Measurement 94 (2016) 344354

Contents lists available at ScienceDirect

Measurement
journal homepage: www.elsevier.com/locate/measurement

Novel electrochemical impedance simulation design via stochastic

algorithms for fitting equivalent circuits
Marco A.A. Kappel a,, Ricardo Fabbri b, Roberto P. Domingos b, Ivan N. Bastos b
a
Centro Federal de Educao Tecnolgica Celso Suckow da Fonseca CEFET/RJ Campus Nova Friburgo, 28635-000 Nova Friburgo, RJ, Brazil
b
Rio de Janeiro State University, Polytechnic Institute, 28625-570 Nova Friburgo, RJ, Brazil

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

Article history: Electrochemical impedance spectroscopy (EIS) is of great value to corrosion studies because it is sensitive
Received 25 October 2014 to transient changes that occur in the metal-electrolyte interface. A useful way to link the results of elec-
Received in revised form 2 August 2016 trochemical impedance spectroscopy to corrosion phenomena is by simulating equivalent circuits.
Accepted 11 August 2016
Equivalent circuit models are very attractive because of their relative simplicity, enabling the monitoring
Available online 13 August 2016
of electrochemical systems that have a complex physical mechanism. In this paper, the stochastic algo-
rithm Differential Evolution is proposed to fit an equivalent circuit to the EIS results for a wide potential
Keywords:
range. EIS is often limited to the corrosion potential despite being widely used. This greatly hinders the
Electrochemical impedance
Impedance measurements
analysis regarding the effect of the applied potential, which strongly affects the interface, as shown, for
Corrosion example, in polarization curves. Moreover, the data from both the EIS and the DC values were used in the
Differential evolution proposed scheme, allowing the best fit of the model parameters. The approach was compared to the
Optimization standard Simplex square residual minimization of EIS data. In order to manage the large amount of
Stochastic methods generated data, the EIS-Mapper software package, which also plots the 2D/3D diagrams with potential,
was used to fit the equivalent circuit of multiple diagrams. Furthermore, EIS-Mapper also computed all
simulations. The results of 67 impedance diagrams of stainless steel in a 3.5% NaCl medium at 25 C
obtained in steps of 10 mV, and the respective values of the fitted parameters of the equivalent circuit
are reported. The present approach conveys new insight to the use of electrochemical impedance and
bridges the gap between polarization curves and equivalent electrical circuits.
2016 Elsevier Ltd. All rights reserved.

1. Introduction There are a number of possibilities for correlating impedance

Electrochemical impedance spectroscopy is a linear technique
applied to the study of corrosion and electrochemical systems
[1]. The superposition of potential or current sine-wave at the
points of the polarization curve produces the electrochemical
impedance diagrams of the surface-electrolyte interaction. This
characterization is of great importance as it allows the kinetic
analysis of electrochemical processes. Polarization curves show
the response of a steady-state interface, and are therefore used in
most investigations on corrosion. Electrochemical phenomena are
strongly dependent on potential. However, in most corrosion stud-
ies, impedance is chiefly measured only at the corrosion potential.
Thus, they hardly correlate the aspects of polarization curves with
the impedance diagrams for a given potential. Essentially, only the
few works that study the specific mechanism measure impedance
at various potentials [2,3].
Corresponding author.
E-mail address: maakappel@gmail.com (M.A.A. Kappel).
spectroscopy to diverse physical systems [4], the most common
being the association of EIS results by analogy with electrical
equivalent circuits. This analogy allows the EIS data to be used in
practical simulations and predictions of corrosion [5]. The equiva-
lent circuit is not a real physical model, but a model that behaves
as closely as possible to the physical system under study along the
frequency span. More than one equivalent circuit can be used [6,7],
representing certain subtleties in the choice of adequate topology.
Impedance spectroscopy is not restricted to electrochemical
systems, but is used in a variety of technological fields in relation
to the following processes: nanostructure functionality of osmosis
membranes [8], quality of porcine meat [9], brain tumor identifica-
tion [10], characterization of muscular tissue [11], etc. Moreover,
the extraordinary properties obtained by metamaterials can be
measured by impedance [12] and equivalent electric circuits are
useful in explaining the resonance frequency of these materials
[13]. Thus, the proper adjustment of equivalent circuits is of prime
importance in several fields, be in electricity, electrochemistry or
other physical applications where this equivalence presents

