You are on page 1of 11

Journal of The Electrochemical Society, 164 (11) E3081-E3091 (2017) E3081

JES FOCUS ISSUE ON MATHEMATICAL MODELING OF ELECTROCHEMICAL SYSTEMS AT MULTIPLE SCALES IN HONOR OF JOHN NEWMAN
The Relationship between Shunt Currents and Edge Corrosion in
Flow Batteries
Robert M. Darling,a,b,∗,z Huai-Suen Shiau,c Adam Z. Weber,c,∗ and Mike L. Perryb,∗
a Joint Center for Energy Storage Research
b United Technologies Research Center, East Hartford, Connecticut 06108, USA
c Lawrence Berkeley National Laboratory, Berkeley, California 94720, USA

Shunt currents occur in electrochemical reactors like flow batteries, electrolyzers, and fuel cells where many bipolar cells that are
connected in series electrically contact a mobile electrolyte through one or more common fluid distribution manifolds. Shunt currents
reduce energy efficiency, and can cause unwanted side reactions including corrosion and gas generation. Equivalent-circuit models
have been widely used to examine shunt currents in multi-cell electrochemical reactors. However, a detailed investigation of the
interesting electrochemical processes occurring at the edges of the active areas has not been presented. In this work, the generation
of shunt currents and their tendency to drive corrosion at the edges of positive electrodes in the most positive cells in a reactor stack
are investigated with a comprehensive numerical model. An analytical model based on the penetration of current into a semi-infinite
electrode, that can be used in conjunction with traditional equivalent-circuit models to assess the tendency for shunt currents to
drive corrosion, is developed and compared to the numerical model. The models provided here can be used to set requirements on
maximum allowable port currents in order to achieve a particular durability goal.
© The Author(s) 2017. Published by ECS. This is an open access article distributed under the terms of the Creative Commons
Attribution 4.0 License (CC BY, http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse of the work in any
medium, provided the original work is properly cited. [DOI: 10.1149/2.0081711jes] All rights reserved.

Manuscript submitted October 25, 2016; revised manuscript received March 20, 2017. Published April 4, 2017. This paper is part
of the JES Focus Issue on Mathematical Modeling of Electrochemical Systems at Multiple Scales in Honor of John Newman.

Shunt currents are an important source of inefficiency in elec-


trochemical reactors like flow batteries, electrolyzers, and fuel cells
where many bipolar cells are connected electrically in series and con-
tact a mobile electrolyte through one or more common fluid distribu-
tion manifolds. Figure 1 shows a flow-battery stack with arrows added
to illustrate how one of the two electrolyte streams enters through a
single inlet manifold, distributes to the individual cells, and recom-
bines and leaves through an exit manifold. The other electrolyte traces
a similar path through the stack in a second set of inlet and exit man-
ifolds. The fluid manifolds act as ionic short circuits between cells in
the reactor stack. The shunt currents that flow through the manifolds
continuously discharge the reactants and can drive parasitic reactions
including corrosion that hastens battery failure and gas generation
that represents a safety hazard. Shunt currents are a particularly acute
concern in typical flow batteries because very conductive electrolytes
circulate through the reactors. Thus, minimizing the deleterious effects
of shunt currents is a primary concern of stack designers. This can be
accomplished by making the liquid paths outside of the active area
long and narrow to increase ionic resistance at the cost of increasing
the pumping work associated with circulation. While shunt currents
can be minimized in this way, they cannot be completely eliminated.
Therefore, understanding how shunt currents emanate and terminate
at the edges of the active areas is critical to the successful operation Figure 1. A flow-battery stack with arrows added to illustrate how flow is
of bipolar electrochemical reactors. distributed from a common inlet manifold at the bottom of the stack, through
Figure 2 is a picture of a portion of the junction between the active individual cells, and then through a common exit manifold.
area of a bipolar plate and its surrounding plastic frame. This picture
was taken during the disassembly of an all-vanadium flow battery,
similar to the one shown in Figure 1, after a lengthy course of cycles. carbon on the tops of the ribs between the channels on the bipolar
The active area is comprised of graphite, or a carbon composite, so plate caused by corrosion is visible in the region bounded by red lines
that it is electrically conductive. The surrounding frame, commonly directly adjacent to the port. The corrosion extends down the sides and
referred to as the port, is electrically insulating to prevent electro- bottoms of the channels. The visibly corroded region extends inwards
chemical reactions and contains labyrinth channels and the reactant approximately 1 mm from the junction with the port on the positive
manifolds (not shown in Figure 2) which are holes in the frame that sides of the most positive cells in the stack. Explaining the operating
allow flow in the direction normal to the page. The labyrinth channels conditions, chemical properties, and design features responsible for
in the port are designed to have high ionic resistance to minimize shunt this localized corrosion are the goals of this paper.
currents and to distribute flow uniformly from the reactant manifold Shunt currents can result in large overpotentials near the edges of
to the inlet edge of the active area. The active area of this bipolar plate bipolar plates as illustrated in Figure 3, which depicts the potentials
contains channels that distribute flow to the electrodes. Discolored in the solid, 1 , and electrolyte, 2 , phases in the vicinity of a port
connected to a positive electrode near the positive end of a cell stack.
∗ Electrochemical Society Member. The potential in the solid phase in the active area is essentially constant
z
E-mail: darlinrm@utrc.utc.com because of the high conductivity of carbon. The electrolyte potential

Downloaded on 2017-11-09 to IP 190.111.30.254 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).
E3082 Journal of The Electrochemical Society, 164 (11) E3081-E3091 (2017)

