DOI: 10.1002/ente.

201600154

Analysis of Lithium-Ion Battery Models Based on

Electrochemical Impedance Spectroscopy
Uwe Westerhoff,*[a, c] Kerstin Kurbach,[a, c] Frank Lienesch,[b] and Michael Kurrat[a, c]

This work is an overview of various equivalent circuits (ECs) trum, a review and assignment of equivalent circuits for po-
containing various degrees of detail. The ECs are evaluated tential applications is given. Simple equivalent circuits with
in terms of model accuracy and parameterization time for a series resistor and a maximum of two resistance–capaci-
the systematic assignment of an equivalent circuit to applica- tance (RC) elements are ideal for simulations with lower dy-
tion fields. For this purpose, impedance spectra were mea- namics. Equivalent circuits with up to five RC elements or
sured using electrochemical impedance spectroscopy at dif- even a constant-phase element (CPE) are promising for sim-
ferent states of charge, health and temperatures. Then the ulating highly dynamic processes. By using RCPE elements
parameters of the EC were extracted using the least-squares the impedance spectrum can be modeled with the highest ac-
method and the Levenberg–Marquardt algorithm. After curacy, which is why this type of model should be used for di-
comparing the simulated to the measured impedance spec- agnostic purposes.

Introduction

In the literature three different approaches of modeling Li- an approach is shown that describes the individual compo-
ion batteries are typically proposed: theoretical quantitative nents of a battery cell with electrical and electrochemical ele-
models (white box),[1] qualitative models with experiment ments. The series connection of the equivalent circuit ele-
(gray box),[1] and experimental quantitative models.[1, 2] Many ments yields the entire model for a battery cell.
parameters are required for the calculation of differential Another approach to determine the optimum equivalent
equations, which are obtained from the literature or identi- circuit configuration is the density function of the distribu-
fied by using complex measurement methods for the quanti- tion of relaxation times (DRT).[9, 10] Using this method, the
tative theoretical models of the white box method.[1] There- high-intensity characteristic frequencies of a measured impe-
fore measurements of the conductivity of the electrolyte or dance spectrum are determined to derive the number of RC-
electrode, porosity, particle radius distribution, tortuosity, or elements. A simplified version to determine the equivalent
diffusion coefficients of the individual materials should be circuit configuration is curve sketching to identify the
conducted and their dependency on temperature and aging minima, maxima, and inflection points in the Nyquist plot.[11]
determined.[3] So-called black modeling is based on purely The frequencies of the distinctive points are also called pro-
mathematical models.[4] In this case the input and output var- cess frequencies as they show in which frequency range of
iables are interconnected using control structures and empiri-
cal data. The quality of the models improves the more data [a] U. Westerhoff, K. Kurbach, Prof. Dr. M. Kurrat
(so-called training data) it receives.[5] Thus the models learn Institute of High Voltage Technology and Electrical Power Systems (elenia)
to simulate the electrical behavior without needing the physi- Technische Universit-t Braunschweig
Schleinitzstraße 23, 38106 Braunschweig (Germany)
cal information of the battery. The use of electrical equiva- E-mail: [email protected]
lent circuits is firmly established and belongs to the gray box [b] Dr. F. Lienesch
category, the qualitative models using experimental data. If Physikalisch-Technische Bundesanstalt
the correct model assumptions and order are fulfilled, the Bundesallee 100, 38116 Braunschweig (Germany)
equivalent circuits can be a quantitative model. In the litera- [c] U. Westerhoff, K. Kurbach, Prof. Dr. M. Kurrat
Battery LabFactory Braunschweig
ture, a great variety of options for modeling Li-ion batteries
Technische Universit-t Braunschweig
with electrical equivalent circuits has been presented.[6] The Langer Kamp 19, 38106 Braunschweig (Germany)
choice of the equivalent circuit depends strongly on the cell The ORCID identification number(s) for the author(s) of this article can
chemistry and the detailed characteristics. Therefore no stan- be found under http://dx.doi.org/10.1002/ente.201600154.
dard model can be used for every battery type,[7, 8] as the T 2016 The Authors. Published by Wiley-VCH Verlag GmbH & Co. KGaA.
probability that the model is over- or under-modeled is high. This is an open access article under the terms of the Creative Commons
Attribution Non-Commercial License, which permits use, distribution and
To obtain a model with an optimal reproduction of the cell
reproduction in any medium, provided the original work is properly cited,
characteristics, the basic structure of a battery has to be and is not used for commercial purposes.
known. Additionally, an approach is needed for the estima- Part of a Special Issue on “Li-Ion Batteries”. To view the complete issue,
tion of the parameters of the equivalent circuit. In Figure 1, visit: http://dx.doi.org/10.1002/ente.v4.12

Energy Technol. 2016, 4, 1620 – 1630 T 2016 The Authors. Published by Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim 1620
 21944296, 2016, 12, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/ente.201600154 by Cochrane Mexico, Wiley Online Library on [24/08/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Figure 1. Description of individual cell components with equivalent circuit elements.