http://dx.doi.org/10.1016/j.measurement.2016.08.008
0263-2241/ 2016 Elsevier Ltd. All rights reserved.
 M.A.A. Kappel et al. / Measurement 94 (2016) 344354
additional advantages to the investigation at hand. In all these
investigations, correctly fitting equivalent electric circuits is essen-
tial in understanding the related phenomena that are ordinarily
very complex. Moreover, the choice of the equivalent circuit needs
to be consistent with the physical characteristics of the studied
phenomenon. However, the best fit can be found when the circuit
itself is adjusted by the optimization technique as in Ramos and
Janeiro [7]. In corrosion studies, the freedom of circuit topologies
is not considered a good choice as it generally lacks the model-
physical correspondence.
Typically, equivalent circuit fitting is performed by proprietary
software that uses only a single isolated EIS data point to minimize
the square residuals using deterministic algorithms [14]. The most
widely used method in electrochemical impedance applications is
Levenberg-Marquardt [15,16], but the use of other deterministic
methods, such as the Nelder-Mead Simplex, can also be found
[14]. However, the optimizing process regarding the equivalent
electrical circuit fitting get trapped in a local minimum, preventing
proper convergence to the global solution [17]. In general,
depending on the complexity and the number of circuit parame-
ters, different solutions can lead to similar responses of the
objective function, characterizing a multimodal problem [18].
In order to perform a robust parameter identification from
impedance spectra, stochastic methods can be applied. Indeed,
there are works exploring the application of stochastic methods
to electrochemical impedance circuit fitting, such as Simulated
Annealing [19], Evolutionary Strategies [20], Genetic Algorithms
[21] and Differential Evolution [22]. Gonzlez et al. [22] recently
proposed the use of Differential Evolution in EIS. All of these works
show advantages of stochastic over the classical approach if the
reliable identification is the objective.
This paper proposes the use of experimental DC (direct current)
values, concerning the steady-state response of an electrode, as a
regularization factor, besides considering the EIS data. This enables
fitting equivalent circuits with excellent agreement, ensuring their
physical analogy to the corrosion phenomena. We emphasize that
this concern is present in works of dissolution mechanisms, where
the kinetic response must be valid also for stationary cases. Some
examples are the papers of Mattos et al. [2] and Barcia et al. [23].
In contradiction to this, the concern is ordinarily not present in
works where equivalent circuits are applied, in spite of being by
far more numerous than the works that investigate the electro-
chemical mechanisms while considering the steady-state behavior.
Experimental impedance measurements of stainless steel were
performed in a wide potential range beyond the corrosion
potential and taken into account in our fitting. This method of mea-
surement differs from dynamical impedance spectroscopy [2426]
whose impedances are measured simultaneously with potential
sweeping.
2. Experimental
Samples of UNS S30400 stainless steel were abraded to #600
emery paper, rinsed with distilled water, dried with alcohol and
hot air. Subsequently, the samples were subsequently immersed
into aerated 3.5% NaCl aqueous solution at 25.0 0.2 C. Table 1
shows the chemical composition of the steel. The solution simu-
lates natural seawater and has a pH near neutrality. The counter
electrode used in the tests was a platinum wire in the form of a
coil. Saturated calomel electrode (SCE) was used as the reference
electrode. Initially, the system was subjected to an open circuit
condition during the 3600 s to achieve its stability. The sequence
of the potentiostatic polarization started at
400 mV up to
250 mV  SCE. The steps of DC potential used for mapping were
10 mV. The impedance was measured after a waiting period of
345
Table 1
Chemical composition of the steel UNS S30400 (% mass).
C Mn Si P S Cr Mo Al Cu
0.018 1.291 0.409 0.035 0.003 18.453 0.067 0.003 0.105
Co V Nb Ti Sn W Ni N Fe
0.135 0.043 0.006 0.002 0.006 0.017 8.017 0.152 Balance
100 s to stabilize at each new potential (step). The average scan
rate of the DC potential was circa 70 lV/s, low enough to allow a
stable measurement for a relatively high minimum frequency
(f P 0.100 Hz). If lower minimum frequencies are used, the scan
rate decreases. Either way, the mapping is obtained at a much
lower rate than those usually employed in the polarization curves
of stainless steel [27]. The sinusoidal perturbation amplitude was
8 mV rms. The electrochemical cell was located in a Faraday cage
to reduce the spurious signals.
3. Differential evolution
Differential Evolution (DE) is a meta-heuristic optimization
method developed by Storn et al. [28]. The main idea is to apply
so-called evolutionary operators to modify a vector population of
real values, whose vector characterizes a possible solution to the
problem. The applied operators are mutation, crossover, and selec-
tion. Differential Evolution shares several features with the basic
cycle of an evolutionary algorithm. Therefore, the algorithmic flow
is practically the same in both cases as depicted in Fig. 1.
The mutation operator is applied to create new individuals from
a group of individuals of the previous generation. It consists of a
weighted addition of the difference between two randomly
selected vectors v1, v2 to a third vector v3, which is also randomly
selected. The weight, named F, is a configuration parameter of the
algorithm, controlling the variation level of vector difference. This
level is a real constant value between 0 and 2. To the modified vec-
tor resulting from the previous operations a crossover is applied. A
fourth random vector v4 is again selected from the population and
its components are mixed to the modified vector, generating a test
vector. An objective function is applied to the test vector and v4 in
order to obtain the best result that is then transferred to the next
generation. This process repeats until the stopping criterion is
reached, usually characterized by the minimization or maximiza-
tion of the objective function.
As a stochastic direct search technique, Differential Evolution is
able to handle non-differentiable, nonlinear and multi-modal
objective functions. This aspect is extremely beneficial in cases of
complex problems. Furthermore, by the fact that stochastic pertur-
bation can be made independently, the method allows paralleliza-
tion and is considered efficient to deal with problems of high
computational cost as in the case of the present work. Although
the objective function used in impedance fitting itself does not
demand high computation costs, the consecutive application of
the optimization method for each potential can increase sensibly
the computation time. The fitting begins at lower potential and
goes to the upper ones. Then, the previous set of values can carry
the memory of the system at the next applied potential.
Differential Evolution is also easy to use, requiring an input of
few configuration parameters. Moreover, the method has good
convergence properties, i.e., good convergence to the global mini-
mum in independent tests [28].
Generally, stochastic optimization methods such as Differential
Evolution are not present among the options provided by propri-
etary impedance fitting tools, at least to EIS applications. These
tools normally offer only deterministic methods, which highly
depends on initial estimates. Thus, despite its advantages, few
346 M.A.A. Kappel et al. / Measurement 94 (2016) 344354