Id,1 Id,2 Id,3 Id,4


Ec Ec Ec Ec

Rc Rc Rc Rc
Ip,1 Ip,2 Ip,3 Ip,4

Rp Rp Rp Rp
Rm Rm Rm

Is,1 Is,2 Is,3

Figure 4. Equivalent circuit representation of an electrochemical reactor con-


sisting of four cells where the negative electrodes connect to a common elec-
trolyte manifold.

manifold to the active area. The manifold resistance depends on the


Figure 2. Picture showing discoloration on the edges of the ribs separating
adjacent channels on a bipolar plate caused by corrosion near the insulating
conductivity of the electrolyte, the cross-sectional area of the mani-
frame that distributes incoming flow. The corroded region is between the red fold, and the thickness of the cells. The currents through the different
lines. The brightness and contrast of the image have been adjusted to accentuate resistors are the drain current, Id , the port current, Ip , and the shunt
the corrosion. current Is . The convention adopted here is that a port current is positive
when it flows from a cell to the manifold. In practice, positive port
currents occur near the positive end of the stack, while negative port
is drawn with a relatively slight decline through the active area and a currents (from the manifold to the active area) occur near the negative
more substantial decline through the port. The port is designed to have end of the stack.
high resistance to minimize parasitic currents in order to ensure high A weakness of most circuit models is that they do not describe the
efficiency. The port current moves down the potential gradient, which electrochemical reactions that occur to support the shunt current. Katz
can be calculated using Ohm’s law, and into the manifold that connects addressed this deficiency by adding an impedance element (a Zener
the positive electrodes of all cells in the reactor. The extension of this diode) between the port and the cell that had the polarization char-
potential gradient into the active area is of particular concern because acteristics of the edge materials, either carbon or platinum supported
deleterious side reactions can occur at the edge of the active area if on carbon, in the relevant chemical environment.12 More recently,
the potential difference (1 -2 )edge becomes sufficiently large. Bennett et al. presented a conceptually similar analysis of a polymer-
Equivalent circuits, like the one depicted in Figure 4, have been electrolyte fuel cell with an electrically conducting (316L stainless
widely used to examine: the magnitudes of shunt currents,1 schemes steel) manifold.13 Equivalent circuits that include reaction elements
for connecting stacks2,3 including microfluidic arrays4 , the associ- have also been explored by White et al.14 and Kuhn and Booth.15 A
ated heat generation5 and capacity loss6 , mitigation schemes7 , and problem with the approach taken by all of these authors is that the
the trades between reactor efficiency and pump work.8–11 Figure 4 area over which the reactions supporting the shunt currents occur must
shows an equivalent-circuit diagram for four cells where the nega- be specified. In a redox flow battery with conductive bipolar plates
tive electrodes of each cell are connected by a common electrolyte surrounded by an insulating plastic frame that distributes flow, for ex-
manifold. This configuration was studied by Katz12 in the context of ample, the portion of the active area near the junction with the frame
phosphoric acid fuel cells, and we adopt his notation in this paper. that supplies current to the manifold is not well delineated. Rather,
The theoretical open-circuit voltage of each cell is Ec . Rc is the inter- the size of this region is determined by interactions among kinetic,
nal resistance (which encompasses ohmic, kinetic, and mass-transport transport, and ohmic losses near the edge.
contributions) in the active area of each cell, Rp is the ionic resistance Various researchers have modeled multi-dimensional current dis-
in the electrically insulating port region of each cell, and Rm is the tributions in stacks containing a few undivided bipolar cells.16–18 Re-
ionic resistance in a manifold section between adjacent cells. The actions near the edges are inherently treated in these models (assuming
values or Rc , Rp , and Rm do not vary with the position of the cell in that all relevant reactions are included in the formulation), at the cost
this example, and therefore do not have distinguishing subscripts. The of increased complexity and computing power when compared to
port resistance depends on the conductivity of the electrolyte and the traditional equivalent-circuit models. Yin et al.19 recently published
length and cross-sectional area of the labyrinth path connecting the a fully rendered three-dimensional model of a vanadium redox flow
battery containing from 5 to 20 cells that included parametric studies
of port and shunt currents. While the scope of this model is impres-
sive, a detailed consideration of the processes occurring at the edge
Active area Port of the active area and how these can cause corrosion, as has been ob-
served experimentally (see Figure 2), was not included. Scaling this
model to commercially relevant active areas and numbers of cells with
Φ1 the resolution necessary to elucidate corrosion mechanisms would re-
quire considerable computing power. To overcome these difficulties,
Φ2 we use numerical simulations of a simplified model system to de-
velop our intuition regarding processes at the edge and to support the
development of an analytical model of the edge that can be used in
conjunction with traditional equivalent-circuit models to understand
Port current when corrosion will occur at an unacceptable rate.
In this paper, numerical simulations of a flow-battery stack with
Figure 3. Schematic depiction of the potentials in the solid and electrolyte two cells at steady state are described and used to assess the impact of
phases near the junction between the active area and the port. The electrolyte the port and manifold resistances on the overpotential at the junction
phase potential drops to support the port current, following Ohm’s law. between the more positive bipolar plate and the adjacent insulating

Downloaded on 2017-11-09 to IP 190.111.30.254 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).
Journal of The Electrochemical Society, 164 (11) E3081-E3091 (2017) E3083

