Journal of The Electrochemical

Society

You may also like

- Physical Modelling of the Slow Voltage
Parametrisation and Use of a Predictive DFN Relaxation Phenomenon in Lithium-Ion
Batteries
Model for a High-Energy NCA/Gr-SiOx Battery Toby L. Kirk, Colin P. Please and S. Jon
Chapman

- Comparison of Electrolyte Transport

To cite this article: Alana Zülke et al 2021 J. Electrochem. Soc. 168 120522 Modelling in Lithium-ion Batteries:
Concentrated Solution Theory Vs
Generalized Nernst-Planck Model
Marco Lagnoni, Cristiano Nicolella and
Antonio Bertei
View the article online for updates and enhancements. - DandeLiion v1: An Extremely Fast Solver
for the Newman Model of Lithium-Ion
Battery (Dis)charge
Ivan Korotkin, Smita Sahu, Simon E. J.
O’Kane et al.

This content was downloaded from IP address 201.71.169.193 on 26/10/2023 at 20:39

 Journal of The Electrochemical Society, 2021 168 120522
1945-7111/2021/168(12)/120522/13/$40.00 © 2021 The Electrochemical Society (“ECS”). Published on behalf of ECS by IOP Publishing Limited

Parametrisation and Use of a Predictive DFN Model for a High-

Energy NCA/Gr-SiOx Battery
Alana Zülke,1,2,z Ivan Korotkin,2,3 Jamie M. Foster,2,4
Mangayarkarasi Nagarathinam,1 Harry Hoster,1,2 and Giles Richardson2,3
1
Department of Chemistry, Lancaster University, Lancaster, LA1 4YB, United Kingdom
2
The Faraday Institution, Quad One, Harwell Campus, Didcot, OX11 0RA, United Kingdom
3
Mathematical Sciences, University of Southampton, SO17 1BJ, United Kingdom
4
School of Mathematics and Physics, University of Portsmouth, PO1 3HF, United Kingdom

We demonstrate the predictive power of a parametrised Doyle-Fuller-Newman (DFN) model of a commercial cylindrical (21700)
lithium-ion cell with NCA/Gr-SiOx chemistry. Model parameters result from the deconstruction of a fresh commercial cell to
determine/confirm chemistry and micro-structure, and also from electrochemical experiments with half-cells built from electrode
samples. The simulations predict voltage profiles for (i) galvanostatic discharge and (ii) drive-cycles. Predicted voltage responses
deviate from measured ones by <1% throughout at least ∼95% of a full galvanostatic discharge, whilst the drive cycle discharge is
matched to a ∼1%–3% error throughout. All simulations are performed using the online computational tool DandeLiion, which
rapidly solves the DFN model using only modest computational resources. The DFN results are used to quantify the irreversible
energy losses occurring in the cell and deduce their location. In addition to demonstrating the predictive power of a properly
validated DFN model, this work provides a novel simplified parametrisation workflow that can be used to accurately calibrate an
electrochemical model of a cell.
© 2021 The Electrochemical Society (“ECS”). Published on behalf of ECS by IOP Publishing Limited. [DOI: 10.1149/1945-7111/
ac3e4a]

Manuscript submitted September 10, 2021; revised manuscript received November 10, 2021. Published December 10, 2021. This
paper is part of the JES Focus Issue on Women in Electrochemistry.
Supplementary material for this article is available online

