Professional Documents
Culture Documents
Contents
1 Introduction 2
7 Demonstration 15
7.1 Demonstration Using Simulated Data . . . . . . . . . . . . . . . 15
7.2 Demonstration Using Real Data . . . . . . . . . . . . . . . . . . . 16
8 Summary 20
1
1 Introduction
Electrochemical impedance spectroscopy (EIS) is a widely used approach for
battery analysis. In EIS, the frequency response of the battery, known as the
Nyquist plot, is studied for insights about battery health. Numerous works in
the literature have employed Nyquist plots to understand and visualize battery
aging. This chapter provides insights about the relationship between EIS and
the ECM parameters of the battery; it is shown how to estimate the ECM
parameters once the Nyquist plot is obtained. The estimated ECM parameters
can then be related to SOH using the ideas provided in Chapter ??.
This chapter also illustrates how the parameter estimation becomes challeng-
ing with measurement noise in the sensors. Measurement noise is a significant
problem in practical systems that are preferred in general to be low-cost; it is
also possible that the measurement noise increases as the system ages. With
measurement noise, the uncertainty about the (estimated) ECM parameters
increases. It is important to have an understanding about the performance
of ECM parameter estimation at various noise levels. This chapter introduces
performance analysis of ECM parameter estimation, based on the Nyquist spec-
trum, at various signal to noise ratio (SNR) levels.
Despite numerous literature that employed EIS to analyze battery charac-
teristics, there is only a sparse work in literature that enlightens the signal
processing aspects and challenges involved in battery analysis. The present
chapter tries to paint a signal processing view of EIS analysis for battery engi-
neers by highlighting the ECM parameter estimation aspects and by presenting
performance analysis at various SNR levels.
2
where the Warburg impedance defined as
σ
Zw (jω) = (1 − j) √ (1)
ω
Z(ω) ,Z(jω)
1 1
=jωL + RΩ + 1 + 1
RSEI + jωCSEI RCT +Zw (jω) + jωCDL
RSEI RCT + Zw (jω)
=jωL + RΩ + + (2)
1 + jωRSEI CSEI 1 + jω (RCT + Zw (jω)) CDL
where the parameters are as indicated in Figure 1(a). Likewise, the qualitative
impedance plot shown in Figure 1(b) has four branches associated with four
specific electrochemical processes. In the first branch, denoted in this chapter
as the ‘RL Arc’, the effect of the inductive behavior at high frequencies (ω > ωk3 )
as well as the ohmic resistance (RΩ ) can be seen. The second branch (ωk2 < ω <
ωk3 ), consists of a semi-circle, denoted in this chapter as the ‘SEI Arc’, related
to the solid electrolyte interface. The effect of the double-layer capacitance
and charge transfer resistance at the electrodes represents the second semi-
circle (denoted as ‘CT Arc’) in the third branch (ωk1 < ω < ωk2 ). Finally,
the constant slope (denoted as ‘Diffusion Arc’) in the impedance plot in the last
branch, represents the diffusion processes in the active material of the electrodes;
it has a significant effect at very low frequencies (ω0 < ω < ωk1 ) only.
3
(a) Adaptive Randles equivalent circuit model (AR-ECM)
4
that zv (k) and zc (k) are the measured voltage and current, respectively, from
the battery over a certain time window L, i.e.,
where k indicates time, v(k) is the true voltage, and i(k) is the true current.
The voltage and current measurement noise nv (k) and nc (k) are assumed to be
zero-mean i.i.d. with standard deviation σv and σc , respectively. The Fourier
transform of the voltage and current measurements in (3) are defined as
L
X −i2πkω
Zc (ω) = FFT(zc (k)) = zc (k)e L = I(ω) + Nc (ω) (4)
k=1
L
X −i2πkω
Zv (ω) = FFT(zv (k)) = zv (k)e L = V (ω) + Nv (ω) (5)
k=1
where Zv (ω) and Zc (ω) are voltage and current measurement in the frequency
domain. Here, V (ω) and I(ω) indicate the Fourier transforms of the noiseless
voltage and current, respectively. The Fourier transforms of the voltage and
current measurement noises are given by Nv (ω) and Nc (ω), respectively. The
impedance at frequency ω is now written as
Zv (ω) V (ω) + Nv (ω) 1
Z(ω) = = = (V (ω) + Nv (ω))
Zc (ω) I(ω) + Nc (ω) I(ω) + Nc (ω)
1 Nc (ω)
= (V (ω) + Nv (ω)) − (6)
I(ω) I(ω)2
V (ω)
Z(ω) = + Nz (ω) (8)
I(ω)
where it can be shown that the noise Nz (ω) is zero-mean. Let us denote the
real and imaginary parts of the frequency response at ωk as
We will make use of this notation to describe the ECM parameter estimation
approaches in the subsequent sections.
