Professional Documents
Culture Documents
H I G H L I G H T S
A R T I C L E I N F O A B S T R A C T
Keywords: Recently, reduced order electrochemical models (ROMs) have been obtained more attention by industries for
Lithium-ion battery real-time applications. Its computational efficiency is maximized by regarding particles in electrodes as an
Reduced order electrochemical model equivalent single particle, named ROM-Single Particle Model (SPM). However, the model has a major drawback
Single particle model
in the inaccuracy of the predicted terminal voltage, particularly at high currents. Therefore, an advanced ROM-
Variable diffusivity
SPM considering variable diffusivity (VD) of lithium ions in solid is proposed that significantly improves the
accuracy of the terminal voltage, named as ROM-SPMVD. The model is firstly developed, and the sensitivity
analysis of the parameters is performed, which reveals that the diffusivity of lithium ions in solid of cathode (Ds,p)
is strongly sensitive and identifiable. Next, the relationship between the Ds,p and state of charge (SOC) and
temperature is determined by a newly developed nondestructive method. Finally, the ROM-SPMVD is validated
against experimental data of a pouch type lithium ion energy cell at different current profiles and at different
temperatures. Results show that consideration of variable Ds,p leads to a significant improvement in the accuracy
of the ROM-SPM at high currents while maintaining the reduced computational time. The voltage error of ROM-
SPMVD is reduced up to 79% compared with that of ROM-SPM.
* Corresponding author. 1418 Wiggins Hall, Auburn University, AL, 36849, USA.
E-mail address: choeson@auburn.edu (S.-Y. Choe).
https://doi.org/10.1016/j.jpowsour.2020.228322
Received 9 January 2020; Received in revised form 29 April 2020; Accepted 7 May 2020
Available online 12 June 2020
0378-7753/© 2020 Elsevier B.V. All rights reserved.
Y. Hu et al. Journal of Power Sources 468 (2020) 228322
equations (PDEs) and nonlinear equations, which presents a full order instrumentations and testing conditions. Therefore, this technique is not
model (FOM). Solving the equations of the FOM is very computational applicable for measuring commercial cells made of several dozens of full
intensive and thus its implementation in BMS is impossible with current cells. In addition, the diffusivity measured from half cells might not be
hardware solution, particularly in real-time. There are several ap identical with the one needed for a model. Thus, a fast and efficient
proaches to reduce the computational time by employing order reduc method that allows for determination of the diffusivity as a function of
tion techniques and simplifying the models, while maintaining accuracy SOC without fabricating a half cell, which can be directly used in the
in acceptable levels. The order reduction is performed by numerous model, is also desired.
order reduction and simplification methods to obtain a reduced order In this work, the ROM-SPMVD is firstly constructed on the basis of
model (ROM) by employing polynomial approximation [9], Pad�e ROM-SPM by using the polynomial approximation method for lithium
approximation [10], residue grouping method [11], proper orthogonal ions concentration in electrodes, the residue grouping method for
decomposition method [12] and others. The ROMs for lithium-ion bat lithium ions concentration in electrolyte, and the analytical solution for
teries can be divided in two categories by the model setup: the potential in electrolyte. Particularly, the equations derived for the
pseudo-two-dimensional (ROM-P2D) model and single particle model lithium ion concentration in solid are numerically solved by the first
(ROM-SPM) [13]. order Runge–Kutta method. Then, the lithium ion diffusivity–SOC
The ROM-P2D model assumes that gradients of ion concentrations, relationship in cathode is predicted from the newly developed nonde
potentials and current density only in the particles and through-the- structive method that is based on the ROM-SPM using gradient-descent
plane are present, which significantly reduces the computational time algorithm, and the result is compared with that obtained from GITT
compared with that of FOM. This simplified models have been experi given in references. Finally, the predicted diffusivity using the proposed
mentally validated and then considered for purpose of controls and method is incorporated in the ROM-SPMVD that is validated against the
analyses of ion dynamics [14,15]. However, the models still require a experimental data at the discharge profiles and HPPC test at 25 � C. In
high computing power that microprocessors commonly used for BMS addition, the model is also validated at 45 � C and 0 � C with the predicted
cannot perform efficiently. Conversely, ROM-SPM assumes that the Ds,p(SOC) at each temperature to examine the validity in other tem
current density in each domain is uniformly distributed, such that all of peratures. The accuracy of the terminal voltage response and compu
particles in composite electrodes can be substituted by a single particle. tation time are compared with those of ROM-SPM.
Since the ROM-SPM has much simplified structure with minimum
number of internal states compared to that of ROM-P2D model, the 2. Model development
computational time can be drastically reduced. However, the accuracy
of the estimated internal variables drops, particularly in the terminal 2.1. Governing equations
voltage. The errors of the terminal voltage are dependent upon ampli
tude of currents and SOC, which presents the technical barriers for BMS Generally, a single cell of the pouch type lithium-ion polymer battery
applications. In order to increase the accuracy and improve the appli is made of microcells that are connected in parallel to increase capacity
cability of ROM-SPM, more aspects of dynamic effect within the cells and power. The model for the cell is approximated by a microcell under
have been taken into consideration within the model, which include the assumption that all of the microcells have the same characteristics. The
incorporation of the electrolyte dynamics [16,17], microcell has a sandwich structure with three domains, a composite
temperature-dependent parameters [18], and so on. anode, a separator and a composite cathode between the two current
When the cell is charged or discharged, the dynamic response of the collectors at the end of each electrodes. The electrochemical behavior of
terminal voltage, available power and actual capacity are affected by the microcell can be described by the principles of diffusion, mass
internal processes including ion transport, chemical reactions and transport, electrochemical kinetics, and Ohm’s laws. The governing
intercalation and deintercalation, such as diffusion process of lithium equations with boundary conditions are listed in Table 1.
ions, electronic conductivity and electrochemical kinetics. The most
predominate process for dynamic behaviors of a cell is the diffusion that
takes place in solid phase [16]. The associated parameter is the solid 2.2. Model simplification
phase diffusivity (or diffusion coefficient), Ds.