Lithium-ion batteries (LiBs) are likely to remain the main energy carefully designed to ensure the overall conservation of lithium
storage technology powering portable electronics, stationary energy within the device.6,7 At an even more detailed level simulations of
storage units and electric vehicles over the next few decades. lithium transport can be performed on electrode particle geometries
Furthermore, the share of the automotive market held by electric extracted from microscopy measurements conducted on real elec-
vehicles is predicted to carry on growing precipitately. The trode geometries, see for example,8 though these are extremely
continued rapid growth in the LiB market has provided the drive computationally expensive and certainly are no better at fitting to
for a swift expansion in lithium battery sub-technologies that are discharge data from real cells than is the DFN model.
being brought from the lab onto the market; these include the In contrast to other modelling approaches, the implementation of
development of new electrode materials, cell designs, and cell DFN models guarantees i) higher accuracy over other traditionally
chemistries. Despite these new developments in materials and cell faster and easier simulation approaches, such as equivalent circuits,
architecture, the physics underpinning the operation and ageing of and ii) does not require large data-sets, a fundamental aspect of
these devices remains essentially unchanged. The efficient design of purely data-driven approaches using machine learning algorithms,
LiBs requires a fundamental understanding of the key physical for instance. P2D models are limited, however, by the user`s access
phenomena, and mathematical models encapsulate the interactions to an extensive list of input parameters describing the internal
between these phenomena within the complex structure of the components of the system in terms of design and geometry,
device. transport, and kinetics properties.9
In this context, accurate physics-based models of battery perfor- Extracting the parameters and functions required by the DFN
mance are extremely valuable and have the potential to yield model is an expensive and time-consuming task but nonetheless
significant competitive advantage to battery developers with an necessary for full physics-based simulations. Studies that conduct
active R&D program. The seminal works of Doyle, Fuller and such parameter estimation can be divided into two broad groups:
Newman in the 90ʼs1–3 led to the development of thermodynamically those that assume linear diffusivity of lithium-ion transport in the
consistent pseudo-two dimensional (P2D) battery models. At the electrode materials (see, for example,10–17) and those that assume a
battery pack scale, or even that of a cylindrical or a pouch cell, nonlinear diffusion law in the electrode materials (see, for
phenomenological equivalent circuit models (EQC)4 are often used example,18,19). Adopting the former approach leads to a relatively
(instead of physics-based models) because of the high computational simple parametrization problem. However, it cannot be expected that
cost of simulating multiple copies of the P2D model in a thermally the DFN (thus parametrised) will agree well with the experiment
heterogeneous three-dimensional structure. However, these phenom- across the full spectrum of battery operation, simply because lithium
enological models often termed “black boxes”, cannot be directly transport in most electrode materials is highly nonlinear. Some
related to the chemistry occurring within the battery micro-structure authors try to overcome this limitation by introducing an element of
and lead to increased errors especially at the extremes of state-of- non-physical fitting; thus, for example,11 uses the concept of a
charge (SoC).5 They are thus restricted to a narrow range of current current dependent electrode particle radius while17 uses current
draws and, furthermore, do not allow the possibility of exploring dependent weighting factors. Despite some success achieved by
how changes to the cell design affect overall battery performance. adopting these strategies, the optimal parametrization techniques
The P2D model, which we hereafter referred to as the Doyle-Fuller- require that lithium diffusivity in the electrode materials is quantified
Newman (DFN) model, is widely employed in commercial tool- as a function of local lithium concentration. This is usually achieved
boxes such as COMSOL Multiphysics or Dassault’s Dymola soft- by performing galvanostatic intermittent titration technique (GITT)
ware and solves the governing equations with algorithms that are or electrochemical impedance spectroscopy (EIS) (see, for example,
Ecker et al.19), potentiostatic intermittent titration technique (see, for
example, Oca et al.18), or by applying the Randle‘s Sevick equation
z
E-mail: a.zulke@lancaster.ac.uk on cyclic voltammetry data obtained at different sweep rates. The
 Journal of The Electrochemical Society, 2021 168 120522