the impedance spectrum the individual electrochemical pro- is highly dependent on the particle size distribution.[32] In ad-
cesses occur.[12] dition, the diffusion behavior (especially the solid-state diffu-
sion) should be modeled properly to consider the slow pro-
cesses occurring in the battery; therefore a Warburg element
Modeling is used. For diffusion processes, the impedance spectrum Ny-
In various publications different equivalent circuits are se- quist diagram extends with a 458 angle upward slope. This
lected as the model approaches for simulating the battery behavior can also be modeled using a CPE element; howev-
characteristics.[5] The quality of an equivalent circuit depends er, the physicochemical description of Li-ion batteries is lo-
on the application and differs essentially by the speed of pa- cated at the CPE for the porosity and particle radius distri-
rameterization and the accuracy. In Figure 2, the applied bution, whereas the physicochemical origin at a Warburg ele-
equivalent circuit models are shown, which differ in terms of ment describes the diffusion behavior of the carriers. The
the two aforementioned criteria. The examined equivalent equivalent circuit configurations in Figure 3 were derived
circuit models can be used to simulate the impedance spectra from accumulated experience with impedance measure-
of circuits consisting of only resistors, inductors, and capaci- ments.
tors. Most battery systems, such as those in full-vehicle simu-
lations, stationary storage in the home, or large-scale electri-
Parameter estimation
cal grid storage, can already be described with sufficient ac-
curacy using these models.[13–15] For more detailed considera- The parameters for the equivalent circuit are, as mentioned
tions of the cell characteristics and the dependence on the above, elements determined from the measured impedance
state of charge (SOC), the temperature (T) or the state of spectra. For this purpose, as a first step curve sketching is
health (SOH), further equivalent circuit elements can be used to estimate the parameters of the components. The esti-
used. This is implemented by using constant-phase elements mated parameters are the initial values for the minimization
(CPEs, Figure 3). With CPEs it is considered that the real function, which is based on the least-squares method in
electrodes are not plates and have no uniform boundary be- Equation (1).[11] It is necessary to set appropriate initial con-
cause they are produced with a high porosity to intercalate ditions for the algorithm so that it can locate the global mini-
the lithium ions.[16] The diffusion behavior of the electrodes mum and does not get stuck in a local minimum. Using the

Figure 2. Equivalent circuit models with resistance, inductor, and a variation in the the number of RC elements; these are the RC models.

Energy Technol. 2016, 4, 1620 – 1630 T 2016 The Authors. Published by Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim 1621
 21944296, 2016, 12, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/ente.201600154 by Cochrane Mexico, Wiley Online Library on [24/08/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Figure 3. Equivalent circuits with resistance, inductance, capacitance, constant-phase-elements, and Warburg elements; these are the CPE models.

least-squares method, the parameters are adjusted so that the function of electrochemical impedance spectroscopy and
the sum of squared deviation (residuals) is minimized. the measurement setup are described in the Experimental
Section. For a description of the physical and electrochemical
X
fmax effects that can be identified in the impedance spectrum, the
@ >2
min f ð xÞ ¼ min ZEC ðR; C; L; CPE; W; fn Þ @ Zmeasure;n impedance spectrum is divided into individual frequency
x R;C;L;CPE;W
n¼fmin ranges.[24, 20] In the very high frequency range > 20 kHz, an in-
ð1Þ ductive behavior is measured in addition to the real part of
the impedance. The behavior is mainly caused by the mea-
For variation of the parameters in the direction of the global surement setup such as the connecting lines and the type of
minimum, there are a number of algorithms. The following cable wiring. That the battery cell does not have an inductive
two examples of these algorithms are discussed in more component cannot be excluded, however this was not explic-
detail:[17–19] itly observed in these measurements. Other processes are, for
example, the charge-transport processes in the electrolyte,
· trust-region algorithm the solid–electrolyte interphase (SEI), and in the active ma-
· Levenberg–Marquardt algorithm terial (including the anode and cathode). The charge-transfer
processes, both from the electrolyte into the SEI and from
With the trust-region expansion the iteration increment is the SEI into the active material of the anode/ cathode, are
increased. In a defined radius around the iterate (the confi- located in the middle frequency range (typically 1 kHz–
dence interval) the algorithm searches for a greater mini- 10 mHz) and are represented in the impedance spectrum in
mum as it is defined by the normal increment. This allows the form of semicircular arches.[25] However, the transitions
the process to converge very fast toward the steepest de- of the processes are often so subtle, that a discretization of
scent. The Levenberg–Marquardt algorithm is based on a sim- impedance ranges for the individual transport and transfer
ilar principle. The iteration step size is increased by a factor processes is very difficult. The separator is produced as
that is recalculated from one step to another. This provides a porous structure so that the lithium ions can run through it.
a robust algorithm that converges despite having poor start Therefore, the separator is shown as a capacitor in parallel
parameters in the direction of the steepest descent. It com- with a resistor. However the capacity effect of the separator
bines the advantages of the steepest descent method and the is so small compared to the other components that this RC
Gauss–Newton method. The difference between the two element can usually be neglected. In the low-frequency
methods is that, in the trust-region method, the radius is de- range (< 10 kHz) diffusion processes dominate in the anode
termined directly whereas, in the Levenberg–Marquardt and cathode. The course of the impedance spectrum corre-
method, it is determined implicitly, through the use of sponds to a 458 rising line in the Nyquist plot. However, not
a damping parameter.[17, 18] only electrochemical processes can be derived from the im-
pedance spectrum but also physical quantities such as the
conductivity of lithium-ions and electrons. The only directly
Electrochemical Impedance Spectroscopy
measured parameter for an equivalent circuit is a purely
Electrochemical impedance spectroscopy has long been used ohmic resistance at abscissa zero crossing for the imaginary
to characterize the condition of a battery and for the descrip- part of the impedance. This resistance is dominated by the
tion of the electrochemical characteristics and processes in conductivity of the electrolyte and thus constitutes a resist-
the cell.[2, 20–22] Before the electrochemical impedance spec- ance for the lithium-ion transport.
troscopy measurement is applied, it is important to take into When recording impedance spectra as a function of vari-
account that a battery is a nonlinear, time-invariant system, ous experimental states (i.e., SOC, T, SOH) the selected step
and therefore a long rest time is needed to ensure that the sizes must be sufficiently large so that a change in the impe-
battery is in electrochemical equilibrium.[23] Information on dance spectrum is caused. Only with the right selection of