In most of recent published papers, the diffusivity in solid is regarded The governing equations include partial differential equations and
as a constant [9–12,15,19]. However, in reality, the diffusivity of nonlinear equations that can be simplified into linear ordinary differ
lithium ions, particularly in cathode particles, varies dependent upon ential equations by mathematical treatments, which results in a reduced
temperature [20,21], cycling numbers [22], as well as ion concentration order model (ROM). When the volume current density maintains uni
or SOC [23–25]. For example, experimental investigation on NMC532 formly distributed in anode and cathode, all of particles can be assumed
(Li[Ni0.5Mn0.3Co0.2]O2) [24] revealed that the diffusivity of lithium ions to be the same and can be replaced by a spherical particle, which is
in spherical cathode materials ranges from ~10 9 cm2 s 1 at lithiated called ROM-SPM [33]. Fig. 1 shows the schematic diagram of the
states to ~10 11 cm2 s 1 at delithiated states. Thus, in this paper, a ROM-SPM.
varying diffusivity is considered to be incorporated into ROM-SPM such The uniformed volume current density in solid phase and electrolyte
that the accuracy of the model can be increased while maintaining the is expressed as:
low computational cost. The model is called single particle model with
variable diffusivity (ROM-SPMVD). I
jLi ¼ (1)
As the key parameter of the ROM-SPMVD, the value of the diffusivity L⋅A
at each SOC should be known. Currently, the values of the diffusivity The terminal voltage of the cell is the difference between two solid
have been attempted to measure at different SOCs and temperatures by potentials of two electrodes and an Ohmic voltage drop in current
employing several techniques [25] that include the incremental titration collectors:
based galvanostatic intermittent titration technique (GITT) [24,26,29,
Vcell ¼ φs ðLÞ φs ð0Þ IRc (2)
30] and potentiostatic intermittent titration technique (PITT) [27,28],
electroanalytical techniques such as electrochemical impedance spec The Fick’s law for ion diffusion in solid phase is simplified using the
troscopy (EIS) [28,29] and cyclic voltammetry (CV) [31]. The GITT is polynomial approximation method [9]. The ion concentrations inside
the one that is widely used because of the possibility to determine the the solid particles are replaced using average concentration cs;ave ,
diffusivity accurately at discrete SOC points. However, this method has average concentration flux qave , and surface concentration cs;surf , which
often been applied to half cells, which requires specialized experimental results in:
2
Y. Hu et al. Journal of Power Sources 468 (2020) 228322
Table 1
Summary of governing equations and boundary conditions.
Description Equation Boundary condition
� � �
Ion concentration in solid phase ∂cs 1 ∂ ∂c s ∂cs ��
¼ 2 Ds r2 ¼0
∂t r ∂r ∂r ∂r �r¼0
�
∂cs �� jLi
Ds � ¼
� � ∂� r r¼Rs as F
Ion concentration in electrolyte ∂εe ce ∂ ∂c e 0
1 tþ ∂ce ��
¼ Deff ⋅ þ jLi ¼0
∂t ∂x e ∂x F ∂x �x¼0
�
∂ce ��
¼0
� � ∂x �x¼L � �
Ohm’s law in solid phase ∂ ∂ eff ∂
� ∂ � I
σeff φs jLi ¼ 0 σ φs �� ¼ σeff φs �� ¼
∂x ∂x ∂x x¼0 ∂x x¼L A
� �
∂ �� ∂ �
φ ¼ φs �� ¼0
� � � � ∂x s ��x¼Ln ∂x � x¼Ln þLs
Ohm’s law in electrolyte ∂ ∂ ∂ ∂ ∂ �� ∂ ��
κeff φe þ κeff ln ce þ jLi ¼ 0 φ ¼ φ ¼0
∂x ∂x� � ∂x �D ∂x � �� ∂x e �x¼0 ∂x e �x¼L
Electrochemical kinetics αa F αc F -
jLi ¼ as i0 exp η exp η
RT RT
η ¼ φs φe U
i0 ¼ kðce Þ ðcs;max cs;surf Þ ðcs;surf Þαc
αa αa
d 3 Li
cs;ave ¼ j
dt Rs Fas
d 30Ds 45 Li
qave þ qave þ j ¼0 (3)
dt Rs 2 2Rs 2 Fas
35 35 jLi
cs;surf 8qave cs;ave ¼
Rs Rs Ds Fas
If the diffusivity in solid phase depends upon SOC, Eq. (3) can be
modified as:
d 3 Li
cs;ave ¼ j
dt Rs Fas
d 30Ds ðSOCÞ 45 Li
qave þ qave þ j ¼0 (4)
dt Rs 2 2Rs 2 Fas
35 35 jLi
cs;surf 8qave cs;ave ¼
Rs Rs Ds ðSOCÞFas
In fact, SOC is defined as a ratio of the releasable charge capacity to
the maximum charge capacity (Qmax ) that can be expressed using the
average lithium ion concentration within the electrode particles:
Rt
IðτÞdτ θ θ0%
SOC ¼ SOC0 0
� 100% ¼ � 100% (5)
Qmax θ100% θ0%
, where SOC0 denotes the initial SOC, θ ¼ cs;ave =cs;max , and the subscripts Fig. 2. (a) Overpotential analysis and (b) parameter sensitivity analysis result
of 100% or 0% denote the states where battery is fully charged (100% of ROM-SPM.
