C3186 Journal of The Electrochemical Society, 166 (11) C3186-C3192 (2019)

JES FOCUS ISSUE ON ELECTROCHEMICAL TECHNIQUES IN CORROSION SCIENCE IN MEMORY OF HUGH ISAACS
Pourbaix Diagrams as a Simple Route to First Principles Corrosion
Simulation
S. C. Perry, 1,∗ S. M. Gateman,1,∗ L. I. Stephens, 1,∗
R. Lacasse,2 R. Schulz,2
and J. Mauzeroll 1,∗∗,z
1 Laboratory for Electrochemical Reactive Imaging and Detection of Biological Systems, McGill University, Montreal,
QC H3A 0B8, Canada
2 Institut de recherche d’Hydro-Quebec, Varennes J3X 1S1, Canada

Simulation of corrosion reactions has become an important tool in the analysis of novel materials for corrosion resistant materials
and applications. A number of analytical, boundary element and finite element methods currently exist in the literature, which use
experimental corrosion parameters as model inputs for the extraction of simulated polarization curves, oxide growth rates, and other
corrosion behaviors. Here, we propose the combination of finite element simulation with thermodynamic data from Pourbaix diagrams
to allow for the simulation of metal corrosion, where individual corrosion reactions can be monitored at the metal surface.
© The Author(s) 2019. Published by ECS. This is an open access article distributed under the terms of the Creative Commons
Attribution Non-Commercial No Derivatives 4.0 License (CC BY-NC-ND, http://creativecommons.org/licenses/by-nc-nd/4.0/),
which permits non-commercial reuse, distribution, and reproduction in any medium, provided the original work is not changed in any
way and is properly cited. For permission for commercial reuse, please email: [email protected]. [DOI: 10.1149/2.0111911jes]

Manuscript submitted January 21, 2019; revised manuscript received March 27, 2019. Published May 2, 2019. This paper is part of
the JES Focus Issue on Electrochemical Techniques in Corrosion Science in Memory of Hugh Isaacs.

In recent years simulation analysis has become commonplace presents a great challenge for the construction of simulations, as un-
in corrosion literature. Novel analytical and numerical expressions certainty around the input parameters makes it difficult to accurately
have been developed to describe uniform corrosion,1–5 as well as lo- validate the model output.
calized corrosion events such as pitting,6–12 crevice13–16 or galvanic Here, we propose the use of thermodynamic material data from
corrosion.17–21 The use of an analytical solution is often preferable Pourbaix diagrams as the input for finite element method (FEM) sim-
due to the significant reduction in computational cost. However the ulations of metal corrosion. Rates of corrosion are determined by ex-
strict symmetry requirements in terms of surface reactivity and so- plicitly simulating all possible electrochemical reactions occurring
lution composition make these unsuitable for highly heterogeneous between the metal surface and the surrounding environment, follow-
systems, such as metals with varying crystallographic orientations, ing Butler-Volmer kinetics. By taking rate constants from metal oxide
composition, impurities or surface inclusions, which have significant thermal decomposition data, we are able to produce an accurate sim-
effects on corrosion behaviors.22 Additionally, finding an analytical ulated PDP. Additionally, by simulating each reaction independently,
solution requires making a number of assumptions to simplify the we are able to model individual corrosion reactions at the metal sur-
analyzed system, including the use of one dimensional geometries, face, rather than working in terms of an all-encompassing corrosion
uniform surface reactivity, and negating the impact of solution species potential or corrosion rate. This enables the model to determine which
on the rate of reaction.18 This severely limits the scope of applicable species will be formed during metal corrosion based on the thermody-
systems. namic data and reaction environment, rather than having the species
Numerical models can function either as model-first or experiment- formed being an input parameter into the model.
first, depending on whether the model starts with theoretical parame-
ters to predict experimental behavior, or uses simulation to rationalize
Experimental
experimental observations. In virtually all cases across currently avail-
able models, input parameters take the form of physical parameters, Standardized polishing procedure.—Iron metal samples
such as species concentration and diffusion coefficient, and experi- (99.98%) were received from Goodfellow. All specimens were sec-
mental parameters, such as Tafel slope (β) corrosion potential (Ecorr ) tioned into 2 cm × 2 cm samples using an abrasive cutter (abrasiMet
and corrosion current (jcorr ). Values for β, Ecorr and jcorr invariably 250, Buehler, USA). Samples were fixed in cold mounting epoxy
come from experimental sources, whether from independent experi- (Epofix from Struers, Canada). The substrate was abraded using a
mentation or sourced from the literature, and so there is currently an series of SiC papers (800, 1200 and 4000 grit, Struers) followed by pol-
inherent requirement for electrochemical experimentation to be per- ishing with 0.05 μm aluminum oxide suspension and a MD Chem cloth
formed on a material before its corrosion can be simulated. This can (Struers, Canada) to obtain a mirrored finish. All polishing was done
be challenging, as determining accurate values for β, Ecorr and jcorr using a TegraPol-25 polishing wheel and an automated TegraForce-5
is far from trivial; the extraction of these from potentiodynamic po- polishing arm (Struers, USA). All samples were sonicated in anhy-
larization curves (PDPs) requires the extrapolation of linear regions drous ethanol for 2 min to remove any alumina residue, and then dried
of the anodic and cathodic regions of the PDP, and real experimental under air.
examples often deviate from ideal linear behavior.
To combat this, a number of works have proposed rigid potential Polarization curves.—To minimize solution resistance and pro-
ranges for the linear fit,23,24 although this is not a complete solution vide a significantly corrosive environment, 3.5 wt% sodium chloride
as many metals exhibit linear regions in very different regions of their electrolyte was chosen as the main test solution for electrochemi-
PDP due to their distinct corrosion behavior,5 which has led to differ- cal measurements and corresponding simulations. Electrolytes were
ent papers using a number of different fitting techniques to analyze made using sodium chloride (99%, Sigma-Aldrich). In order to pro-
the same type of data.24–31 Such variance in the experimental analysis duce solutions with specific pH values, the electrolyte was titrated
with either hydrochloric acid (38%, ACP) or sodium hydroxide (98%,
∗ Electrochemical Society Student Member. ACP) in the presence of universal indicator (Fluka). All solutions were
∗∗ Electrochemical Society Member. made using deionized water from a Milli-Q water purification system
z
E-mail: [email protected] (18.2 M cm resistivity). A Faraday cage and a vibration isolation