Energy Technol. 2016, 4, 1620 – 1630 T 2016 The Authors. Published by Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim 1622
 21944296, 2016, 12, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/ente.201600154 by Cochrane Mexico, Wiley Online Library on [24/08/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
step sizes can the electrochemical processes be considered This dependence changes the entire characteristic curve
separately and expressed in response to the experimental behavior, which also becomes visible in all parameters of the
conditions. equivalent circuit. The value of the abscissa zero crossing
and thus the value of the internal resistance increases greatly
with decreasing temperature. Furthermore, the first semicir-
Results and Discussion cle increases slightly, whereas the second semicircle is
spreading much stronger. The internal resistance increases
To evaluate the equivalent circuits, first the characteristics of
even further with these two expansions of the semicircle. If
a Li-ion battery cell were determined by means of electro-
the temperature rises again, the impedance spectrum revers-
chemical impedance spectroscopy in various metrological
es back to its original state, and thus the temperature behav-
studies. For this purpose, various states of the battery were
ior is a reversible process. The requirement is that the cell is
prepared to investigate the changes of the characteristics in
operated within the prescribed temperature range, which was
the impedance spectrum. For the metrological tests, the vari-
the case. In Figure 5, the impedance spectrum expands with
ables with the greatest influence on the battery performance
decreasing state of charge in the measured mean frequency
and characteristics have been varied. These variables are the
range of 1 Hz > f > 50 Hz. This area is often associated in the
ambient temperature, the change in the state of charge, and
literature with the charge-transfer process, which describes
the continuous state of aging. The investigations were per-
the intercalation process of the lithium ions into the active
formed in the following areas:
material of the electrodes.
· temperature range: 50 to @20 8C
· state of charge range: 100 to 0 %
· state of health range: 100 to 86 %

The different states of health have been set by the de-

crease in capacity due to a 1 C cyclization of cells. In the lab-
oratory, the battery cells were charged at a depth of dis-
charge (DOD) of 100 % with constant current–constant volt-
age (CC–CV) and discharged with CC. The percentage de-
crease in capacity relative to the initial capacity defines the
SOH. The temperature was first investigated at @40 8C.
However, the frequency range was not changed during the
experiment, which proved to be disadvantageous afterwards.
In Figure 4 an abscissas zero crossing could not be measured
at a temperature of @20 8C because of the shift of the process
frequencies which will be discussed in Figure 7. Nevertheless,
the temperature-dependent behavior in the impedance spec-
trum is clearly visible. Figure 5. Change of the impedance spectrum showing the dependence on the
state of charge at T = 25 8C, SOH = 100 %, frequency range 100 kHz–5 mHz,
and AC amplitude 1/20 C.

With decreasing state of charge, the resistive parts of the

impedance rise correspond to an increase in the internal re-
sistance. Due to the increase of the internal resistance,
a higher voltage drop arises, and the temperature of the cell
increases at the same current. This behavior is observable in-
fluences the open-circuit voltage characteristics of a Li-ion
battery cell, as the battery voltage has a very strong nonlin-
ear decrease shortly before reaching the final discharge volt-
age. The change in the state of charge is a reversible process
when the prescribed voltage range is maintained. The impe-
dance spectra of the aged cell in Figure 6 also enlarge with
time, at constant temperature and state of charge. In addi-
tion, the internal resistance of the cell increases almost con-
tinuously. These aging processes are irreversible, and there-
Figure 4. Change of the impedance spectrum showing dependence on the
temperature at SOH = 100 %, SOC = 50 %, frequency range 100 kHz–10 mHz, fore attention was paid to a careful treatment of the cells
and AC amplitude 1/20 C. throughout the investigations.[26]

Energy Technol. 2016, 4, 1620 – 1630 T 2016 The Authors. Published by Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim 1623
 21944296, 2016, 12, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/ente.201600154 by Cochrane Mexico, Wiley Online Library on [24/08/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Figure 6. Change of the impedance spectrum showing dependence on the
state of health at T = 25 8C, SOC = 100 %, frequency range 100 kHz–5 mHz,
and AC amplitude 1/20 C.

All of the influencing variables (temperature, state of

charge, aging) result in the expansion of the second semicir-
cle and thus lead to an increase in charge-transfer resistance.
To be able to state which of the three factors is responsible
for the increase, the other parameters have to be included as
well. For this reason, the process frequencies were also deter-
mined to deduce the original influence factor. Figure 7 a illus-
trates in which process parts these special frequencies have
to be determined in the impedance spectrum. At the zero
crossing of the imaginary part of the impedance is the fre-
quency fZIM,0, which indicates the frequency point of a purely
ohmic resistance. The semicircle in the middle of the impe-
Figure 7. Shifts in the process frequencies depending on various factors of an
dance spectrum is dominated the most by the charge-transfer
impedance spectrum: a) impedance spectra, b) state of charge, c) state of
process and has its characteristic frequency fZIM,max at the health, d) temperature.
maximum of the imaginary part of the impedance. From the
frequency fZIM,min the diffusion processes start in the battery.
This frequency is determined at the minimum of the imagi- with a nonlinear, continuous behavior to lower frequencies,
nary part of the impedance. For the range of the frequency, as shown in Figure 7 d. Only the frequency fZIM,0 is greater,
the following boundary condition applies: fZIM,0 @ fZIM,max > which simply indicates an increase in the internal resistance.
fZIM,min. Before the evaluation of the equivalent circuits it was ex-
In Figure 7 b it is shown that the frequencies in the course amined which algorithm is the most suitable for parameter
of the discharge increase only slightly and then fall continu- estimation. Only the five equivalent circuits of 1 RC up to
ously. Upon reducing the state of charge, the frequency 5 RC were used to be able to eliminate the influence of dif-
changes, which describes significantly the charge-transfer ferent equivalent circuit elements within the CPE. The
process. This shows that an impedance measurement at equivalent circuit equation of the impedance, generally writ-
a fixed frequency (e.g., 1 kHz for AC resistance measure- ten, forms Equation (2).
ment) cannot always measure a specific process frequency. In
the course of aging (Figure 7 c) there is a continuous, perma-
X
5
Ri
nent shift in the process frequencies in addition to the ZEC;iRC ðR; C; f Þ ¼ R0 þ ð2Þ
change in frequency caused by the state of charge and tem- i¼1
1 þ jw ? Ri ? Ci
perature. With continuous aging, the impedance of a cell is
increased by depletion of the electrolyte of free lithium ions, The parameters for these equivalent circuits were deter-
which results in growth of the SEI layer. Thus, the entire im- mined by the least-squares method in Equation (1) and the
pedance spectrum shifts further into the capacitive range of Levenberg–Marquardt algorithm and trust-region algorithm.
impedance. The result is an increase of the zero crossing fre- The numerically calculated parameters for the respective
quency. At low temperatures, the process frequencies shift equivalent circuit elements were parameterized to simulate