3
Y. Hu et al. Journal of Power Sources 468 (2020) 228322
� � �
Ecell Vcell ¼ Ep En Up Un
�
ðϕe ðLÞ ϕe ð0ÞÞ ηp ηn þ IRc
¼ Ep U p ðaÞ solid concentration overpotential in positive electrode
ðEn Un Þ ðbÞ solid concentration overpotential in negative electrode
ðϕe ðLÞ � ϕe ð0ÞÞ ðcÞ electrolyte Ohmic and concentration overpotential
ηp ηn ðdÞ difference between activation overpotential of both electrodes
þIRc ðeÞ Ohmic voltage drop at current collectors
� �
RT jLi
η¼ sinh 1
(10) Contribution of individual overpotential to the total cell over
0:5F 2as i0
potential is also depicted in Fig. 2 (a). The overpotentials caused by solid
The terminal voltage of the cell results in as: phase concentration (a) and resistance of current collectors (e) are the
�
Vcell ¼ Up Un þ ðφe ðLÞ φe ð0Þ Þ þ ηp ηn
�
IRc (11) dominant parts, which are around 0.024 V and 0.020 V that make a
percentage of 91.5% of the total overpotential, while (b), (c) and (e)
Here, Un and Up denote the equilibrium potential of the solid in together contribute with a small portion. These results are similar to
negative (anode) and positive (cathode) electrodes, and are expressed as those by ROM-P2D [32].
a function of stoichiometric numbers and the associated surface ion The first one (a) is the potential difference between the cathode
concentration: particle at equilibrium state Ep and transient state Up . Since Ep and Up
are a function of cps;ave and cps;surf , the overpotential is directly affected by
Up ¼ Up ðxÞ; and Un ¼ Un ðyÞ
variation of the lithium ion concentration within the particles in the
cps;surf cns;surf cathode. In fact, the variation is induced by the diffusion of ions in the
, where x ¼ cp and y ¼ cns;max .
s;max solid in cathode governed by the Fick’s law.
Finally, the model was implemented and solved with the custom- The second one (e) is the pure Ohmic voltage drop caused by the
made code developed in MATLAB environment. resistance of the current collectors. The voltage drop increases linearly
as the current increases, which is governed by Ohm’s law. Since the
4
Y. Hu et al. Journal of Power Sources 468 (2020) 228322
Fig. 3. (a) Flowchart of diffusivity determination procedure. (b) Termination tolerance versus SOC, combined with an OCV-SOC curve as a reference. (c) Training
and fitting path of Ds,p and Vcell to show the convergence of the algorithm.
5
Y. Hu et al. Journal of Power Sources 468 (2020) 228322
others. � Electrode plate area: 89.1 mm (W) � 258.0 mm (L) for cathode and
91.1 mm (W) � 260.5 mm (L) for anode,
� Thickness of electrodes and separator: 77.5 μm for cathode, 84.0 μm
3.2. ROM-SPM based diffusivity determination algorithm
for anode and 8.9 μm for separator,
� Dimension: 99.7 mm � 301.5 mm � 13.2 mm.
One of the most sensitive parameter for the proposed ROM-SPMVD is
the Ds,p. Its determination is challenging because most of the measure
Cells are placed in a custom made thermostat system and in a thermal
ments for the diffusivity can be only carried out with a half cell, which is
chamber with a preset constant temperature. The terminal current,
not applicable for commercial cells that consist of stacked or folded
voltage and surface temperatures are measured. During the operation,
microcells. In addition, the measured value is usually inappropriate for
the cells’ temperature is actively controlled by the thermostat system
the simplified diffusion equation used in ROM. Therefore, we propose a
with a maximum temperature rise of 0.25 � C for all the experiments,
new method for determination of the diffusivity based on a gradient
such that the effects of the temperature on the discharging character
descent (GD) method (loop) [38], where an optimal value can be found
istics are limited. The details on the design of the thermostat system can
by minimizing the voltage error between the simulation and the
be found in Ref. [39].
experiment. The simulation is performed using ROM-SPM, where the
During the tests, the Ds,p(SOC) of the cathode NMC materials is
input is the current and the output is voltage during discharge from
estimated using the aforementioned method. Firstly, the cell was fully
100% SOC to 0%. Initial values of the parameters used for the simulation
charged at 25 � C and rested for 1 h and then discharged with a 0.8C
is listed in Appendix A. The selected current was 0.8C that corresponds
(46.4A) constant current until the voltage reaches 2.5 V. The current and
to 46.4A and should minimize degradation. An initial value of Ds,p is set
voltage are processed using the method outlined in section 3.2. The
as 4.0 � 10 10 cm2 s 1 at 25 � C.
initial Ds,p and learning rate α are listed in Table 3.
Fig. 3 (a) shows a flowchart of the method for determination of Ds,p,
Fig. 3 (c) shows the detailed training and fitting path of Ds,p and Vcell
which is carried out using an iterative GD optimization loop. In the GD
with time steps. The training steps are included at the bottom of the
loop, a cost function is formulated as an absolute error between the
figure. At the beginning of each time step, an initial value of Ds,p is
simulated and experimentally measured terminal voltage. Then, the cost
assumed, and the corresponding Vcell is calculated. Then the Ds,p con
function is minimized by calculating the gradient of the cost function
verges to the optimal value (green circle) following the training path
with regard to the diffusivity (rVerr ) and updating the value of Ds,p with
(solid purple line) by the gradient descent iteration, while the Vcell ap
a specified learning rate (α). This GD method requires a minimum error
proaches to the experimental measured value (red circle) along with the
bound that is called the termination tolerance, ε, which represents the
fitting path (solid blue line). Finally, the Ds,p(SOC) relationship curve at
maximum admissible discrepancy between the predicted and measured
25 � C is obtained and plotted in Fig. 4 (a).
voltage. In fact, the terminal voltage decreases rapidly at low SOC. As a
Similarly, the Ds,p(SOC) is estimated at 45 � C and 0 � C. For simplicity,
result, any small change of the Ds,p largely affects the terminal voltage,
which can lead to the algorithm not converging. Consequently, the
determination tolerance should be increased, which is approximated
with Eq. (16) for the given OCV-SOC curve. The value 1 mV indicates the
tolerance for the SOC range of 100%~20%, while the hyperbolic
tangential function represents an increase of the error bound when SOC
is smaller than 20%, which is plotted in Fig. 3 (b). At each SOC point, the
GD loop is repeated until the condition of the termination tolerance is
satisfied.
