Journal of Power Sources 468 (2020) 228322

Contents lists available at ScienceDirect

Journal of Power Sources

journal homepage: www.elsevier.com/locate/jpowsour

A control oriented reduced order electrochemical model considering

variable diffusivity of lithium ions in solid
Yang Hu , Yilin Yin , Yalan Bi , Song-Yul Choe *
Department of Mechanical Engineering, 1418 Wiggins Hall, Auburn University, AL, 36849, USA

H I G H L I G H T S

� A single particle model considering variable diffusivity is developed.

� Overpotential and parameter sensitivity analysis of the model are conducted.
� A model based method is used to predict the diffusivity in cathode.
� The model improves the accuracy and maintains the reduced computational time.

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.

1. Introduction undercharging, overcharging or thermal runaway to name a few [3].

The enhanced concerns about environmental pollution, global
warming, and depletion of fossil fuels have triggered development of
alternative powertrains for vehicles in addition to increased electric
power needed for actuators of autonomous vehicles. These vehicles
include battery electric vehicle (BEVs), hybrid electric vehicles (HEVs)
and fuel cell electric vehicles (FCEVs) [1]. Preferred energy storage for
the xEVs is the lithium-ion batteries due to their high energy density,
high specific energy and power, and electrochemical and thermal sta­
bility [2].
The lithium-ion battery systems are protected by battery manage­
ment system (BMS) from any abused operations that include
The BMS primarily measures the current, voltage and temperature that
are used to estimate state of charge (SOC) and state of health (SOH) and
control the temperature. Particularly, the SOC and SOH are accurately
estimated using a model that resides in the BMS. One of such models is
physics-based electrochemical model developed by Doyle et al. [4,5]
and Fuller et al. [6] that has obtained more and more attention for BMS
applications because of the benefits that the electrochemical changes
and variation of kinetic and transport quantities of lithium ions can be
estimated [7,8]. The physics based electrochemical models are con­
structed based on the principle of mass transport, electrochemical ki­
netics, diffusions, Ohm’s laws, and the associated potentials. The
governing equations consist of a set of coupled partial differential

