questionable at low pump rate and the end of VO+ 2
/VO2+ and V3+ /V2+ in the positive and negative electrodes, which
discharging [34]. Such a limitation is attributed to the large concen- can be mathematically quantified by Eq. (2).
tration gradient of active species at electrode interfaces at a low flow ( 𝑐𝑤 ) ( 𝑐𝑤 )
𝑅𝑇 𝑅𝑇
rate, which affects both the mass transport coefficient and discharging 𝑎
𝐸𝑒𝑞 = ln 5𝑤 − ln 3𝑤 (2)
𝐹 𝑐4 𝐹 𝑐2
time [34], and therefore the cell performance. Besides, existing 0D
models usually manifest large arbitrariness in parameter calibration. where 𝑅, 𝑇 , and 𝐹 denote ideal gas constant, temperature, and Faraday
For example, [31] fitted a value of 420 m−1 for the specific area to constant. 𝑐5𝑤 , 𝑐4𝑤 , 𝑐3𝑤 , and 𝑐2𝑤 represent concentrations of VO+
2
, VO2+ ,
match the measured cell voltage. Such a value does not affect the 3+ 2+
V , and V at the electrode surface (see Section 2.2).

2
Y. Chen et al. Journal of Power Sources 506 (2021) 230192

Fig. 1. A sketch of a unit flow battery cell. The cell voltage is measured as the voltage difference at the two red dots at current collector–electrode interfaces. c and s denote
microchannels and adjacent solids. 𝑐5𝑤𝑖 , 𝑐5𝑤 , and 𝑐5𝑐 denote the concentration of vanadium VO+2 at the wall of cell entrance, and at the wall and centerline of the cell outlet. The
solid black arrow lines denote charge and the dashed black arrows denote discharge.

As proton and water are generated or consumed in the positive by Eq. (7) [41]. The formulas for mass transfer rate (𝑘𝑚𝑁 and 𝑘𝑚𝑃 ) are
electrode, the resultant chemical potentials should be included in cell introduced in Section 2.2.
voltage and can be quantified by Eq. (3) with 𝑐𝐻𝑃 and 𝑐𝑤𝑃 denot-
|𝛥𝑐𝑁 | = |𝐼|∕(𝐹 𝑘𝑚𝑁 ); |𝛥𝑐𝑃 | = |𝐼|∕(𝐹 𝑘𝑚𝑃 ) (7)
ing the concentrations of proton and water in the positive electrode,
respectively [46]. The ohmic potentials are attributed to electron transport resis-
( 𝑐2 ) tances in current collectors and electrodes and proton transport in the
𝑅𝑇
𝐸𝑝𝑤 = ln 𝐻𝑃 (3) membrane, which can be estimated by Eq. (8) [31,46,47].
𝐹 𝑐𝑤𝑃
2𝐿𝑐 𝐿 2𝐿𝑒
The activation potentials are potential losses caused mainly by 𝐸𝑜ℎ𝑚 = 𝐼0 + 𝐼0 𝑚 + 𝐼0 (8)
charge transfer resistance, whose net contribution to the cell voltage 𝜎𝑐 𝜎𝑚 𝜖 3∕2 𝜎𝑒
can be quantified by Eq. (4). where 𝐼0 denotes a nominal current density, calculated by 𝐼0 = 𝐼𝑡 ∕(𝐵𝐻)
( ) ( ) during charging and 𝐼0 = −𝐼𝑡 ∕(𝐵𝐻) during discharging. 𝐵 and 𝐻
𝑅𝑇 𝐼 𝑅𝑇 −𝐼
𝐸𝑎𝑐𝑡 = sinh−1 √ − sinh−1 √ (4)
𝛼𝐹 𝛼𝐹 denote the width and height of the electrode (see Fig. 1). Symbols
2𝐹 𝑘𝑝𝑜𝑠 𝑐4𝑤 𝑐5𝑤 2𝐹 𝑘𝑛𝑒𝑔 𝑐2𝑤 𝑐3𝑤
𝐿𝑐 , 𝜎𝑐 , 𝐿𝑚 , 𝜎𝑚 , and 𝜎𝑒 represent current collector thickness, current
where 𝛼, 𝑘𝑝𝑜𝑠 , 𝑘𝑛𝑒𝑔 denote reaction transfer coefficient, standard reac- collector electronic conductivity, membrane thickness, membrane elec-
tion rate constant in positive electrode, and that in negative electrode, tronic conductivity, and electrode conductivity, respectively. Previous
respectively. 𝐼 is the local current density calculated by 𝐼 = 𝐼𝑡 ∕(𝑉𝑒 𝐴𝑠 ) work shows that 𝜎𝑚 = (0.5139𝜆 − 0.326) exp[1268 ∗ (1∕303 − 1∕𝑇 )] with
for charging and 𝐼 = −𝐼𝑡 ∕(𝑉𝑒 𝐴𝑠 ) for discharging [34]. Symbols 𝐼𝑡 , 𝜆 = 22 [48], while electrode electronic conductivity (𝜎𝑒 ) is provided by
𝑉𝑒 , and 𝐴𝑠 represent the applied current, the volume of positive or the manufacture (www.sglcarbon.com). See the values of parameters in
negative electrode, and the specific area of electrodes, respectively. For Table 1.

a given porosity 𝜖, the specific area is estimated as 𝐴𝑠 = 4𝛼𝐴 ( 𝜖 − 𝜖)∕𝑏𝑠
with 𝛼𝐴 and 𝑏𝑠 denoting a correction factor and average fiber size, 2.2. Concentrations
respectively [34]. The correction factor accounts for the effects of
simplifying a real electrode into multiple microchannels on specific To compute the cell voltage, this section derives solutions to the
area. A value of 1 for 𝛼𝐴 means no correction. concentrations and mass transfer rate as required in Section 2.1. It is
The concentration potentials are attributed to the concentration important to mention that the following equations are derived without
gradient at the electrode interface. As analytical solutions to bulk considering cross-over of active ions and side reactions, which assumes
concentrations of VO+ 2
(𝑐5𝑐 ), VO2+ (𝑐4𝑐 ), V3+ (𝑐3𝑐 ), and V2+ (𝑐2𝑐 ) inside indi- the total concentrations of vanadium ions in positive and negative are
vidual pores have been derived in our previous work (see Section 2.2), independent of time. With such an assumption, the concentrations of
the concentration potential can be estimated by Eqs. (5) and (6) for VO+ 2
, VO2+ , V3+ , and V2+ at electrode interface are given by Eqs. (9)
charging and discharging, respectively [41]. and (10) [34].
( |𝛥𝑐𝑁 | )( |𝛥𝑐𝑃 | ) [ ]
𝑐 𝑅𝑇 5 𝐼𝑃
𝐸𝑐𝑜𝑛 =− ln 1 − 𝑐 1− (5) 0
𝑐5𝑤 = 𝑐45 SoC + ( 𝑃
− 2𝛾)𝐼𝑐𝑑 ∕𝛽 + 2 𝑐𝑢 𝐻∕𝛽 ; 𝑐4𝑤 = 𝑐45
0
− 𝑐5𝑤 (9)
𝛼𝐹 𝑐3 𝑐4𝑐 16 𝑏
( |𝛥𝑐𝑁 | )( |𝛥𝑐𝑃 | ) [ 𝐼 𝑁 ]
𝑑 𝑅𝑇 5
𝐸𝑐𝑜𝑛 = ln 1 − 𝑐 1− (6) 0
𝑐2𝑤 = 𝑐23 SoC + ( 𝑁
− 2𝛾)𝐼𝑐𝑑 ∕𝛽 + 2 𝑐𝑢 𝐻∕𝛽 ; 𝑐3𝑤 = 𝑐23
0
− 𝑐2𝑤 (10)
𝛼𝐹 𝑐2 𝑐5𝑐 16 𝑏
0 and 𝑐 0 denote the total initial concentrations of active
where |𝛥𝑐𝑃 | and |𝛥𝑐𝑁 | are used to estimate the concentration difference where 𝑐45 23
at the interface. In general, such a difference is determined by the species in positive and negative electrode. SoC and 𝛾 represent the state
mass transfer rate near the electrode interface and can be calculated of charge and a constant accounting for the entrance effect. 𝑏 denotes an