linear diffusivity approximation can work well without ad-hoc non- microscopy data. DFN is a multiscale modelling framework that
physical fitting procedures but only where comparison is made to a circumvents some of these difficulties by posing the model on
restricted range of the batteries state of charge, for example for drive representative (usually spherical) electrode particles and using an
cycles which operate in a constrained region (e.g.10,13,16). averaged (or homogenised) representation of the electrolyte. The
Physics-based modelling has often been claimed to be “too microscale of this multiscale model is associated with lithium
computationally expensive”, and therefore impractical, for battery transport taking place within individual electrode particles while
management systems (BMS). However, the recent development of the macroscale is associated with transport processes taking place on
linear scaling algorithms, with compiled language and cloud-based the cell scale, in particular in the electrolyte. The complexity of the
data management to solve the DFN model means that this is no pore geometry is retained in the multi-scale equations in the form of
longer true. In this context, we mention the package DandeLiion20,21 “effective” coefficients which are determined by the microstructure.
which overcomes the aforementioned limitations and provides a In the original papers by Newman and coworkers,1–3 the multi-scale
standalone solution to solve the DFN model in a user-friendly way. system was posited on an ad-hoc basis, but the model equations have
Besides, P2D models can be potentially combined with other since been systematically derived from much more rigorous argu-
modelling techniques creating corrugated models of even faster ments in asymptotic homogenisation.26,27 In the remainder of this
performance. The research has seen substantially increased interest Section we shall outline the model and briefly discuss the physical
for the hybridisation of modelling techniques since physics-based principles that underpin each of the equations.
models are known to improve accuracy and do not require large data
pools, ultimately yielding decreased degradation rates of the Macroscopic equations.—There are two important charge trans-
batteries with obvious financial and environmental implications.22 port processes that must be captured on the macroscopic scale,
It has been recently reported the coupling of P2D solvers to namely: (i) ionic conduction in the electrolyte, and (ii) electron
electrochemical observers in algorithms for state of charge and conduction in the solid network of electrode particles, binder, and
remaining useful life (SOC/RUL) , evidencing the enormous additives. The former process is captured in the following equations:
potential of physics-based modelling for the battery industry.23
The purpose of this paper is twofold: firstly, to demonstrate how ∂c ∂N−
ϵl (x ) + = 0,
the DFN model parameters can be determined efficiently by ∂t ∂x
experiments conducted on a real cell; and, secondly, to show that ∂c j
solutions obtained from a DFN model parametrised in this fashion N− = − (x ) De (c) − (1 − t 0+)
∂x F
are capable of accurately predicting the cell’s dynamic behavior for in L 1 < x < L 4. [1]
different discharge protocols, including both constant current dis-
charges and full drive cycles, which encompass both fully charged
and fully discharged states of the battery. Our work thus extends that ∂j
= b (x ) jn ,
of Ecker et al.19,24 which performs a careful parametrisation but does ∂x
not make the comparison to full drive cycles. In order to accomplish ⎛ ∂Φ 2RT 1 − t 0+ ∂c ⎞
these goals we parametrised a 21700 cell composed of lithium- j = −  (x ) κ (c ) ⎜ − ⎟
⎝ ∂x F c ∂x ⎠
nickel-cobalt-aluminum oxide (NCA) as cathode and graphite-
silicon oxide (Gr-SiOx) as anode material and then used our in L 1 < x < L 4, [ 2]
parameter set in DandeLiion to make predictions of galvanostatic
and drive-cycle discharges. The accuracy of the parametrised model in which c is the concentration of both the positive lithium and
is quantified via the absolute differences between simulated and negative anions, j is the (averaged) ionic current density, jn is the
experimentally measured voltage profiles during discharge. To our transfer current density (from electrode material to electrolyte per
knowledge, no other DFN parametrisation of a complete NCA/Gr- unit surface area of electrode material), Φ is the electrolyte potential
SiOx system exists in the literature, despite these batteries being measured with respect to a lithium reference electrode and N− is the
widely employed in the automotive sector, e.g. on Tesla’s model 3 flux of the negatively charged anions. Our choice to work with N− is
cars.25 We also discuss the main approximations made in our model motivated by the fact that anions do not undergo reactions and
and introduce a simplified experimental workflow for the required therefore the anion conservation equation requires no source/sink
parametrisation process. By providing relevant data and a set of terms as would be the case for the Li-ion conservation equation; the
comprehensive guidelines on the parametrisation and usage of resulting PDEs are thus more readily discretised, using a finite
DandeLiion, we hope to catalyse the widespread use of electro- element method, so as to ensure no artificial loss of lithium occurs in
chemical models in both academic research and industry alike. the cell. We note that the DFN model is often stated in an equivalent
form in terms of N+, the flux of Li+ ions. The parameters ϵl, , b, t0+
and F are the electrolyte volume fraction, the inverse MacMullin
The DFN Model
number, the Brunauer-Emmett-Teller (BET) surface area of the
The internal structure of LiB electrodes are intricate and the electrode particles (i.e. the reactive surface area of electrode particles
processes that allow them to store, and release, energy effectively per unit volume of electrode), the transference number and
comprises of various length scales. As shown in Fig. 1, both the Faraday’s constant, respectively. In addition, the functions De(c)
anode and cathode are comprised of relatively small (∼10−6 m) and κ(c) are the electrolyte diffusivity and the electrolyte conduc-
electrode particles bound together by a polymer binder network, into tivity, respectively. Notably ϵl,  and b are effective coefficients that
which carbon black particles are embedded to improve their are determined by the properties of the microstructure. Thus, for
electrical conductivity. The resulting porous material (comprised instance, more tortuous microscale geometries lead to smaller values
of electrode particles and binder) is bathed by a liquid electrolyte. of  which captures the retardation of the transport process caused
Both electrodes have typical thicknesses on the order of ∼10−4 m, by the tortuosity of the electrolyte paths. Physically, (1a) is a
and are therefore many times wider than the dimensions of the conservation equation for anions whilst (1b) is a constitutive
electrode particles that are incorporated into their structure, which equation for the anions flux N−. Moreover, (2a) is a charge
have typical diameters of a few microns or less. Obtaining solutions conservation equation within the electrolyte while (2b) is the
to partial differential equations posed on a geometry with such constitutive equation for the electrolyte current density, which, in
disparate length scales is computationally expensive and, even the electrolyte, plays an analogous role to Ohm’s Law.
where this is possible, is hindered by the difficulty in obtaining a Electron conduction in the solid parts of the electrode matrices
precise specification of the geometry of the electrode structure from are modelled by charge conservation equations and Ohm’s Law
 Journal of The Electrochemical Society, 2021 168 120522

Figure 1. Schematic of a planar Li-ion battery cell.

there is no flux of anions, no ionic current, and the current in the

∂ja solid network is equal to that provided/extracted by the external
= −b (x ) jn , circuit. At the internal interfaces where the anode and cathode meet
∂x
the separator the current density in the solid network must be zero
∂Φ
ja = −σa a in L 1 < x < L 2, [3] because the separator is electronically insulating. As such the
∂x macroscopic equations are supplemented by the following boundary
conditions
∂jc
= −b (x ) jn ,
∂x I (t )
∂Φ ja ∣x = L1 = ,
jc = −σc c in L 3 < x < L 4, [ 4] A
∂x N−∣x = L1 = 0, j∣x = L1 = 0, [ 7]
Here, ja is the current density in the solid anode matrix, Φa is the
electric potential in the anode matrix, and σa is the effective ja ∣x = L 2 = 0, [ 8]
conductivity in the anode. The analogous variables in the cathode
are distinguished by the subscript c. jc∣x = L 3 = 0, [ 9]
The transfer current density jn induced by the redox reactions on
the interfaces between the electrode particles and electrolyte are, as
usual, given by Butler-Volmer kinetics I (t )
jc∣x = L 4 = ,
A
⎧ 2Fk a c1 2 (ca∣r = R )1 2 (c max − ca∣r = R )1 2 N−∣x = L 4 = 0, j∣x = L 4 = 0. [10]
⎪ a a a