Fig. 1. Flow chart for differential evolution.

studies make use of these techniques in modeling EIS equivalent

circuits.

4. Equivalent electrical circuit

Various kinds of physical systems are well represented by

transfer functions. Thus, the behavior of distinct systems can be
found to be similar when their transfer functions are equivalent.
This motivates comparing the behavior of electrochemical systems
with known or proposed electrical circuits, so that a physical anal-
ogy between the phenomena involved can be found [1]. In this
work, the parameters of the equivalent circuit shown in Fig. 2 were
estimated for EIS data. For the two constant phase elements, and
the necessary resistors, seven parameters have to be fitting to a cir-
cuit as shown in Fig. 2 for each potential.
In theory, each circuit element models a specific physical
phenomenon of the corrosion system. Marcelin et al. [29] describe
the circuit (Fig. 2) in two main parts: resistance R1 and CPE1
(Q 1 and a1 ) are related to the passive film, in high frequency
domain; moreover, R2 and CPE2 (Q 2 and a2 ) are responsible for
expressing the charge transfer reactions (oxidation substrate/
oxygen reduction), in the low frequency region. This circuit was
Fig. 2. Equivalent circuit model used to simulate electrochemical impedance.
applied with success to simulate EIS spectra of stainless steel in a
neutral chloride solution at corrosion potential [27,28]. The param-
eters a and Q are independent of frequency and are associated with experimental impedance at each potential, taking the frequency
either surface or normal time-constant distribution [6]. RX as input parameter. The resulting circuit is then used to simulate
represents the resistance of the electrolyte. One aim of the present experiments computationally. The equivalent circuit impedance
work is to adjust the parameters of the equivalent circuit to the equation is given by Eq. (1).
 M.A.A. Kappel et al. / Measurement 94 (2016) 344354 347

Z1 tainty levels of EEC estimated parameters were studied by Ciucci

Z eq RX a ; 1
Z 1 Q 1 ix 1 1 [31]. He presented a procedure devoted to the generation of a sta-
tistical analysis, including the generation of confidence regions.
where The typical approach when dealing with ill-posed problems
R2 consists in the so-called regularization techniques, such as Tikho-
Z 1 R1 a : 2
R2 Q 2 ix 2 1 nov regularization [32]. Here, we propose the use of additional
experimental data approximation beyond impedance, as a regular-
In this equation, the parameters to be adjusted are RX , R1 , R2 , Q 1 , Q 2 , ization technique. Creating constraints of physical nature on the
a1 , and a2 . The input data are the applied potential, angular fre- adjustment, the optimization process has a better chance of finding
quencies x and the output is the impedance Z eq , having real and sound results. In the present work, the values of the steady-state
imaginary components. For each applied potential, experimental response of the electrode, i.e., the values of the DC currentpoten-
impedance values are obtained. Consequently, there are a large tial plot, obtained as the response of interface regulation, were
number of optimization problems to solve. The impedance Z is a used for such purpose. The equivalent impedance at the null fre-
complex quantity which can be represented in Eqs. (3)(6): quency allows the evaluation of steady-state current using Eqs.
(8) and (9). If the circuit is properly adjusted, the current values
Z Z real jZ imag ; 3
calculated in this way should be closer to DC values measured
or, equivalently: experimentally. Eq. (8) is an assumption that the polarization resis-
tance Rp must equal the reciprocal of the slope of the steady-state
Z jZjejh ; 4 current-polarization plot, as shown by Harrington et al. [33] to be
where valid since the system is stable (Eq. (8)). To perform the regulariza-
q tion methods, some possibilities exist. In the present paper, we use
jZj Z 2real Z 2imag ; 5 a direct relationship between applied potential and current, but
the use of polarization resistance or low frequency impedance
modulus could also be used.
Z imag
h tg
1 : 6  
Z real dE
Rp limx!0 Zx ss 8
Initially, the objective function to be used would be: dI
!2 !2 Since it is not possible to measure the impedance values when
X Z calc
Z exp X Z calc exp
imag
Z imag the frequency tends to zero, the equivalent circuit model is extrap-
f obj x real real
: 7
Z exp
real Z exp
imag
olated to fulfill this gap:

Eq. (7) is the sum of the squares of the relative deviations of the real limx!0 Zx limx!0 Z eq x Z eq 0 9
and imaginary parts of the impedance and is common in EIS fitting. Taking the following approximations,
Usually, commercial software products only use the electrochemi-
cal impedance spectrum to adjust the experimental data to the cir- dE DE E
Ecorr E
Ecorr
ss ; 10
cuit. In addition, they usually employ deterministic algorithms, dI DI Iss
Icorr Iss
such as Levenberg-Marquardt or Simplex, to minimize the objective Icorr 0, x 2pf, Eq. (8) [29] can be replaced by Eq. (11), expliciting
function (Eq. (7)) [15,16,14]. This enforces a unique answer,
the evaluation of the calculated current, Icalc ,
although it may not be the global optimum. As can be seen in
[22], the use of deterministic optimization methods can result in E
Ecorr
Icalc ; 11
erroneous values, potentially leading to misguided corrosion previ- Z eq 0
sions. Moreover, commercial software coupled to potentiostat only
fits individual EIS diagram. These facts may lead to wrong estimates where E is the applied potential and Iss the measured steady-state
of circuit parameters. Due to the characteristics of the circuit, for current at that potential. The superscript eq refers to the equivalent
example, it is possible to obtain different combinations of parame- electric circuit. The difference E
Ecorr is the overpotential in rela-
ters that result in exactly the same result of equivalent impedance tion to corrosion potential and expresses the trend of the electro-
for the frequency span. In other words, when an individual diagram chemical process; positive values are related to net anodic
is adjusted for a given limited frequency range, the uniqueness of processes, and negative values, net cathodic ones. Therefore, to
the parameter value set is not assured, albeit the EIS experimental ensure that the adjusted parameters are actually representing the
and simulated diagrams agree very well. As an equivalent model, best analogy of the electrochemical phenomenon with the electric
the fitted EEC must be able to reproduce both transient and station- circuit, the new objective function is proposed in this work as
ary responses of the system. This concern commonly exists in works expressed in Eq. (12).
of dissolution mechanisms. In these cases, the steady-state data is !2 !2
considered in the calculations of kinetic response. Works of Mattos X Z calc
Z exp X Z calc exp
imag
Z imag
f obj x real real