3
Y. Chen et al. Journal of Power Sources 506 (2021) 230192

average pore size in the electrode, which is determined by 𝑏 = 4𝜖∕𝐴𝑠 if starting the discharging process. The discharging stage stops at time 𝑡𝑑
simplifying a real electrode as multiple microchannels [34]. 𝛽 is a fitted when one of the concentrations of ions reaches 0 (i.e., a limiting current
parameter to account for the effects of simplifying an actual electrode density scenario). It is necessary to mention that the ending SoC (SoC𝑒 )
into multiple channels [34] and 𝛽 = 1 means no correction. The none- and the waiting time 𝑡𝑤 need to be determined from experiments or a
dimensional parameters 𝐼𝑐𝑑 𝑃 = 𝐼𝑏∕(𝐹 𝑐 0 𝐷 ), 𝐼 𝑁 = 𝐼𝑏∕(𝐹 𝑐 0 𝐷 ), 𝐼 𝑃 = battery’s typical working conditions.
45 5 𝑐𝑑 23 2 𝑐𝑢
0 𝑢𝑃 ), 𝐼 𝑁 = 𝐼∕(𝐹 𝑐 0 𝑢𝑁 ) denote the relative importance of the
𝐼∕(𝐹 𝑐45 0 𝑐𝑢 23 0
driving current density to the diffusion and advection in controlling 3. Experiments and parameters
species concentrations in positive and negative electrodes. Symbols 𝐷5
and 𝐷2 represent diffusivities of VO+ 2
and V2+ , respectively. Symbols To test the efficacy of the proposed cell voltage model under various
𝑢𝑃0 and 𝑢𝑁0
are the average electrolyte velocity and can be calculated conditions, two experiments were performed at PNNL, a computational
by 𝑢𝑃0 = 𝜔𝑃 ∕(𝐵𝐿𝑒 𝜖) and 𝑢𝑁 0
= 𝜔 𝑁 ∕(𝐵𝐿 𝑒 𝜖) with the pump rate for fluid dynamics (CFD) model [26] and experiment data from existing
positive and negative electrode denoted by 𝜔𝑃 and 𝜔𝑁 (see details literature [31], were selected for validation. For convenience these
in [34]). Usually, the pump rates in positive and negative electrode are cases are denoted by A, B, C, and D, respectively. Case A uses graphite
identical and thus 𝜔 = 𝜔𝑃 = 𝜔𝑁 is used to represent the pump rates for felt electrodes and a Nafion® 115 cation exchange membrane with
convenience. vanadium mixed acid electrolyte (2 M VOSO4 + 5 M HCL). Case B uses
Similarly, the bulk concentrations of VO+ 2
, VO2+ , V3+ , and V2+ inside air-heated graphite felt electrodes and a Nafion® 112 cation exchange
individual pores can be determined as Eqs. (11) and (12) [34]. membrane; the electrolyte is 2 M VOSO4 + 5 M HCL plus a small
5 𝑃 0 𝑐 amount of acetic acid. Case C uses carbon felt electrodes and a Nafion®
0
𝑐5𝑐 = 𝑐5𝑤 − 𝐼 𝑐 ; 𝑐 = 𝑐45 − 𝑐5𝑐 (11) 114 cation exchange membrane with vanadium mixed electrolyte (1 ∼
16𝛽 𝑐𝑑 45 4
5 𝑁 0 𝑐 0
1.5 M vanadium + 4 M H2 SO4 ). Case D uses the same electrodes and
𝑐2𝑐 = 𝑐2𝑤 − 𝐼 𝑐 ; 𝑐 = 𝑐23 − 𝑐2𝑐 (12) electrolyte composition as case C though the amount of vanadium is 1
16𝛽 𝑐𝑑 23 3
∼ 1.6 M and the membrane is Nafion® 115. Details of cases C and D
For proton (𝑐𝐻𝑃 ) and water (𝑐𝑤𝑃 ) concentrations in the positive
can be found in [20,26], and [31].
electrode, we assume that their concentrations are uniform within the
To drive the present model, two types of parameters are required.
electrode [31]. With such an assumption, their solutions can be given
The first type can be determined from the battery sizes, operation
as Eqs. (13) and (14) (see derivations in Appendix).
conditions, and basic physical properties of electrode and electrolyte
0 (2 − 𝑛𝐻 )𝐼𝑡 [ (1 + 𝜖𝛿)𝜔𝑃 𝑡 (1 + 𝜖𝛿)𝜔𝑃 𝑡 ] (parameters without superscripts 𝑎 and 𝑏 in Table 1). The second
𝑐𝐻𝑃 = 𝑐𝐻𝑃 + + 1 − exp(− ) (13)
2
𝜔𝑃 (1 + 𝜖𝛿) 𝐹 𝑉𝑡 𝜖𝑉𝑒 type, including reaction rate constant in positive (𝑘𝑝𝑜𝑠 ) and negative
0 (1 + 𝑛𝑑 )𝐼𝑡 [ (1 + 𝜖𝛿)𝜔𝑃 𝑡 (1 + 𝜖𝛿)𝜔𝑃 𝑡 ] (𝑘𝑛𝑒𝑔 ) electrodes, mass transport coefficient (𝛽), and proton transport
𝑐𝑤𝑃 = 𝑐𝑤𝑃 − + 1 − exp(− ) (14)
2
𝜔𝑃 (1 + 𝜖𝛿) 𝐹 𝑉 𝜖𝑉𝑒 coefficient (𝑛𝐻 ), is difficult to measure in experiments and thus needs
𝑡
0 0 calibration (parameters with superscript 𝑎 in Table 1). The model and
where 𝑐𝐻𝑃 and 𝑐𝑤𝑃 denote the initial concentrations of proton and
parameters used for the 4 cases have been implemented in Matlab and
water in the positive electrode. 𝑛𝐻 is a fitted parameter to estimate
the code is available upon request.
proton flux through membrane. 𝑛𝑑 is the water drag coefficient. 𝛿 =
𝑉𝑒 ∕𝑉𝑡 is the ratio of electrode volume (𝑉𝑒 ) to the tank volume (𝑉𝑡 ).
4. Results and discussion
Regarding the mass transfer rate, some work suggested it is mainly
determined by average flow velocity [16,29,37,38,40–42], others argue
4.1. Model performance
that it is affected by both flow velocity and pore size [39]. However, our
previous work suggests flow velocity, diffusion, pore size, and electrode
In order to examine the model’s capability in predicting the charg-
height all affect the mass transfer rate for a realistic flow battery [34].
ing and discharging voltage curve, Fig. 2 shows a comparison of the cell
To evaluate the effect of mass transport rate on cell voltage, this work
voltage curves from our model (red lines) and those from experiments
calculates the mass transfer rate as 𝑘𝑚𝑃 = 𝛽𝑘𝑇𝑚𝑃 and 𝑘𝑚𝑁 = 𝛽𝑘𝑇𝑚𝑁 for
(black diamonds) and CFD modeling (black squares). The results show
its value in the positive and negative electrode. The formula of 𝑘𝑇𝑚𝑃
that the average difference in cell voltage over all cases is less than
and 𝑘𝑇𝑚𝑁 denote a theoretically derived mass transport rate as shown
0.014 V. This proves that our model is able to accurately reproduce
in Eqs. (15) and (16) [34].
the cell voltage curve with different electrode, membrane, electrolyte,
𝑢0
𝑘𝑇𝑚𝑃 = (15) and operation conditions.
(5∕16 − 2𝛾)𝑃 𝑒𝑃 + 2𝐻∕𝑏 − 4∕(3𝑃 𝑒𝑃 ) More importantly, the fitted parameters for the four distinct cases
𝑢0
𝑘𝑇𝑚𝑁 = (16) are in consistent ranges. For instance, the calculated specific area
(5∕16 − 2𝛾)𝑃 𝑒𝑁 + 2𝐻∕𝑏 − 4∕(3𝑃 𝑒𝑁 )
is 57.8 ∼ 59.4 mm−1 for all cases (𝐴𝑏𝑠 in Table 1), which falls in
Here 𝑢0 = 𝑢𝑃0 = 𝑢𝑁 0
= 𝜔∕(𝐵𝐿𝑒 𝜖), 𝑃 𝑒𝑃 = 𝑢0 𝑏∕𝐷5 , and 𝑃 𝑒𝑁 = the commonly measured range of electrodes [7]. The mass transport
𝑢0 𝑏∕𝐷2 . The parameter 𝛽 is a fitted parameter that quantifies the effects coefficients fall between 0.69 ∼ 1.22 (𝛽 𝑎 in Table 1), which means
of simplifying an actual electrode to multiple channels. It is worth that estimating the mass transport coefficient using Eqs. (15) and (16)
mentioning that the mass transport formula provided by [34] does not is reliable for real electrodes even though these formulas are derived
require fitted parameters; however, it can still predict similar values to from a simplified condition. For the reaction rate constants, the fitted
those by [41] and [40] for flow-through cells with a velocity range 0 parameters are around a 2 ∼ 5 times difference for cases A, C, and D.
∼ 5 cm/s. The differences between B and A (ca. 6 ∼ 22 times) are higher, but
Noting that the concentrations for vanadium ions (Eqs. (9) through still smaller than the diverse ranges from observations [50]. The ratio
(12)) are functions of SoC, while concentrations of proton and water of the reaction rate constants between positive and negative electrodes
(Eqs. (13) and (14)) are dependent of time (𝑡), Eq. (17) links SoC to (ca. 6 ∼ 18) also falls in the range (ca. 0.24 ∼ 44) of observations [50].
time based on charge and mass conservation [31]. Though the proton transfer coefficient is treated as a fitted parameter,
𝐼𝑡 𝐼𝑡 it has little effect on the overall accuracy of cell voltage and battery
SoC𝑐 = SoC𝑠 + 𝑡; SoC𝑑 = SoC𝑒 − (𝑡 − 𝑡𝑒 − 𝑡𝑤 ) (17) performance because the proton-water potential accounts for less than
0
𝐹 𝑉𝑡 𝑐23 0
𝐹 𝑉𝑡 𝑐23
1% of the total energy losses. Details are discussed in Section 4.2.
where SoC𝑐 and SoC𝑑 denote the SoC for charging and discharging. In addition, the fitted parameters agree with the commonly ac-
The charging stage starts at SoC𝑠 and stops at SoC𝑒 at the time 𝑡𝑒 = cepted understanding of how different physical and chemical treat-
0 (SoC − SoC )∕𝐼 . The battery then waits for time 𝑡
𝐹 𝑉𝑡 𝑐23 𝑒 𝑠 𝑡 𝑤 before ments affect the battery properties. For example, with a small amount