ε ¼ 5 � ð1 þ tanhð 20 � SOCÞ Þ þ 1 ½mV� (16)
At the first, the D*s,p is taken as the predicted diffusivity of cathode
materials at certain SOC, and is used as the initial guess for the next time
step. The process of algorithm will stop when all the experimental data
are used for determination, and out put the Ds,p(SOC) relationship at the
end of the process.
Table 3
Initial Ds,p and learning rate at 0, 25 and 45 � C.
Temperature/� C Initial Ds,p/cm2 s 1
Learning rate α
Fig. 4. (a) Predicted Ds,p(SOC) at 0, 25 and 45 � C. The black dashed line rep
11 22
0 3.0 � 10 1.0 � 10 resents the experimental measured diffusivity using GITT method at 25 � C
10 21
25 4.0 � 10 1.0 � 10
10 21 taken from Ref. [41]. (b) Logarithmic scale of averaged diffusivity versus in
45 9.0 � 10 5.0 � 10
verse temperature at 0, 25 and 45 � C.
6
Y. Hu et al. Journal of Power Sources 468 (2020) 228322
Table 4
� � Ea
Measured lithium ion diffusivity of NMC622 materials from literatures. ln Ds;p ¼ ln D0;p (17)
RT
Ref. Cell Temperature/ Ds range/cm2 s 1
Measurement
�
C method , where Ea is the activation energy for lithium ion diffusion, and D0;p is
[23] NMC622/ 25 1 � 10 11
–9 � GITT the diffusivity at infinite temperature.
Li 10 11 The logarithmic scale of the averaged diffusivity versus inverse
11
[41] NMC622/ 25 4 � 10 –1 � GITT temperature at 0, 25 and 45 � C is plotted in Fig. 4 (b), where, the red
Li 10 9
[42] NMC622/ 25 1 � 10 10
–1 � PITT
dashed line indicates the fitting using the Arrhenius equation. The
Li 10 9 determined activation energy for lithium ion diffusion is 19.71 kJ mol 1,
[20] NMC622/ 25 7 � 10 11
–8 � GITT which is comparable to the range from 4.82 kJ mol 1 to 45.83 kJ mol 1
Li 10 11
11
50 9 � 10 –1 �
10 10
11
0 3 � 10 –6 �
10 11
we only consider the temperature effect on Ds,p, and neglect its effect on
the current collector resistance and other parameters. The initial values
of Ds,p at 45 � C and 0 � C were set to be greater and smaller than that at
25 � C respectively, as listed in Table 3, which could be verified from the
tendency of the experimental measured diffusivity [20]. With this
approach, the convergence of the method is accelerated. In addition, the
learning rates were also adjusted to improve the convergence speed at
each temperature. The predicted Ds,p(SOC) relationship at 0 � C and 45
�
C are also depicted in Fig. 4 (a).
As shown in Fig. 4 (a), the values of predicted diffusivity of NMC622
cathode at 25 � C ranges between 2.53 � 10 11 cm2 s 1 and 2.26 � 10 10
cm2 s 1. When SOC increases from 0% to 50% SOC, the Ds,p values tend
to decrease from 2.21 � 10 10 cm2 s 1 to 4.23 � 10 11 cm2 s 1, and then
reach to a peak value of 1.46 � 10 10 cm2 s 1 at around 68% SOC. Then,
the Ds,p gradually increases from 1.29 � 10 10 cm2 s 1 to 2.26 � 10 10
cm2 s 1 as the SOC increases, and then slightly decreases at 100% SOC.
The result of the predicted Ds,p(SOC) was compared with a few
experimentally measured ones of the same materials, as listed in Table 4.
These experiments were performed using the cells that consist of
NMC622 (Li[Ni0.6Mn0.2Co0.2]O2) and metal lithium. The diffusivity of
lithium ions were measured by galvanostatic intermittent titration
technique (GITT) or potentiostatic intermittent titration technique
(PITT) method. The experimental measured lithium ion diffusivity at 25
�
C in those references were in the range of 1 � 10 11–1 � 10 9 cm2 s 1,
which is consistent with the range of the predicted values of Ds,p using
the proposed method. In addition, the shape of the predicted Ds,p(SOC)
curve at 25 � C shows similar to that obtained using GITT method [41]
with respect to a decrease at low SOC range, two peaks at 68% SOC and
95% SOC and the overall tendency.
Moreover, the predicted lithium ion diffusivity of NMC622 is
dependent upon the temperature, which can be described using Arrhe Fig. 6. Comparison of (a) RMSE and (b) computation time between ROM-SPM
nius equation: and ROM-SPMVD at 1C, 0.6C and 0.2C CC discharge at 25 � C.