Energy Technol. 2016, 4, 1620 – 1630 T 2016 The Authors. Published by Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim 1624
 21944296, 2016, 12, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/ente.201600154 by Cochrane Mexico, Wiley Online Library on [24/08/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
the impedance spectrum in a function of the previously mea-
sured frequencies. This was followed by the comparison of
the measured with the simulated impedance spectrum. The
relative deviation of the individual impedances was separate-
ly calculated for the real and imaginary parts of the impe-
dance in Equation (3).

ZRE;xRC ð f Þ ZIM;xRC ð f Þ
rRE ð f Þ ¼ @ 1; rIM ð f Þ ¼ @1 ð3Þ
ZRE;measure ð f Þ ZIM;measure ð f Þ

From the relative deviation of the complex impedance

components, the relative deviation of the total impedance is
calculated by means of the absolute value function in Equa-
tion (4).

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Figure 8. Deviations of the simulated from the measured impedance spec-
rZ ð f Þ ¼ ðrRE Þ2 ?ðrIM Þ2 ð4Þ trum at a state of charge of 50 %, temperature of 25 8C, and no aging: L–
M: Levenberg–Marquardt, T-R: trust-region, (solid line) standard deviation,
(dashed) average deviation.
As rz is still a function of frequency, this value is further eval-
uated to derive a characteristic value for the quality of the
whole impedance spectrum. The quality criteria are the aver- Evaluating the equivalent circuit models
age deviation [Eq. (5)] and the standard deviation [Eq. (6)],
which are determined by the relative deviation of the mod- The reproduction of a measured impedance spectrum with
eled to the measured impedance spectrum. At the average the examined equivalent circuits is presented graphically in
deviation, the sum of all relative deviations is divided by the Figure 9 and as data values in Table 1. In Figure 9 only the
number of frequencies: frequency range 20 kHz < f < 20 mHz is shown because the
diffusion section and the inductive area of the impedance
spectrum are not shown in the Nyquist diagram. But in the
1 X
Nmeasure
r¼ ? rZ;i ð f Þ ð5Þ Bode diagrams, Figure 10–Figure 12, the complete frequency
Nmeasure i¼1 range can be observed.[27] This is particularly true as the
middle frequency range has the decisive influence on the
quality of the equivalent circuit.[28]
For the calculation of the standard deviation, the average de-
The impedance spectrum in Figure 9 was measured at
viation is also used. These two deviations are the quality pa-
a charge state of 10 %, an ambient temperature of 25 8C, and
rameters in this investigation.
a state of health of 98 %. With the simplest equivalent circuit
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u
u 1
NXmeasure
E C2
s¼t ? rZ;i ð f Þ @ rZ ð6Þ
Nmeasure @ 1 i¼1

In Figure 8, the results of the average deviation and stan-

dard deviation are shown by the selected equivalent circuits
from 1 RC to 5 RC. As shown in Figure 8 the blue and red
dashed lines deviate only slightly from each other, and there-
fore the difference between the algorithms is difficult to see.
Only for the equivalent circuits with four and five RC ele-
ments was the Levenberg–Marquardt algorithm 0.1 % better
than the trust-region algorithm. The standard deviation
shows a much greater dependence on the selected algorithm.
With a difference of not more than 0.14 %, the difference is
relatively large compared to the deviation of the average de-
viation. Consequently the Levenberg–Marquardt algorithm
has been selected for further parameterization of the equiva-
lent circuit models. The Levenberg–Marquardt algorithm is
Figure 9. Comparison of measured impedance spectrum (red) at 25 8C,
particularly robust towards the choice of a poor start param- SOC = 10 %, and SOH = 98 %, with all simulated equivalent circuits of this in-
eter. vestigation (blue), RC-EC’s: with marking, CPE EC’s: without marking.