Table I. Species descriptors. ion-exchange membrane separator are:


 
k
Species V2+ V3+ VO2+ VO2 + H+ SO4 2− Momentum conservation : ∇ · −

v = ∇ · − ∇p = 0 [5]
μ
Vanadium oxidation state 2 3 4 5 - -
Charge number, zi 2 3 2 1 1 −2
Acronym and subscript V2 V3 V4 V5 H SO4 Species conservation :
 
z i F Di ci −
→ a j+or −
∇ · −Di ∇ci − ∇2 + v ci = [6]
port. The model is later expanded to include a negative manifold RT F
and more cells in order to understand how these variables influence
the behavior near the edges of the most positive cell. Simplifications
are made that allow one to focus on the conditions that cause corro- Charge conservation :
sion. For example, open-circuit conditions are considered throughout, −
→ −

∇ · i 1 = −∇ · i 2 = −σ∇ 2 1 = κ∇ 2 2 = a j+or − [7]
although extension to the behavior under load should be straightfor-
ward. The active area is artificially small in the initial simulations The subscript i identifies the species from Table I, + and – refer
for the sakes of exposition and visualization. Later, simulations are to the positive and negative electrodes, − →v is the mass-averaged fluid
conducted to demonstrate how the the behavior at the edges depends velocity, μ is fluid viscosity, k is the permeability of the porous elec-
on the size of the active area. Finally, an analytical model for the trode, p is the fluid pressure, Di , ci , and zi , are the is the diffusivity,
edge based on the penetration of current into a semi-infinite electrode concentration, and charge number of species i, respectively, F is the
is presented and compared to the numerical model. The analytical Faraday constant, R is the universal gas constant, T is the absolute
model can be used in conjunction with traditional equivalent circuits temperature, a is the interfacial area per unit volume of electrode, jn
to make predictions regarding the propensity to corrode in large stacks is the interfacial transfer current density from Butler-Volmer kinetics,
with many cells, where using a detailed numerical model would be −
→ −

i 1 is the superficial electronic current in the solid phase, i 2 is the
impractical. All calculations in this paper pertain to vanadium flow superficial ionic current in the electrolyte phase, σ is the conductivity
batteries, but the approach and findings are applicable to a broad range of the solid phase, κ is the conductivity of the electrolyte phase, and
of electrochemical reactors. 1 and 2 are the solid and electrolyte potentials. The species conser-
vation equations rely on dilute-solution theory (Nernst-Planck), with
the familiar diffusion, migration, and convection terms to describe the
Background
flux of species i, and the Nernst-Einstein equation to relate mobility
The all-vanadium flow battery invented by Skyllas-Kazacos in and diffusivity. Material balances are written for protons and all of
the 1980s20,21 is the exemplar investigated in this work. However, the vanadium ions: V2, V3, V4 and V5. Electroneutrality dictates the
the approach should be valid for other types of flow batteries and concentration of the only anion, SO4 2− . Ohm’s law with a constant
electrochemical systems with similar reactors. The main reactions at conductivity is applicable to the electrolyte solutions because of the
the positive and negative electrodes in an all-vanadium flow battery high concentrations of supporting electrolyte and open-circuit con-
are:22 ditions. Reaction source terms are present in the two electrodes and
absent from the membrane separator.
Positive : VO2+ + H2 O ↔ VO+ +
2 + 2H + e

Uθ = 1.004 V Butler-Volmer expressions describe the electrochemical kinetics
[1] on the interfacial surfaces in the porous electrodes, and form the
source terms in the conservation equations. The subscripts + and −
Negative : V2+ ↔ V3+ + e− Uθ = −0.255 V [2] denote the positive and negative electrodes, respectively.
    
Table I shows oxidation states, charge numbers, and acronyms αa,+ Fη+ −αc,+ Fη+
j+ = i 0,+ exp − exp [8]
for the four vanadium species present in a flow battery, as well as RT RT
protons and sulfate. The equilibrium speciation between sulfate and
bisulfate is ignored in this work for the sake of simplicity. All standard     
αa,− Fη− −αc,− Fη−
potentials are due to Pourbaix and co-workers.22 j− = i 0,− exp − exp [9]
This paper focuses on the impact of shunt currents on the poten- RT RT
tial profiles near the edges of bipolar plates in an electrochemical All charge transfer coefficients, α, are assumed to be 0.5. The
reactor. Abnormally large interfacial overpotentials near the edges of porous electrode is assumed to consist of long cylindrical carbon fibers
the positive sides of cells near the positive end of reactor stacks can of uniform diameter, df . The exchange current densities, i0,+ and i0,− ,
lead to oxygen evolution and corrosion of carbon bipolar plates and depend on reference values and the prevailing states of charge,
electrodes:  
cH
2H2 O → O2 + 4H+ + 4e− i 0,+ = i 0,+ X + α+,a (1 − X + )α+,c
ref
O2 evolution : U = 1.229 V [3] [10]
c H,r e f

Corrosion : C + 2H2 O → CO2 + 4H+ + 4e− U = 0.207 V i 0,− = i 0,− X − α−,a (1 − X − )α−,c
ref
[11]
[4]
Oxygen evolution is hindered by poor kinetics on carbon in acid The positive and negative states of charge are:
and large overpotentials are needed to drive the reaction at high rates cV 5 cV 2
in the absence of a noble metal catalyst. The low thermodynamic X+ = ; X− = [12]
cV 5 + cV 4 cV 2 + cV 3
potential for carbon oxidation belies its stability as the kinetics of
Reaction 4 are quite sluggish, and carbon components have achieved The overpotentials in Eqs. 8 and 9 are:
decades of durability in flow batteries and fuel cells under controlled
η+ = 1 − 2 − U+ [13]
conditions.23

Numerical model.—Governing equations.—The governing dif-


ferential equations in the positive and negative electrodes and the η− = 1 − 2 − U− [14]

Downloaded on 2017-11-09 to IP 190.111.30.254 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).
E3084 Journal of The Electrochemical Society, 164 (11) E3081-E3091 (2017)

Figure 5. A schematic diagram of the modeling domain for two cells showing the boundary conditions. The thicknesses of the various regions are not shown to
scale.