et al. [2] and Barcia et al. [23] are some examples of this type of Z exp
real Z exp
imag
application. On other hand, studies involving EEC fitting do not con- !2
cern the stationary data, despite being much more numerous than X Icalc
Iss
12
works that applies electrochemical mechanism models. Iss
In addition, the objective function normally has multiple local
minima, different from the global minimum. Thus, depending on The parameter Icalc is the current of the steady-state response of the
the initial estimates, the deterministic methods, in general, cannot equivalent circuit, calculated using Eq. (11) when the angular fre-
find the desired global minimum and may even estimate quency x is equal to zero, and Iss is the experimental DC value. This
parameters quite distant from the real ones, jeopardizing the inter- new objective function optimizes the transient (electrochemical
pretation of the electrochemical system. In general, the direct impedance) and steady-state (curve potential-current) simulated
application of nonlinear least squares to EIS models can give rise behaviors. Experimental EIS impedance does not be measured at
to ill-posed problems [30] and high uncertainty levels. The uncer- null frequency, but the associated impedance of the equivalent elec-
348 M.A.A. Kappel et al. / Measurement 94 (2016) 344354
Table 2
Parameter setting for Differential Evolution.
Parameter Value
F (variation vector of difference) 0.5
CR (crossover constant) 0.9
Population size 700
Max. number of generations 1000
tric circuit can be evaluated in zero frequency. Indeed, we use this
fact to improve the computational fitting. This procedure is nor-
mally used in EIS equivalent circuit adjustment. The settings used
in the Differential Evolution algorithm are listed in Table 2.
5. Results and discussion
Using EIS-Mapper software, developed at our research group, it
is possible to interactively visualize all Bode plots of experimental
data collectively gathered in figures corresponding to impedance
modulus and phase for all frequencies and applied potential
(Fig. 3). These figures can be seen both in two (2D) and three
dimensions (3D) and can be contrasted with the usual visualization
of diagrams limited to corrosion potential; 67 diagram EISs are
simultaneously present in these results. Furthermore, it is also pos-
sible to display the Nyquist diagrams and polarization curve con-
taining the DC values measured at each potential (Fig. 4). For this
study, a complementary program was also developed with the
same functionality of the EIS-Mapper, but instead of using as input
spreadsheets with experimental data, it uses the parameters of the
fitted equivalent circuit to computationally simulate the impe-
dance data. In the present case, the predominant phenomena that
take place are dependent on the applied potentials: cathodic, near
the corrosion potential (mixed) and anodic processes. As they are
homogeneous processes, they can be adjusted by a single electrical
equivalent circuit. If pitting was present, it would take more than
one circuit to simulate the data [34], but up to the higher applied
potential, it was not observed.
The potential range from
0.4 up to
0.14 V  SCE corresponds
to cathodic region, and for this aerated solution the main reaction
is oxygen reduction. At corrosion potential,
0.14 V  SCE, the
cathodic and anodic currents are equal; therefore, the net mea-
sured current is null. In spite of the extreme importance of study-
ing the phenomena at the corrosion potential simply because it is
the potential of almost all real metallic structures in service the
measurement of electrochemical properties outside this potential
is also fundamental to the understanding of the electrode pro-
cesses, as done, for instance, in polarization curve tests and
demonstrated by our experiments. Two drawbacks occur when
the impedance measurement is just restricted to corrosion poten-
tial: (i) the global aspect of electrochemical reactions is not avail-
able, (ii) the use of DC current density-potential cannot be used
to properly adjust the equivalent electric circuit parameters, as
employed in the objective function (Eq. (12)). For the present stain-
less steel, from
0.14 V  SCE up to 0.25 V  SCE a slight improve-
ment of impedance modulus with potential is observed, such as
depicted in Fig 3a. This behavior is somewhat expected, since the
current density is considered approximately constant over a given
potential range. Then, the impedance of an interface must increase,
but it was not clearly revealed as in the Fig. 3c. The high angle
spreads out a wide frequency range, typically 0.1300 Hz. In this
case, the use of the constant phase element, instead of pure
capacitor, is a standard choice.
Figs. 3 and 4 show the maps of experimental impedance and the
computational simulation by equivalent circuit in each potential.
The 3D maps enable a detailed comparison between the
experimental impedance and the simulation computed using the
adjusted equivalent circuit. All 67 diagrams were fitted, and not
simply one, as usually performed in corrosion studies at fixed
potential. The 2D maps with overlays of DC values allow even a
comparison of measured and calculated steady-state current den-
sity as the application of Eqs. (8)(11). By using the new objective
function, both results, impedance data (AC-alternate current), and
polarization curve (DC) are very close, indicating that the adjust-
ment was successful (see Fig. 6).
An example of a deceptive adjustment can be seen in Fig. 5. The
experimental and adjusted data using the standard Simplex algo-
rithm to minimize the objective function of Eq. (7) were taken into
account. In this case, the modulus and phase characteristics of
experimental impedance are quite similar to those simulated using
the adjusted equivalent circuit. However, when the polarization
curves are analyzed, there is a big difference amongst them (see
Fig. 6). This means that although the circuit parameters have been
fitted to the EIS data, the adjustment does not reflect the reality of
a physical analogy, because the DC values (here, an approximation
of polarization curve) are not the same for both simulations. Obvi-
ously, the electrochemical impedance measurement cannot be per-
formed at zero frequency, but the extrapolation of equivalent
circuit impedance to zero frequency allows the comparison to
the steady-state polarization curve. The approximate method
expressed by Eqs. (8)(10) along with the new objective function
(Eq. (12)) is a simple way to correlate these findings. Indeed, Har-
rington et al. [33] demonstrated that Eq. (8) is always valid if the
system is stable. Then, as the DC values were obtained for an
extra-low potential rate (70 lnV/s) this condition was satisfied in
our experiments.
As in any spectroscopic technique, EIS is experimentally fre-
quencylimited. Consequently, the fitting can be very good for
simulating at a frequency interval, but the necessity of extrapola-
tion, such as for null frequency, makes it clear whether the adjust-
ment is robust or not. Moreover, in the context of corrosion, this
interplay is a real necessity, but not employed to find the best
set of parameters. Thus, in Fig. 6, the computed current density
exhibits certain spikes that do not correspond to the experimental
curve. Additionally, the average Simplex curve does not agree as
well as the DE does, even when the objective function takes into
account the steady-state behavior (Eq. (12)), as can be seen in
Fig. 7.
Fig. 7 shows the behavior of objective functions from Eqs. (12)
and (7), with and without the use of stationary data, respectively,
using the parameters found by Differential Evolution and Simplex.
It is noticed that the Differential Evolution, using Eq. (12), even
with the deviation of the stationary data, results in objective func-
tion values similar to those found by the Simplex, the latter using
only transient data (Eq. (7)). When the parameters found by Sim-
plex are evaluated by the objective function of Eq. (12), including
in this case the deviation of stationary data, it is clear that the error
increases significantly, as shown by the green1 curve in Fig. 7. Thus,
despite the parameters estimated by the conventional method being
able to get good adhesion with the transient response of the system,
they are not capable of consistently simulating the stationary
response.
Commercial software products employ deterministic methods
to always produce the same set of results, since the stochastic ones
can produce a different solution at each run. However, this unique
answer provided by the deterministic method can be misguided,
depending on the initial estimate provided by an ill-advised oper-
ator. Furthermore, the computational cost of the fitting procedure
1
For interpretation of color in Fig. 7, the reader is referred to the web version of
this article.
 M.A.A. Kappel et al. / Measurement 94 (2016) 344354 349

Fig. 3. 3D impedance map: (a) experimental phase, (b) phase fit by equivalent circuit, (c) experimental modulus, (d) fitted modulus. Nyquist diagram: (e) experimental, (f)
simulated loops.