5
4 ECM Parameter Estimation Problem
The ECM parameter estimation in the frequency domain can be formally stated
as follows: given the frequency response of the system Z(ω) at the frequencies
ω1 , ω2 , . . . , ωL , estimate the ECM parameters.
Consequently, the real part of Z(ω), when ω < ωk1 , can be written as
From (16), σ can be calculated as the slope of zr (k) versus √1ω such that k < k1 .
k
To compute σ, two low frequencies are chosen and the corresponding resistance
value is taken from the impedance plot. Let us select these two frequencies as
follows
ωa = ω0 (17)
ωb = ω s.t. ω0 < ω < ωk1 (18)
6
Then, the Warburg coefficient can be written as follows:
√
( ωa ωb )(zr (a) − zr (b))
σ̂ = √ √ (19)
ωb − ωa
Now, let us consider the two arcs (SEI Arc and CT Arc) in the Nyquist
plot to determine the value of RSEI , CSEI , RCT and CDL . First, consider the
CT Arc which occurs in lower frequencies, i.e., ωk1 < ω < ωk2 . The (Faradaic)
impedance due to RCT and CDL , in this region is
1
ZF (ω) = 1
RCT +Zw (jω) + jωCDL
1
RCT +Zw (jω) − jωCDL
= 1 ωk1 < ω < ωk2 (20)
( RCT +Z w (jω)
)2 + ω 2 C 2 DL
7
Based on the above observation, we have
1
|Im(ZSEI (ω))| = RSEI at ω = ωSEI,peak (28)
2
and the following two estimates can be obtained:
where the unit of SNR is decibels (dB). Figure 3 shows the Nyquist plots at four
different SNR values (0, 5, 15, 30 dB). The algorithm summarized in Section
5 is applied to estimate the AR-ECM parameters in each case. The estimated
parameters are then used to generate the Nyquist plot. Ideally, both plots
should coincide. It can be noticed that with increasing noise, the discrepancies
become prominent.
8
0.16
With Warburg Impedance
0.14 Without Warburg Impedance
0.12
0.1
0.08
0.06
0.04
0.02
-0.02
0.55 0.6 0.65 0.7 0.75 0.8 0.85
• Warburg:
[zr (0), zi (0)], [zr (1), zi (1)], . . . , [zr (k1 ), zi (k1 )] (32)
• CT:
[zr (k1 + 1), zi (k1 + 1)], [zr (k1 + 2), zi (k1 + 2)], . . . , [zr (k2 ), zi (k2 )] (33)
• SEI:
[zr (k2 + 1), zi (k2 + 1)], [zr (k2 + 2), zi (k2 + 2)], . . . , [zr (k3 ), zi (k3 )] (34)
9
0.15 0.15
Approximate Approximate
Measurements 0.1 Measurements
0.1
0.05
0.05
0
0
-0.05
0.5 0.6 0.7 0.8 0.5 0.6 0.7 0.8 0.9
0.15 0.15
Approximate Approximate
Measurements Measurements
0.1 0.1
0.05 0.05
0 0
0.5 0.6 0.7 0.8 0.9 0.5 0.6 0.7 0.8
• RL:
[zr (k3 + 1), zi (k3 + 1)], [zr (k3 + 2), zi (k3 + 2)], . . . , [zr (k4 ), zi (k4 )] (35)
We will use the above notations to discuss an improved version of the approxi-
mate parameter estimation approach discussed in Section 5.
10
where n < k1 /2. The observations in (36) were selected in such a way that the
quantity zr (i) − zr (j) could be as high as possible — this strategy is designed
to reduce the effect of noise in the observations. The observations (36) can be
written in matrix form as
z̃ = bσ (37)
where
√1 − √
1
zr (0) − zr (k1 ) ω0 ωk1
zr (1) − zr (k1 − 1) √1 1
− √
ωk1 ωk1 −1
z̃ = , b = (38)
..
..
.
.
zr (n) − zr (k1 − n)
1 1
ωn − ωk −n
√ √
1
11
where â and b̂ are estimates of a and b, respectively.
In order to estimate a and b, the pairs of impedance measurements shown
in (33) can be substituted in (41) to get the following sets of equations
−(zr (k1 + 1)2 + zi (k1 + 1)2 ) = azr (k1 + 1) + b
−(zr (k1 + 2)2 + zi (k1 + 2)2 ) = azr (k1 + 2) + b
.. (45)
.
−(zr (k2 )2 + zi (k2 )2 ) = azr (k2 ) + b
the above can be written in matrix form as
z = BxCT (46)
where
−(zr (k1 + 1)2 + zi (k1 + 1)2 )
zr (k1 + 1) 1
−(zr (k1 + 2)2 + zi (k1 + 2)2 ) zr (k1 + 2) 1
a
z= , B = , xCT = (47)
.. .. b
. .