4
Y. Chen et al. Journal of Power Sources 506 (2021) 230192

Table 1
Parameters used for each case.
Symbol Meaning Unit A B C D
𝑇 Temperature K 298 298 300 297
𝐻 Cell height m 0.05 0.05 0.1 0.1
𝐵 Cell width m 0.02 0.02 0.1 0.1
𝐿𝑐 Collector thickness m 0.015 0.015 0.006 0.005
𝐿𝑚 Membrane thickness m 1.27e−4 5.08e−5 1.8e−4 1.25e−4
𝜖 Porosity 1 0.67 0.67 0.68 0.67
𝛼𝐴 Specific area coefficient 1 1 1 1 1
𝜔 Pump rate m3 s−1 3.33e−7 3.33e−7 1.00e−6 1.00e−6
𝐼𝑡 Applied current A 0.75 0.5 10 10
𝐼0 Nominal current density A m−2 750 500 1000 1000
𝑐20 Initial concentration of V2 mol m−3 0 0 27 60
0
𝑐23 Total concentration of V2/V3 mol m−3 2000 2000 1080 1200
𝑐50 Initial concentration of V5 mol m−3 0 0 27 60
0
𝑐45 Total concentration of V4/V5 mol m−3 2000 2000 1080 1200
0
𝑐𝐻𝑃 Positive initial H+ concentration mol m−3 5000 5000 1200 4200
0
𝑐𝐻𝑁 Negative initial H+ concentration mol m−3 3000 3000 1200 4200
0
𝑐𝑊 𝑃
Positive initial H2 O concentration mol m−3 47530 47530 4200 42300
𝑉𝑡 Tank volume m3 4.5e−5 4.0e−5 2.5e−4 2.232e−4
SoC𝑠 Charging start SoC 1 0 0 0.025 0.05
SoC𝑒 Charging end SoC 1 0.8 0.95𝑐 0.8015 0.7934
𝑘𝑎𝑝𝑜𝑠 Positive rate constant m s−1 9.0e−8 2.0e−6 3.2e−8 1.65e−8
𝑘𝑎𝑛𝑒𝑔 Negative rate constant m s−1 5.0e−9 3.2e−8 2.5e−9 3.0e−9
𝛽𝑎 Mass transport coefficient 1 1.1 1 1.22 0.69
𝑛𝑎𝐻 Proton transport coefficient 1 1 2 3 5
𝑘𝑏𝑚𝑃 Positive mass transport rate m s−1 2.85e−6 2.60e−6 1.03e−6 5.67e−7
𝑘𝑏𝑚𝑁 Negative mass transport rate m s−1 2.73e−6 2.48e−6 1.01e−6 5.59e−7
𝐴𝑏𝑠 Specific area mm−1 59.4 59.4 57.8 59.4
𝑏𝑏 Pore size μm 45.1 45.1 47.0 45.1

Values of superscript 𝑎 are fitted parameters; values with superscript 𝑏 are calculated from 𝛽 𝑎 and 𝛼𝐴 ; and the value with superscript 𝑐 is 0.8
for case B when calculating data in Tables 2 and 3. Other symbols used but not listed include: fiber size 𝑏𝑠 = 10 μm [7,31], cell thickness
𝐿𝑒 = 4 mm, conductivity of current collector (𝜎𝑐 = 91 000 S m−1 ) and electrode (𝜎𝑒 = 333 S m−1 ), pump efficiency 𝛼𝑃 = 0.8 [36], formal
0
potentials in positive and negative electrode 𝐸𝑃0 = 1.004 V, 𝐸𝑁 = −0.255 V [20], ideal gas constant 𝑅 = 8.314 J mol−1 K−1 , Faraday constant
𝐹 = 96485.33 C mol−1 , reaction transfer coefficient 𝛼 = 0.5, entrance effect constant 𝛾 = 0.033 [49], water drag coefficient 𝑛𝑑 = 2.5 [46],
diffusivities of V2/V3 (𝐷2 = 𝐷3 = 2.4e−10 m2 s−1 ), V4/V5 (𝐷4 = 𝐷5 = 3.9e−10 m2 s−1 ), water (𝐷𝑤 = 2.3e−9 m2 s−1 ), and proton (𝐷𝐻 =
1.4e−9 m2 s−1 ), water viscosity 𝜇 = 1e-3 Pa s [20], Kozeny–Carman constant 𝐾𝑐 = 5.55 [20].

of acetic acid added to the electrolyte, case B shows larger reaction rate 𝐸𝑥 denotes 𝐸𝑒𝑞 , 𝐸𝑎𝑐𝑡 , 𝐸𝑐𝑜𝑛 , 𝐸𝑜ℎ𝑚 , and 𝐸𝑝𝑤 ; and the 𝑒𝑥 denotes 𝑒𝑒𝑞 , 𝑒𝑎𝑐𝑡 ,
constants compared to those in case A. Further, the usage of different 𝑒𝑐𝑜𝑛 , 𝑒𝑜ℎ𝑚 , and 𝑒𝑝𝑤 , respectively. During charging, 𝑡𝑎 = 0, 𝑡𝑏 = 𝑡𝑒 , and
acid in cases A/B and cases C/D results in larger differences in the 𝑖 = 𝐼𝑡 ; during discharging, 𝑡𝑎 = 𝑡𝑒 +𝑡𝑤 , 𝑡𝑏 = 𝑡𝑑 , and 𝑖 = −𝐼𝑡 (see definition
reaction rate constants which are affected by electrolyte composition. of 𝑡𝑒 and 𝑡𝑑 in Section 2.2). The energy consumed by pump during
Due to these agreements, the present model can likely provide clues for charging and discharging is 𝑒𝑃 = 2𝑃𝑤 (𝑡𝑏 − 𝑡𝑎 ) with the pump power and
better electrolyte molecule synthesis and composition design. electrode permeability calculated as 𝑃𝑤 = 𝜔2 𝜇𝐻∕(𝐿𝑒 𝐵𝐾𝛼𝑃 ) [36] and
𝑏2 𝜖 3
𝐾 = 16𝐾 𝑠(1−𝜖)2 [35], respectively. Therefore, the total energy consumed
4.2. Sources of cell voltage and energy losses 𝑐
(positive value) or generated (negative value) in the battery system is