Downloaded on 2019-05-02 to IP 139.81.202.21 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, 166 (11) C3186-C3192 (2019) C3187

Figure 1. A: Schematic of the corrosion cell used for recorded all polarization
curves. The metal sample (grey disc) is sealed in an epoxy pick and loaded into
the corrosion cell (red arrow), where a 1 cm2 area is exposed via an O-ring.
The RE and CE are loaded into set positions, and a luggin capillary is used to
bring the RE close to the WE to minimize iR drop. B: Simulated cell geometry,
showing WE (red) CE (green) and rotational symmetry (blue). These areas are Figure 2. Replicated Pourbaix diagram for the iron-water system, αFe2+ |t=0
also overlaid in Fig 1A to show the corresponding regions in the experimental = 10−6 M. Bold solid lines show boundaries between two solid species, thin
cell. WE radius = 0.564 cm to give 1 cm2 area with rotational symmetry. CE solid lines show boundaries between one solid and one aqueous species and
is given as a series of rings to approximate the large surface area of a platinum dotted lines show boundaries between two aqueous species. The two dashed
mesh. Cell wall sizes and the angle of the lower boundary were all selected to lines show the boundaries for the oxidation and reduction of water. Figure
match the experimental cell geometry. Mesh density was chosen by increasing constructed using data from Reference 33.
mesh density until no change in simulated PDPs was seen.
using an existing boundary, the counter electrode was modelled as a
series of concentric rings within the geometry to ensure that its area
table were used during all PDP measurements. All polarization tests was significantly larger than the working area, as is the experimental
were repeated at least three times to ensure reproducibility. case when using a platinum mesh. This also allowed the counter to be
All PDP measurements were performed using a multi-channel positioned at the same distance from the working electrode in both the
VSP-300 potentiostat (BioLogic Science Instruments, USA). Experi- simulation and the model. The real 3D geometry has been simplified
mental parameters were based on the ASTM standard test method for to its 2D axisymmetric equivalent (Figure 1B), taking advantage of
conducting polarization measurements on Fe-based alloys in aqueous the plane of rotational symmetry seen through the center of the metal
environments.32 The cell was assembled in a three-electrode setup, sample.
with the metal sample working electrode, saturated calomel electrode
(SCE) reference electrode (ATS Scientific) placed in a luggin capil-
Construction of the model.—Most existing corrosion models are
lary, and platinum mesh counter electrode. A bench top corrosion cell
built upon experimental corrosion parameters such as Ecorr , jcorr and β,
(K0235 Flat Cell, Princeton Applied Research, AMETEK Scientific
which are determined using PDP analysis of the real samples. Record-
Instruments) was used as the main corrosion cell for all experiments
ing polarization curves can be challenging for metals that readily
(Figure 1A). The metal sample was first immersed in the working so-
corrode, as corrosion begins on immersion in solution or even im-
lution for one hour in order to determine Ecorr , before being biased
mediately after polishing, which provides inherent uncertainty in the
∼300 mV more negative than the Ecorr and then ramped positively at
extracted corrosion parameters. Here, we propose a simulation built
0.167 mV/s until a current density limit of ∼0.5 mA/cm2 was reached.
upon easily accessible thermodynamic parameters, as extracted from
The exposed surface area of the test material was kept constant by us-
Pourbaix diagrams. This has the advantage that simulation analysis
ing a 1 cm2 Teflon ring to isolate a standard geometric area.
can be conducted in the absence of preceding experimental analysis,
allowing the simulation to act as a means of predictive analysis. The
Pourbaix diagram for the iron water system (Figure 2) shows the most
Results
thermodynamically stable species occurring at various potentials and
Building the model.—Geometry.—All considerations for the sim- pH, when the system is under equilibrium.
ulation geometry were made in order to provide a simulated environ- The Pourbaix diagram shows Fe, Fe3 O4 , Fe2 O3 , Fe2+ and Fe3+ as
ment that resembles a real corrosion cell as closely as possible. Real the predominant species during any aqueous PDP experiment. How-
corroding metal surfaces are highly heterogeneous, with oxidation ever, FeO, whilst not the most stable, may be an important rate deter-
processes being highly dependent on crystal orientation, scratches or mining intermediate, and so will also be simulated. It is worth noting
defects, and other surface features, particularly when more extreme that, at higher pH values, additional anionic corrosion products may
corrosion conditions lead to the onset of pitting or crevice corrosion, be present. Since this work focuses on neutral to acidic environments
which cause significant changes to the sample topography. In all lab- it is not necessary to model these species currently, although this could
oratory corrosion experiments, great care is taken to polish the metal easily be incorporated into future models. The simulation will inde-
surface until smooth, in order to provide as even a metal surface as pendently look at the reactions possible between all of these species
possible prior to PDP recording. The corrosion cell used in this work and, by running them concurrently, use the relative rates of reactions
is displayed in Figure 1A. A 1 cm2 area of the working electrode is to predict which species will be present during a PDP current, and use
exposed to solution via an O-ring. All simulations were performed us- the rate of this oxidation to extract a simulated anodic current. This
ing the COMSOL 5.3 Multiphysics finite element simulation package. method will result in some simulated reaction rates being negligibly
The simulation geometry was built according to this exposed surface small, where they could be excluded altogether, such as in strongly
area, taking into account the 1 cm2 active area, the curved geometry acidic electrolytes where the simulations of solid electrolytes could be
of the cell and the volume of solution within the cell. Rather than omitted without significantly impacting the output. This has been done