The equilibrium half-cell potentials are given by Nernst equations: The numerical coefficient was modified from the original because
  the published measurements were made in a bed of packed spheres
RT cV 5 c H 2 while typical carbon papers and felts consist of long cylindrical fibers.
U+ = U+θ + ln [15]
F cV 4 Hence, the coefficient was corrected for the change in the ratio of
surface area to volume.
  We used the commercial finite element software COMSOL Mul-
RT cV 3
U− = U−θ + ln [16] tiphysics 5.1 (COMSOL, Inc., Palo Alto, CA) to solve the equations
F cV 2 presented above simultaneously. We used the MUMPS parallel sparse
direct solver with free triangular meshing. The mesh consisted of
Knehr et al. discussed the accuracy of the Nernst approximation
115484 domain elements and 3170 boundary elements for the simu-
in the vanadium system.24 Kinetic expressions for carbon oxidation
lation of two cells in series (Fig. 5).
and oxygen and hydrogen evolution are excluded from the model.
Our approach is to include only the main reactions and evaluate the
tendencies to drive possible side reactions by examining the local Boundary conditions.—Figure 5 shows the modeling domain and
overpotentials that develop to support the shunt currents. The partial the associated boundary conditions for the governing equations. At
current densities associated with carbon oxidation and gas evolution the inlet of the positive manifold, species concentrations were set
should be small fractions of the local current density in order to ensure in accordance with the prevailing state of charge (SOC), and the
longevity. Carbon corrosion and oxygen evolution are predominant at fluid pressure was set to ambient. The electronic and ionic currents
the positive end of the stack, while hydrogen evolution is predominant and the diffusive flux components of all species were set to zero at
at the negative end. Hydrogen evolution is sluggish on typical carbon the other side of the positive electrodes (or far edge), which derives
papers and felts, and a significant overpotential is needed to drive the from the simplifying assumption that the inlet and exit manifolds
reaction at appreciable rates.25 behave similarly. The far edge, away from the port, acts like a line
The concentrations of the vanadium ions at the reacting surfaces of symmetry. The fluid velocity was specified at the far edge. Only
in the porous electrodes differ from those in the bulk of the pores protons are permitted to transport through the membrane, so there is
due to limited rates of mass transport. A mass-transfer coefficient, no concentration gradient in the membrane. A continuity condition
km , was introduced to capture this effect. The interfacial and pore was applied at the two sides of each membrane to ensure that the total
concentration are related by the flux equation: flux of ions across the positive and negative electrodes is equal to the
  proton flux through the membrane. The solid potential was set to zero
Nis = km ci − cis [17] (ground) at the negative end of the stack and the current was set to
zero at the positive end of the stack to simulate open circuit.
where ci is the bulk concentration and csi is the surface concentration. Table II summarizes key physical parameters used in the simula-
A correlation due to Wilson and Geankoplis26 was used to calculate tions. Many of the parameters were taken from a study of the V4/V5
the mass-transfer coefficient. electrode by Darling and Perry,27 and identical properties were as-
0.832D 2/3 v 1/3 sumed for the V2/V3 electrode for the sake of simplicity.
km = 2/3
[18] Table III gives component dimensions used in the simulations.
εd f Initial simulations were done with unrealistically small active areas

Downloaded on 2017-11-09 to IP 190.111.30.254 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).
Journal of The Electrochemical Society, 164 (11) E3081-E3091 (2017) E3085

Table II. Physical properties.

Parameter Symbol Value Reference


Thermodynamically reversible potential Uθ 1.259 V 22
Temperature T 25◦ C Assumed
Fiber diameter df 8.3 μm 27
Thickness of electrode L 200 μm Assumed
Porosity of electrode ε 97% 27
Roughness aL 39.4 Assumed
Specific interfacial area a 1970 cm−1 Calculated
Ionic conductivity κ 307 mS/cm 30
Solid conductivity σ 2.2 S/cm 27
Membrane conductivity κm 43 mS/cm 27
Exchange current density i0 ref 12 mA/cm2 c 27
Diffusivity of all vanadium ions D 1.7 × 10−6 cm2 /s 31
Diffusivity of protons DH 9.6 × 10−6 cm2 /s Fit to conductivity
Diffusivity of sulfate ions DSO4 1.1 × 10−6 cm2 /s Fit to conductivity
Total vanadium concentration c 1.5 mol/L Assumed
Reference proton concentration cH,ref 5.2 mol/L Assumed
State of charge X+ , X- 0 to 100% Assumed
Thickness of separator Ls 50 μm Assumed
Thickness of bipolar plates Lplate 500 μm Assumed
Length of electrodes Le 2 mm Assumed
Port length Lp 0.6 mm Assumed