In order to understand the relative contribution of each potential 𝑒𝑠 = 𝑒𝑃 + 𝑒𝑥 for charging and discharging. Table 2 lists the value
component to the cell voltage as discussed in Section 2.1, Fig. 3 shows of duration time (𝑡𝑏 − 𝑡𝑎 ), 𝑒𝑥 , 𝑒𝑃 , and 𝑒𝑠 for charge, discharge, and full
how individual components, such as equilibrium potential, activation cycle for various cases. Though these values are determined from the
potential, concentration potential, ohmic losses, and proton-related model in the present work, most of them are measurable and could
potential, vary with time during charging and discharging. The results be used to verify the model’s predictive capability. For instance, by
show that the equilibrium potential (solid green lines) is the biggest timing the charge/discharge time and measuring electronic resistance
contribution of the total cell voltage (red lines), and it shows approx- from electrodes, current collectors, and membrane under a constant
imately a symmetric behavior during charging and discharging. The applied current, the ohmic energy losses can be estimated. Meanwhile,
proton and water potential (magenta lines) is the second important a measurement of the time series of open-circuit potential can ap-
contribution for cases A and B, while activation potential (solid blue proximately determine the energy losses associated with equilibrium
lines) stands for the second place for cases C and D. The concentration potential considering that proton energy loss is negligible. Further,
potential (solid black lines) is small during charging and most of the measurements of the energy provided or consumed by external power
time during discharging, but becomes highly important at the end of source, load device, and pumps can determine the system energy loss.
discharging. The ohmic potential (cyan lines) is relatively small com- Nevertheless, measurements of activation and concentration energy
pared to equilibrium and activation potentials for the current densities losses are still challenging.
(500 ∼ 1000 A/m2 ) considered here. Though not shown in Fig. 3, the The cases A to D in Table 2 show that the net (full cycle) energy
open-circuit cell voltage (𝐸𝑜𝑐𝑣 = 𝐸𝑒𝑞 + 𝐸𝑝𝑤 ) has a similar trend as the loss due to equilibrium potentials is much smaller than that due to
equilibrium potential. However, its magnitude is always lower than cell activation loss even though their absolute values are the highest during
voltage (𝐸𝑐 ) during charging and higher than that during discharging. charging and discharging. We argue that such a low net energy loss
The energy consumed or generated from individual potentials is from equilibrium potential is due to the symmetric behavior of the
another way to understand their relative importance. Here we calculate equilibrium potential curve. The ratio of discharging time to charging
the energy from each component by Eq. (18). time, equivalent to the Coulombic efficiency (CE) for a constant current
𝑡𝑏 density, is a proper parameter to quantify the symmetric behavior. To
𝑒𝑥 = 𝑖 ⋅ 𝐸𝑥 𝑑𝑡 (18) further understand the role of a CE, Table 3 lists the relative ratio of
∫𝑡𝑎

5
Y. Chen et al. Journal of Power Sources 506 (2021) 230192

Fig. 2. A comparison of cell voltage from the present model with those obtained from PNNL experiments (A and B), CFD modeling (C), and experiment reported in the literature
(D). Parameters used in each case are listed in Table 1. More details of C and D can be found in [20,26], and [31].

Table 2
Energy consumption and generation from individual components.
No. 𝛽 𝐴𝑠 Type Time 𝑒𝑒𝑞 𝑒𝑎𝑐𝑡 𝑒𝑐𝑜𝑛 𝑒𝑜ℎ𝑚 𝑒𝑝𝑤 𝑒𝑃 𝑒𝑠
Unit h kJ kJ kJ kJ kJ kJ kJ
Charge 2.57 8.54 0.83 0.01 0.29 1.18 0.65 11.50
A 1.1 59.4 Discharge 2.54 −8.41 0.82 0.03 0.29 −1.16 0.64 −7.80
Full cycle 5.11 0.13 1.65 0.04 0.58 0.02 1.28 3.70
Charge 3.43 7.59 0.14 0.01 0.15 1.00 0.86 9.75
B 1.0 59.4 Discharge 3.40 −7.50 0.14 0.02 0.15 −0.99 0.85 −7.34
Full cycle 6.83 0.09 0.27 0.03 0.30 0.01 1.71 2.41
Charge 0.56 25.21 5.05 0.21 1.45 2.77 0.46 35.15
C 1.22 57.8 Discharge 0.52 −23.15 4.72 0.42 1.35 −2.62 0.42 −18.84
Full cycle 1.08 2.06 9.78 0.63 2.80 0.15 0.88 16.30
Charge 0.53 24.12 4.86 0.35 1.08 2.54 0.48 33.34
D 0.69 59.4 Discharge 0.48 −21.20 4.38 0.57 0.97 −2.32 0.43 −17.18
Full cycle 1.01 2.92 9.24 0.92 2.05 0.22 0.91 16.25
Charge 0.53 24.12 4.86 0.35 1.08 2.54 0.48 33.43
D1 1.0 59.4 Discharge 0.48 −21.20 4.38 0.57 0.97 −2.32 0.43 −17.18
Full cycle 1.01 2.92 9.24 0.92 2.05 0.22 0.91 16.25
Charge 0.53 24.02 4.87 0.22 1.08 2.54 0.48 33.20
D2 2.0 59.4 Discharge 0.51 −22.45 4.67 0.44 1.02 −2.48 0.46 −18.32
Full cycle 1.04 1.57 9.54 0.66 2.10 0.07 0.94 14.88
Charge 0.53 23.90 4.88 0.10 1.08 2.54 0.48 32.98
D3 5.0 59.4 Discharge 0.53 −23.64 4.96 0.23 1.08 −2.62 0.48 −19.52
Full cycle 1.07 0.26 9.84 0.33 2.16 −0.08 0.97 13.46

Unit of 𝐴𝑠 is 1/mm.

6
Y. Chen et al. Journal of Power Sources 506 (2021) 230192

Fig. 3. The variations of total cell voltage (𝐸𝑐 ), equilibrium potential (𝐸𝑒𝑞 ), activation potential (𝐸𝑎𝑐𝑡 ), concentration potential (𝐸𝑐𝑜𝑛 ), ohmic potential (𝐸𝑜ℎ𝑚 ), and proton and water
potential (𝐸𝑝𝑤 ), with respect to time for cases A to D.

Table 3
Relative contribution of individual components to total energy loss.
No. 𝛽 𝐴𝑠 𝑒𝑒𝑞 ∕𝑒𝐿 𝑒𝑎𝑐𝑡 ∕𝑒𝐿 𝑒𝑐𝑜𝑛 ∕𝑒𝐿 𝑒𝑜ℎ𝑚 ∕𝑒𝐿 𝑒𝑝𝑤 ∕𝑒𝐿 𝑒𝑃 ∕𝑒𝐿 CE EE SE
A 1.1 59.4 0.04 0.45 0.01 0.16 0.00 0.35 0.99 0.78 0.68
B 1.0 59.4 0.04 0.11 0.01 0.13 0.00 0.71 0.99 0.92 0.75
C 1.22 57.8 0.13 0.60 0.04 0.17 0.01 0.05 0.93 0.56 0.54
D 0.69 59.4 0.18 0.57 0.06 0.13 0.01 0.06 0.90 0.53 0.51
D1 1.0 59.4 0.11 0.64 0.04 0.14 0.00 0.06 0.95 0.57 0.55
D2 2.0 59.4 0.02 0.73 0.02 0.16 −0.01 0.07 1.00 0.62 0.59
D3 5.0 59.4 0.01 0.75 0.01 0.17 −0.01 0.07 1.00 0.63 0.60

Unit of 𝐴𝑠 is 1/mm.

each energy loss relative to the total energy loss in a charge–discharge the equilibrium potential itself, but attributed to the concentration
cycle. The results of cases A to D in Table 3 show that the CE decreases gradients within the electrode. The reason for a lower CE in cases C
from 0.99 to 0.90 with the value of 𝑒𝑒𝑞 ∕𝑒𝐿 increasing from 4% to 18%. and D will be further discussed in Section 4.3.
This result together with the equilibrium potential curves (solid green Though the proton and water potential ranks second or third in
lines) in Fig. 3 means that, with increasingly obvious asymmetry from cell voltage, its contribution to the total energy loss is negligible (see
cases A to D, the CE decreases and the net energy loss from equilibrium 𝑒𝑝𝑤 ∕𝑒𝐿 in Table 3 cases A to D) due to a good symmetry in its potential
potential increases. Though the net equilibrium potential is non-zero, curve and a small magnitude compared to the equilibrium potential
it is necessary to mention that such an energy loss is not caused by (magenta lines in Fig. 3). In contrast, the activation potential ranks