⎪ sinh
⎪
⎪
( )
F η
2RT
a
in L 1 ⩽ x < L 2 , where I is the current leaving/entering the cell and A is the area of
the contact between the anode/cathode and the current collector.
jn = ⎨ 0 in L 2 < x < L 3, [5]
⎪ 12 1 2 max 12
⎪ 2Fk c c (cc∣r = Rc ) (cc − cc∣r = Rc ) Microscopic equations and boundary conditions.—The inter-
⎪ facial current density jn depends strongly on the concentration of Li
( )
Fη
⎪ sinh 2RTc in L 3 ⩽ x < L 4,
⎩ on the surface of the electrode particles necessitating that at every
location in the macroscopic dimension, x, we solve an appropriate
transport model for the Li within the particles. The transport within
ηa = Φa − Φ − Ueq, a (ca∣r = Ra ),
the particles occurs on a much smaller microscopic scale and we
ηc = Φc − Φ − Ueq, c (cc∣r = Rc ), [6] denote radial position within the particle by r. There is considerable
debate about the most appropriate equations for describing solid-
where R is the universal gas constant, T is absolute temperature, ka is state transport. In the original papers by Newman and coworkers1–3 a
the reaction rate constant in the anode,ca is the concentration of Li linear (Fickian) diffusion model was assumed, but since most
within the anode material, camax is the maximum concentration of sophisticated models, including Cahn-Hilliard,28–31 and nonlinear
lithium within the anode material and we assume that the anode diffusion models32 have all been proposed. Here we opt to take the
particles are spherical with radius Ra. The overpotential and latter approach and use a nonlinear diffusion equation noting that
equilibrium overpotential of the anode particles are denoted by ηa multiple studies have demonstrated the effectiveness of this model in
and Ueq,a respectively. Notably, the condition in which ηa = 0 is fitting experimental trends.33 Here we opt for the very simplest
equivalent to equality between the electrochemical potential of a microscopic model by assuming that all particles in the anode (or
lithium-ion in the anode material and that in the electrolyte and so, in cathode) are identical spheres of equal radius Ra (or Rc). We argue
turn, is equivalent to the Nernst equation as would be expected when that the SiOx particles in the anode occur in sufficiently small
the redox reaction has reached an equilibrium. This may, at first, numbers (4.6%) that they do not significantly undermine this
appear counter-intuitive until one recalls that Φ is not the (Maxwell) assumption. The microscopic equations and boundary conditions are
electric potential, in the electrolyte, but rather one measured with
respect to a lithium reference electrode, from which it follows that
FΦ is the electrochemical potential of a lithium-ion in the electro- ∂ca 1 ∂ ⎛ ∂c ⎞ ⎫
= 2 ⎜r 2Da (ca ) a ⎟ in 0 < r < R a ⎪
lyte. Counterparts to the anode variables in the cathode are denoted ∂t r ∂r ⎝ ∂r ⎠ ⎪
similarly, but with a subscript c. ⎬ in L 1 < x < L 2,
∂ca ∂ca jn ⎪
r=0 = 0, − D (
a ac ) = ⎪
Macroscopic boundary and interface conditions.—At the ex- ∂r ∂r r = R a F ⎭
tremities of the cell where the electrodes meet the current collectors [11]
 Journal of The Electrochemical Society, 2021 168 120522