* 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.
equations (PDEs) and nonlinear equations, which presents a full order
model (FOM). Solving the equations of the FOM is very computational
intensive and thus its implementation in BMS is impossible with current
hardware solution, particularly in real-time. There are several ap­
proaches to reduce the computational time by employing order reduc­
tion techniques and simplifying the models, while maintaining accuracy
in acceptable levels. The order reduction is performed by numerous
order reduction and simplification methods to obtain a reduced order
model (ROM) by employing polynomial approximation [9], Pad�e
approximation [10], residue grouping method [11], proper orthogonal
decomposition method [12] and others. The ROMs for lithium-ion bat­
teries can be divided in two categories by the model setup:
pseudo-two-dimensional (ROM-P2D) model and single particle model
(ROM-SPM) [13].
The ROM-P2D model assumes that gradients of ion concentrations,
potentials and current density only in the particles and through-the-
plane are present, which significantly reduces the computational time
compared with that of FOM. This simplified models have been experi­
mentally validated and then considered for purpose of controls and
analyses of ion dynamics [14,15]. However, the models still require a
high computing power that microprocessors commonly used for BMS
cannot perform efficiently. Conversely, ROM-SPM assumes that the
current density in each domain is uniformly distributed, such that all of
particles in composite electrodes can be substituted by a single particle.
Since the ROM-SPM has much simplified structure with minimum
number of internal states compared to that of ROM-P2D model, the
computational time can be drastically reduced. However, the accuracy
of the estimated internal variables drops, particularly in the terminal
voltage. The errors of the terminal voltage are dependent upon ampli­
tude of currents and SOC, which presents the technical barriers for BMS
applications. In order to increase the accuracy and improve the appli­
cability of ROM-SPM, more aspects of dynamic effect within the cells
have been taken into consideration within the model, which include the
incorporation of the electrolyte dynamics [16,17],
temperature-dependent parameters [18], and so on.
When the cell is charged or discharged, the dynamic response of the
terminal voltage, available power and actual capacity are affected by
internal processes including ion transport, chemical reactions and
intercalation and deintercalation, such as diffusion process of lithium
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
phase diffusivity (or diffusion coefficient), Ds.
In most of recent published papers, the diffusivity in solid is regarded
as a constant [9–12,15,19]. However, in reality, the diffusivity of
lithium ions, particularly in cathode particles, varies dependent upon
temperature [20,21], cycling numbers [22], as well as ion concentration
or SOC [23–25]. For example, experimental investigation on NMC532
(Li[Ni0.5Mn0.3Co0.2]O2) [24] revealed that the diffusivity of lithium ions
in spherical cathode materials ranges from ~10 9 cm2 s 1 at lithiated
states to ~10 11 cm2 s 1 at delithiated states. Thus, in this paper, a
varying diffusivity is considered to be incorporated into ROM-SPM such
that the accuracy of the model can be increased while maintaining the
low computational cost. The model is called single particle model with
variable diffusivity (ROM-SPMVD).
As the key parameter of the ROM-SPMVD, the value of the diffusivity
at each SOC should be known. Currently, the values of the diffusivity
have been attempted to measure at different SOCs and temperatures by
employing several techniques [25] that include the incremental titration
based galvanostatic intermittent titration technique (GITT) [24,26,29,
30] and potentiostatic intermittent titration technique (PITT) [27,28],
electroanalytical techniques such as electrochemical impedance spec­
troscopy (EIS) [28,29] and cyclic voltammetry (CV) [31]. The GITT is
the one that is widely used because of the possibility to determine the
diffusivity accurately at discrete SOC points. However, this method has
often been applied to half cells, which requires specialized experimental
2
Journal of Power Sources 468 (2020) 228322
instrumentations and testing conditions. Therefore, this technique is not
applicable for measuring commercial cells made of several dozens of full
cells. In addition, the diffusivity measured from half cells might not be
identical with the one needed for a model. Thus, a fast and efficient
method that allows for determination of the diffusivity as a function of
SOC without fabricating a half cell, which can be directly used in the
model, is also desired.
In this work, the ROM-SPMVD is firstly constructed on the basis of
ROM-SPM by using the polynomial approximation method for lithium
ions concentration in electrodes, the residue grouping method for
lithium ions concentration in electrolyte, and the analytical solution for
the potential in electrolyte. Particularly, the equations derived for the
lithium ion concentration in solid are numerically solved by the first
order Runge–Kutta method. Then, the lithium ion diffusivity–SOC
relationship in cathode is predicted from the newly developed nonde­
structive method that is based on the ROM-SPM using gradient-descent
algorithm, and the result is compared with that obtained from GITT
given in references. Finally, the predicted diffusivity using the proposed
method is incorporated in the ROM-SPMVD that is validated against the
experimental data at the discharge profiles and HPPC test at 25 � C. In
addition, the model is also validated at 45 � C and 0 � C with the predicted
Ds,p(SOC) at each temperature to examine the validity in other tem­
peratures. The accuracy of the terminal voltage response and compu­
tation time are compared with those of ROM-SPM.
2. Model development
2.1. Governing equations
Generally, a single cell of the pouch type lithium-ion polymer battery
is made of microcells that are connected in parallel to increase capacity
and power. The model for the cell is approximated by a microcell under
assumption that all of the microcells have the same characteristics. The
microcell has a sandwich structure with three domains, a composite
anode, a separator and a composite cathode between the two current
collectors at the end of each electrodes. The electrochemical behavior of
the microcell can be described by the principles of diffusion, mass
transport, electrochemical kinetics, and Ohm’s laws. The governing
equations with boundary conditions are listed in Table 1.
2.2. Model simplification
The governing equations include partial differential equations and
nonlinear equations that can be simplified into linear ordinary differ­
ential equations by mathematical treatments, which results in a reduced
order model (ROM). When the volume current density maintains uni­
formly distributed in anode and cathode, all of particles can be assumed
to be the same and can be replaced by a spherical particle, which is
called ROM-SPM [33]. Fig. 1 shows the schematic diagram of the
ROM-SPM.
The uniformed volume current density in solid phase and electrolyte
is expressed as:
I
jLi ¼ (1)
L⋅A
The terminal voltage of the cell is the difference between two solid
potentials of two electrodes and an Ohmic voltage drop in current
collectors:
Vcell ¼ φs ðLÞ φs ð0Þ IRc (2)
The Fick’s law for ion diffusion in solid phase is simplified using the
polynomial approximation method [9]. The ion concentrations inside
the solid particles are replaced using average concentration cs;ave ,
average concentration flux qave , and surface concentration cs;surf , which
results in:
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

SOC) or discharged (0% SOC).