Energy Technol. 2016, 4, 1620 – 1630 T 2016 The Authors. Published by Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim 1625
 21944296, 2016, 12, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/ente.201600154 by Cochrane Mexico, Wiley Online Library on [24/08/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Table 1. Estimated parameter values of the equivalent circuit components of the least-squares method and the Levenberg–Marquardt algorithm to a mea-
sured impedance spectrum at 25 8C, SOC = 10 % and SOH = 98 %

EC-component/ EC R 1 RC 2 RC 3 RC 4 RC 5 RC 1 RCPE 2 RCPE 3 RCPE

R0 [W] 0.846 0.399 0.399 0.399 0.399 0.399 0.399 0.399 0.399
L0 [mH] – 0.148 0.166 0.173 0.174 0.175 0.176 0.176 0.176
R1 [W] – 1.295 0.980 0.272 0.154 0.108 0.087 0.107 0.043
C1 j CPE1[a] [F] – 0.040 0.012 0.002 0.001 0.001 0.002 0.009 0.002/0.959
R2 [W] – – 9.157 0.694 0.256 0.205 1.041 0.098 0.186
C2 j CPE2[a] [F] – – 10.580 0.091 0.018 0.008 0.243/0.509 0.004/0.829 0.017/0.788
R3 [W] – – – 8.077 0.602 0.483 – 0.819 0.789
C3 j CPE3[a] [F] – – – 9.983 0.190 0.099 0.782/0.879 0.212/0.664 0.217/0.680
R4 [W] – – – – 8.910 0.324 – – –
C4 j CPE4[a] [F] – – – – 10.160 2.538 – 6.975/0.850 6.923/0.848
R5 [W] – – – – – 11.640 – – –
C5 j CPE5[a] [F] – – – – – 10.750 – – –

[a] If the field has one value, it is a capacitor. When the field has two values, the first value is the CPE parameter for the imperfect capacitor with the unit
Farad, and the second value is the CPE-parameter for the exponent, which sets the phase angle.[11]

Figure 11. Magnitude (blue) and phase response (red) of the measured impe-
dance spectrum at 25 8C, SOC = 10 %, and SOH = 98 %, compared to the
magnitude and phase responses of the simulated RCPE-equivalent circuit
models (gray).

Figure 10. Magnitude (a) and phase (b) responses of the measured impe-
dance spectrum a = blue; b = red at 25 8C, SOC = 10 %, and SOH = 98 %, and
the simulated RC equivalent circuit models (gray).

Figure 12. Standard deviation (left) and average deviation (right) of the rela-
tive deviation from the simulated to the measured impedance spectra.

Energy Technol. 2016, 4, 1620 – 1630 T 2016 The Authors. Published by Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim 1626
 21944296, 2016, 12, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/ente.201600154 by Cochrane Mexico, Wiley Online Library on [24/08/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
R, only one point on the real axis can be generated. This is due to the calculations of the relative deviation of the total
equivalent circuit can represent any capacitive or inductive impedance according to Equations (3) and (4). The relative
effects and is also not dependent on the frequency. The 1 RC deviation of rIM is constant at 100 % due to the missing imag-
equivalent circuit is essentially based on fZIM,0 and fZIM,max be- inary part of the equivalent circuit, whereas the value of the
cause it does not require fZIM,min, and an element for simulat- real part rRE at the maximum is 160 %. Thus, the statistical
ing the diffusion behavior does not exist in this equivalent spread of the deviation is very low, which is reflected in the
circuit. The situation is different for the 2 RC equivalent cir- standard deviation. The average deviation in Figure 12
cuit, which is based on fZIM,0 and fZIM,min and thus can simu- (right) decreases with an average of D(r = 0.12 % with increas-
late the diffusion behavior. For that case, fZIM,max cannot be ing complexity of the equivalent circuit model. Both quality
considered. The simplest and most suitable RC equivalent parameters reveal a difference between the 5 RC and
circuit is the 3 RC model, which is based on all three process 1 RCPE < 1 %. Moreover, both quality parameters result in
frequencies. With the equivalent circuits 4 RC and 5 RC, an increasing accuracy at low states of charge. This is due to
good reproductions of the impedance spectra were also pro- the maximum expansion of the second semicircle at low
duced. The focus in the models lies on the consideration of states of charge. The superimposed electrochemical processes
fZIM,0 and fZIM,min, which is the same for the 2 RC model. In in the semicircles can be better modeled because the individ-
the magnitude response of the Bode diagram in Figure 10 a, ual physicochemical processes are discernible.
it was observed that the process frequencies mark the area As the third and final quality characteristic of the equiva-
of the inflection point in the measured profile. In the three lent circuit models, the number of iteration steps which are
equivalent circuits 3 RC, 4 RC, and 5 RC the inflection points needed to achieve the global optimum of the parameter esti-
are in the immediate proximity of the process frequencies. mation has been counted. The results are shown in Figure 13.
In Figure 10 b, the measured phase response of the Bode The exact duration and therefore the parameterization speed
diagram is plotted in red. It is noticeable that no RC equiva- depends on the respective computing performance of the
lent circuit simulates the phase at 08 very accurately. Further- simulation computer. As described the least-squares method
more, the 5 RC equivalent circuit exhibits an unusual drop in with Levenberg–Marquardt algorithm was used for the pa-
phase at the process frequency fZIM,min. Because the parame- rameter estimation.
ters are derived from a numerical estimate and not from
a physical computation with material constants, this phenom-
enon has to be caused by the estimation method. As shown
in Figure 10 b, the profile of the phase response cannot be si-
mulated with any RC equivalent circuit very well. Based on
the Nyquist diagram in Figure 9, the capacitor is essential for
the RC models and not the resistor. A capacitor is not appro-
priate for modeling the compression of the semicircle, so
that equivalent circuits with CPE elements show an advant-
age by modeling this behavior, which is shown in Figure 11.
The magnitude responses show only very small differences
between the simulated and measured values. Deviations
occur only in the very high-frequency range > 20 kHz. This is
the inductive area of the impedance spectrum, and due to
the influence of the measurement setup they are not further
investigated. The phase responses of the CPE models show
the same deviation as the RC models at a phase of 0 8C. Fur-
thermore, the phase response from 1 RCPE in the middle
frequency range is similar to the 5 RC. This is also the reason Figure 13. Number of iterations until the global optimum of the parameter
why a constant phase element is often converted into a series estimation.
circuit of a multitude of RC elements.[29] The difference be-
tween 2 RCPE and 3 RCPE is very small over the entire fre-
quency range. As a function of state of charge, the simulated
Conclusions
impedance spectra are analyzed in terms of the standard de-
viation and the average deviation to indicate a quality char- The standard deviation, the average deviation, and the
acteristic in Figure 12. These deviations are calculated from number of iterations were used as quality criteria for assess-
the relative deviation of the simulated impedance of an ing the quality of an equivalent circuit simulation for battery
equivalent circuit to the measured impedance versus the fre- modeling. If all criteria are considered together, the follow-
quency [see Eq. (2)–(5)]. ing statements about the quality of the equivalent circuits by
The standard deviation in Figure 12 (left) is reduced by an the examining the impedance spectra in Figure 9–11 are de-
average of Ds = 0.14 % with increasing complexity of the scribed in this work. The use of an equivalent circuit of R is
equivalent circuit. The very low standard deviation for R-EC completely unsuitable for simulating an impedance spectrum.