for all potential range can reach a hundred times higher than a
deterministic technique applied in only one potential. On the other
hand, the fitting itself does not have to be performed rapidly in this
type of problem. The simulation processing time does not depend
on the fitting algorithms, only its quality does which is what we
target. Furthermore, the stochastic nature of the DE algorithm is
made predictable by slight human intervention during the fitting
phase while evaluating the results, which is not a problem since
the simulation is still predicable. Each figure is the adjustment of
4020 points (67 diagrams  10 points/decade  6 frequency dec-
ades) for the seven parameters, regardless the chosen objective
function. It is important to notice that the main reason for the
increase of the computational cost is the consecutive application
of the optimization method for each of 67 measured diagrams.
Both the stochastic method and regularization technique can be
applied in only one chosen impedance diagram, if the stationary
350 M.A.A. Kappel et al. / Measurement 94 (2016) 344354

Fig. 4. Impedance phase maps: (a) experimental, (b) fit with equivalent circuit. Modulus map: (c) experimental, (d) fit Nyquist diagrams: (e) experimental, (f) simulated
loops.

data is available. However, it will not be possible to simulate the
complete stationary response in the form of the DC value curves,
as shown in the present paper.
The parameters obtained by both Simplex and DE methods
were good enough to fit the equivalent circuit to the impedance
data, but only those estimated using Differential Evolution with
the proposed Eq. (12) as the objective function were able to also
approximate the DC values, as shown in Fig. 6. The DC current
density-potential plot is closer to the experimental curve, but it
is computationally intensive. Both methods obtained similar val-
ues of RX , around 5.5 X cm2. This specific parameter is negligible
to the others since it has lower magnitude, and can be easily found
in the impedance modulus at high frequency. The same equivalent
circuit used here has previously been employed to model superdu-
plex stainless in high chloride content [35]. Some dependencies of
parameters with potential can be shown in Figs. 810. For some
 M.A.A. Kappel et al. / Measurement 94 (2016) 344354 351

Fig. 5. Phase map: (a) experimental, (b) fitted without the use of DC values. Modulus map: (c) experimental, (d) fitted without the use of DC values.

Fig. 6. Steady-state curves for the equivalent circuit fitted by both methods, and
experimental DC values. Fig. 7. Behavior of the objective functions of Eqs. (12) and (7), with and without the
use of stationary data, respectively, using the parameters found by Differential
Evolution and Simplex.

parameters the difference between Simplex and Differential Evolu-

tion produces similar values, as is the case of R1 and a1 of CPE1 in deterministic and stochastic methods, with small differences. For
the Figs. 8 and 9. The CPE1 components are sensitive to high fre- CPE2, which is related to the low frequency range, the fitting is
quency, then the stability is high, and the results are good for more difficult because some drift can perturb the system. Low
352 M.A.A. Kappel et al. / Measurement 94 (2016) 344354

Fig. 8. Evolution of parameter R1 throughout the potentials, using DE and Simplex methods.

Fig. 9. Evolution of parameter a1 throughout the potentials, using DE and Simplex methods.