Downloaded on 2019-05-02 to IP 139.81.202.21 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).
 C3188 Journal of The Electrochemical Society, 166 (11) C3186-C3192 (2019)

intentionally so that the same model can be used regardless of the reac- The second exponential term in Equations 7 and 8 takes into account
tion environment, and so the model can determine which species shall the fact that, as an oxide layer starts to form, the activation energy for
be formed, rather than this being a user input. The cathodic current is further oxidation will progressively increase. Here, γox is an activation
assumed to come from the oxygen reduction reaction (ORR).5 Fluxes energy gradient4 and Lox (t) is the oxide thickness at time ‘t’, so as
of all species were calculated assuming mass transport by both diffu- the existing oxide layer thickens the second exponential term will
sion and convection mechanisms, assuming that the initial convective decrease, reducing kox and slowing the rate of further oxidation. The
flux is zero. overpotential term (η) is defined by
Ni = −Di ∇ci + uci [1] η = E − E 0 − Vox (t ) [10]
where N i is the molar flux of species ‘i’, −Di ∇ci is the diffusive term where E is the applied potential, E0 is the standard redox potential for
and u ci is the convection term; Di and ci are the diffusion coefficient the reaction in question, and Vox (t) is a potential drop term. Vox (t)
and concentration of species ‘i’ respectively, and u is the fluid velocity. is necessary, as the oxide layer formed at the metal/solution interface is
Migration is not considered due to the presence of excess electrolyte in a more resistive material, and so a potential gradient will exist between
most practical experimental set ups. The net transport at the symmetry the potential applied at the metal WE and the potential at the oxide
boundaries and the cell walls was set at zero solution interface. Vox (t) is therefore also dependent on Lox (t) as
defined by
n · Ni = 0 [2]
Vox (t ) = εox Lox (t ) [11]
where n is the normal unit vector, which prevents any solution species
interacting with or passing through the specified boundary. To aid in where ɛox is a measure of potential drop per unit distance. This means
simulation convergence, the concentrations of species in solution at that as the oxide layer thickens the overpotential will be decreased,
boundaries sufficiently far from the WE were fixed at initial values hindering the formation of further oxide. It is worth noting that not
all oxides are uniformly resistive at the electrode surface. Electro-
ci = c0,i [3] chemically formed oxides may form porous structures, which are less
where c0,i is the concentration of species ‘i’ at time t = 0. The working passivating than homogeneous films of the same material. Values of
electrode boundary was defined as the only boundary with a non-zero ɛox used in this model take this into account, where a larger ɛox reflects
flux both a more resistive material and a less porous structure.
In this way, by monitoring Lox (t) over the course of the simulation,
n · Ni = N0,i [4] the model is able to constantly reduce and refine the rate of oxide
where N0,i is the flux expression as defined from the difference between formation not just based on the applied experimental conditions, but
the rate of oxidation and reduction at the electrode surface. For all solid also on the state of the electrode and its surrounding environment based
– solid oxide reactions, reaction rates are determined by Butler-Volmer on the previous degree of corrosion. Values for E° can be simply taken
kinetics, from Pourbaix diagrams by using the equation of the lines between the
  appropriate neighboring species. In the case of species which are not
(1 − α) nF η shown on the Pourbaix diagram, equations for boundary lines can be
jox = kox · exp [Fe] [5]
RT calculated in the same way, so that standard potentials can be calculated
for all possible reactions. In depth discussions on the procedures for
  deriving Pourbaix diagrams can be found in the literature.33,38 Briefly,
−αnF η
jred = kred · exp [FeOx] [6] for the reaction aA + wH2 O  bB + hH+ + ne− , equations of the
RT lines come from the Nernst equation,
where jox and jred are the rates of oxidation and reduction, kox and  
kred are the oxidation and reduction rate constants, α is the transfer G0 (αB )b (αH+ )h
E =
0
+ log  w [12]
coefficient, n is the number of electrons transferred, F is Faraday’s nF (αA )a αH2 O
constant, η is the applied overpotential, R is the ideal gas constant and
T is the temperature. The concentration of Fe is taken as the number where G0 is the standard free energy change for the reaction, and αi
of Fe atoms found in a square centimeter of a Fe, which was taken as is the activity of species ‘i’. The model uses concentration as activity
the molar volume of Fe, divided by the unit cell length. Values for kox with the assumption that the activity coefficient is unity. The simulated
and kred come from the Arrhenius equation, where αFen+ was from the sum of the concentration of dissolved iron from
    previous steps in the corrosion model and the starting concentration
−Ea,ox −γox Lox (t ) of dissolved iron (αFen+ |t=0 ), which was assumed to be 10−6 M.
kox = A · exp · exp [7] 
RT RT αFen+ = Fen+ + αFen+ |t=0 [13]
    Solving this for a specific reaction, such as for 3Fe2+ + 4H2 O 
−Ea,red −γox Lox (t )
kred = A · exp · exp [8] Fe3 O4 + 8H+ + 2e− , gives the Pourbaix equation of the line.
RT RT  
E 0 = 0.98 − 0.024pH − 0.089 log αFe2+ V vs. SHE [14]
This expression is made up of two exponential terms. The first, where
A is the pre-exponential factor and Ea is the activation energy, gives Equation 12 shows how the standard potential for this oxidation
the initial value of kox according to the activation energy for the ox- reaction is dependent on the applied potential, the local pH, and the
idation reaction, assuming no product is present. Activation energies activity of dissolved iron in solution. By using equations such as this
for reduction reactions were approximated from isothermal kinetic as the main input for the simulation, the model will be able to adjust
studies of iron oxide reduction under hydrogen atmospheres.34–37 Al- the rate of corrosion based on the amount of iron already dissolved and
though real activation energies in the aqueous phase are expected to based on the localized pH change that results from continued metal ox-
differ from those determined experimentally from those recorded in idation. Contributions to pH change from hydrolysis are not included
the gas phase, these values provided a starting point for further refine- to minimize computational costs. In the case of species existing in
ment, where the relative magnitudes of the rate of reduction reactions solution away from the WE, rates of reaction are determined from the
between various oxides is known. For oxidation reactions, activation relative concentration gradients of species involved
energies were approximated by the difference between the reduction ∂ci ∂ci
activation energy and the enthalpy of formation for the species in- Ri = + ∇ · (−Di ∇ci ) + u · ∇ci = + ∇Ni [15]
∂t ∂t
volved, as given by Equations 9.
where Ri is the reaction rate for species ‘i’, -Di ∇ci is the diffu-
Ea,ox = Ea,red + H f ,ox − H f ,red [9] sion term and u·∇ci is the convection term as previously defined