Energy Technol. 2016, 4, 1620 – 1630 T 2016 The Authors. Published by Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim 1627
 21944296, 2016, 12, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/ente.201600154 by Cochrane Mexico, Wiley Online Library on [24/08/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
However, with the element1 RC the relevant factors of the
charge-transfer resistance can be simulated, such as tempera-
ture, state of charge and state of health. Along with 2 RC,
which can also represent the range of the diffusion processes
because of the additional RC element, these two equivalent
circuit models are very good for larger simulation models be-
cause it requires only a small effort for parameterization and
selection of the equivalent circuit parameters. This could be,
for example, a battery model with a high number of cells,
which is integrated into a distribution grid simulation (or
smart grid) or in a battery management system (BMS) for
photovoltaic storage. The accurate simulation of battery
characteristics is not necessary for either simulation models
in dynamic supply or load cases. Perhaps, the application of
the R model would be suitable in a simulation with a very Figure 14. Recommended assignment of equivalent circuit models for simula-
tions of different application scenarios of Li-ion batteries.
large number of cells as is the case for MWh-scale storage
that is integrated in the high-voltage grid because the assess-
ment of the power loss and the sudden change in voltage are
essentially important. Considering the process frequencies, tion), the model can be simplified to establish any network
the 3 RC model is the simplest RC equivalent circuit model need. Should the focus be more on the derivation of physico-
that simulates an impedance spectrum accurately and re- chemical processes to identify safety-critical features in real
quires only a low number of iteration steps until the determi- time, which adjust themselves due to battery aging, at least
nation of the optimal parameters. When using equivalent cir- 3rd-order models are necessary for the home storage and
cuits with a higher number of RC elements, the deviations electric-mobility models with time constants. With these
are minimized, but the number of required iterations is equivalent circuit models, it is possible to obtain information
greatly increased. Due to the higher accuracy and the consid- on the SEI growth, increase in the charge transfer resistance,
eration of different time constants, these models can be used and diffusion behavior. In the cell production field, a variety
for simulations of mobility applications. Thus, the rates of of R k CPE elements should be used additionally to deter-
current change for a traction battery of a hybrid electric ve- mine the contact resistance between the arrester and active
hicle (HEV) or a battery electric vehicle (BEV) are highly material as well as the impact of calendering.
dynamic and rarely continuous. When comparing the 5 RC Due to the long duration of parameterization when using
and 1 RCPE circuits, the discrepancies between the two CPE elements, the simulation speed will be greatly reduced
models are hardly visible. But they differ significantly in the with CPE models for a battery cell simulation. It is still a cur-
number of necessary iterations, which is why the 1 RCPE rent topic of investigation as to what effect this would have
should be chosen. Small differences are observed in the Ny- on a simulation model of a battery cell. The created simula-
quist plot (Figure 9) and Bode diagram (Figure 11) between tion model is used here to predict the battery status in vary-
the most complex models 2 RCPE and 3 RCPE. Only the ing load scenarios and does not reflect the value of impe-
standard deviation and average deviation parameters show dance, but rather the battery voltage and many other sizes as
the trend that the 2 RCPE has a lower deviation average of an output value. As this model is an addition to the equiva-
2.4 % and will only use one third of the number iterations for lent circuit block, other blocks are required such as a temper-
optimal parameter estimation. These two models should not ature, open circuit, and lifetime models to achieve a complete
be used as part of a higher-level simulation, but as a diagnos- battery model. The other mentioned blocks are only meant
tic tool for deriving the physical factors that influence the as an outlook and could be incorporated into future work.
state of function of a battery cell. By convention, a higher
number of equivalent circuit elements causes no improved
significance. Using too many elements does not allow for the Experimental Section
assignment of additional electrochemical processes. Figure 14 For setting up a battery cell model, measurements on real cells
summarizes the evaluation to derive a recommendation for were necessary to incorporate the characteristics of the model.
the use of equivalent circuit models in the simulations of dif- The measured pouch battery cells in Figure 16 were self-made in
ferent applications. the Battery LabFactory Braunschweig (BLB). They consist of
A battery system that is made up of many individual cells LiNi1/3Mn1/3Co1/3O2 cathodes with an electrode formulation of
can be modeled by an equivalent circuit network with x ser- 4 %:2 %:4 %:90 % (binder/additive/carbon black/active material)
coated onto an aluminum foil and a graphite anode with an elec-
ies equivalent circuits. For MWh-scale energy storage, which
trode formulation of 4 %:1 %:2 %:93 % coated on a copper foil.
consists of thousands of individual cells, it is recommended The electrodes had a size of 50 X 50 mm2, and the cell had a nomi-
to use the simplest possible equivalent circuit to keep the nal capacity of approximately 50 mAh. The anode was contacted
simulation effort low. By multiplication or division of the with a welded nickel conductor and the cathode with an alumi-
equivalent circuit parameters (depending on the interconnec- num conductor. Between the two electrodes a ceramic-coated