7
Y. Hu et al. Journal of Power Sources 468 (2020) 228322
shows the input current profile of the HPPC test. The HPPC profile
consists of a sequence of a 30-s discharge pulse at Imax (112A, manu
facture’s absolute maximum allowable pulse discharge current), a 40-s
rest, and a 10-s regen pulse at 0.75 Imax (84A). Between each pair of
the discharge and regen pulses, the cell was discharged to 10% of ca
pacity at C/3 (19.3A). This sequence was repeated until a cutoff voltage
reaches 2.5 V.
Fig. 7 (b) compares the terminal voltage obtained from the simula
tion and experiment, where both curves match well at charge, discharge
and resting period over full range of SOC, and the RMSE is 0.012 V.
Likewise, the response of the ROM-SPMVD are further investigated
at two other different operating temperatures of 45 � C and 0 � C, where
the Ds,p (SOC) at 45 � C and 0 � C, as shown in Fig. 4 (a) , was incorporated
into the model. The results were plotted in Fig. 8, where the performance
of ROM-SPM and ROM-SPMVD are compared each other.
For the ROM-SPM, the value of the Ds,p at 45 � C and 0 � C was so
chosen as to best fit to the experimental curve, which results in 4.5 �
10 10 cm2 s 1 and 5.2 � 10 11 cm2 s 1. At 45 � C, the ROM-SPM tends to
deviate from the experimental data under 1C discharge, while ROM-
SPMVD keeps following the voltage drops. The RMSE of the voltage
by ROM-SPM and ROM-SPMVD is 0.074 V and 0.018 V, respectively.
At 0 � C, a similar tendency has been observed. The discrepancy of the
ROM-SPM becomes larger, while ROM-SPMVD follows the voltage fairly
well. The RMSE of voltage by ROM-SPM and ROM-SPMVD is 0.143 V
and 0.049 V, respectively.
4. Conclusion
8
Y. Hu et al. Journal of Power Sources 468 (2020) 228322
real time applications. However, the inaccuracy of the terminal voltage In future, advanced SOC estimation techniques such as adaptive
predicted by the ROM-SPM, especially at high C rates, limits wide ap Kalman filter will be applied to increase the accuracy of the model. In
plications. Therefore, a reduced order electrochemical model consid addition, aging effect will also be incorporated for further analysis.
ering variable diffusivity of lithium ions in solid is proposed and
developed based on ROM-SPM scheme in order to improve the accuracy. Declaration of competing interest
Experimental validation has revealed that by considering of variable
diffusivity in solid phase in cathode, the voltage error of ROM-SPMVD The authors declare that they have no known competing financial
can be reduced up to 79% compared with that of ROM-SPM. interests or personal relationships that could have appeared to influence
Here is a summary of the major outcomes: the work reported in this paper.
� A control oriented ROM-SPMVD of lithium-ion battery is developed CRediT authorship contribution statement
that minimizes terminal voltage errors at high currents.
� Overpotential and parameter sensitivity analysis of the model pa Yang Hu: Methodology, Software, Validation, Formal analysis,
rameters leads to a conclusion that the parameter Ds,p is sensitive and Writing - original draft, Conceptualization. Yilin Yin: Methodology,
stronly idenfiable. Resources. Yalan Bi: Methodology, Resources. Song-Yul Choe: Super
� A nondestructive ROM based method for determination of the Ds,p as vision, Project administration, Writing - review & editing.
a function of SOC and temperature is proposed.
� ROM-SPMVD is validated against experimental data of a pouch type
lithium-ion energy cell at different current profiles and different
temperatures.
Appendix
Appendix A
Parameters for lithium-ion battery electrochemical model (25� C)
(1) Ds
Solid phase diffusivity governs the surface concentration of lithium ion in particles, cs,surf, and affect the terminal voltage through the equilibrium
potential U. Thus, by applying the chain rule of differentiation, the sensitivity of Ds can be calculated as:
∂Vcell ∂U ∂cs;surf
¼� (18)
∂Ds ∂cs;surf ∂Ds
In Eq. (18), ∂c∂s;surf
U
is the slope of the equilibrium potential, and is obtained from the fitted curve. It is noted that cs,surf would also affect Vcell through
the overpotential term, ηcell, however the correlation was found to be minimal and is thus neglected [37].
∂cs;surf
In order to find the term ∂Ds on the right hand side, take the derivative of Ds to the third equation in Eq. (3), we have:
Li
∂cs;surf 8Rs ∂ Rs j 1
¼ qave þ (19)
∂Ds 35 ∂Ds 35 Fas D2s
To obtain the ∂D∂ s qave , we firstly take the Laplace to the second equation in Eq. (3),
45 1
qave ðsÞ ¼ �⋅IðsÞ (20)
2Fas LA sRs 2 þ 30Ds
9
Y. Hu et al. Journal of Power Sources 468 (2020) 228322
∂ 675 1 1
qave ðsÞ ¼ � �2 ⋅IðsÞ (21)
∂Ds Fas LA Rs
s þ 30D
Rs 2
s
(2) De
Diffusivity of electrolyte affects the terminal voltage through the potential difference in electrolyte phase as in Eq. (8), since it governed the lithium
ion concentration distribution in electrolyte, ce. Therefore, the sensitivity variable for De is expressed as:
� �
∂Vcell ∂ 2RT 1 tþ0 ∂
¼ ðφe ðLÞ φe ð0Þ Þ ¼ ðc ðL; tÞ ce ð0; tÞ Þ (24)
∂De ∂De ce;0 F ∂De e
Here we also neglect the effect of ce to the Vcell through the overpotential term, η, since the correlation was found to be minimal. Now, the key step is
to derive the expression for ∂D∂ e ce ð0; tÞand ∂D∂ e ce ðL; tÞ. By assuming that the cell operates under galvanostatic discharge, an approximate analytical so
lution for the electrolyte concentration distribution is obtained based on the parabolic profile approximation [40]. The analytical expressions for
ce ð0; tÞ and ce ðL; tÞ are:
IðtÞ
ce ð0; tÞ ¼ P2 ð1 expð t=τn Þ Þ þ c0 (25)
A
, where
1 P2
τn ¼
De ε0:5
n P1
1 tþ0 1
P1 ¼
De F ε1:5
n Ln
!