7
Y. Chen et al. Journal of Power Sources 506 (2021) 230192

second or third among all potentials, its contribution to the net energy different mass transport coefficients are designed to understand how
loss is up to 45% ∼ 60% (see 𝑒𝑎𝑐𝑡 ∕𝑒𝐿 in Table 3 of A, C, and D). Such a the mass transport coefficient changes CE and cell performance. Fig. 4
dominant contribution is attributed to the anti-symmetric behavior in illustrates how the cell voltage curve changes with SoC and time at
its potential curve and its moderate high magnitude compared to other four mass transport coefficients based on case D. The results show that
potential components. Though the activation energy loss dominates the the minimum discharge SoC shifts from around 0.13 to 0.05 when
total energy losses in cases A, C, and D, case B is dominated by pump 𝛽 increases from 0.69 to 5, and the end discharging time increases
energy consumption (see 𝑒𝑃 in Table 2 and 𝑒𝑃 ∕𝑒𝐿 in Table 3). For the from around 64 min to 67 min. With such a change, it was observed
concentration and ohmic potentials, though their values are relatively that the CE increases from 0.90 to 1.0, the equilibrium energy loss
smaller compared to other potentials during charging and most of the decreases from 2.92 kJ to 0.26 kJ, and the concentration energy loss
time during discharging, they together can contribute to 14% ∼ 21% decreases from 0.92 kJ to 0.33 kJ, while ohmic- and proton-related
of the total energy losses. energy losses have little changes (see full cycle of cases D and D1 to D3
Based on the above results and discussion, we propose a guideline in Table 2). These changes also increase the SE from 51% to 60% (see
to identify the key sources of energy losses from a cell voltage curve SE in Table 3). Though increasing the mass transport coefficient reduces
as follows. First, a CE value less than 0.93 (a battery similar to cases C the energy loss from equilibrium and concentration potentials, the
and D) suggests the energy loss associated with concentration gradients results also show that increasing 𝛽 from 2.0 to 5.0 does not significantly
may be important and optimization should be conducted to improve improve the energy loss and SE. Therefore, a good guideline for mass
the CE and thus reduce the overall energy loss. Second, if CE is already transport optimization is trying to increase the mass transport but not
close to 1 (a battery similar to A and B), the energy loss due to pump too much.
consumption should be examined and pump optimization should be Though it is straightforward to increase the mass transport coef-
conducted if the pump loss is important or if the system efficiency (SE) ficient using our model, obtaining a better flow battery design such
is much lower than the energy efficiency (EE). Finally, if both a high that the mass transport coefficient is higher is another story. As the
CE and low pump consumption are confirmed, optimization of reaction improvement of energy losses due to increased mass transport is mainly
kinetics should be improved in order to reduce the contribution of attributed to improved CE, increasing 𝛽 is equivalent to decreasing
𝑃 , 𝐼 𝑃 𝐻∕𝑏, 𝐼 𝑁 , and 𝐼 𝑁 𝐻∕𝑏 according to Eqs. (9) to (12). Therefore,
𝐼𝑐𝑑
activation loss. Optimization of ohmic losses is the final option if 𝑐𝑢 𝑐𝑑 𝑐𝑢
reducing electrode electronic resistance and membrane thickness is decreasing the local current density and average pore size and/or
possible. Sections 4.3 to 4.5 further discuss potential approaches to increasing total active species concentration, active species diffusivity
optimize energy losses from equilibrium potential, pump consumption, and pump rate can likely improve the mass transport rate. Our model
can act as a tool to design how these parameters should change in order
and activation potential using our model.
to achieve a desired cell performance.
It is necessary to mention that the above guidelines are proposed
based on a battery operation perspective. Such a viewpoint emphasizes
4.4. Cell performance optimization: pump rate
that the overall battery performance depends on both operation condi-
tions and material properties. For given battery design and materials,
As the pump energy occupies 71% of the total energy loss for case B
optimization for operation conditions to improve CE and minimize
(Table 3), pump optimization is the most effective approach to improve
pump consumption is relatively easier than changing key material prop-
the cell performance of B. The pump optimization usually involves
erties such as reaction rate constant and electronic resistance. However,
changing pump rate 𝜔. A higher pump rate means higher pump energy
if the energy losses associated with material properties dominate the
consumption because it is proportional to 𝜔2 , while a higher pump rate
overall energy losses, the order of optimization should be changed
also means a faster mass transport, which results in higher CE and lower
accordingly. For example, optimization for ohmic losses should be the
energy losses from equilibrium potential and concentration potential.
priority at high current density because ohmic losses are shown as
Pump optimization is designed to identify the pump rate that minimizes
dominant loss [51]. the total energy losses and thus the maximum SE.
Using the parameters of case B, Fig. 5 shows the absolute (a) and
4.3. Cell performance optimization: mass transport relative (b) energy losses from individual components with a pump
rate varying between 0 and 20 mL/min. The results show that the
As mentioned in Section 4.2, a lower CE means a higher asymmetry SE peaks at around 7 mL/min with a value of around 87%, which is
in the equilibrium potential, which results in a higher energy loss due 12% higher than that in case B where the pump rate is 20 mL/min
to the high magnitude of equilibrium potential. Therefore, the main (Fig. 5b). Fig. 5a further shows that the maximum SE is reached
approach to reduce the equilibrium potential energy is to increase the when the total energy loss, a summation of total cell energy loss 𝑒𝑐
CE. Here we further argue that CE is controlled by the concentration (blue diamond line) and pump energy consumption 𝑒𝑃 (cyan line),
gradients along flow direction and the normal direction of electrode– reaches the trough. The trough is mainly attributed to the reduced
electrolyte interface. By definition, the CE is the ratio of discharging energy losses from the equilibrium potential (blue solid line), although
time to charging time for a constant current density. For a given energy losses from concentration and proton potentials also contribute
start and end charging SoC (i.e., SoC𝑠 and SoC𝑒 ), the charging time to it. Meanwhile, Fig. 5b suggests the reduced equilibrium energy
is determined as 𝑡𝑐 = 𝐹 𝑉𝑡 𝑐230 (SoC - SoC )/𝐼 , while the discharging
𝑒 𝑠 𝑡 loss is mainly attributed to the increased CE (blue solid line), which
time is 𝑡𝑑𝑐 = 𝐹 𝑉𝑡 𝑐230 (SoC - SoC∗ )/𝐼 . Here SoC∗ is the minimum
𝑒 𝑠 𝑡 𝑠 significantly reduces the percentage of equilibrium potential loss in the
SoC, which is calculated as SoC∗𝑠 = max[SoC𝑠 , SoC2𝑚 , SoC5𝑚 ] with SoC2𝑚 total energy losses. As the CE signifies the effect of mass transport,
and SoC5𝑚 denoting the SoC that makes 𝑐2𝑤 = 0 and 𝑐5𝑤 = 0 during the above analyses suggest it is the improved mass transport due to
discharge, respectively. According to Eqs. (9) and (10), SoC2𝑚 and SoC5𝑚 increased pump rate that improves the cell performance.
are further limited by the mass transport coefficient 𝛽. Therefore, the However, with increasing pump rate, the improvement of cell per-
CE is determined by the mass transport coefficient. Though this work formance due to improved mass transport becomes less important as the
does not consider cross-over and side reactions, it is necessary to pump energy consumption gradually dominates the overall energy loss
mention that these processes affect the total vanadium concentrations, (see cyan lines in Fig. 5a–b). This suggests that a pump optimization is
and thus the charge/discharge time and CE. With such a perspective, especially useful when the mass transport coefficient is low like case D
the calibrated coefficient 𝛽 represents the combined effects from mass (see calculated mass transport rate 𝑘𝑏𝑚𝑃 and 𝑘𝑏𝑚𝑁 in Table 1), which is
transport, cross-over, and side-reactions. equivalent to a scenario of low CE or low pump rate. This conclusion
As case D (Table 3) has the lowest CE and highest energy loss from can be demonstrated by Fig. 5b where the SE increases from 75% to
equilibrium potential, three extra cases, named D1, D2, and D3, with 87% with the pump rate increasing from 1 mL/min to 7 mL/min.

8
Y. Chen et al. Journal of Power Sources 506 (2021) 230192

Fig. 4. The variations of cell voltage with respect to SoC (a) and time (b) at four mass transport coefficients (𝛽). Parameters used for 1.0, 2.0, and 5.0 cases are identical to those
used in 0.69 case (Table 1 case D) except for the value of 𝛽.

Fig. 5. The variations of absolute and relative energy losses predicted by the present model for case B at different pump rates (a, b), the present model for case A at different
porosities (c, d), and a 0D model for case B at different pump rates (e, f). 𝑒𝑥 , 𝑒𝑐 , and 𝑒𝐿 denote six energy loss sources, total cell energy loss, and total system energy loss,
respectively.

4.5. Cell performance optimization: activation loss and electrode structure is to develop new active materials and electrolyte to make the reaction
rate constants higher. Though developing a new active material is time
When energy losses from both equilibrium potential and pump
consuming, changing the composition of electrolyte is feasible and
consumption have been optimized to a low level, like case A in Table 3,
reducing the energy loss from activation potential becomes important. could achieve higher reaction rate constants as demonstrated in case
The most straightforward approach to reduce the activation energy loss B when compared to case A (see rate constants in Table 1).