Energy Technol. 2016, 4, 1620 – 1630 T 2016 The Authors. Published by Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim 1628
 21944296, 2016, 12, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/ente.201600154 by Cochrane Mexico, Wiley Online Library on [24/08/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
separator was inserted for electrical insulation of the electrodes
from each other. The entire electrode assembly was packed in
a laminated aluminum bag with the addition of an electrolyte
based upon LiPF6 and then vacuum sealed.

To record the impedance spectrum, the battery was connected

using a four-wire measuring system to a galvanostat, and then
the measuring program was started. The cell was first charged to
an initial charge level of 100 % and then discharged gradually.
Each short discharge was followed by a one-hour break so that
the voltage could relax, and the cell was set in electrochemical
equilibrium before the impedance spectrum was measured. The
galvanostat generated a sinusoidal alternating current of a small
Figure 16. Schematic of the test setup for the electrochemical impedance
amplitude (1/20 C) and measured the sinusoidal voltage re-
spectroscopy measurements.
sponse. From the difference between the magnitude and phase,
the frequency-dependent impedance could be determined by
using the complex alternating current calculation. The impe-
List of Symbols
dance spectrum was determined for using 10 frequency measure-
ments per decade with a stepwise variation of the frequency AC alternating current
starting at the highest frequency of 500 kHz to the lowest fre- BEV battery electric vehicle
quency of 5 mHz. During the measurement, the position of the BLB Battery LabFactory Braunschweig
cell was not changed and no mechanical stress was exerted onto
BMS battery management system
the cell. The cells are very sensitive to mechanical stress, which
DRT distribution of relaxation time
has an effect on the result of the impedance spectroscopy mea-
surement.[30, 31] For this reason an apparatus was constructed that EC equivalent circuit
fixed the cells and enabled a reproducible measurement. During HEV hybrid electric vehicle
the measurement the battery was placed in a climate-controlled L–M Levenberg–Marquardt algorithm
chamber to adjust constant ambient conditions (Figure 15). The LiMxOx lithium–metal oxide
list of equipment used in the measurement setup is presented in LiNiMnCoO2 lithium–nickel–manganese–cobalt dioxide
Table 2 and shown schematically in Figure 16. LiPF6 lithium hexafluorophosphate
RC parallel circuit resistor and capacitor
SEI solid–electrolyte interface
SOC state of charge
SOH state of health
T-R trust-region algorithm
Cx [F] capacitor of the equivalent circuit
CPEx [F] constant-phase element
f [Hz] frequency
fmax [Hz] maximum frequency
fmin [Hz] minimum frequency
fZIM,0 [Hz] frequency at ZIM = 0
fZIM,max [Hz] frequency at ZIM = ZIM,max
fZIM,min [Hz] frequency at ZIM = ZIM,min
Figure 15. Left: measurement setup consisting of a climate chamber, the Gal-
IB [A] battery current
vanostat, and a power booster to improve performance. Right: self-built labo-
ratory Pouch cell with a nominal capacity of approximately 50 mAh. L0 [H] inductivity in the impedance spectrum
Nmeasure [1] number of measured values
w [1/s] angular frequency
fZ [8] phase response of the impedance spectrum
Table 2. List of the important devices at measurement setup.
rIM [%] relative deviation of the imaginary part of im-
equipment label pedance
galvanostat VersaSTAT 3 M1
rRE [%] relative deviation of the real part of impe-
shielded sensor cable to the front-end L1 dance
current cable to the back-end L2 rZ [%] relative deviation of the impedance
PC with the VersaStudio software P1 (rz [%] average deviation of the impedance
climate chamber VT 4030 T1
R0 [W] purely ohmic resistance in the impedance
spectrum
Rx [W] resistance of the equivalent circuit
s [%] standard deviation of the impedance
t [s] time