and thin electrodes in order to facilitate presentation of results. The The negative electrodes are not connected in this simulation in order
unusual potentials of primary interest in this work occur in a narrow to simplify the presentation and analysis. Self-discharge results in net
region near the edge of the active area, and the path that the current reduction of V5 to V4 and oxidation of V2 to V3 in the battery, thus
takes through connected cells can be displayed more clearly when lowering X+ and X− . The shunt current from the positive (top) cell to
the active area is small and the electrodes are thin. Neither of these the negative (bottom) cell gives rise to an oxidation reaction adjacent
choices affects general trends, rather they help to elucidate the phe- to the top port, Figure 6a, and a reduction reaction adjacent to the
nomena. The sensitivities of the edge region to active area follow the bottom port, Figure 6b. This agrees with the convention that a current
explanation of the basic phenomenon. is positive leaving an anode. The oxidation reaction near the port on
The port resistance is calculated from parameters in Tables II and the top cell oxidizes V4 to V5. However, carbon corrosion or oxygen
III: evolution could substitute for the benign V4 to V5 reaction in adverse
Lp circumstances, such as when the battery approaches full charge and
Rp = [19] the V4 concentration goes to zero. Reduction of V5 to V4 occurs
κp Ap
over the remainder of the positive electrode of the top cell. There is
where κ p is the ionic conductivity in the port, L p and A p are the no net reduction of V5 to V4 in the top cell, and, consequently, no
length and the cross-sectional area of the port. The ionic conductivity net contribution to self-discharge. Oxidation of V2 to V3 occurs on
in the port channels is slightly higher than in the electrodes due to the all portions of the negative side of the top cell resulting in net self-
absence of solid material. discharge of the negative electrolyte. Reduction of V5 to V4 occurs
on all portions of the positive electrode of the bottom cell. The rate
is highest near the port, and this distribution essentially mirrors the
Results and Discussion
negative electrode of the top cell. Balanced oxidation of V2 to V3 and
At open circuit, no current passes through the positive and negative reduction of V3 to V2 occurs at the negative electrode of the bottom
ends of the reactor to an external circuit, but self-discharge of the cell, resulting in no net contribution to self-discharge. Recall that no
battery occurs due to the internal parasitic currents. This is shown in net current crosses either the positive or negative end plates because
Figure 6, where the top cell is positive, as denoted on the figure, and the battery is at open circuit. The reactions on the negative side of
the bottom cell is negative. The positive electrodes of the two cells the bottom cell can be regarded as the consequence of the potential
are connected via electronically insulating ports to a single manifold. profile imposed by the positive electrode on this cell.
The behavior of this simple stack of just two small cells is fairly
complicated, but the essential points are that oxidation occurs near
Table III. Design adjustable parameters. the ports of positive electrodes near the positive end of a stack, and
reduction occurs near the ports of positive electrodes near the negative
Parameter Value end of a stack. This combination ensures that positive currents leave
anodes and enter cathodes. The oxidation reactions tend to be confined
Length of active area 2 mm to 2 cm to small regions because there is a net tendency to self-discharge by
Length of port (Lp ) 0.6 mm to 10 cm reduction of V5 to V4 in the positive electrolyte. These small regions
Cross-sectional area of port∗ 200 μm (electrode thickness) where oxidation occurs are susceptible to carbon corrosion and oxygen
× 1.56 mm (port width).
evolution, as will be shown more clearly by the potential profiles.
Manifold length 1650 μm
Figure 7 shows the electrolytic potential in the positive electrode of
Manifold width 100 μm
Conductivity in port (κ p ) κ p = κ = 307 mS/cm
the top cell and the adjacent port. The potential drops monotonically
Conductivity in electrode (κe ) κe = κ ε 1.5 = 293 mS/cm
from the far edge of the electrode to the junction with the port. Thus,
because the potential in the solid phase is essentially constant, the
∗ Variations in the direction of the port width are excluded because the driving force for oxidation reactions, like carbon corrosion and oxygen
model is two dimensional. The port length is used in the calculation of evolution, is largest near the port. The potential profile through the
the port resistance. port is linear, following Ohm’s law, and the drop is nearly 200 mV.

Downloaded on 2017-11-09 to IP 190.111.30.254 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).
E3086 Journal of The Electrochemical Society, 164 (11) E3081-E3091 (2017)

Figure 6. (a) Oxidative and (b) reductive current densities in the edge region of a stack with two cells at open circuit. The active area is 2 mm long and the port is
0.6 mm long.

In the example considered here, an identical potential drop occurs in


the positive port adjacent to the bottom cell, because the same current
passes through both ports and the resistances of the two ports are
equal.
Figure 8 shows how the overpotential in the positive electrode of
the positive cell varies with distance from the edge. The junction be-
tween the port and the active area is at 2 mm. Oxidation of V4 to V5
occurs when the overpotential is positive, while reduction of V5 to
V4 occurs when the overpotential is negative. Oxidation of vanadium
is confined to the 500 μm nearest to the port. The overpotential im-
mediately adjacent to the port is approximately 80 mV greater than
the reversible potential. Therefore, the driving force for deleterious
reactions is significantly greater at the interface with the port than in
the majority of the active area. The Tafel slope for carbon oxidation is
typically ∼200 mV,28 so an additional 80 mV of overpotential should
increase the corrosion rate by approximately 2.5X.
Figure 9 shows the overpotential at the junction between the pos-
itive electrode of the top cell and the adjacent port as a function of Figure 8. Overpotential in the positive electrode of the top (positive) cell
depicted in Figure 5 near the junction between the active area and the insulating
port resistance at 20%, 50%, 90% and 99% SOC. The overpotential at frame (located at 2 mm). The overpotential is approximately 80 mV at the
junction between the port and electrode.

Figure 9. Overpotential at the junction between the positive electrode of the


Figure 7. Electrolyte potential (2 ) near the port on the positive side of the top cell depicted in Figure 5 and the adjacent port as a function of port resistance
top cell depicted in Figure 5. The potential drops ∼100 mV near the edge of at various states of charge. The port resistances are always equal on the two
the electrode and a further ∼200 mV through the port. cells.