Downloaded on 2019-05-02 to IP 139.81.202.21 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, 166 (11) C3186-C3192 (2019) C3189

in Equation 1. Reaction rates are determined from the equilibrium

constants (Keq ), which are also sourced from the Pourbaix diagram.
So, for Fe2+ + H2 O  FeO + 2H+ ,

  
2
RFe2+ = −RFeO = k 0 Keq H+ [FeO] − Fe2+ [16]

where Ri is the reaction rate for species ‘i’ and k0 is a fast rate constant
that keeps the system at equilibrium. The cathodic current is assumed
be solely from the reduction of oxygen at the metal surface.5 For
simplicity, the reduction is assumed to proceed via the direct four
electron reduction pathway.
O2 + H2 O + 4e−  4OH− [17]
The current density is then calculated from the flux of species
towards the working electrode surface
  
isim = ni F jox,i − jred,i [18]
i

where isim is the simulated current density, and jox,i and jred,i are the
oxidative and reductive fluxes respectively for all species being simu-
lated. A full list of all reactions studied along with rate constants used
is given in the supplementary information (Table SI-1).

Modelling the counter electrode.—The function of the counter

electrode (CE) is to pass the oppositely signed current with respect
to the working electrode (WE) in order to maintain charge balance
within the cell. In our corrosion cell, we use a platinum gauze CE. In
an aerated cell, the predominant reaction at the counter electrode will
be the reduction of dissolved oxygen, as defined by Equation 16, and
also the reduction of water as part of the hydrogen evolution reaction
(HER) according to Equation 19.
2H2 O + 2e−  H2 + 2OH− [19]
Through a combination of these reactions, net reduction or oxidation
at the WE will result in water oxidation or oxygen reduction at the CE
respectively. Calculating which of these reactions is dominant at any
given time would require the calculation of an effective overpotential
for the counter electrode that would give the correct relative fluxes of Figure 3. A: Simulated (red) and experimental (black) polarization curves for
all possible reactions. As the concentration of oxygen became severely iron in 3.5% NaCl, scan rate = 0.167 mV/s, following ASTM standards.32 B:
depleted, it may then be necessary to include the calculation of the ex- Weighting function used in percentage difference analysis (red) and percent-
change of oxygen gas at the liquid – air interface, which adds further age differences between simulated and experimental polarization curves at all
computational costs. Rather than applying such a computationally ex- potentials (black).
pensive calculation, the model focused on the pH change at the CE, as
this is likely to have the most significant impact on the overall behav-
ior of the corrosion cell. A probe was therefore added to the model to the average percentage difference across the whole curve being 3.3%.
calculate the total current being passed at any given point during the Figure 3B shows the calculated percentage difference across the whole
simulation. This live probe was then used to calculate the total flux of graph, as well as the weighting function used in the percentage dif-
OH− towards or away from the CE, based on whether the CE should ference analysis. The merits of the weighting waveform have been
provide a reductive or oxidative environment, respectively. Since in discussed in previous works.5 Briefly, the largest changes in current
all proposed CE reactions the stoichiometry of OH− and e− are equal, are found around Ecorr , where a small shift in Ecorr corresponds to a
the flux of OH− can be given by massive deviation in the current, which would result in a disproportion-
isim ately large percentage error. The weighting function therefore reduces
jOH, CE = [20] the impact of the larger percentage differences about Ecorr , without
F ACE impacting the percentage difference calculated at points further than
where jOH,CE is the flux OH towards the CE, and ACE is the electroactive ±10 mV from Ecorr . Extracted values of Ecorr from the experimental
surface area of the CE. and simulated PDPs also agree within 6 mV. This is remarkably close
given that no experimental electrochemical information went into the
Comparison with experimental.—Experimental vs. simulated model, yet we can extract high precision electrochemical information
PDPs.—Explicitly simulating all aspects of the reaction environment from the simulated currents.
gives this model the distinct advantage that all components of the The anodic current density of around 10 mA cm−2 is characteristic
corrosion experiment can be individually observed from the simula- of a diffusion limited response. This may seem counter intuitive as
tion data. This allows for the traditional PDP to be extracted from this model begins with an initial oxidation of iron to solid iron oxide
the model, along with concentration profiles for individual species at products. However, since the formation of aqueous iron is thermo-
any given point during the potential sweep. First, the precision of the dynamically favored, the model will simulate a rapid dissolution of
model with regards to the real corrosion cell was inspected by com- the formed solid oxide to give aqueous iron, resulting in a diffusion
paring an experimental PDP to a simulated PDP recorded under the layer containing a relatively high concentration of Fe2+ . High con-
same conditions. centrations of Fe2+ at the electrode surface will push the dissolution
Comparison of theoretical and experimental polarization curves equilibrium towards the solid products. At this point, the simulated
in Figure 3 shows good agreement in all regions of the PDP, with rate of corrosion will become limited by the diffusion of Fe2+ away

Downloaded on 2019-05-02 to IP 139.81.202.21 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).
 C3190 Journal of The Electrochemical Society, 166 (11) C3186-C3192 (2019)