Table I. A primary focus of this work is to obtain these parameters

and functions for the cylindrical lithium-ion cell (21700), which has
∂cc 1 ∂ ⎛ ∂c ⎞ ⎫ a mixed graphite-SiOx anode and an NCA cathode. Indeed, the
= 2 ⎜r 2Dc (cc) c ⎟ in 0 < r < R c ⎪
∂t r ∂r ⎝ ∂r ⎠ ⎪ parameter values and input functions determined by the experi-
⎬ in L 3 < x < L 4, mental component of this study are also given in Table I. While
∂cc ∂cc jn ⎪
r = 0 = 0, −Dc (cc) = ⎪ some of the parameters are relatively straightforward to obtain, such
∂r ∂r r = R c F ⎭ as Anode/Cathode/Separator thicknesses, others such as the inverse
[12] MacMullin number  are more difficult to ascertain. Accurate
parametrisation of the input functions, which represent variables
where Da(ca) and Dc(cc) are the lithium diffusivities in the anode and that vary importantly with lithium concentration, such as lithium
cathode particles respectively. Equations 11 and 12 are the nonlinear diffusivities in the anode and cathode particles Da(ca) and Dc(cc),
diffusion equations for the lithium-ion concentration in the electrode presents an even greater challenge.
particles with boundedness conditions on the concentrations which
ensures symmetry and zero lithium-ion flux at the particle core, and Solution of the DFN model using DandeLiion.—We use the
the Neumann boundary conditions on the surfaces of the particles highly accurate (second-order) solver DandeLiion,21 which is avail-
that ensures the fluxes of lithium leaving the electrolyte balance able online at,20 to solve the DFN model. This software readily
those entering the electrode particles. allows the user to simulate the discharges that we investigate here,
namely constant current, GITT and drive-cycle discharges.
Initial conditions.—The initial conditions are required for the Furthermore, it is straightforward to re-parametrise and to output
ionic concentration in the electrolyte as well as the Li concentration simulation data in a suitable form to use for comparison plots against
in the anode and cathode particles. We set experiments.
c∣t = 0 = c0, ca∣t = 0 = ca,0, cc∣t = 0 = cc,0. [13] Experimental Characterisation
Cylindrical cells (21700) with 4.8 Ah rated capacity at 25 °C
The full cell potential.—The results from the solution to the full upon C/3 discharge rate were used in the present investigation. The
cell DFN model can be used to compute the potentials at the anode main technical specifications as given by the supplier are shown in
and cathode current collectors Va and Vc, respectively via the Table II and the experimental workflow used for parametrisation is
relations illustrated by the diagram in Fig. 2.
A thorough parametrisation is time-consuming and expensive.
Va (t ) = Φa ∣x = L1 , Vc (t ) = Φc∣x = L 4 . [14] Hence it is sensible to first identify the critical model parameters in
the particular operating scenario that is to be simulated. By
and hence V(t), the potential drop across the full cell, performing a one-at-a-time parameter sensitivity analysis Li et
V (t ) = Vc (t ) − Va (t ). [15] al.34 concluded that simulated drive cycle curves generated by
DFN are very sensitive to parameters related to the overall capacity
It is this cell voltage that we shall use to make comparisons of the battery, such as cathode electrode thickness and active
between the model and experimental data. In the equation above we material volume fraction, but are nearly insensitive to electrolyte
neglect the ohmic drop between the current collectors since it is and separator-related parameters. In contrast, for the simulation of
relatively small in comparison with the potential difference given by constant current curves, the authors reported different sensitivity
Eq. 15. depending on the voltage window and C-rates simulated. For
instance, the thickness of the cathode has a high impact on the
Model parameters.—The DFN model described above requires simulation of curves at low C-rates, but separator lengths and
an extensive list of input parameters and functions. Other than particle radius are more important at higher C-rates. Motivated by
universal gas constant R and Faraday’s constant F, these are listed in these observations, the experimental workflow proposed here is

Table I. The DFN model parameters: (*) measured, (**) fitted, (†) derived analytically, (‡) provided by the manufacturer. The subscript s should be
read as s = a in the anode and s = c in the cathode.

Parameter Units Anode Cathode Separator

−5 −5
Anode/Cathode/Separator thickness (*) m 7.5 × 10 5.0 × 10 1.65 × 10−5
Effective particle radius Rs (*) m 11.2 × 10−6 5.0 × 10−7 —
Cell cross-sectional area A () m2 0.14112
Electrolyte volume fraction ϵl (*) — 0.2082 0.2753 0.4914
Inverse MacMullin number  (†) — 0.0950 0.1444 0.3445
Transference number t0+ 24 — 0.26
Particle surface area per unit volume electrode b (†) m−1 212 089 4 348 000 —
Diffusivity of electrolyte De(c)19 m2s−1 Function (Fig. 3)
Conductivity of electrolyte κ(c)19 S m−1 Function (Fig. 3)
Conductivity in solid σs24 S m−1 14 91 —
Maximum Li concentration in solid csmax (**) mol m−3 29400 43100 —
Reaction rate constant ks (**) m5/2 mol−1/2s−1 1.40 × 10−11 5.0 × 10−10 —
Equilibrium potential Ueq,s(cs) (*) V Fig. 4 (top) —
Lithium diffusivity in
electrode particles Ds(cs) (*) m2s−1 Fig. 4 (bottom) —
Absolute temperature T (*) K 298.15
Initial concentration of Li in electrolyte c024 mol m−3 1000
Initial concentration of Li in particles cs,0 (**) mol m−3 24800 400 —
 Journal of The Electrochemical Society, 2021 168 120522

Table II. Technical specification.

Properties Specification

Cell type Cylindrical 21700

Cathode chemistry Nickel cobalt aluminum oxide (NCA) [20%–50%g/ga)]
Anode chemistry Graphite/Silicon [10%–30% g/ga)]
Electrolyte salt Lithium Hexafluorophosphate [0.05%–5%g/ga)]
Ethylene carbonate,
Electrolyte solvent Ethyl methyl carbonate,
Dimethyl carbonate [5%–20%g/ga)]
Typical capacity (4.2 V, C/3 discharge) 4.8 Ah
Typical energy (4.2 V, C/3 discharge) 17.4 Wh
Nominal voltage 3.62 V
Lower voltage limit, Vmin 2.5 V
Upper voltage limit, Vmax 4.2 V
Energy density 256 Wh/kg; 717 Wh/L
Standard charging current rate C/3
Maximum charging current rate 1C
Standard discharging current rate C/5
Maximum discharging current rate 2C
Peak discharging current rate (30 s, 10 s) at 50% SOC and BOL 42 A/54 A
Cell weight 68 g
Max. height to top: 70.15 mm
Cell dimension Max. height to crimping: 70.15 mm
Max. upper diameter: 21.1 mm

a) Mass range as a percentage of the total cell mass (g/g) stated in the safety data sheet provided by the supplier.