9
Y. Chen et al. Journal of Power Sources 506 (2021) 230192

For a given active material and electrolyte composition, adjusting the pump rate. For these reasons, a uniform concentration distribution
the electrode structure properties, e.g., porosity and specific area, could used by the 0D model results in very different energy losses, CE, and SE
also improve the cell performance. However, such an improvement features compared to those from a 2D model which considers the con-
is not straightforward due to three competing processes. Specifically, centration gradients along flow direction and at electrolyte–electrode
increasing the specific area reduces the local current density and thus interface.
reduces the activation energy loss. However, a higher specific area usu- In addition, due to the inability to predict the equilibrium energy
ally means smaller porosity which results in higher pressure drop [29, loss, the 0D model significantly underestimates the overall energy loss
35,36,52] and ohmic potential [31]. Therefore, an optimization of at low flow rate, which results in significant overestimation of the SE
electrode porosity or specific area is required to minimize total energy at low flow rate. For example, the 0D model predicts the total energy
losses. Considering
√ that the specific area is linked to porosity through loss and the SE at a flow rate 1 mL/min as around 0.6 kJ and 93%
𝐴𝑠 = 4( 𝜖 − 𝜖)∕𝑏𝑠 [34], we only study how porosity affects the cell
(Fig. 5e,f), while the present model gives values as 2.3 kJ and 75%
voltage and overall performance. It is worth mentioning that the above
(Fig. 5a,b). With the increasing of the flow rate, the 0D model tends to
relationship is similar to 𝐴𝑠 = 4(1−𝜖)∕𝑏𝑠 [52] when the porosity is close
give similar prediction as the present model.
to 1 and its applicability for real electrodes has been demonstrated by
a pore-scale study [29].
Following the above formula, Fig. 5c shows that the cell voltage 5. Conclusions
increases with increasing porosity during charge, while decreasing with
porosity during discharge. This is consistent with the results from a This work proposed an analytical model for evaluating cell voltage
2D numerical model [20]. In addition, it shows little change to the and battery performance based on a 2D analytical solution to concen-
cell voltage when increasing the porosity from 0.5 to 0.67, but varies trations for RFBs. Such a model demonstrates a good agreement of
around 0.05 V when the porosity increases from 0.8 to 0.9. This the cell voltage with experimental and CFD modeling data and the
suggests that the porosity affects cell voltage in a non-linear manner calibrated model parameters are consistent in various conditions of
and such a non-linearity is likely attributed to the competing effects of electrode structure, electrolyte composition, and operation conditions.
porosity on activation potential and ohmic potential.
With the validated results, the equilibrium, activation, and pro-
To further understand the role of porosity in changing cell perfor-
ton/water potentials are identified as the key components of cell volt-
mance, Fig. 5d shows the variations of energy loss components (left
age, while the activation energy loss, equilibrium energy loss, and
axis) and battery inefficiencies (CE and SE in right axis) with respect
pump energy consumption are identified as the dominant energy losses
to porosity. As reducing the activation energy loss is the optimization
of the battery system. A guideline of optimizing the key energy losses
goal, we analyze the reversed 𝑥-axis, i.e., from 0.9 to 0.5. The result
is proposed. Further analyses show that the mass transport process
shows that the activation energy loss (green line) is decreased when
decreasing the porosity. Meanwhile, both pump consumption (cyan controls the CE and the energy loss from equilibrium potential. A pump
line) and ohmic energy loss (black line) increase. Due to these three rate optimization can improve the SE of the cell through improving the
competing processes, the overall battery energy loss (red line) and the mass transport, increasing the CE, and significantly reducing energy
SE (dashed blue line) reach their extrema at an optimal porosity of loss from equilibrium potential. Further improvement of cell perfor-
0.8. It is quite clear that the improvement of the battery performance mance can be achieved by using either new active species/electrolyte
is mainly achieved by reducing the activation energy loss through composition or better electrode structure design. Such a battery de-
increasing the specific area when the electrode porosity is higher than sign depends on the relative magnitude of the electrode porosity to
the optimal porosity. By contrast, the battery performance could be im- an optimal porosity obtained from the present model. Specifically,
proved through reducing the pump consumption and ohmic energy loss increasing the specific area is recommended if the electrode porosity
by increasing the electrode porosity when it is lower than the optimal is higher than the optimal porosity, otherwise increasing the porosity
value. As the optimal porosity determines the strategy of improving the is a better option. The comparison of the present model with its 0D
battery performance, the present model, which can estimates its value, version demonstrates a very different prediction of the energy losses
is of great practical worth for better electrode structure design. when a uniform concentration assumption is invoked and a significant
overestimation of the SE is observed using a 0D model.
4.6. Comparison with a 0D cell voltage model Based on the accuracy of the present model, the insights of sources
of energy losses, and the guideline for cell performance optimization,
As 0D modeling has been widely used to predict the cell voltage the present model provides a reliable and efficient tool for flow battery
and evaluate cell performance [31,46], we evaluate the difference design, evaluation, and optimization.
between the present model and a 0D model. A 0D model assumes that
the concentrations of active species are uniform everywhere, which
is equivalent to set 𝛽 = +∞ in the present model. Fig. 5e–f shows CRediT authorship contribution statement
the absolute and relative contribution of each energy loss by setting
𝛽 = 1010 for case B. The results show that the activation and ohmic Yunxiang Chen: Conceptualization, Methodology, Software, Vali-
energy losses do not change with pump rate, and energy losses from dation, Analyses, Investigation, Writing. Jie Bao: Discussion, Supervi-
equilibrium, concentration, and proton potentials are almost 0. There- sion, Project administration, Funding acquisition, Data curation. Zhijie
fore, an optimized pump rate is not observed in a 0D model compared Xu: Discussion, Supervision, Project administration, Funding acquisi-
to what is observed in a 2D model (Fig. 5a–b). Such a difference can tion, Data curation. Peiyuan Gao: Analyses, Discussion. Litao Yan: Ex-
be explained as follows: (a) the uniform distributions of concentrations periments, Discussion. Soowhan Kim: Discussion. Wei Wang: Project
mean that the equilibrium potential depends only on SoC, which results administration, Funding acquisition.
in a fully symmetric potential curve during a charge–discharge cycle;
(b) the uniform distributions of concentrations also mean that the
charging time equals the discharging time, which means CE = 1; (c) the Declaration of competing interest
uniform distributions of concentrations mean that the concentrations
at the electrode interface and in the bulk are identical, which means The authors declare that they have no known competing finan-
concentration loss is zero; and (d) as all concentrations depend only cial interests or personal relationships that could have appeared to
on SoC, the activation potential only changes with SoC but not with influence the work reported in this paper.