from the electrode to allow for further dissolution, and therefore fur- to give a plot of pH vs. x distance for a series of time points throughout
ther corrosion. In this way, although the model initially forms a solid the video.
oxide layer, it is simulating a diffusion limited corrosion process. Figure 4A shows the advancement of high pH electrolyte from the
counter electrode (x = 6 cm) towards the working electrode (x = 0)
Localized pH change.—During PDPs, anodic corrosion reactions as the PDP proceeds. Of course, there is a degree of uncertainty to the
at the working electrode will be balanced by the reduction of oxygen or extraction of absolute pH from RGB, such as the detection of optical
water at the counter electrode. This results in the solution environment artifacts in the image interfering with the RGB value, and the limited
becoming more acidic at the working electrode due to hydrolysis, and size of the RGB to pH database offering resolution to approximately
more basic around the counter. The degree of this pH change is easy to the nearest 0.1 pH unit. The clear experimental change in pH around
track with the simulation through the analysis of the concentration of the CE revealed an interesting feature of the simulation regarding the
protons within the solution domain. However, tracking the localized movement of species between the CE and WE. Simulations did also
change in pH during the experimental PDP proved more of a challenge. show that the reactions at the CE had a greater impact on the pH change
Traditional options such as the use of a glass pH probe do not offer compared to the WE. However, the extent to which reactions at the CE
the spatial resolution needed to track the pH gradient that is expected increased the pH in bulk solution was far greater in the experimental
within the cell when moving from the working to counter electrodes. example than was seen in the simulation (Figure 4B).
Instead, the corrosion cell was assembled as previously described with Inspection of the final section of the video (supplementary informa-
a solution containing universal indicator along with 3.5 wt % NaCl. tion, highlighted frames in Figure SI-2C and 2D) shows the presence
This allowed for a continuous inspection of the pH through color of bubbles forming at the CE when the WE is being significantly ox-
changes, ranging from red at pH < 4 to blue at pH > 10. The starting idized. These bubbles can be seen to move away from the CE after
pH of the solution was measured at 6 with a glass pH probe to act as they become significantly large. This would generate a convective flux
a reference point for subsequent color changes. away from the CE, and is likely a significant contributor to the asym-
A digital camera was fixed opposite the corrosion cell, and oriented metrical solution front that is seen in the movement of basic species
so that it could record the contents of the entire cell, covering both the away from the CE (Figure SI-2C). While it would be possible to add a
working and counter electrode environments. A video of the changing flow term from the counter electrode to the simulation to take this into
cell environment was then recorded to show the solution becoming account, the large degree of turbulence from this convective flow mean
more acidic or basic as the solution changed to red or blue respectively that a truly accurate simulation will be difficult and computationally
(available in supplementary information, accelerated 100x). In order expensive.
to easily analyze the observed color changes, the video was processed With or without the convective terms, the increase in pH due to
in order to extract approximate values for the pH at any given point the ORR or HER at the CE can be seen to substantially outweigh
in the solution volume at any given time during the PDP. To do this, the decrease in pH due to the corrosion of the iron WE. This can be
the video was first divided into 100 image frames to be individually attributed to the incomplete hydrolysis of iron cations, which results
analyzed. A script was then written in Python 2.7 to calculate mean red, in the total proton generation at the WE being less than the number
green and blue (RBG) intensities for any chosen pixel in the solution of electrons transferred. At the CE the proton consumption is equal
volume (SI Section 5). This was done for 100 rows of pixels from the to the number of electrons transferred (Equation 19) and so the net
area within the corrosion cell volume, which were then averaged to pH within a closed corrosion system will be expected to increase.39
give the final color change vs the corrosion cell length, starting at the The solution around the corroding iron during a PDP becomes more
left of the cell (x = 0, working electrode) moving towards the right of complex after extended polarization, as eventually the basic solution
the cell x = 6 cm, counter electrode) (Figure SI-1). A database of RBG front will reach the acidic environment around the WE. This highlights
to pH conversion was then assembled by recording the RGB values in the importance of precisely modelling the corrosion cell volume, as
from solutions of 3.5 wt % NaCl and universal indicator for a range this will impact if and when the basic solution from the CE will reach
of known pH, which were captured using the same digital camera in the WE surface, which has the potential to impact the recorded PDP
the same laboratory environment (Figure SI-3). After correcting all and the observed corrosion behavior over the course of the experiment.
image frames to match the brightness level of the images used for the
pH database, these recorded RGB values for each frame were cross Oxide thickness.—Since the simulated oxide thickness, Lox (t), is
referenced against the RGB to pH data base for all 100 frames, in order time dependent, the rate of oxide growth over the course of the PDP can