Figure 2. Simplified experimental workflow used for parametrisation of the DFN model.

designed to obtain parameters to a sufficient degree of precision to present double-side coating of total geometric surface area of 1411.2
enable accurate simulations to be carried out over a wide range of cm2 and 1562.4 cm2, respectively. Anode electrode foils are
conditions, such as low-to-moderate constant current discharges and typically oversized with regards to the cathode in order to achieve
drive-cycles. the energy requirement and to limit risks of lithium plating. In this
way, anode-to-cathode ratios in commercial cells are always larger
Geometric parameters and physical characterization.—The first than 1. The anode region which has no cathode counter-part is often
stage of the parametrisation consists of the teardown of a represen- called overhang area, and lithium movement from/to these regions is
tative battery sample to reverse-engineer its materials properties. known to cause reversible capacity effects taking place at varied
One full cell was completely discharged to 2.5 V, using constant time scales.35
current at C/25, followed by disassembly in a dedicated argon-filled The chemistry of the active material in the cathode was
glove box with oxygen content and humidity below 1 ppm (MBraun, confirmed to be lithium nickel-cobalt-aluminum oxide by using
MB-Unilab Plus SP). Tear down process begins with the careful powder X-ray diffraction (PXRD, Rigaku Smartlab, 9 Kw Cu source
removal of the top and insulation caps. Subsequently, the can is cut generator, λ ∼ 1.540 51 Å). The resulting diffraction spectrum is
with a pipe cutter near the groove. The top tabs are disconnected first displayed against a reference spectrum for NCA-80
and carefully peeled off using ceramic pliers. After peeling only, the (LiNi0.8Co0.15Al0.05O2) in the Supplementary Information Fig. S1
bottom tabs are disconnected. The electrode foils are then separated (available at stacks.iop.org/JES/168/120522/mmedia).
from the separator film by unfolding the jelly-roll, having their Results from elemental mapping using energy-dispersive X-ray
lengths and masses measured. In our sample, both cathode and anode spectroscopy(EDX)/scanning electron microscopy (SEM, JEOL
 Journal of The Electrochemical Society, 2021 168 120522

Figure 3. Conductivity (left) and ionic diffusivity (right) of the electrolyte using data provided from Refs. 19, 24.

Figure 4. Equilibrium potentials of graphite anode (top left) and NCA cathode (top right), and diffusion coefficients in solid particles as a function of full cell
SOC for anode (bottom left) and cathode (bottom right).

JSM-7800F) confirm that the anode is a mixture of graphite and crucible for thermal gravimetrical analysis (NETZSCH STA 449 F3
(SiOx). Association with other elements allows us to also infer about Jupiter) in N2/O2 from 0 to 1100 at 5 °C/min. The content of silicon
the presence of (LiPF6) and carboxymethylcellulose, commonly was estimated as 4.6% w/w from the residual mass at T > 800 °C
used as electrolyte salt and binder material, respectively. In order to inferred by TGA (Fig. S2). In terms of moles proportionality, this
estimate the (SiOx) content, 15 mg of the anode material was analysis suggests that there are around 45 moles of graphite (carbon
mechanically scraped off from the copper current collector foil, graphite, MM = 12 g/mol) to every mol of Si in the anode.
using a ceramic scalpel, and placed in a cleaned and tared Al2O3 Electrodes and separator thicknesses, in addition to particle sizes,
 Journal of The Electrochemical Society, 2021 168 120522

Figure 5. Constant current discharge curves for slow (C/24) discharge (a) and relatively fast (C/3.2) discharge (b). Simulation (red lines), experiment (black
lines), and relative deviation of simulation results from the experiment (orange lines) are shown in time.