When SOC in Eq. (4) is substituted by Eq. (5), the second equation in
Eq. (4) becomes a first order ODE with a variable coefficient and can be
numerically solved using the first order Runge-Kutta method. The

Fig. 1. Schematic diagram of ROM-SPM.

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

equation is iteratively solved by: 3. Determination of diffusivity in solid phase of cathode

Z t
IðτÞdτ! Determination of the diffusivity in solid phase of cathode of a cell is
30Ds 0
the most crucial and challenging task because only the terminal voltage
Qmax
ki ¼ qave ðti Þ
45 Li
j ðti Þ (6) of a cell can be measured. Therefore, a model based diffusivity deter­
R2s 2Rs 2 Fas mination algorithm is proposed, which is developed under two as­
sumptions: (1) the overpotential caused by the diffusion in cathode
qave ðtiþ1 Þ ¼ qave ðti Þ þ ki Δt
particles is predominant, and is separable from the terminal voltage; and
, where Δt is time step, and i ¼ 1,2,…,N is the iteration index. (2) the diffusivity of cathode is a sensitive parameter to the terminal
The PDE of lithium ion concentration in the electrolyte phase is voltage. These assumptions are examined by the overpotential analysis
approximated using the residue grouping method that groups the states and the parameter sensitivity analysis as follows.
according to similarity of eigenvalues [11]. The detailed description can
be found in a previously published paper [34].
3.1. Overpotential and parameter sensitivity analysis
The equation for Ohm’s law in electrolyte is a PDE with a nonlinear
term that is linearized as follows:
3.1.1. Overpotential analysis
!
�
∂ eff ∂
�
∂ κeff The relationship between the diffusion and overpotential is analyzed
D ∂
κ φe þ ce þ jLi ¼ 0 (7) using the experimentally validated ROM-SPM, whose parameters are
∂x ∂x ∂x ce;0 ∂x
listed in Appendix A. All the parameters including the diffusivity in
By integrating the linearized equation twice, an analytical expression active materials are set to be constant.
for the potential difference in electrolyte phase can be obtained [35]: Simulation results of the terminal voltage during discharge with 0.4C
! � � (23.2A) from 100% SOC and the open circuit voltage (OCV) are plotted
φe ðLÞ φe ð0Þ ¼
I Ln 2Ls Lp
þ þ þ
2RT 1 tþ0
ðce ðL; tÞ in Fig. 2 (a). The OCV (indicated as black line in Fig. 2 (a)) is the po­
tential differences between cathode and anode at equilibrium states,
eff eff eff
2A κn κs κp ce;0 F
ce ð0; tÞ Þ (8) where no concentration gradient exists in both particles, and is
expressed as:
The electrochemical kinetics are described by the Butler-Volmer
equation that is a function of concentration, potential and electro­ Ecell ¼ Ep ðxÞ En ðyÞ (12)
chemical reaction rate, where the anodic and cathodic charge transfer
cp cn
coefficients are set to equally 0.5 [36]. Then, the equation is used to get , where E denotes the OCV, x ¼ cps;ave and y ¼ cns;ave .
s;max s;max
the overpotential as follows: The total cell overpotential is the difference between the OCV and
� �
0:5F
� �
0:5F
�� �
0:5F
� the terminal voltage:
jLi ¼ as i0 exp η exp η ¼ 2as i0 sinh η (9)
RT RT RT ηcell ¼ Ecell Vcell (13)
Then, the total overpotential can be decomposed into those for
electrodes, electrolyte and current collectors:

� � �
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

Table 2 3.1.2. Sensitivity analysis of parameters

Mathematical expression of sensitivity variables of the parameters for ROM- In order to find out the sensitivity of parameters on the terminal
SPM. voltage, following analysis of the key parameters in ROM-SPM was
Parameters Sensitivity variables performed. The sensitivity variables can be obtained by the partial de­
Ds � � ! rivative of the terminal voltage over the parameters:
∂Vcell ∂U 1080 30Ds Rs
¼� t exp t þ IðtÞ
∂Ds ∂cs;surf 7Fas LA Rs 2 35Fas D2s LA ∂Vcell
De ∂Vcell ∂
�
2RT 1 tþ 0�
∂ Si ¼ (14)
¼ ðφ ðLÞ φe ð0Þ Þ ¼ ðc ðL; tÞ ce ð0; tÞ Þ ∂Xi
∂D e ∂De e ce;0 F ∂De e
� �
∂ ∂ t ∂ IðtÞ
ce ð0; tÞ ¼ ð1 expð t =τn ÞÞ P2 P2 expð t =τn Þ 2 τn , where Xi and Si represent the parameters and the corresponding
∂De ∂De τn ∂De A
∂
�
∂ t ∂
�
IðtÞ sensitivity variables.
ce ðL; tÞ ¼ ð1 expð t =τp ÞÞ P6 P6 expð t =τp Þ 2
∂De ∂De ! τn ∂De
τp
A From the expression for the terminal voltage given in Eq. (11), the
κeff
i ∂Vcell 1 ∂ Ln 2Ls Lp parameters that have been considered for the sensitivity analysis are: (1)
¼ þ þ IðtÞ
∂κeff 2A ∂κeff κeff
n κeff
s κeff
p diffusivity in solid phase in anode (Ds,n) and cathode (Ds,p), (2) diffu­
i i