Figure 4. The pH change within the corrosion cell during a PDP. The x axis represents the length of the corrosion cell, where x = 0 corresponds to the WE and
x = 6 corresponds to the CE. A shows the experimental pH change as determined from a pixel by pixel analysis of images recorded of the color change during a
PDP for a corrosion cell containing universal indicator. B shows the simulated pH change during a PDP due to the consumption or formation of protons during
corrosion. The simulated value was capped at pH 10 since this is the upper range of the indicator. The times elapsed correspond to 0, 63, 69, 72, 75, 78, 81, 84, 87
and 98 minutes after the start of the PDP.

Downloaded on 2019-05-02 to IP 139.81.202.21 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, 166 (11) C3186-C3192 (2019) C3191

Figure 5. A) Simulated corrosion distance (Lox (t)) at the iron surface during a PDP. The total Lox (t)) is shown in the black curve, and the red and blue curve show
contributions from iron dissolution and solid oxide formation respectively. B) SEM cross sectional image of an iron electrode after corrosion during a PDP. The
grey portion is Fe metal, the black portion is empty space created during the PDP measurement and has been filled with epoxy resin.

be easily tracked. In order to compare this to experimental examples, an baix diagrams are built, rather than being built from the diagrams
iron electrode was taken immediately after a PDP was recorded, dried themselves. As such, a number of the factors that limit the application
under nitrogen flow, and then sealed in an epoxy puck. The sample of Pourbaix diagrams to experiment systems are not passed on to this
was then cut in half and polished side-on to reveal a cross section of model, or can be taken into account through the modification of the
the iron oxide, with the non-corroded iron beneath. Unfortunately it model.
was not possible to record SEM images of the oxide itself, as the oxide One limitation that is particularly applicable to this model is that
was too thin to observe or had poor adhesion and so did not remain Pourbaix diagrams assume the system is at steady state, which is never
attached to the sample after polarization. Instead, the total iron loss the case for real experimental systems. By using a previously defined
due to corrosion can be easily observed. pre-treatment and slow sweep rate (following ASTM standards) we
Since the profile of the corroded sections after polarization is highly are able to record a PDP whilst staying reasonably close to steady
non-uniform, an average thickness of the corroded portion was deter- state, although the dynamic nature of a PDP means that a true steady
mined by calculating the area of iron lost due to corrosion, and then state is not really reached. The use of the ASTM standard is essential
dividing by the sample length. The area of the iron loss was calcu- in allowing deviations from steady state in the experimental PDP to
lated using an image analysis protocol (IsoCalc ‘Get Area’ Macro,40 be kept to a minimum in order to make the Pourbaix diagram more
CorelDraw2017). The average thickness of the corroded section was applicable.
found to be 73.5 μm. Since this is a measure of iron lost to corrosion The Pourbaix diagrams also assume a homogeneously corroding
by all reactions, the experimental distance must be compared to the surface, and so do not take into account other corroding phenomena,
entirety of the simulated iron loss. The total length of simulated iron such as pitting or crevice corrosion. This problem is not specific to
loss (Lcorr (t)) was taken from the model by combining the simulated Pourbaix diagrams, but affects many finite element models, particular
oxide thickness with the volume of iron lost to aqueous Fe2+ when a 2D simulated geometry is rotated to model a 3D reaction envi-
ronment. Just as the Pourbaix diagram represents an average reaction
Vm · ∫∫∫ Fe2+ dx dy dz environment at equilibrium, the 2D FEM represents an average corrod-
Lcorr (t ) = LFeO (t ) + LFe3 O4 (t ) + LFe2 O3 (t ) +
Aede ing environment, which gives an average behavior of a substrate that
[21] in reality will likely be partly active and partly passivated. However,
where Vm is the molar volume of iron, Aede is the area of the electrode, whilst the Pourbaix diagram itself assumes a homogeneously active
and ∫∫∫ Fe2+ gives the total number of moles of Fe in the simulated surface, the equations used to build the model could be used to de-
domain. The simulated loss of iron during a PDP and the experimen- scribe a heterogeneously reacting surface though FEM. For example,
tal cross section of the Fe electrode after polarization are shown in if a complex 3D geometry were built that included a bulk domain of
Figure 5. iron with a thick oxide layer, where there was a scratch that revealed
After the simulation ran under the same conditions as the experi- the bare iron surface, this model would predict a slow rate of corro-
ment, the simulated oxide loss was 14.9 μm, with most of the mass sion at the bulk oxide, whilst simultaneously predicting a rapid rate of
loss due formation of aqueous iron cations, rather than a solid oxide reaction at the depassivated surface.
phase. The difference between the simulated and experimental values A further limitation in the Pourbaix diagram comes with the treat-
can be attributed to the fact that the simulation gives a value that treats ment of aqueous and solid corrosion products. As well as only show-
the entire simulated surface as equal, whereas the experimental cross ing the most thermodynamically stable product, Pourbaix diagrams
section represents only a small section of a heterogeneous surface. assume that all solid corrosion products are passivating. The top right
Once dissolution begins, this would reveal step, edge and similar high portion of Figure 2 is often described as the ‘passivated region’, and
energy defect sites on the iron surface, which exhibit different corro- so it might be assumed that the additional corrosion is slowed by many
sion behavior to the bulk substrate. Additionally, continual corrosion orders of magnitude in this region. Real experimental systems may not
will gradually increase the surface area of the iron electrode, resulting exhibit uniform passivity due to porosity in the passive film allowing
in a greater overall corrosion rate. Together these may explain why a certain amount of access to the metal surface. This is before men-
the simulation underestimates the total depth of the iron corrosion. tioning the added complexity in trying to predict the point at which
the passive film may break down, giving a greatly enhanced rate of
Limitations of the Pourbaix diagram.—As with most approxi- corrosion at the exposed surface. This model aims to encompass the
mations of complex systems there are a number of limitations to using effect of porosity in the film within the ɛox parameter, where a porous
Pourbaix diagrams to describe corroding metal systems, which have oxide would be reflected in a smaller value of ɛox . Prediction of pas-
been well discussed in the literature.41–43 However, it is important to sive film breakdown is a field in corrosion symmetry on its own and
note that this model is built on the same equations upon which Pour- remains challenging.44,45

Downloaded on 2019-05-02 to IP 139.81.202.21 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).
 C3192 Journal of The Electrochemical Society, 166 (11) C3186-C3192 (2019)