1 tþ0 1 Ln2 Ls2 Lp2 εs Ln Ls εp Ls Lp εp Ln Lp
P2 ¼ þ 0:5 þ 0:5 þ þ 1:5 þ
De F εn Ln þ εs Ls þ εp Lp 6εn
0:5 2εs 3εp 2ε1:5
n εs 2ε1:5
n
; and
� � � IðtÞ
ce ðL; tÞ ¼ P6 1 exp t τp þ c0 (26)
A
, where
1 P6
τp ¼ ;
De ε0:5
p P5
1 tþ0 1
P5 ¼ ;
De F ε1:5
p Lp
!
1 tþ0 Ln Ls Lp
P6 ¼ P2 þ 1:5 þ 1:5
De F 2ε1:5
n εs 2εp
where
!
∂ 1 tþ0 1 Ln2 L2s L2p εs Ln Ls εp Ls Lp εp Ln Lp
P2 ¼ þ 0:5 þ 0:5 þ þ 1:5 þ ;
∂De D2e F εn Ln þ εs Ls þ εp Lp 6ε0:5
n 2εs 3εp 2ε1:5
n εs 2ε1:5
n
10
Y. Hu et al. Journal of Power Sources 468 (2020) 228322
!
∂ 1 ε1:5
n Ln L2n L2s L2p εs Ln Ls εp Ls Lp εp Ln Lp
τn ¼ þ 0:5 þ 0:5 þ þ 1:5 þ
∂De D2e ε0:5
n εn Ln þ εs Ls þ εp Lp 6εn
0:5 2εs 3εp 2ε1:5
n εs 2ε1:5
n
; and
� �
∂ � �� ∂ � � t ∂ IðtÞ
ce ðL; tÞ ¼ 1 exp t τp P P6 exp t τp 2 τ ; (28)
∂De ∂De 6 τn ∂De p A
where
!
∂ 1 t0 ∂ Ln Ls Lp
P ¼ P þ 2 þ þ 1:5 þ 1:5 ;
∂De 6 ∂De 2 De F 2ε1:5
n εs 2εp
!
∂ 1 ε1:5
p Lp L2n Ls2 L2p εs Ln Ls εp Ls Lp εp Ln Lp
τp ¼ þ 0:5 þ 0:5 þ 1:5 þ 1:5 þ
∂De D2e ε0:5
p εn Ln þ εs Ls þ εp Lp 6εn
0:5 2εs 3εp 2εn εs 2ε1:5
n
!!
Ln Ls Lp
ε1:5
p Lp þ 1:5 þ 1:5
2ε1:5
n εs 2εp
Thus, bring Eq. (27) and Eq. (28) into Eq. (24), the analytical expression of the sensitivity variable for De is obtained.
eff
(3) κi
Effective electrolyte ionic conductivity affects the terminal voltage Vcell through the potential difference in electrolyte phase as in Eq. (8). Hence, its
sensitivity is obtained by applying chain rule of differentiation to Eq. (11), as:
!
∂Vcell 1 ∂ Ln 2Ls Lp
¼ þ eff þ eff IðtÞ
∂κeff
i
2A ∂κeff
i κeff
n κs κp
8 Ln 1
> � IðtÞ; i ¼ n
eff 2
>
> 2A κn
>
>
>
>
>
>
>
> (29)
>
< Ls 1
¼ � IðtÞ; i ¼ s
> A κeff 2
>
>
s
>
>
>
>
>
>
>
> Lp 1
>
: � � IðtÞ; i ¼ p
2A eff 2
κp
(4) k
According to Eq. (10), the kinetic rate constant, k, governs the activation overpotential, η, which then affects Vcell by Eq. (11). Therefore, taking the
differentiation to Eq. (11), the sensitivity of k can be obtained as:
∂Vcell RT 1 IðtÞ
¼ sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (30)
∂k Fi0 � �2 k
IðtÞ 2
2i0
þ ðas LAÞ
(5) Rc
The expression of the sensitivity variable of Rc is simple and straightforward. Take the derivative of Rc to Eq. (11), we have:
∂Vcell
¼ IðtÞ (31)
∂Rc
References [2] M. Hannan, M. Hoque, A. Mohamed, A. Ayob, Review of energy storage systems for
electric vehicle applications: issues and challenges, Renew. Sustain. Energy Rev. 69
(2017) 771–789.
[1] A. M Andwari, A. Pesiridis, S. Rajoo, R. Martinez-Botas, V. Esfahanian, A review of
[3] M. Hannan, M. Lipu, A. Hussain, A. Mohamed, A review of lithium-ion battery state
Battery Electric Vehicle technology and readiness levels, Renew. Sustain. Energy
of charge estimation and management system in electric vehicle applications:
Rev. 78 (2017) 414–430.
11
Y. Hu et al. Journal of Power Sources 468 (2020) 228322
challenges and recommendations, Renew. Sustain. Energy Rev. 78 (2017) [33] D. Zhang, B.N. Popov, R.E. White, Modeling lithium intercalation of a single spinel
834–854. particle under potentiodynamic control, J. Electrochem. Soc. 147 (3) (2000) 831.