frequency implies long time to complete the experimental
measurement. Then, a less adaptive method as Simplex can exhibit
invalid results (Fig. 10), as is the case of a2 of CPE2. Some diagrams,
at specific potential, reach very high values, well beyond unity, sig-
naling a scarcity of physical significance. Conversely, for the
stochastic method, all values are compatible with the expected val-
ues for the constant phase element [1]. This fact implies that mul-
tiple solutions can be found fitting EIS data for the same equivalent
circuit, but neither always consistent with the underlying theory,
nor the DC behavior.
The impedance is an extension of the polarization curve and the
same model must account for both, as described by Keddam [36].
Potentiostatic EIS for wide potential range, as done in the present
paper, gives the possibility of filling the gap between impedance
spectra and polarization curves. However, the fitted model has
always to obey the steady-state as well as the transient behavior.
For higher applied potentials, as present in Ref. [37], it is neces-
sary to use an equivalent electric circuit that takes into account the
occurrence of localized corrosion. Most studies have relied on
steady-state measurements to determine parameters such as pit-
ting potential and protection potential [34], but the use of EIS is
possible in the presence of localized corrosion. Indeed, Mansfeld
[34] has proposed a circuit that considers the surface fraction occu-
pied by the pits coupled by the impedance of passive film. In this
case, the circuit is different from that used here, and consequently
the parameters have diverse physical interpretation. However, the
dependence on equivalent circuit also changes with potential,
more specifically in pitting potential.
 M.A.A. Kappel et al. / Measurement 94 (2016) 344354
Fig. 10. Evolution of parameter a2 throughout the potentials, using DE and Simplex methods.
6. Conclusions
In this paper, electrochemical impedance diagrams of stainless
steel UNS S30400 were obtained in a potentiostatic manner for a
wide range of applied potential. Furthermore, a stochastic algo-
rithm was employed to fit the equivalent electrical circuits to all
data obtained by electrochemical impedance spectroscopy, using
additionally the DC values from current density-potential
response. The results show that the combination of a stochastic
method with the use of steady-state data in the optimization pro-
cess, together with the comprehensive set of impedance data, is a
great way to fit equivalent electric circuits that has a valid physical
analogy with the phenomena of electrochemical corrosion.
Results were also obtained for the Simplex algorithm and the
traditional cost function, showing the advantages of the proposed
approach, which can produce better results, and success in cases
where the former is not sufficiently precise in DC behavior. By
applying the proposed procedure to all impedance diagrams, both
the impedance and DC values were correctly simulated. In contrast,
using the traditional approach lead to improper deviations of the
DC curve, exhibiting incorrect parameters to represent an equiva-
lence between the EEC and the electrochemical system.
An argument in favor of deterministic techniques is the unique-
ness of the results informed to end users, although this consistent
response greatly depends on the provided initial estimate. The
usual commercial EEC fitting software systems offer a series of
tools for guessing initial estimates, e.g., geometrical correlation
and multiple randomly generated values. Using a meta-heuristic
such as Differential Evolution involves an intrinsic and well-
tested logic that distinguishes the technique from simple educated
guesses.
The main limitation of wide potential range EEC fitting is com-
puter hardware. In this work, 67 EIS diagrams were measured, rep-
resenting an equal number of optimization problems for the
evolutionary algorithm. Depending on the available computational
capacity, using such an approach to estimate the EEC parameters
for a wide potential range can become unwieldy. However, the
specific application of a stochastic method and regularization with
stationary values does not significantly increase the computational
cost, and can be performed for any common uni-potential
353
impedance fitting problem. Furthermore, Differential Evolution is
massively parallelizable, so there are direct ways to reduce com-
pute time for applications.
It must be noted that, in actual practice, the fitting of equivalent
circuits does not have to be solved on a recurrent or emergencial
basis, so that execution time is not a critical problem; the main
objective is to obtain reliable parameters for the chosen equivalent
electric circuit, which can then efficiently simulate the experi-
ments. The visualization of EIS over a wide potential range also
produces new insights to experimenters.
Acknowledgments
The authors acknowledge the support of Capes/Brazil, whose
research incentive through scholarship is essential to the develop-
ment of scientific work like this.
References
[1] M.E. Orazen, B. Tribollet, Electrochemical Impedance Spectroscopy, John Wiley
& Sons Inc., USA, 2008.
[2] O.R. Mattos, O.E. Barcia, The role of chloride and sulphate anions in the iron
dissolution mechanism studied by impedance measurement, Electrochim. Acta
35 (1990) 10031009.
[3] J.-M. Hu, J.-Q. Zhang, C.-N. Cao, I.-M. Hsing, Kinetics investigation of H2/CO2
electrooxidation in PEFCs by the combined use of equivalent circuit fitting and
mathematical modeling of the Faradaic impedance, Electrochim. Acta 49
(2004) 52275234.
[4] R.T. Coverdale, H.M. Jennings, E.J. Garboczi, An improved model for simulating
impedance spectroscopy, Comp. Mater. Sci. 3 (1995) 465474.
[5] D.C. Silverman, Corrosion prediction in complex environments using