Finally, most Pourbaix diagrams only consider the reactions of a 10. P. Marcus, V. Maurice, and H. H. Strehblow, Corros. Sci., 50, 2698 (2008).
single metal, which is not applicable to most industrial applications 11. G. T. Gaudet, W. T. Mo, T. A. Hatton, J. W. Tester, J. Tilly, H. S. Isaacs, and
R. C. Newman, AlChE J., 32, 949 (1986).
that require more complex metal alloys. This is a trend that is starting 12. N. J. Laycock and R. C. Newman, Corros. Sci., 39, 1771 (1997).
to change, with recent works developing Pourbaix diagrams for CrC,46 13. A. Keating and S. Nešić, Corrosion, 57, 621 (2001).
FeCrNi47 and NiTi48 alloys to name a few. Calculations involved can be 14. F. M. Song, D. A. Jones, and D. W. Kirk, Corrosion, 61, 145 (2005).
computationally expensive, as would be the resultant simulation given 15. J. C. Walton, Corros. Sci., 30, 915 (1990).
16. J. C. Walton, G. Cragnolino, and S. K. Kalandros, Corros. Sci., 38, 1 (1996).
the large number of possible reaction occurring at a corroding alloy 17. A. Stenta, S. Basco, A. Smith, C. B. Clemons, D. Golovaty, K. L. Kreider, J. Wilder,
surface. Further work would be needed to determine the feasibility of G. W. Young, and R. S. Lillard, Corros. Sci., 88, 36 (2014).
this type of model towards metal alloys. 18. G.-L. Song, Corros. Sci., 52, 455 (2010).
19. A. Atrens, Z. Shi, and G. L. Song, in Corrosion of Magnesium Alloys, p. 455, Wood-
head Publishing (2011).
Conclusions 20. R. S. Munn and O. F. Devereux, Corrosion, 47, 618 (1991).
21. J. X. Jia, G. Song, and A. Atrens, Corros. Sci., 48, 2133 (2006).
A finite element model has been constructed for the simulation 22. S. M. Gateman, L. I. Stephens, S. C. Perry, R. Lacasse, R. Schulz, and J. Mauzeroll,
of iron corrosion during a PDP experiment. The basis of this model npj Mat. Deg., 2, 1 (2018).
23. W. S. Tait, An Introduction to Electrochemical Corrosion Testing for Practicing En-
comes from the Pourbaix diagram, where all possible corrosion reac- gineers and Scientists, PairODocs Publications (1994).
tions between the iron starting material and between various corrosion 24. E. Poorqasemi, O. Abootalebi, M. Peikari, and F. Haqdar, Corros. Sci., 51, 1043
products are considered and explicitly simulated. The model gives a (2009).
good agreement with the simulated current, and also predicts the lo- 25. G. Song, A. Atrens, and M. Dargusch, Corros. Sci., 41, 249 (1998).
26. M.-C. Zhao, P. Schmutz, S. Brunner, M. Liu, and G.-l Song, and Atrens, A. Corros.
calized pH around the counter electrode, and mass loss from the iron Sci., 51, 1277 (2009).
surface due to corrosion, where aqueous iron appears to be the domi- 27. Z. Shi, M. Liu, and A. Atrens, Corros. Sci., 52, 579 (2010).
nant corrosion product. During a simulation, the state of the electrode 28. A. J. Bard, M. Stratmann, and G. S. Frankel, Encyclopedia of Electrochemistry, Cor-
surface is considered, where increased oxide thickness at the electrode rosion and Oxide Films, Wiley (2003).
29. L. Y. Xu and Y. F. Cheng, Corros. Sci., 73, 150 (2013).