−(zr (k2 )2 + zi (k2 )2 ) zr (k2 ) 1
The least-square estimate of xCT is
−1 T
x̂CT = BT B B z (48)
and the estimates of a and b are
â = x̂CT (1), b̂ = x̂CT (2) (49)
which will be substituted in (44) to estimate RCT . Then, for estimating CDL ,
the inverse of (20) is calculated as:
1 1 1
= + jωCDL = + jωCDL
ZCT (ω) RCT + Zw (jω) RCT + σω − j √σω
√
Hence,
√σ
1 ω
Im = 2 + ωCDL (51)
ZCT (ω) √σ σ2
RCT + ω
+ ω
k = k1 + 1, k1 + 2, . . . , k2 (52)
12
Finally, all estimates of C̄DL (k) are averaged to obtain
k2
ˆ 1 X
ĈDL = C̄DL (k) (53)
k2 − k1
k=k1 +1
where it was assumed that the center of the circle lies on the real axis (see Figure
1). The center of the circle (41) can be denoted as (xSEI , 0) where
c
xSEI = − (56)
2
and the radius of the circle (41) is
r
c2
rSEI = −d (57)
4
By the same reasoning in Subsection 6.2, the argument of the square root can
be shown to be always positive. It now is easy to see that the estimate of the
resistance RCT is
r
ˆ ĉ2
R̂SEI = 2 − dˆ (58)
4
y = AxSEI (60)
13
where
−(yr (k2 + 1)2 + yi (k2 + 1)2 )
yr (k2 + 1) 1
−(yr (k2 + 2)2 + yi (k2 + 2)2 ) yr (k2 + 2) 1
c
y= , A = , xSEI =
.. .. d
. .
−(yr (k3 )2 + yi (k3 )2 ) yr (k3 ) 1
(61)
which will be substituted in (58) to estimate RSEI . Then, CSEI is simply driven
from the inverse of (26):
1 1
= + jωCSEI (64)
ZSEI (ω) RSEI
k4
ˆ 1 X
R̂Ω = zr (k) (67)
k4 − k3
k=k3 +1
k4
ˆ 1 X zi (k)
L̂ = (68)
(k4 − k3 ) ωk
k=k3 +1
14
6.5 Feature Point Extraction
It should be stressed that the least squares based (approximate) parameter esti-
mation algorithm presented in this section needs to know the critical frequency
values ωk0 , . . . , ωk4 . It is easy to know the values of ωk0 and ωk4 as these are
the lowest and highest frequencies, respectively, in the Nyquist spectrum. An
approach to estimate ωk1 , ωk2 , and ωk3 is presented in [1]. In this approach,
a straight line is fitted to the Diffusion Arc with progressively increasing data
starting from the lower frequency. As the Diffusion Arc turns into CT Arc,
the correlation of fitting starts to drop. The critical point k1 is detected by
observing the drop in k1 relative to a predefined threshold. Similarly, a circular
curve is fitted to the data starting from the (detected) critical point k1 and the
correlation coefficient is monitored to detect the critical point k2 based on a
predefined threshold. The critical point k3 can be detected based on the fact
that the imaginary part of the Nyquist curve changes its sign at k3 .
7 Demonstration
In this section, computational demonstration of the proposed ECM parame-
ter identification techniques are presented using simulated and real-world data.
[selected codes will be added later.]
where Idc is the DC current, Im is the peak of perturbation current and each tk
is selected such that tk − tk−1 = f1k i.e., each frequency is made to have one full
cycle of data.
For measuring the noise effects it is assumed that the measured current and
voltage are corrupted with Gaussian noise of the same noise variances, i.e.,
σv = σi = Im 10(− )
SNR
20 (70)
15
where SNR varies from 0 to 50 dB (0, 10, 20, 30, 40, 50 dB).
The Nyquist plot is derived for each case (each level of noise) and algorithms
explained in sections 4, 5, and 6 are applied to estimate ECM parameters.
The performance of each algorithm is quantified in terms of the normalized
percentage mean square error, simply referred hereafter as Error%. For example,
the Error (%) of estimating the Ohmic resistance by the proposed approach is
defined as
ˆ
RΩ − R̂Ω
Error(%) = 100 (71)
RΩ
Each reported Error measure is averaged over 100 Monte-Carlo runs.