10
Y. Chen et al. Journal of Power Sources 506 (2021) 230192

Acknowledgments [6] B.K. Chakrabarti, E. Kalamaras, A.K. Singh, A. Bertei, J. Rubio-Garcia, V. Yufit,
K.M. Tenny, B. Wu, F. Tariq, Y.S. Hajimolana, N.P. Brandon, C.T. John Low, E.P.
Roberts, Y.M. Chiang, F.R. Brushett, Modelling of redox flow battery electrode
This research was supported by Energy Storage Materials Initiative
processes at a range of length scales: A review, Sustainable Energy Fuels 4 (2020)
(ESMI), which is a Laboratory Directed Research and Development 5433–5468, http://dx.doi.org/10.1039/d0se00667j.
Project at Pacific Northwest National Laboratory (PNNL). PNNL is a [7] A.Z. Weber, M.M. Mench, J.P. Meyers, P.N. Ross, J.T. Gostick, Q. Liu, Redox
multiprogram national laboratory operated for the U.S. Department of flow batteries: A review, J. Appl. Electrochem. 41 (2011) 1137–1164, http:
Energy (DOE) by Battelle Memorial Institute under Contract no. DE- //dx.doi.org/10.1007/s10800-011-0348-2.
AC05-76RL01830. The authors thank the two anonymous reviewers [8] L. Grande, Redox Flow Batteries 2018-2028: Markets, Trends, Applications,
Technical Report, IDTechEx, 2018, www.idtechex.com.
and editor for their constructive feedback of this work.
[9] S. Vorrath, Australia’s first grid-scale vanadium flow battery to be built in south
Australia, 2020.
Appendix. Concentrations of proton and water [10] S. Kim, E. Thomsen, G. Xia, Z. Nie, J. Bao, K. Recknagle, W. Wang, V.
Viswanathan, Q. Luo, X. Wei, A. Crawford, G. Coffey, G. Maupin, V. Sprenkle,
As the concentrations of proton and water in the positive electrode 1 kW/1 kWh advanced vanadium redox flow battery utilizing mixed acid
electrolytes, J. Power Sources 237 (2013) 300–309, http://dx.doi.org/10.1016/
are required to predict the potentials from water and proton, their
j.jpowsour.2013.02.045.
analytical solutions are derived in this section. The processes that [11] Y.K. Zeng, T.S. Zhao, L. An, X.L. Zhou, L. Wei, A comparative study of all-
control the transport of proton include the proton exchange between vanadium and iron–chromium redox flow batteries for large-scale energy storage,
electrolyte tank and the electrode, the generation or consumption of J. Power Sources 300 (2015) 438–443, http://dx.doi.org/10.1016/j.jpowsour.
proton from chemical reaction in positive electrode, and the transport 2015.09.100.
of proton through membrane. Similar to [31], we assume the proton [12] K. Mongird, V.V. Viswanathan, P.J. Balducci, M.J.E. Alam, V. Fotedar, V.S.
Koritarov, B. Hadjerioua, Energy Storage Technology and Cost Characteriza-
concentration is uniform within the electrode. Denoting 𝑐𝐻𝑃 and 𝑐𝐻𝑅 tion Report, Technical Report, Pacific Northwest National Laboratory (PNNL),
as the proton concentration in the positive electrode and tank, their Richland, WA, USA, 2019.
mass conservation can be written as Eqs. (19) and (20). [13] B. Huskinson, M.P. Marshak, C. Suh, S. Er, M.R. Gerhardt, C.J. Galvin, X.
Chen, A. Aspuru-Guzik, R.G. Gordon, M.J. Aziz, A metal-free organic–inorganic
𝑑(𝜖𝑉𝑒 𝑐𝐻𝑃 ) 𝐼 𝐼
= 𝑢𝑃0 𝐵𝐿𝑒 𝜖(𝑐𝐻𝑅 − 𝑐𝐻𝑃 ) + 2 𝑡 − 𝑛𝐻 𝑡 (19) aqueous flow battery, Nature 505 (2014) 195–198, http://dx.doi.org/10.1038/
𝑑𝑡 𝐹 𝐹 nature12909.
𝑑(𝑉𝑡 𝑐𝐻𝑅 ) [14] P. Leung, T. Martin, M. Liras, A.M. Berenguer, R. Marcilla, A. Shah, L. An,
= −𝑢𝑃0 𝐵𝐿𝑒 𝜖(𝑐𝐻𝑅 − 𝑐𝐻𝑃 ) (20)
𝑑𝑡 M.A. Anderson, J. Palma, Cyclohexanedione as the negative electrode reaction
for aqueous organic redox flow batteries, Appl. Energy 197 (2017a) 318–326,
where the last two terms in Eq. (19) denote the amount of proton
http://dx.doi.org/10.1016/j.apenergy.2017.04.023.
generated from chemical reaction and transported from the membrane. [15] P. Leung, A.A. Shah, L. Sanz, C. Flox, J.R. Morante, Q. Xu, M.R. Mohamed,
A reference value for 𝑛𝐻 is 1 if the concentration gradient of proton at F.C. Ponce de León, Recent developments in organic redox flow batteries: A
the membrane–electrolyte interface is zero and water velocity through critical review, J. Power Sources 360 (2017b) 243–283, http://dx.doi.org/10.
the membrane is zero. For an actual flow battery, such a value requires 1016/j.jpowsour.2017.05.057.
measurements or fitting form model. [16] X. Zhou, T. Zhao, L. An, Y. Zeng, L. Wei, Critical transport issues for improving
the performance of aqueous redox flow batteries, J. Power Sources 339 (2017)
Dividing Eq. (20) by 𝑉𝑡 and Eq. (19) by 𝜖𝑉𝑒 and then subtracting 1–12, http://dx.doi.org/10.1016/j.jpowsour.2016.11.040.
the resulted Eq. (19) from Eq. (20), an equation for 𝑐𝐻𝑅 − 𝑐𝐻𝑃 is given [17] Q. Zheng, X. Li, Y. Cheng, G. Ning, F. Xing, H. Zhang, Development and
by Eq. (21). perspective in vanadium flow battery modeling, Appl. Energy 132 (2014)
𝑃 254–266, http://dx.doi.org/10.1016/j.apenergy.2014.06.077.
𝑑(𝑐𝐻𝑅 − 𝑐𝐻𝑃 ) 𝑢 𝐵𝐿𝑒 𝑉 (2 − 𝑛𝐻 )𝐼𝑡 [18] J.L. Barton, F.R. Brushett, A one-dimensional stack model for redox flow
=− 0 (1 + 𝜖 𝑒 )(𝑐𝐻𝑅 − 𝑐𝐻𝑃 ) − (21)
𝑑𝑡 𝑉𝑒 𝑉𝑡 𝜖𝑉𝑒 𝐹 battery analysis and operation, Batteries 5 (25) (2019) http://dx.doi.org/10.
3390/batteries5010025.
An analytical solution to Eq. (21) is given by Eq. (22) with an initial [19] I. Aramendia, U. Fernandez-Gamiz, A. Martinez-San-Vicente, E. Zulueta, J.M.
0
condition (𝑐𝐻𝑅 − 𝑐𝐻𝑃 )𝑡=0 = 𝑐𝐻𝑅 0
− 𝑐𝐻𝑃 = 0. Lopez-Guede, Vanadium redox flow batteries: A review oriented to fluid-dynamic

(2 − 𝑛𝐻 )𝐼𝑡 [ 𝜔𝑃 (1 + 𝜖𝛿)𝑡 ]
optimization, Energies 14 (176) (2020) http://dx.doi.org/10.3390/en14010176.
𝑐𝐻𝑅 − 𝑐𝐻𝑃 = 1 − exp(− ) (22) [20] A.A. Shah, M.J. Watt-Smith, F.C. Walsh, A dynamic performance model for redox-
𝜔𝑃 (1 + 𝜖𝛿)𝐹 𝑉𝑒 𝜖 flow batteries involving soluble species, Electrochim. Acta 53 (2008) 8087–8100,
http://dx.doi.org/10.1016/j.electacta.2008.05.067.
where 𝜔𝑃 = 𝑢𝑃0 𝐵𝐿𝑒 𝜖 and 𝛿 = 𝑉𝑒 ∕𝑉𝑡 . Substituting Eq. (22) to Eq. (19),
[21] H. Al-Fetlawi, A.A. Shah, F.C. Walsh, Non-isothermal modelling of the all-
a solution to 𝑐𝐻𝑃 can be given as Eq. (13) with an initial condition vanadium redox flow battery, Electrochim. Acta 55 (2009) 78–89, http://dx.
𝑐𝐻𝑃 |𝑡=0 = 𝑐𝐻𝑃
0 .
doi.org/10.1016/j.electacta.2009.08.009.
The solution to water in the positive electrode can be derived sim- [22] D. You, H. Zhang, J. Chen, A simple model for the vanadium redox battery,
ilarly, although the value 2 and 𝑛𝐻 are replaced by -1 and 𝑛𝑑 because Electrochim. Acta 54 (2009) 6827–6836, http://dx.doi.org/10.1016/j.electacta.
2009.06.086.
1 mole of water is consumed when 2 mole of proton is generated and
[23] A.A. Shah, H. Al-Fetlawi, F.C. Walsh, Dynamic modelling of hydrogen evolution
the water transport flux through the membrane is different from that effects in the all-vanadium redox flow battery, Electrochim. Acta 55 (2010)
of proton. The water drag is selected as 𝑛𝑑 = 2.5 [46,48]. 1125–1139, http://dx.doi.org/10.1016/j.electacta.2009.10.022.
[24] H. Al-Fetlawi, A.A. Shah, F.C. Walsh, Modelling the effects of oxygen evolution
References in the all-vanadium redox flow battery, Electrochim. Acta 55 (2010) 3192–3205,
http://dx.doi.org/10.1016/j.electacta.2009.12.085.
[1] I. Tsuda, K. Nozaki, K. Sakuta, K. Kurokawa, Improvement of performance in [25] K.W. Knehr, E. Agar, C.R. Dennison, A.R. Kalidindi, E.C. Kumbur, A transient
redox flow batteries for PV systems, Sol. Energy Mater. Sol. Cells 47 (1997) vanadium flow battery model incorporating vanadium crossover and water
101–107, http://dx.doi.org/10.1016/S0927-0248(97)00030-5. transport through the membrane, J. Electrochem. Soc. 159 (2012) A1446–A1459,
[2] M. Skyllas-Kazacos, M.H. Chakrabarti, S.A. Hajimolana, F.S. Mjalli, M. Saleem, http://dx.doi.org/10.1149/2.017209jes.
Progress in flow battery research and development, J. Electrochem. Soc. 158 [26] J.R. Clausen, V.E. Brunini, H.K. Moffat, M.J. Martinez, Numerical Modeling of an
(2011) R55–R79, http://dx.doi.org/10.1149/1.3599565. All Vanadium Redox Flow Battery, Sandia National Laboratories, Albuquerque,
[3] X. Li, H. Zhang, Z. Mai, H. Zhang, I. Vankelecom, Ion exchange membranes for New Mexico, USA, 2014, p. 39.
vanadium redox flow battery (VRB) applications, Energy Environ. Sci. 4 (2011) [27] G. Qiu, A.S. Joshi, C.R. Dennison, K.W. Knehr, E.C. Kumbur, Y. Sun, 3-D pore-
1147–1160, http://dx.doi.org/10.1039/c0ee00770f. scale resolved model for coupled species/charge/fluid transport in a vanadium
[4] G. Strbac, M. Aunedi, D. Pudjiant, P. Djapic, F. Teng, A. Staurt, D. Jackravut, redox flow battery, Electrochim. Acta 64 (2012b) 46–64, http://dx.doi.org/10.
R. Sansom, V. Yufit, N. Brandon, Strategic Assessment of the Role and Value of 1016/j.electacta.2011.12.065.
Energy Storage Systems in the UK Low Carbon Energy Future, Technical Report, [28] G. Qiu, C.R. Dennison, K.W. Knehr, E.C. Kumbur, Y. Sun, Pore-scale analysis of
Energy Futures Lab, Imperial College, London, 2012. effects of electrode morphology and electrolyte flow conditions on performance
[5] DOE, Energy Storage Grand Challenge Roadmap, Technical Report, Research of vanadium redox flow batteries, J. Power Sources 219 (2012a) 223–234,
Technology Investment Committee, DOE, US, 2020. http://dx.doi.org/10.1016/j.jpowsour.2012.07.042.