[4] M. Doyle, Modeling of galvanostatic charge and discharge of the lithium/polymer/ [34] Y. Zhao, S.-Y. Choe, A highly efficient reduced order electrochemical model for a
insertion cell, J. Electrochem. Soc. 140 (6) (1993) 1526–1533. large format LiMn2O4/Carbon polymer battery for real time applications,
[5] M. Doyle, J. Newman, The use of mathematical modeling in the design of lithium/ Electrochim. Acta 164 (2015) 97–107.
polymer battery systems, Electrochim. Acta 40 (13–14) (1995) 2191–2196. [35] S. Yuan, L. Jiang, C. Yin, H. Wu, X. Zhang, A transfer function type of simplified
[6] T. Fuller, Simulation and optimization of the dual lithium ion insertion cell, electrochemical model with modified boundary conditions and Pad�e
J. Electrochem. Soc. 141 (1) (1994) 1–10. approximation for Li-ion battery: Part 2. Modeling and parameter estimation,
[7] W. Sung, D. Hwang, B. Jeong, J. Lee, T. Kwon, Electrochemical battery model and J. Power Sources 352 (2017) 258–271.
its parameter estimator for use in a battery management system of plug-in hybrid [36] P. Bai, M.Z. Bazant, “Charge transfer kinetics at the solid–solid interface in porous
electric vehicles, Int. J. Automot. Technol. 17 (3) (2016) 493–508. electrodes, Nat. Commun. 5 (1) (2014) 3585.
[8] W. Sung, C. Shin, Electrochemical model of a lithium-ion battery implemented into [37] Q. Lai, S. Jangra, H.J. Ahn, G. Kim, W.T. Joe, X. Lin, Analytical Sensitivity Analysis
an automotive battery management system, Comput. Chem. Eng. 76 (2015) 87–97. for Battery Electrochemical Parameters, ” American Control Conference (ACC),
[9] V. Subramanian, V. Diwakar, D. Tapriyal, Efficient macro-micro scale coupled 2019, pp. 890–896.
modeling of batteries, J. Electrochem. Soc. 152 (10) (2005) A2002–A2008. [38] Y. Nesterov, Introductory Lectures on Convex Optimization: a Basic Course, ”
[10] J. Forman, S. Bashash, J. Stein, H. Fathy, Reduction of an electrochemistry-based Springer Science & Business Media, 2013.
Li-ion battery model via quasi-linearization and Padé approximation, [39] Y. Yin, Z. Zheng, S.-Y. Choe, Design of a calorimeter for measurement of heat
J. Electrochem. Soc. 158 (2) (2011) A93–A101. generation rate of lithium ion battery using thermoelectric device, SAE Int. J.
[11] K. Smith, C. Rahn, C. Wang, Model order reduction of 1D diffusion systems via Alternative Powertrains 6 (No. 2) (2017) 252–260.
residue grouping, J. Dyn. Syst. Meas. Contr. 130 (1) (2008), 011012. [40] W. Luo, C. Lyu, L. Wang, L. Zhang, An approximate solution for electrolyte
[12] L. Cai, R. White, Reduction of model order based on proper orthogonal concentration distribution in physics-based lithium-ion cell models, Microelectron.
decomposition for lithium-ion battery simulations, J. Electrochem. Soc. 156 (3) Reliab. 53 (6) (2013) 797–804.
(2009) A154–A161. [41] H.J. Noh, S. Youn, C.S. Yoon, Y.K. Sun, Comparison of the structural and
[13] A. Jokar, B. Rajabloo, M. D� esilets, M. Lacroix, Review of simplified Pseudo-two- electrochemical properties of layered Li [NixCoyMnz]O2 (x¼ 1/3, 0.5, 0.6, 0.7, 0.8
Dimensional models of lithium-ion batteries, J. Power Sources 327 (2016) 44–55. and 0.85) cathode material for lithium-ion batteries, J. Power Sources 233 (2013)
[14] K.A. Smith, C.D. Rahn, C.-Y. Wang, Control oriented 1D electrochemical model of 121–130.
lithium ion battery, Energy Convers. Manag. 48 (9) (2007) 2565–2578. [42] Q. Wang, C.-H. Shen, S.-Y. Shen, Y.-F. Xu, C.-G. Shi, L. Huang, J.-T. Li, S.-G. Sun,
[15] X. Li, M. Xiao, S.-Y. Choe, Reduced order model (ROM) of a pouch type lithium Origin of structural evolution in capacity degradation for overcharged NMC622 via
polymer battery based on electrochemical thermal principles for real time operando coupled investigation, ACS Appl. Mater. Interfaces 9 (29) (2017)
applications, Electrochim. Acta 97 (2013) 66–78. 24731–24742.
[16] S.G. Marquis, V. Sulzer, R. Timms, C.P. Please, S.J. Chapman, An asymptotic [43] R. Amin, Y.-M. Chiang, Characterization of electronic and ionic transport in Li1-
derivation of a single particle model with electrolyte, J. Electrochem. Soc. 166 (15) xNi0.33Mn0.33Co0.33O2(NMC333) and Li1-xNi0.50Mn0.20Co0.30O2(NMC523) as a
(2019). function of Li content, J. Electrochem. Soc. 163 (8) (2016).
[17] A. Subramaniam, S. Kolluri, C.D. Parke, M. Pathak, S. Santhanagopalan, V. [44] P. Lyu, Y. Huo, Z. Qu, Z. Rao, Investigation on the thermal behavior of Ni-rich NMC
R. Subramanian, Properly lumped lithium-ion battery models: a tanks-in-series lithium ion battery for energy storage, Appl. Therm. Eng. 166 (2020) 114749.
approach, J. Electrochem. Soc. 167 (1) (Feb. 2020), 013534.
[18] T.R. Tanim, C.D. Rahn, C.-Y. Wang, A temperature dependent, single particle,
lithium ion cell model including electrolyte diffusion, J. Dyn. Syst. Meas. Contr. Nomenclature
137 (1) (2014).