were measured from cross-sectional and top-view scanning electron for anode, cathode and separator, respectively. We stress that these
microscopy (SEM). Representative micrographs are displayed in values should be cross-checked whenever possible, with values from
Fig. S3. Raw SEM images were turned into binary files and then porosimetry analysis.
processed using a conventional image processing software (ImageJ)
in order to obtain the distribution of particle sizes by choosing the Electrolyte.—The inverse MacMullin number  of the the
proper sphericity level for the material being analysed. As the anode electrode matrix, which is the reciprocal of the MacMullin number
is a mixture of graphite and silicon oxide, the average diameter of and appears in the electrolyte equations, is estimated from the
the Gr-SiOx composites was chosen as the effective (average) anode Bruggeman relation.36 It is thus related to the volume fraction of
particle size assuming the Gr/SiOx particles ratio to be 95/5. As part electrolyte via
of the set of sensitive information about the product composition, we
do not have access to the exact proportion of binder conductive  (x ) = (ϵl (x ))1.5.
carbon and active material present on the electrodes, so we assume a
commonly reported recipe, as 95% of active material for both Particle surface area per unit volume can then be derived
electrodes. The volume fraction of the electrolyte in the materials analytically assuming spherical particles with radius R(x),
was estimated from a combination of top-view and cross-sectional
SEM micrographs. This is accomplished by first identifying and b (x ) = 3 (1 − ϵl (x ))(R (x ))−1.
confirming the active particle content using the elemental mapping
information obtained from the EDX/SEM (Fig. S4). EDX/SEM can We note that the use of the Bruggeman relation to estimate the
only be used qualitatively to confirm which elements are present. MacMullin number  −1 has been criticised, e.g. in Ref. 37, and in
Subsequently, a threshold range analysis is applied to the binary particular it tends to overestimate the real value of .
images to distinguish active particles from the background, to reveal It is experimentally inaccessible to harvest the electrolyte from
the porosity. Active content of NCA in the cathode inferred by three the full cell during the tear-down process. Conductivity and
different images shows results varying from 56 up to 73 percent. diffusivity in the electrolyte (Fig. 3) were obtained from the data
Image-based analysis for active content is not a quantitative method available in Ecker et al.19,24 by making the assumption that both
and should therefore be used only as an estimate. Using the average systems use electrolyte of very similar composition (in both cases
background from three different micro-graphs we estimate the LiPF6 dissolved in a carbonate-based solvent). The complete
volume fraction of electrolyte ϵl to be 0.2082, 0.2753, and 0.4914 electrolyte composition including the ratio of different solvents

Figure 6. Full cell GITT simulation in comparison with the measurements. Full discharge profile and the current.
 Journal of The Electrochemical Society, 2021 168 120522

Figure 7. Four fragments of the full discharge curve shown in Fig. 6. Beginning of discharge (a), middle of discharge (b), near the end of discharge (c), and at
the end of the experiment (d).

and additives are proprietary information, and therefore not avail- counter-electrode. All electrode and separator discs were cut using a
able. small manual die-cutting punching machine and assembled by
hydraulic crimping (TOB Energy). Fiberglass was used for the
Electrochemical characterization.—In order to extract data on separator and a battery-grade 1 M LiPF6, in ethylene carbonate/
the electrode thermodynamics, kinetics and transport properties we dimethyl carbonate (LP30, Sigma Aldrich), was employed as the
fabricated half-cells in coin cell configurations. This was accom- electrolyte. No solvent-based cleaning of the electrode foils was
plished by mechanically punching out anode and cathode circles of conducted in order to avoid removal of the solid electrolyte
diameter d = 19 mm from the electrode foils (double-side coating) interphase (SEI) and guarantee minimum interference with the
and re-assembling them into CR2032 cells with a metallic lithium resulting electrochemical response. After a 12-hour rest period for

Figure 8. A single pattern of the total current that mimics a drive cycle of an electric vehicle.
 Journal of The Electrochemical Society, 2021 168 120522

Figure 9. Drive cycle simulation full discharge curve in comparison with the experiment (a), and two zoomed-in fragments in the middle of discharge (b) and at
the very end of discharge (c). Orange line is the relative deviation of simulation results from the experiment.

proper wetting of the internal components, three consecutive cycles the first five, consecutive and reproducible cycles after conditioning
at C/12 (determined with regards to the active material mass) and of the coin cells were used in this work. We show the comparison of
25 °C were undertaken for conditioning between 0.05 V and 1.0 V pseudo-OCV curves obtained at the same very low C-rate and
for the anode half-cell and from 3.0 to 4.3 V for the cathode half- temperature for cathode, anode and full cell as Supplementary
cell. Pseudo-OCV curves were then obtained at sufficiently slow Material (Fig. S6). Analytical fits for the half-cell anode and cathode
constant current and at 25 °C. We confirm that a current rate of C/25 (dis)charge voltage profiles are obtained by averaging the pseudo-
yields near-equilibrium voltage profiles by comparing it with the OCV curves for charge and discharge and are provided in
curves at even slower rates, such as C/60 where identical electro- Supplementary Material as well (Eqs. SE1–SE2, Tables STI and
chemical signatures are obtained. (Fig. S5) The half-cells made from STII). We note that the hysteresis implied by the presence of
intact harvested material can be used for DFN parametrisation but different pseudo-OCVs for charge and discharge cannot be ex-
show poor cyclability, hence can only be used for a few cycles. Only plained by the nonlinear diffusion model for lithium transport in the
 Journal of The Electrochemical Society, 2021 168 120522

Table III. A summary of all heat dissipation terms used in the energy conservation law 16.