Figures 4-7 presents the percentage error for each of the approaches discussed
in this chapter. Figure 4 compares the performance in estimating the Warburg
coefficient (σ). An explanation of the performance loss by the approximate
approach (Section 5) could be that it used only two points to find the slope of
a line to estimate σ. On the other hand, the improved approach (Section 6)
used many pairs of points and resulted in better estimation error. The reason
for the failure of the non-linear LS approach could be attributable to the severe
non-linearity in the model when it comes to estimating the Warburg coefficient.
Figures 5(a) and 5(b) show the performance comparison of the three ap-
proaches presented in this chapter for RCT and CDL estimation. And figures
6(a) and 6(b) show similar comparison for RSEI and CSEI estimation. These two
figures exhibit the performance trade-off of the different approaches presented
in this chapter and highlight the need to develop robust approaches to estimate
ECM parameters.
Figure 7(a) shows the estimation errors corresponding to the Ohmic resis-
tance, RΩ . All the algorithms have an excellent performance (error rate lower
than 1%) in estimating RΩ . This is due to the linear relationship of RΩ to the
measurements in (2). Figure 7(b) shows the comparison of different estimators
in the estimation of stray inductance L.
For a comprehensive analysis of the performance of each three approaches
presented in this chapter, the extracted parameters from each method are used
to generate the Nyquist plot. Figure 8(a) compares the performance of the
previous approach and the proposed algorithm in estimating the Nyquist plot
at SNR = 30 dB. Figure 8(b) shows the performance comparison of all three
approaches in a severe but practical case (low SNR). In the presence of a high
level of noise, the approximate LS approach is observed to outperform both the
non-linear LS approach and the approximate approach, as shown in Figure 8(b).
16
70
Approximate
Approximate LS
60 Non-linear LS
50
40
30
20
10
0
0 5 10 15 20 25 30 35 40 45 50
40 50
Approximate Approximate
Approximate LS 45 Approximate LS
35
Non-linear LS Non-linear LS
40
30
35
25
30
20 25
20
15
15
10
10
5
5
0 0
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
EIS device were used. Fig. 9 shows these devices in an experimental setup —
the computer screen shows the interface of the Arbin software that allows to
collect Nyquist data from the Gamry device. A relatively new cylindrical Li-ion
17
400 2000
Approximate Approximate
Approximate LS 1800 Approximate LS
350
Non-linear LS Non-linear LS
1600
300
1400
250
1200
200 1000
150 800
600
100
400
50
200
0 0
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
(a) Solid electrolyte interface resistance (b) Solid electrolyte interface capacitance
1 25
Approximate Approximate
0.9 Approximate LS Approximate LS
Non-linear LS Non-linear LS
0.8 20
0.7
0.6 15
0.5
0.4 10
0.3
0.2 5
0.1
0 0
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
18
0.16 0.18
0.14 0.16
0.12 0.14
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
Approximate 0.02 Approximate
Approximate LS Approximate LS
0 Non-Linear LS 0 Non-Linear LS
Measurements Measurements
-0.02 -0.02
0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95
19
ECM parameters of the battery.
0.02
0.01
-0.03
-0.04
-0.05
-0.06
-0.07
0.405 0.41 0.415 0.42 0.425 0.43 0.435
Figure 10: Nyquist plot for real battery. The nyquist plot in blue depicts the
impedance response of the battery with 0.2 A DC current. Impedance response
of the battery with 0 DC Current is shown in green.
8 Summary
This chapter introduced frequency domain approaches to battery equivalent cir-
cuit model parameter estimation using EIS. In EIS, an excitation signal (either
20
Table 2: Estimated parametes (Nonlinear LS)
voltage or current) is applied to the battery and its response (current or voltage)
is measured. This procedure is repeated and the amplitude and phase of the
frequency response is computed at various (fixed) frequencies spanning very low
frequency in fractions of Hz and very high frequency in several MHz. Based
on the obtained responses at wide ranging frequencies the ECM parameters
can be estimated. This chapter outlines three different approaches to estimate
the ECM parameters based on the frequency response. The first approach is
based on the non-linear least squares estimation which requires significant com-
putational resources. Also, the non-linear least squares approach is shown to
be susceptible local convergence even at high SNR regions. It is also shown
that distinct portions of the Nyquist plot can be extracted to estimate ECM
parameters.
The EIS approach to battery analysis is a time consuming process. Es-
pecially, measurements at low frequencies incurs significant delay. Significant
recent research work is focused on reducing the experimental time of the EIS
approach for battery analysis.
References
[1] Marzieh Abaspour, “Performance Analysis and Improvement of Electro-
chemical Impedance Spectroscopy for Online Estimation of Battery Param-
eters,” Dissertation, University of Windsor (Canada), 2021.
21
[2] Mark E. Orazem and Bernard Tribollet, “Electrochemical Impedance Spec-
troscopy,” John Wiley & Sons, Inc., New Jersey, 2008.
22