Downloaded on 2017-11-09 to IP 190.111.30.254 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).
Journal of The Electrochemical Society, 164 (11) E3081-E3091 (2017) E3087

Figure 12 provides the distributions of oxidative and reductive


current densities in a stack of two cells at open circuit, where one
manifold connects the positive electrodes of both cells and a sec-
ond manifold connects the negative electrodes. The positive manifold
is on the right of the stack, while the negative manifold is on the
left. Different boundary conditions were employed for these simula-
tions. The ionic currents exiting the electrodes on the sides opposite
the manifolds were forced to be equal in magnitude and opposite in
direction to the currents on the manifold side. Thus, the simulated
region includes complete cells with inlets and exits (although the exit
manifold is treated simplistically) instead of cells cut in the middle.
Oxidation occurs on both edges of the positive electrode at the posi-
tive end. Mirroring this, reduction reactions occur at the edges of the
negative electrode at the negative end of the stack. The magnitudes of
the port currents in the top and bottom cells were 3.5 mA when only
the positive electrodes were connected via a single manifold. These
Figure 10. The effect of electrode length on port current. The length of the currents decreased by 6% in magnitude when the second manifold
oxidizing region (η+ > 0) is plotted on the secondary y-axis. connecting the negative sides of the cells was added. This indicates
that the reactions associated with shunt currents in the positive and
negative electrolyte manifolds are only weakly coupled.
the junction increases with decreasing port resistance and increasing Both of the cells in a two-cell reactor are unusual in that the net
SOC. Thus, a large port resistance is necessary to avoid localized cor- current is zero one of the electrodes and non-zero on the other. At
rosion, as well as to minimize efficiency losses due to shunt currents, least three cells are required in order to have a cell that is free of
and operating at high SOC significantly increases the risk of localized these peculiar end effects. Figure 13 shows the distributions of (a)
corrosion. oxidative and (b) reductive current densities in a stack with five cells
Figure 10 shows how port current varies with the length of the where the positive electrodes area connected by a common manifold
active area. The port current increases from 3.54 to 3.61 mA when on the right. Oxidation occurs near the edge on the positive sides of the
the active area lengthens from 2 mm to 2 cm. This corresponds to an two most positive cells. The rate of oxidation is highest on the most
increase of < 2% in current when the length increases by a factor of positive cell. Rapid reduction occurs near the edges on the positive
10. Thus, the port current asymptotes as the active area gets longer. sides of the two most negative cells. The rate of reduction is highest
The length of the region over which oxidation occurs (η+ > 0) is on the most negative cell. The edge of the middle cell appears to be
plotted against the right vertical axis. The lengths of this region are indistinguishable from the rest of the middle cell. The port currents
0.5 mm and 1 mm when the lengths of the active area are 2 mm and from bottom (negative) to top (positive) are: −5.1 mA, −1.1 mA, 0
2 cm, respectively. Thus, the length of the oxidizing region increases mA, 1.1 mA, and 5.1 mA. The currents are very nearly symmetric
by 2x while the length of the active area increases 10x. The monotonic about the center cell because the kinetics of the V2/V3 and V4/V5
increase in the size of the oxidizing region appears to contradict the reaction are symmetric since all charge transfer coefficients are 0.5.
asymptotic behavior of the port current. This apparent discrepancy Including the side reactions like carbon corrosion, oxygen evolution,
can be reconciled by examining the overpotential profiles for different and hydrogen evolution would break this symmetry provided they
electrode lengths in Figure 11. The overpotential at the extreme edge occurred at appreciable rates. The port currents at the end cells are
increases slowly as the electrode gets longer, while the magnitude higher in the five-cell stack than the two-cell stack because the overall
of the slope of the overpotential where it crosses η+ = 0 decreases voltage driving force is larger because of the additional cells.
sharply as the port gets longer. Thus, the curves pick up a long tail that
contributes comparatively little to the current due to the exponential
Analytical Model
weighting introduced by the Butler-Volmer equation. The net result
is that the contribution of the extended oxidation region to the total Modeling a large multi-cell stack at the resolution necessary to
amount of oxidation becomes less significant as the active area be- observe the behavior near the edges by the methods discussed in the
comes larger, as shown in the right graph of Figure 11. Therefore, the previous section is inconvenient and, therefore, an analytical model
average amount of oxidation found by integrating along the electrode of the edge region that can be used in conjunction with traditional
length exhibits asymptotic behavior like the port current. equivalent-circuit models is desirable. Describing the overpotential in

Figure 11. Left: The effect of electrode length on the overpotential distribution in the oxidizing region. The extreme left of the x-axis is the edge overpotential.
Right: The average rate of oxidation along the electrode [A/cm2 ] found by integrating over the oxidizing region versus electrode length.

Downloaded on 2017-11-09 to IP 190.111.30.254 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).
E3088 Journal of The Electrochemical Society, 164 (11) E3081-E3091 (2017)

Figure 12. (a) Oxidative and (b) reductive current densities in a stack with two cells at open circuit. One manifold connects the positive electrodes, while a second
manifold connects the negative electrodes.

Downloaded on 2017-11-09 to IP 190.111.30.254 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).
Journal of The Electrochemical Society, 164 (11) E3081-E3091 (2017) E3089

Figure 13. (a) Oxidative and (b) reductive reaction rates near the edge of a reactor with five cells at open circuit.