surface results in a decreased rate of further oxidation at that specific 30. J. Ge and O. B. Isgor, Mater. Corros., 58, 573 (2007).
point. This opens the possibility of expanding this model for more het- 31. X. L. Zhang, Z. H. Jiang, Z. P. Yao, Y. Song, and Z. D. Wu, Corros. Sci., 51, 581
erogeneous surfaces, such as a scratch in a passivated metal surface, (2009).
where the model would give a greater rate of corrosion at the scratch 32. ASTM Standard G61-86, 2014, “Standard Test Method for Conducting Cyclic Po-
compared to the bulk passivated surface. tentiodynamic Polarization Measurements for Localized Corrosion Susceptibility of
Iron-, Nickel-, or Cobalt-Based Alloys,” ASTM International, West Conshohocken,
PA
ORCID 33. E. D. Verink, in Uhlig’s Corrosion Handbook, p. 93, John Wiley & Sons, Inc. (2011).
34. B. Hou, H. Zhang, H. Li, and Q. Zhu, Chin. J. Chem. Eng., 20, 10 (2012).
35. H.-Y. Lin, Y.-W. Chen, and C. Li, Thermochim. Acta, 400, 61 (2003).
S. C. Perry https://orcid.org/0000-0002-6263-6114 36. I. I. J. Moon, C.-H. Rhee, and D.-J. Min, Steel Res., 69, 302 (1998).
L. I. Stephens https://orcid.org/0000-0002-0064-7555 37. K. Piotrowski, K. Mondal, H. Lorethova, L. Stonawski, T. Szymański, and
J. Mauzeroll https://orcid.org/0000-0003-4752-7507 T. Wiltowski, Int. J. Hydrogen Energy, 30, 1543 (2005).
38. L. L. Pesterfield, J. B. Maddox, M. S. Crocker, and G. K. Schweitzer, J. Chem. Educ.,
89, 891 (2012).
References 39. R. G. Kelly, J. R. Scully, D. Shoesmith, and R. G. Buchheit, Electrochemical Tech-
niques in Corrosion Science and Engineering, Taylor & Francis (2002).
1. R. E. Melchers, Corros. Sci., 45, 923 (2003). 40. N. Wilkinson, Get Area Version C, IsoCalc (2004).
2. W. Sun and S. Nešić, Corrosion, 65, 291 (2009). 41. Z. Ahmad, in Principles of Corrosion Engineering and Corrosion Control, Z. Ahmad,
3. A. Kahyarian, M. Singer, and S. Nesic, J. Nat. Gas Sci. Eng., 29, 530 (2016). Editor, p. 9, Butterworth-Heinemann, Oxford (2006).
4. M. Momeni and J. C. Wren, Faraday Discuss., 180, 113 (2015). 42. L. L. Shreir, Corrosion: Metal/Environment Reactions, Elsevier Science (2013).
5. L. I. Stephens, S. C. Perry, S. M. Gateman, R. Lacasse, R. Schulz, and J. Mauzeroll, 43. E. McCafferty, Introduction to Corrosion Science, Springer New York (2010).
J. Electrochem. Soc., 164, E3576 (2017). 44. G. S. Frankel and N. Sridhar, Mater. Today, 11, 38 (2008).
6. S. M. Sharland, Corros. Sci., 27, 289 (1987). 45. G. S. Frankel, T. Li, and J. R. Scully, J. Electrochem. Soc., 164, C180 (2017).
7. D. R. Gunasegaram, M. S. Venkatraman, and I. S. Cole, Int. Mater. Rev., 59, 84 (2014). 46. V. Marimuthu, I. Dulac, and K. Kannoorpatti, J. Bio. Tribo. Corros., 2, 17 (2016).
8. N. Sridhar, Corrosion, 73, 18 (2016). 47. B. Beverskog and I. Puigdomenech, Corrosion, 55, 1077 (1999).
9. J. Srinivasan, C. Liu, and R. G. Kelly, J. Electrochem. Soc., 163, C694 (2016). 48. R. Ding, J.-X. Shang, F.-H. Wang, and Y. Chen, Comput. Mater. Sci., 143, 431 (2018).

Downloaded on 2019-05-02 to IP 139.81.202.21 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).