11
Y. Chen et al. Journal of Power Sources 506 (2021) 230192

[29] L. Chen, Y.L. He, W.Q. Tao, P. Zelenay, R. Mukundan, Q. Kang, Pore-scale study [41] S.K. Murthy, A.K. Sharma, C. Choo, E. Birgersson, Analysis of concentration
of multiphase reactive transport in fibrous electrodes of vanadium redox flow overpotential in an all-vanadium redox flow battery, J. Electrochem. Soc. 165
batteries, Electrochim. Acta 248 (2017) 425–439, http://dx.doi.org/10.1016/j. (2018) A1746–A1752, http://dx.doi.org/10.1149/2.0681809jes.
electacta.2017.07.086. [42] M.D. Kok, R. Jervis, T.G. Tranter, M.A. Sadeghi, D.J. Brett, P.R. Shearing, J.T.
[30] M.A. Sadeghi, M. Aganou, M. Kok, M. Aghighi, G. Merle, J. Barralet, J. Gostick, Mass transfer in fibrous media with varying anisotropy for flow battery
Gostick, Exploring the impact of electrode microstructure on redox flow battery electrodes: Direct numerical simulations with 3D X-ray computed tomography,
performance using a multiphysics pore network model, J. Electrochem. Soc. 166 Chem. Eng. Sci. 196 (2019) 104–115, http://dx.doi.org/10.1016/j.ces.2018.10.
(2019) A2121–A2130, http://dx.doi.org/10.1149/2.0721910jes. 049.
[31] A.A. Shah, R. Tangirala, R. Singh, R.G.A. Wills, F.C. Walsh, A dynamic unit [43] M. Li, T. Hikihara, A coupled dynamical model of redox flow battery based
cell model for the all-vanadium flow battery, J. Electrochem. Soc. 158 (2011) on chemical reaction, fluid flow, and electrical circuit, IEICE Trans. Fundam.
A671–A677, http://dx.doi.org/10.1149/1.3561426. Electron. Commun. Comput. Sci. E91-A (2008) 1741–1747, http://dx.doi.org/
[32] A. Tang, J. Bao, M. Skyllas-Kazacos, Thermal modelling of battery configuration 10.1093/ietfec/e91-a.7.1741.
and self-discharge reactions in vanadium redox flow battery, J. Power Sources [44] J. Newman, K.E. Thomas-Alyea, Electrochemical Systems, third ed., Wiley-
216 (2012) 489–501, http://dx.doi.org/10.1016/j.jpowsour.2012.06.052. Interscience, Hoboken, NJ, USA, 2004.
[33] A.K. Sharma, C.Y. Ling, E. Birgersson, M. Vynnycky, M. Han, Verified reduction [45] K.W. Knehr, E.C. Kumbur, Open circuit voltage of vanadium redox flow batteries:
of dimensionality for an all-vanadium redox flow battery model, J. Power Sources Discrepancy between models and experiments, Electrochem. Commun. 13 (2011)
279 (2015) 345–350, http://dx.doi.org/10.1016/j.jpowsour.2015.01.019. 342–345, http://dx.doi.org/10.1016/j.elecom.2011.01.020.
[34] Y. Chen, Z. Xu, C. Wang, J. Bao, B. Koeppel, L. Yan, P. Gao, W. Wang, Analytical [46] D.E. Eapen, S.R. Choudhury, R. Rengaswamy, Low grade heat recovery for power
modeling for redox flow battery design, J. Power Sources 482 (2021) 228817, generation through electrochemical route: vanadium redox flow battery, A case
http://dx.doi.org/10.1016/j.jpowsour.2020.228817. study, Appl. Surf. Sci. 474 (2019) 262–268, http://dx.doi.org/10.1016/j.apsusc.
[35] M.M. Tomadakis, T.J. Robertson, Viscous permeability of random fiber struc- 2018.02.025.
tures: comparison of electrical and diffusional estimates with experimental and [47] J. Bao, V. Murugesan, C.J. Kamp, Y. Shao, L. Yan, W. Wang, Machine learning
analytical results, J. Compos. Mater. 39 (2005) 163–188, http://dx.doi.org/10. coupled multi-scale modeling for redox flow batteries, Adv. Theory Simul. 3
1177/0021998305046438. (2019) 1900167, http://dx.doi.org/10.1002/adts.201900167.
[36] A. Tang, J. Bao, M. Skyllas-Kazacos, Studies on pressure losses and flow rate [48] T.E. Springer, T.A. Zawodzinski, S. Gottesfeld, Polymer electrolyte fuel cell
optimization in vanadium redox flow battery, J. Power Sources 248 (2014) model, J. Electrochem. Soc. 138 (1991) 2334–2342.
154–162, http://dx.doi.org/10.1016/j.jpowsour.2013.09.071. [49] A.F. Mills, Basic Heat and Mass Transfer, second ed., Prentice Hall, Upper Saddle
[37] D. Schmal, J. Van Erkel, P.J. Van Duin, Mass transfer at carbon fibre elec- River, NJ, 1999.
trodes, J. Appl. Electrochem. 16 (1986) 422–430, http://dx.doi.org/10.1007/ [50] A. Bourke, M.A. Miller, R.P. Lynch, X. Gao, J. Landon, J.S. Wainright, R.F.
BF01008853. Savinell, D.N. Buckley, Electrode kinetics of vanadium flow batteries: contrasting
[38] B. Delanghe, S. Tellier, M. Astruc, Mass transfer to a carbon or graphite felt responses of V II -V III and V IV -V V to electrochemical pretreatment of
electrode, Electrochim. Acta 35 (1990) 1369–1376, http://dx.doi.org/10.1016/ carbon, J. Electrochem. Soc. 163 (2016) A5097–A5105, http://dx.doi.org/10.
0013-4686(90)85008-B. 1149/2.0131601jes.
[39] Q. Xu, T.S. Zhao, Determination of the mass-transport properties of vana- [51] Y. Song, X. Li, C. Yan, A. Tang, Uncovering ionic conductivity impact towards
dium ions through the porous electrodes of vanadium redox flow batteries, high power vanadium flow battery design and operation, J. Power Sources 480
Phys. Chem. Chem. Phys. 15 (2013) 10841–10848, http://dx.doi.org/10.1039/ (2020) 229141, http://dx.doi.org/10.1016/j.jpowsour.2020.229141.
c3cp51944a. [52] R. Carta, S. Palmas, A.M. Polcaro, G. Tola, Behaviour of a carbon felt flow by
[40] J.L. Barton, J.D. Milshtein, J.J. Hinricher, F.R. Brushett, Quantifying the impact electrodes Part I: mass transfer characteristics, J. Appl. Electrochem. 21 (1991)
of viscosity on mass-transfer coefficients in redox flow batteries, J. Power Sources 793–798, http://dx.doi.org/10.1007/BF01402816.
399 (2018) 133–143, http://dx.doi.org/10.1016/j.jpowsour.2018.07.046.

12

You might also like