the edge regions that support oxidation reactions in the positive elec- surface that the electrolyte encounters when leaving the port should
trodes at the positive end of the stack is of particular interest because have high surface area (large a) and be active for the main battery
of their susceptibility to corrosion. These regions are generally small reactions (large i0 ) in order to avoid corrosion. Equation 20 does not
compared to the active area of the stack and we will make use of this explicitly include or exclude flow channels, rather this detail enters in
fact in our approximation. Newman29 provides the equation: the calculation of the active area per unit volume at the edge.
      The resistances associated with the desired reactions at the edge
2ai 0 κRT αa Fη αc Fη should be small in comparison to the port and manifold resistances in a
i 2p = αc exp + αa exp − − αa − αc
αa αc F RT RT successful reactor design. Therefore, as a first approximation, one can
[20] estimate the port current from the port, manifold, and cell resistances
for the gross current-potential behavior of a semi-infinite electrode and the voltage across the stack, while neglecting the impedance of the
with constant potential in the solid phase. This equation addresses edge region. This is exactly the approach taken in conventional circuit
kinetic and ohmic effects, but neglects mass transport. This one- models. Equation 20 can then be used to calculate an upper bound on
dimensional treatment (treating variations in the x-direction in Fig. 6) the overpotential at the edge as a function of the port current density.
should provide an upper bound on the overpotential because the move- This procedure indicates that the overpotential at the edge will be
ment of ions in the orthogonal direction through the membrane should higher when the battery is charging because the higher stack voltage
act to shrink the size of the oxidizing region and reduce the associated will cause larger port currents, and, conversely, the overpotential will
ohmic drop. Because the area of an edge region is a small fraction of be lower when the battery is discharging.
the total active area, changes in the concentrations of active species In practice, the port and manifold are designed to achieve accept-
by consumption and generation in the edge region can be neglected able port and shunt currents. If we therefore regard the port current as a
to a good approximation when the flow is large enough to support a fixed quantity, then the current density in the port will tend to increase
typical operating current in the preponderance of the active area. This as the electrode is made thinner. A higher current density will polar-
assumption of excess flow allows us to neglect the material balances ize the edge more according to Equation 20. Consequently, thinner
in our analytical treatment and use Equation 20 without modification. electrodes should be more prone to corrosion than thick electrodes.
Mass transport may limit the rate of vanadium oxidation if the carbon However, thin carbon paper electrodes tend to have higher interfacial
surface area is low or when the concentration of VO2+ is low. The area per unit volume than thick carbon felt electrodes, which will tend
first condition may occur over flat carbon surfaces, while the second to counteract this tendency. Finally, cursory inspection of Equation 20
condition may occur at high SOC. The first electrically conducting indicates that the conductivity should be large in order to reduce the

Downloaded on 2017-11-09 to IP 190.111.30.254 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).
E3090 Journal of The Electrochemical Society, 164 (11) E3081-E3091 (2017)

0.1 in flow batteries. A formula useful for bounding the maximum port
Numerical corrected for far edge potential current that avoids excessive corrosion at the edge of the active area
Numerical model of electrochemical cells stacked in series and connected to a com-
0.08 Analytical model mon manifold was developed. The formula depends on the design of
the edge region, the kinetics of the corrosion reactions, the intended
Edge overpotential [V]

service life, and operating conditions. Operating with highly charged


0.06 electrolyte and at high potentials drives corrosion and limits the allow-
able port current. The analytical formula can be used in conjunction
with equivalent-circuit models of shunt currents in electrochemical
0.04 reactors to size port and manifold resistances.

Acknowledgments
0.02
Work by Robert Darling was supported as part of the Joint Cen-
ter for Energy Storage Research, an Energy Innovation Hub funded
0
by the U.S. Department of Energy, Office of Science, Basic Energy
0 1 2 3 4
Sciences. LBNL acknowledge financial support by EERE, Fuel Cell
Technologies Office of the U.S. DOE under contract number DE-
Port current density [mA]
AC02-05CH11231. UTRC would like to recognize their colleagues at
Vionx Energy for incorporating these concepts into practical battery
Figure 14. Overpotentials (1 −2 −U+ ) at the edge of the active area versus
port current densities calculated with the numerical and analytical models. designs.

overpotential. This conclusion is erroneous because a large conductiv- List of Symbols


ity will reduce the port and manifold resistances, which will increase a interfacial area per unit volume, cm−1
the port current, which has a greater impact on the overpotential than A area, cm2
the conductivity appearing explicitly in Equation 20. c concentration, mol/cm3
Figure 14 compares the overpotentials, the solid potential minus df fiber diameter, μm
the electrolyte and equilibrium potentials, at the port junction as a D diffusion coefficient, cm2 /s
function of port current densities predicted by the analytical and nu- E reversible voltage, V
merical models when the active area is 2 mm long. The green line F Faraday constant, C/mol
is the numerical model corrected for the potential at the far edge. i current density, A/cm2
The analytical model yields systematically larger overpotentials than j interfacial current density, A/cm2
the numerical model because the numerical model is run with a fi- k permeability, cm2
nite, not infinite electrode, and because the analytical model is one km mass-transfer coefficient, cm/s
dimensional and ignores the movement of ions through the membrane L electrode thickness, cm
which shortens the path that charge must travel in the electrode to N flux, mol/cm2 · s
support the port current. Nevertheless, the analytical model provides p pressure, Pa
a useful upper bound on the corrosion risks associated with shunt R universal gas constant, J/mol · K
currents in a stack. For example, one could calculate the edge overpo- R resistance, 
tentials associated with the port currents calculated by a conventional T absolute temperature, K
equivalent-circuit model and then judge whether the design should U reversible voltage, V
achieve the intended durability. This decoupling tacitly assumes that v velocity, cm/s
the resistance elements are limiting and that the impedance offered X state of charge
by the reactions at the edge is of secondary importance. Alternatively, z charge number
non-linear elements containing the I-V relationship of Equation 20
could be added to the equivalent-circuit model ahead of the port resis- Greek
tance in analogy with the approach recommended by Katz,12 although
without the requirement of specifying the area available for reaction. α transfer coefficient
The analytic treatment includes only a single reaction near the ε porosity
edge. Conceptually, we envision that the user would set a maximum  potential, V
allowable augmentation to the interfacial potential drop and then deter- η overpotential, V
mine a port current that yielded this value. Thus, we have a statement κ ionic conductivity, S/cm
for the maximum allowable potential at the edge that depends on μ viscosity, Pa · s
electrolyte properties, the geometry of the edge including electrode σ solid conductivity, S/cm
thickness, the limiting current for reaction of the redox couple, and the
port current. Overall, the potential at the edge should be set below a Superscripts
threshold that gives acceptably low carbon corrosion over the required θ standard state
life of the battery. ref reference state
s surface
Conclusions
Subscripts
The electrochemical behavior near the junction between the active
area and the insulating frame, or port, in a flow battery was studied + positive electrode
with the aid of numerical and analytical models. It was demonstrated − negative electrode
that the overpotential for oxidation on the positive side of the posi- 0 at zero current
tive cells can be significantly higher at the edge than in the bulk of 1 solid phase
the active area. This can result in undesirable side reactions, such as 2 electrolyte phase
localized carbon corrosion, which has been experimentally observed a anodic