8 Ln 1 sivity in electrolyte (De), (3) effective electrolyte ionic conductivity in

IðtÞ; i ¼ n
anode (κeffn), separator (κeffs) and cathode (κeffp), (4) kinetic rate constant
>
> 2A ðκeff Þ2
>
>
>
> n
>
>
>
>
> in anode (kn) and cathode (kp), as well as (5) Ohmic resistance (Rc).
>
< Ls 1 The mathematical expression for the sensitivity variables of the pa­
>
IðtÞ; i ¼ s
> A ðκeff
¼ 2
>
> s Þ rameters are listed in Table 2. Detailed derivation can be found in
>
Appendix B.
>
>
>
>
>
> Lp 1
>
>
>
: IðtÞ; i ¼ p For comparisons, all the sensitivity variables are normalized by
2A ðκeff Þ2
k ∂Vcell RT
p
1 IðtÞ
Ref. [37]:
¼ sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
∂k Fi0 �IðtÞ�2 k ∂Vcell
2
2i0
þ ðas LAÞ Si ¼ X: (15)
∂Xi i
Rc ∂Vcell
¼ IðtÞ
∂Rc The normalized sensitivities of each parameter are analyzed by
applying a set of constant current discharge profiles at 25 � C. The dis­
charging profile is composed of 1C, 0.8C, 0.6C, 0.4C and 0.2C, from
100% SOC until the terminal voltage reaches the cut off voltage. The
normalized sensitivities for each parameter are calculated at five input
current profiles, averaged and plotted in Fig. 2 (b). Results show that the
resistance is easily calculated, the Ohmic voltage drop can be decoupled,
Rc and Ds,p are the most sensitive parameters that affect the terminal
and thus, the diffusion-induced overpotential in cathode particles can be
voltage of the model, which become strongly identifiable compared with
estimated and separated.

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.

3.3. Prediction of Ds,p and model validation

3.3.1. Experiments and Ds,p prediction

For experiments, a large format pouch type NMC622/Graphite
lithium-ion energy cell with the following key specifications has been
used:

� Active materials: Li[Ni0.6Mn0.2Co0.2]O2 (NMC622) for cathode and

graphite for anode,
� Nominal capacity: 58Ah,
� Voltage range: 2.5 V–4.2 V,

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.

Fig. 5. Validation result: 1C, 0.6C and 0.2C CC discharge at 25 � C.

7
Y. Hu et al. Journal of Power Sources 468 (2020) 228322

that is typically seen for NMC materials [20,43,44].