Energy Technol. 2016, 4, 1620 – 1630 T 2016 The Authors. Published by Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim 1629
 21944296, 2016, 12, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/ente.201600154 by Cochrane Mexico, Wiley Online Library on [24/08/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
[12] “Combination of battery model and test method to determine the
T [8C] ambient temperature
battery state of function”: U. Westerhoff, M. Kurrat in 13. Symposi-
UB [V] battery voltage um: Hybrid- und Elektrofahrzeuge (Braunschweig, Germany), 2016.
UOC [V] open-circuit voltage [13] Z. Chen, C. C. Mi, Y. Fu, J. Xu, X. Gong, J. Power Sources 2013, 240,
Wx [W/s@1/2] Warburg element of the equivalent circuit 184 – 192.
[14] “Li-ion batteries and Li-ion ultracapacitors: Characteristics, Model-
jZj [W] magnitude of the complex impedance Z
ing and Grid Applications”: S. A. Hamidi, E. Manla, A. Nasiri in
jZjEC [W] complex impedance of the equivalent circuit Energy Conversion Congress and Exposition (ECCE), 2015 IEEE
ZIM [W] imaginary part of the impedance (Montreal, Canada), 2015.
jZjmeasure [W] measured impedance [15] “Evaluation of the entire battery life cycle with respect to lithium
ion batteries”: U. Westerhoff, K. Kurbach, D. Unger, H. Loges, D.
ZRE [W] real part of the impedance
Hauck, F. Lienesch, M. Kurrat, B. Engel in International ETG Con-
gress 2015 (Bonn, Germany), 2015.
[16] S. E. Li, B. Wang, H. Peng, X. Hu, J. Power Sources 2014, 258, 9 – 18.
Acknowledgements [17] C. Fleischer, W. Waag, H.-M. Heyn, D. U. Sauer, J. Power Sources
2014, 262, 457 – 482.
[18] “Online state and parameter estimation of the Li-ion battery in
The authors thank the Nds. Ministerium fgr Wissenschaft und
a Bayesian framework”: M. F. Samadi, S. M. Mahdi Alavi, M. Saif in
Kultur of the State of Lower Saxony for the financial support American Control Conference (ACC), 2013 IEEE (Washington D.C.,
of this work with Graduiertenkolleg Energiespeicher und USA), 2013.
Elektromobilit-t Niedersachsen (GEENI). We thank col- [19] V. Ramadesigan, P. W. C. Northrop, S. De, S. Santhanagopalan, R. D.
Braatz, V. R. Subramanian, J. Electrochem. Soc. 2012, 159, R31 – R45.
leagues of the Battery LabFactory Braunschweig (BLB) for [20] D. Andre, M. Meiler, K. Steiner, C. Wimmer, T. Soczka-Guth, D. U.
the excellent cooperation and the use of equipment infrastruc- Sauer, J. Power Sources 2011, 196, 5334 – 5341.
ture for cell production. [21] M. Galeotti, L. Cin/, C. Giammanco, S. Cordiner, A. Di Carlo,
Energy 2015, 89, 678 – 686.
[22] U. Westerhoff, T. Kroker, K. Kurbach, M. Kurrat, submitted, 2015.
[23] “Real-time state of charge and electrical impedance estimation for
Keywords: electrochemical impedance spectroscopy · energy lithium-ion batteries based on a hybrid battery model”: T. Kim, W.
Qiao, L. Qu in Applied Power Electronics Conference and Exposition
storage · equivalent circuits · lithium-ion batteries · modeling
(APEC), 2013 Twenty-Eighth Annual IEEE (Long Beach, CA) 2013.
[24] A. Farmann, W. Waag, D. U. Sauer, J. Power Sources 2015, 299, 176 –
188.
[1] A. Seaman, T.-S. Dao, J. McPhee, J. Power Sources 2014, 256, 410 – [25] S. Rodrigues, N. Munichandraiah, A. Shukla, J. Solid State Electro-
423. chem. 1999, 3, 397 – 405.
[2] E. Samadani, S. Farhad, W. Scott, M. Mastali, L. E. Gimenez, M. [26] D. Zhang, B. S. Haran, A. Durairajan, R. E. White, Y. Podrazhansky,
Fowler, R. A. Fraser, Electrochim. Acta 2015, 160, 169 – 177. B. N. Popov, J. Power Sources 2000, 91, 122 – 129.
[3] M. Guo, G.-H. Kim, R. E. White, J. Power Sources 2013, 240, 80 – 94. [27] F.-M. Wang, J. Rick, Solid State Ionics 2014, 268, 31 – 34.
[4] P. M. Gomadam, J. W. Weidner, R. A. Dougal, R. E. White, J. Power [28] J. Zhu, Z. Sun, X. Wei, H. Dai, Appl. Electrochem. 2016, 46, 157 –
Sources 2002, 110, 267 – 284. 167.
[5] X. Hu, A. Li, H. Peng, J. Power Sources 2012, 198, 359 – 367. [29] W. Waag, S. K-bitz, D. U. Sauer, Measurement 2013, 46, 4085 – 4093.
[6] D. Andre, M. Meiler, K. Steiner, H. Walz, T. Soczka-Guth, D. U. [30] M. J. Brand, S. F. Schuster, T. Bach, E. Fleder, M. Stelz, S. Gl-ser, J.
Sauer, J. Power Sources 2011, 196, 5349 – 5356. Mgller, G. Sextl, A. Jossen, J. Power Sources 2015, 288, 62 – 69.
[7] P. Gao, C. Zhang, G. Wen, J. Power Sources 2015, 294, 67 – 74. [31] “Coupled Mechanical and Electrochemical Characterization Method
[8] “A virtual Li/S battery: Modeling, simulation and computer-aided de- for Battery Materials”: J. Schmitt, F. Treuer, F. Dietrich, K. Drçder,
velopment”: D. N. Fronczek, W. G. Bessler in Next Generation Batter- T.-P. Heins, U. Schrçder, U. Westerhoff, M. Kurrat, A. Raatz in Con-
ies 2012 (Boston, USA), 2012 http://elib.dlr.de/76757/. ference on Energy Conversion (CENCON), 2014 IEEE (Johor
[9] J. P. Schmidt, P. Berg, M. Schçnleber, A. Weber, E. Ivers-Tiff8e, J. Bahru, Malaysia), 2014.
Power Sources 2013, 221, 70 – 77. [32] J. Song, M. Z. Bazant, J. Electrochem. Soc. 2013, 160, A15 – A24.
[10] Y. Zhang, Y. Chen, M. Li, M. Yan, M. Ni, C. Xia, J. Power Sources
2016, 308, 1 – 6. Received: March 18, 2016
[11] J. Huang, Z. Li, B. Y. Liaw, J. Zhang, J. Power Sources 2016, 309, 82 – Revised: June 19, 2016
98. Published online on August 18, 2016

Energy Technol. 2016, 4, 1620 – 1630 T 2016 The Authors. Published by Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim 1630