Downloaded on 2017-11-09 to IP 190.111.30.254 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).
Journal of The Electrochemical Society, 164 (11) E3081-E3091 (2017) E3091

c cathodic 12. M. Katz, J. Electrochem. Soc., 125, 515 (1978).


c cell 13. W. R. Bennett, M. A. Hoberecht, and V. F. Lvovich, J. Electroanal. Chem., 737, 162
(2015).
d drain 14. R. E. White et al., J. Electrochem. Soc., 133, 485 (1986).
i species identifier 15. A. T. Kuhn and J. S. Booth, J. Appl. Electrochem., 10, 233 (1980).
p port 16. E. R. Henquin and J. M. Bisang, J. Appl. Electrochem., 35, 1183 (2005).
s shunt 17. E. C. Dimoault-Darcy and R. E. White, J. Electrochem. Soc., 135, 656 (1988).
18. J. Divisek, R. Jung, and D. Britz, J. Appl. Electrochem., 20, 186 (1990).
19. C. Yin, S. Guo, H. Fang, J. Liu, Y. Li, and H. Tang, Applied Energy, 151, 237
References (2015).
20. E. Sum, M. Rychcik, and M. Skyllas-Kazacos, J. Power Sources, 16, 85 (1985).
1. E. A. Kaminiski and R. F. Savinell, J. Electrochem. Soc., 130, 1103 (1983). 21. E. Sum and M. Skyllas-Kazacos, J. Power Sources, 15, 179 (1985).
2. F. T. Wandschneider, S. Rohm, P. Fischer, K. Pinkwart, J. Tubke, and H. Nirschl, J. 22. E. Deltombe, N. De Zoubov, and M. Pourbaix, in Atlas of Electrochemical Equilibria
Power Sources, 261, 64 (2014). in Aqueous Solutions, 2nd ed., M. Pourbaix, Editor, p.234, National Association of
3. S. Konig, M. R. Suriyah, and T. Leibfried, J. Powers Sources, 281, 272 (2015). Corrosion Engineers, Houston (1974).
4. H. Wang, S. Gu, D. Y. C. Leung, H. Xu, M. K. H. Leung, L. Zhang, and J. Xuan, 23. K. Kinoshita, Carbon Electrochemical and Physiochemical Properties, p. 316, John
Electorchimica Acta, 135, 467 (2014). Wiley & Sons, New York (1988).
5. A. Tang, J. McCann, J. Bao, and M. Skyllas-Kazacos, J. Powers Sources, 242, 349 24. K. W. Knehr and E. C. Kumbar, Electrochemistry Communications, 13, 342
(2013). (2011).
6. F. Xing, H. Zhang, and X. Ma, J. Powers Sources, 196, 10753. (2011). 25. K. Kinoshita, Carbon Electrochemical and Physiochemical Properties, p. 372, John
7. J. A. Schaeffer, L. Chen, and J. P. Seaba, J. Power Sources, 182, 599 (2008). Wiley & Sons, New York (1988).
8. Z. Yu, J. Zhao, P. Wang, M. Skyllas-Kazacos, B. Xiong, and R. Badrinarayanan, J. 26. E. J. Wilson and C. J. Geankoplis, I&EC Fundamentals, 5, 9 (1966).
Powers Sources, 290, 14 (2015). 27. R. M. Darling and M. L. Perry, J. Electrochem, Soc., 161, A1381 (2014).
9. Q. Ye, J. Hu, P. Cheng, and Z. Ma, J. Powers Sources, 296, 352 (2015). 28. K. Kinoshita, Carbon Electrochemical and Physiochemical Properties, p. 317, John
10. A. Crawford, V. Viswanathan, D. Stephenson, W. Wang, E. Thomsen, D. Reed, B. Li, Wiley & Sons, New York (1988).
P. Balducci, and M. Kintner-Meyer, J. Power Sources, 293, 388. (2015). 29. J. S. Newman, Electrochemical Systems, 2nd ed., p. 465, Prentice-Hall, Inc.,
11. V. Viswanathan, A. Crawford, D. Stephenson, S. Kim, W. Wang, B. Li, G. Coffey, Englewood-Cliffs (1991).
E. Thomsen, G. Graff, P. Balducci, M. Kintner-Meyer, and V. Sprenkle, J. Power 30. M. Skyllas-Kazacos and M. Kazacos, J. Power Sources, 196, 8822 (2011).
Sources, 247, 1040 (2014). 31. G. Oriji, Y. Katayama, and T. Miura, Electrochimica Acta, 49, 3091 (2004).

Downloaded on 2017-11-09 to IP 190.111.30.254 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).

You might also like