electrochemical impedance spectroscopy, Electrochim. Acta 38 (1993) 2075
2078.
[6] B. Hirschorn, M.E. Orazem, B. Tribollet, V. Vivier, I. Frateur, M. Musiani,
Determination of effective capacitance and film thickness from constant-
phase-element parameters, Electrochim. Acta 55 (2010) 62186227.
[7] P.M. Ramos, F.M. Janeiro, Gene expression programming for automatic circuit
model identification in impedance spectroscopy: performance evaluation,
Measurement 46 (2013) 43794387.
[8] S.Y. Yeo, Y. Wang, T. Chilcott, A. Antony, H. Coster, G. Leslie, Characterizing
nanostructure functionality of a cellulose triacetate forward osmosis
membrane using electrical impedance spectroscopy, J. Membr. Sci. 467
(2014) 292302.
[9] Y. Yang, Z.-Y. Wang, Q. Ding, L. Huang, C. Wang, D.-Z. Zhu, Moisture content
prediction of porcine meat by bioelectrical impedance spectroscopy, Math.
Comput. Model. 58 (2013) 819825.
354 M.A.A. Kappel et al. / Measurement 94 (2016) 344354
[10] H.-G. Jahnke, A. Heimann, R. Azendorf, K. Mpoukouvalas, O. Kempski, A.A.
Robitzki, P. Charalampaki, Impedance spectroscopy an outstanding method
for label-free and real-time discrimination between brain and tumor tissue
in vivo, Biosens. Bioelectron. 46 (2013) 814.
[11] F. Clemente, M. Romano, P. Bifulco, M. Cesarelli, EIS measurements for
characterization of muscular tissue by means of equivalent electrical
parameters, Measurement 58 (2014) 476482.
[12] D.R. Smith, J.B. Pendry, M.C.K. Wiltshire, Metamaterials and negative refractive
index, Science 305 (2004) 788792.
[13] P.V. Tuong, J.W. Park, V.D. Lam, K.W. Kim, H. Cheong, W.H. Jang, Y.P. Lee,
Simplified perfect absorber structure, Comp. Mater. Sci. 61 (2012) 243247.
[14] J.-L. Dellis, J.-L. Carpentier, Nelder and Mead algorithm in impedance spectra
fitting, Solid State Ion. 62 (1993) 119123.
[15] J.R. Macdonald, J. Schoonman, A.P. Lehnen, The applicability and power of
Nonlinear Least Squares for the analysis of impedance and admittance data, J.
Electroanal. Chem. 131 (1982) 7795.
[16] Y.T. Tsai, D.H. Whitmore, Nonlinear Least squares analyses of complex
impedance and admittance data for solid electrolytes, Solid State Ion. 7
(1982) 129139.
[17] J.R. Macdonald, W. Thompson, Strongly heteroscedastic nonlinear regression,
Commun. Stat-Simul. C. 20 (1991) 843886.
[18] I. Sgura, B. Bozzini, Numerical issues related to the modelling of
electrochemical impedance data by non-linear least-squares, Int. J. Nonlinear
Mech. 40 (2005) 557570.
[19] P. Buschel, U. Troltzsch, O. Kanoun, Use of stochastic methods for robust
parameter extraction from impedance spectra, Electrochim. Acta 56 (2011)
80698077.
[20] O. Kanoun, U. Troltzsch, H.R. Trankler, Benefits of evolutionary strategy in
modeling of impedance spectra, Electrochim. Acta 51 (2006) 14531461.
[21] S. Sharifi-Asl, M.L. Taylor, Z. Lu, G.R. Engelhardt, B. Kursten, D.D. Macdonald,
Modeling of the electrochemical impedance spectroscopic behavior of passive
iron using a genetic algorithm approach, Electrochim. Acta 102 (2013) 161
173.
[22] F. Gonzlez, D. Greiner, J.J. Asnrez, V. Mena, R.M. Souto, J.J. Santana,
Simulacin de processos qumicos mediante algoritmos evolutivos:
aplicacinal ajuste de parmetros de impedncia em sistemas
electroqumicos, Afinidad 72 (2015) 278283 (in Spanish).
[23] J.A. Caldern, O.E. Barcia, O.R. Mattos, Reaction model for kinetic of cobalt
dissolution in carbonate/bicarbonate media, Corros. Sci. 50 (2008)
21012109.
[24] R.L. Sacci, F. Sealand, D.A. Harrington, Dynamic electrochemical impedance
spectroscopy for electrocatalytic reactions, Electrochim. Acta 131 (2014) 13
19.
[25] K. Darowicki, Theoretical description of the measuring method of
instantaneous impedance spectra, J. Electroanal. Chem. 486 (2000) 101105.
[26] J.E. Garland, C.M. Petit, D. Roy, Analysis of experimental constraints and
variables for time resolved detection of Fourier transform electrochemical
impedance spectra, Electrochim. Acta 49 (2004) 26232635.
[27] I.N. Bastos, S.S.M. Tavares, F. Dalard, R.P. Nogueira, Effect of microstructure on
corrosion behavior of superduplex stainless steel at critical environment
conditions, Scripta Mater. 57 (2007) 913916.
[28] R. Storn, K. Price, Differential Evolution a simple and efficient heuristic for
global optimization over continuous spaces, J. Glob. Optim. 11 (1997) 341
359.
[29] S. Marcelin, N. Pbre, S. Rgnier, Electrochemical characterization of a
martensitic stainless steel in a neutral chloride solution, Electrochim. Acta
87 (2013) 3240.
[30] I. Sgura, B. Bozzini, Numerical issues related to the modelling of
electrochemical impedance data by non-linear least-squares, Int. J. Non-
Linear Mech. 40 (2005) (2005) 557570.
[31] F. Ciucc, Revisiting parameter identification in electrochemical impedance
spectroscopy: weighted least squares and optimal experimental design,
Electrochim. Acta 87 (2013) 532545.
[32] D. Maier, P. Hug, M. Fiederle, C. Eiche, D. Ebling, J. Weese, High resolution
method for the analysis of admittance spectroscopy data, J. Appl. Phys. 77
(1995) 38513857.
[33] D.A. Harrington, P. van den Driessche, Mechanism and equivalent circuits in
electrochemical impedance spectroscopy, Electrochim. Acta 56 (2011) 8005
8013.
[34] F. Mansfeld, Analysis and Interpretation of EIS Data for Metals and Alloys,
Technical Report N. 26, Solartron Analytical, USA, 1999.
[35] E.V. Costa, I.N. Bastos, T.J. Mesquita, A. Ferreira, R.P. Nogueira, Effect of carbon
dioxide and temperature on passive film parameters of superduplex stainless
steel, Mater. Res. 16 (2013) 929936.
[36] M. Keddam, W.R. Whitney, Award lecture: application of advanced
electrochemical techniques and concepts to corrosion phenomena, Corrosion
62 (2006) 10561066.
[37] I.N. Bastos, M.P.M. Carvalho, R. Fabbri, R.P. Nogueira, Visualization of EIS at
Large Potential Range New Insights. Available from: <1310.1629v2>. [cond-
mat.mtrl-sci] 29 Oct 2013.