3.3.2. ROM-SPMVD validation

The predicted Ds,p(SOC) is incorporated into the ROM-SPMVD and
analyzed with experimental data. The fully charged cell was discharged
with different C rates, 1C, 0.6C and 0.2C to the cutoff voltage of 2.5 V at
25 � C. Experimental and simulated terminal voltage are plotted in Fig. 5,
where the ROM-SPM and ROM-SPMVD are compared. At low C rate
(0.2C), the effects of dependence of the diffusivity on the discharge
characteristic are minimum, so that both the simulated terminal voltage
curves match well with the experimental data. When the applied current
increases, the simulated voltage curve from ROM-SPM gets deviated
from the experimental data, while the ROM-SPMVD performs well in
high currents. The Root-Mean-Square Errors (RMSEs) of the voltage at
0.6C and 1C are 0.049 V and 0.117 V, respectively that are reduced to
0.032 V and 0.024 V by ROM-SPMVD.
The RMSEs of the voltage by ROM-SPM and ROM-SPMVD are plotted
in Fig. 6 (a) and the computational time of the ROM-SPM and ROM-
SPMVD is plotted in Fig. 6 (b), where the models are simulated on a
PC that has a 3.4 GHz Intel Quad Core processor with 16 GB RAM.
It took 27.96 s for ROM-SPMVD to calculate the discharging char­
acteristic with 0.2C, which is around 2.4 times than the ROM-SPM. This
extended calculation time is mainly caused by the embedded
interpolation-based diffusivity update script in ROM-SPMVD. However,
it is still very efficient compared with ROM-P2D that take 5 times more
[15] and the FOM that takes 50 times more [22], that seems to be
acceptable for real time applications. In addition, the ROM-SPMVD
takes the advantage of the simplified structure of ROM-SPM that re­
places the high order matrix calculation with analytical equations for
solid phase concentration and electrolyte potential calculation, which is
friendly for implementation of the ROM into the BMS.
In order to verify the long-term performance under dynamic opera­
tions, the HPPC test at 25 � C is carried out. Before the test, the cell was
fully charged at the rate of C/3 (19.3A) and rested for 1 h. Fig. 7 (a) Fig. 8. Validation result: 1C CC discharge at (a) 45 � C and (b) 0 � C.

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

ROM-SPM of lithium-ion batteries have advantages over other

models because of short calculation time and the associated potentials in
Fig. 7. Validation result: (a) HPPC current; (b) HPPC voltage curves.

8
Y. Hu et al. Journal of Power Sources 468 (2020) 228322

real time applications. However, the inaccuracy of the terminal voltage
predicted by the ROM-SPM, especially at high C rates, limits wide ap­
plications. Therefore, a reduced order electrochemical model consid­
ering variable diffusivity of lithium ions in solid is proposed and
developed based on ROM-SPM scheme in order to improve the accuracy.
Experimental validation has revealed that by considering of variable
diffusivity in solid phase in cathode, the voltage error of ROM-SPMVD
can be reduced up to 79% compared with that of ROM-SPM.
Here is a summary of the major outcomes:
In future, advanced SOC estimation techniques such as adaptive
Kalman filter will be applied to increase the accuracy of the model. In
addition, aging effect will also be incorporated for further analysis.
Declaration of competing interest
The authors declare that they have no known competing financial
interests or personal relationships that could have appeared to influence
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)

Parameter Negative electrode Separator Positive electrode Unit

Electrode plate area, A 18637 18637 cm2

Thickness, δ 84.0 � 10 4 77.5 � 10 4 cm
Particle radius, Rs 17.5 � 10 4 10.7 � 10 4 cm
Active material volume fraction, εs 0.703 0.675
Electrolyte volume fraction, εp 0.247 0.5 0.243
3
Maximum solid phase concentration, cs,max 0.031 0.050 mol cm
3
Average electrolyte concentration, ce 0.012 0.012 0.012 mol cm
Diffusivity in solid phase, Ds 1.5 � 10 9 1.3 � 10 10 cm2 s 1
Diffusivity in electrolyte, De 3.0 � 10 6 3.0 � 10 6
3.0 � 10 6 cm2 s 1
Solid phase conductivity, σ 1 0.01 S cm 1
Electrolyte ionic conductivity, κ 0.057 0.057 0.057 S cm 1
Charge transfer coefficient, αa, αc 0.5 0.5
2
Kinetic rate constant, k 12.9 6.28 (A cm )(cm3 mol 1)1.5
Ohmic resistance, Rc 0.95 � 10 3 0.95 � 10 3 Ω

Appendix B. Derivation of the sensitivity variables of parameters in ROM-SPM

(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

Take the partial derivative of the equation to Ds, we have:

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

Then, take the inverse Laplace transform to Eq. (21), we obtain:

� �
∂ 675 1 30Ds
qave ðtÞ ¼ 2
t exp 2
t ⋅IðtÞ (22)
∂Ds Fas LA Rs Rs

Thus, the analytical expression of the sensitivity equation of Ds becomes:

� � � �
∂Vcell ∂U 1080 30Ds Rs
¼� t exp t þ IðtÞ (23)
∂Ds ∂cs;surf 7Fas LA Rs 2 35Fas D2s LA

(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

Take the derivative of De on Eq. (25) and Eq. (26), we have:

� �
∂ ∂ t ∂ IðtÞ
ce ð0; tÞ ¼ ð1 expð t=τn Þ Þ P2 P2 expð t=τn Þ 2 τn ; (27)
∂De ∂De τn ∂De A

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