[19] Y. Yin, Y. Hu, S.-Y. Choe, H. Cho, W.T. Joe, New fast charging method of lithium-
A: surface area of active material, cm2
ion batteries based on a reduced order electrochemical model considering side
as : specific surface area of electrode, cm 1
reaction, J. Power Sources 423 (2019) 367–379.
cs : concentration of lithium ion in solid phase, mol cm 3
[20] S. Cui, Y. Wei, T. Liu, W. Deng, Z. Hu, Y. Su, H. Li, M. Li, H. Guo, Y. Duan, 3
cs;ave : volume-averaged concentration of lithium ion in solid phase, mol cm
W. Wang, M. Rao, J. Zheng, X. Wang, F. Pan, Optimized temperature effect of Li-
cs;max : maximum concentration of lithium ion in solid phase, mol cm-3
ion diffusion with layer distance in Li(NixMnyCoz)O2 cathode materials for high
cs;surf : surface concentration of lithium ion in solid phase, mol cm 3
performance Li-ion battery, Adv. Energy Mater. 6 (4) (2015) 1501309.
ce : concentration of lithium ion in electrolyte phase, mol cm 3
[21] A. Tang, X. Wang, G. Xu, R. Peng, H. Nie, Chemical diffusion coefficient of lithium
Ds : diffusivity of lithium ion in solid phase, cm2 s 1
ion in Li3V2(PO4)3 cathode material, Mater. Lett. 63 (27) (2009) 2396–2398. eff
[22] X. Tang, L. Li, B. Huang, Y. He, Phenomenon of enhanced diffusion of lithium-ion De : effective diffusivity of lithium ion in electrolyte phase, cm2 s 1
in LiMn2O4 induced by electrochemical cycling, Solid State Ionics 177 (7–8) (2006) E: open circuit voltage, V
687–690. F: Faraday constant, 96485 C mol 1
[23] Y. Wei, J. Zheng, S. Cui, X. Song, Y. Su, W. Deng, Z. Wu, X. Wang, W. Wang, I: applied current, A
M. Rao, Y. Lin, C. Wang, K. Amine, F. Pan, Kinetics tuning of Li-ion diffusion in i0 : exchange current density, A cm 2
layered Li(NixMnyCoz)O2, J. Am. Chem. Soc. 137 (26) (2015) 8364–8367. jLi : reaction rate, A cm 3
[24] S. Yang, X. Wang, X. Yang, Y. Bai, Z. Liu, H. Shu, Q. Wei, Determination of the k: kinetic rate constant, (A cm 2)(cm3 mol 1)1.5
chemical diffusion coefficient of lithium ions in spherical Li[Ni0.5Mn0.3Co0.2]O2, L: thickness of cell unit, cm
4
Electrochim. Acta 66 (2012) 88–93. qave : volume-averaged concentration flux of lithium ion in solid phase, mol cm
[25] C. Delacourt, M. Ati, J.M. Tarascon, Measurement of lithium diffusion coefficient in R: universal gas constant, 8.3145 J mol 1 K 1
Li y FeSO4F, J. Electrochem. Soc. 158 (6) (2011) A741–A749. Rc : Ohmic resistance, Ω
[26] M.A. Caba~ nero, N. Boaretto, M. R€ oder, J. Müller, J. Kallo, A. Latz, Direct Rs : particle radius, cm
determination of diffusion coefficients in commercial Li-ion batteries, r: radial coordinate
J. Electrochem. Soc. 165 (5) (2018). T: temperature, K
[27] P.-C. Tsai, B. Wen, M. Wolfman, M.-J. Choe, M.S. Pan, L. Su, K. Thornton, t: time, s
0
J. Cabana, Y.-M. Chiang, Single-particle measurements of electrochemical kinetics tþ : transference number
in NMC and NCA cathodes for Li-ion batteries, Energy Environ. Sci. 11 (4) (2018) U: equilibrium potential, V
860–871. V: terminal voltage, V
[28] M. Wang, M. Luo, Y. Chen, Y. Su, L. Chen, R. Zhang, Electrochemical x: Cartesian coordinate across the cellGreek
deintercalation kinetics of 0.5Li2MnO3⋅0.5LiNi1/3Mn1/3Co1/3O2 studied by EIS and αa : αc : anodic and cathodic charge transfer coefficient
PITT, J. Alloys Compd. 696 (2017) 907–913. εe : volume fraction of electrolyte
[29] S. Yang, X. Wang, X. Yang, Y. Bai, Z. Liu, H. Shu, Q. Wei, Determination of the η: overpotential, V
chemical diffusion coefficient of lithium ions in spherical Li[Ni0.5Mn0.3Co0.2]O2, κeff : effective electrolyte conductivity, S m 1
Electrochim. Acta 66 (2012) 88–93. eff
κD : effective diffusive conductivity, S m 1
[30] P. Prosini, M. Lisi, D. Zane, Determination of the chemical diffusion coefficient of
σeff : effective solid conductivity, S m 1
lithium in LiFePO4, Solid State Ionics 148 (2002) 45–51.
φs : potential in solid phase, V
[31] S.-I. Pyun, H.-C. Shin, “The kinetics of lithium transport through Li1 δCoO2 thin
φe : potential in electrolyte phase, VSubscript
film electrode by theoretical analysis of current transient and cyclic
n: negative electrode
voltammogram, J. Power Sources 97–98 (2001) 277–281.
s: separator
[32] D.M. Bernardi, J.-Y. Go, Analysis of pulse and relaxation behavior in lithium-ion
p: positive electrode
batteries, J. Power Sources 196 (2011) 412–427.
12