No. Term Description Formula

1 ̇
(elyte) Dissipative effects in the electrolyte 17
2 (a)
̇ part. The heat of mixing in the anode particles 18
3 (c)
̇ part. The heat of mixing in the cathode particles 19
4 (a) Ohmic dissipation in the solid parts of the anode L2 ∂Φa 2
̇ ohm (a)
̇ ohm = ∫ σa
L 1
( ) dx ∂x
5 (c) Ohmic dissipation in the solid parts of the cathode L4 ∂Φc 2
̇ ohm σ ( ) dx
(c)
̇ ohm = ∫ c ∂x
L 3
6 (a)
̇ pol. The polarization losses at the surfaces of the anode particles (a) L2
̇ pol. = ∫ b (x ) ηa j¯n dx
L
1
7 (c)
̇ pol. The polarization losses at the surfaces of the cathode particles (c) L4
̇ pol. = ∫ b (x ) ηc j¯n dx
L3

25 °C (Spec PU2J) having their surface temperature monitored via

K-type thermocouples in all experiments. The equilibrium potential
is taken as the voltage obtained at C/25 (pseudo-OCV) and the full
cell response is compared with the half cells for an evaluation of the
individual electrode’s utilization in the full cell. In Fig. 4, the
equilibrium potentials and diffusion coefficients for both electrodes
are shown as a function of the SoC. By identifying the central
graphite peak on the differential voltage analysis,38 the anode-to-
cathode ratio was estimated as 1.2. The initial and maximum
concentrations of lithium on the particles were derived by fitting
the full cell voltage during low-current (C/25) discharge, and the
reaction rate constants were obtained from fitting to the voltage
profile obtained from the GITT experiment. Additional constant
parameters taken from the literature are the transference number and
the conductivity of the solid, which are not expected to vary
considerably in systems of similar chemical composition. The
Figure 10. Cumulative irreversible energy losses (per unit area of the cell) complete list of parameters used in our simulations can be found in
as a percentage of Etot(t) the cumulative energy use (per unit area of the cell) Table I.
for constant current (C/3.2) discharge.
Model Fitting
For the purposes of the model, maximum and initial lithium
concentration on the individual electrodes are independent inputs.
Effectively, they represent an absolute quantity in moles of lithium
per unit volume of active material and can be estimated theoretically
if the true crystal density and stoichiometries are known. The
maximum theoretical concentration of lithium in the cathode
(ccmax ) can be calculated using the formula ccmax = ρz μ, where z
(the fraction of lithium per mole of cathode material) is 1 and the
molar mass of the lithiated compound is found to be μ = 95.96 g/
mol. If we assume a crystal density of ρ = 4.45 g cm−3 (NCA 80,
CAS No. 193 214-24-3), this yields a maximum lithium concentra-
tion of 46 371 mol m−3. For the anode material, we disregard the
presence of SiOx and take the crystal density and molar mass of the
lithiated graphite only, yielding a maximum lithium concentration in
the anode of 25 348 mol m−3.
Figure 11. Cumulative irreversible energy losses (per unit area of cell) as a As disparities between the theoretical and practical values are to
percentage of Etot(t) the cumulative energy use (per unit area of cell) for be expected, the identification of parameters is made unequivocally
drive cycle simulation. by fitting the DFN model to the pseudo-OCV experimental curves
obtained at low current (C/25). This is accomplished by fitting initial
electrode particles but could be adequately modelled by a phase-field and maximum concentrations so that the shape of the voltage vs the
model as propounded in Ref. 28. However, this adds very significant total capacity curve for the battery corresponds to the measured
extra computational complexity which is not justified by the values as closely as possible. Using a good initial guess for the
relatively small degree of hysteresis. initial/maximum concentrations we performed a set of simulations
Galvanostatic intermittent titration technique (GITT) was em- with different ca,0, cc,0, camax and ccmax and by comparing the output
ployed using 5 min titration pulses of 1.25 mA and 1.30 mA for the voltage to the full cell measurement (with the aid of a root mean
anode and cathode, respectively, with subsequent 45 min of relaxa- square deviation formula) adjusted the values of these parameters
tion. Diffusion coefficients were calculated based on Sand’s equa- appropriately. The values obtained from such fitting (See Table I)
tion as derived by Chang-Hui et al.12 All electrochemical tests are remarkably similar and were used on our simulations. Initial
including the validation experiments were performed using a concentration of Li in the particles are calculated in the same way—
Biologic BCS 815 system with the cells kept under air-cooling at for simulating a discharge, initial Li concentration in the anode will
 Journal of The Electrochemical Society, 2021 168 120522

t
Figure 12. Cumulative electrical work (per unit area of cell) ∫ IV Adt as a percentage of Etot(t), the cumulative energy use (per unit area of cell), for (a)
0
constant current (C/3.2) discharge and (b) drive cycle simulation.

be close to its maximum Li concentration, reflecting the almost fully voltage limit, of 2.5 V, at which point it was considered fully
lithiated state. discharged. While the C/24 discharge takes 24 hours, and is clearly
Likewise, initial guesses for the Butler-Volmer reaction rate slow, the C/3.2 discharge would still be considered relatively fast
constants for delithiation in the anode and lithiation in the cathode when compared to the standard discharge current rate, of C/5
were taken from the COMSOL library of materials (Graphite (960 mA), which is found in the specification document (Table II).
MCBM: 2 × 10−11 and NCA-80: 5 × 10−10 m s−1; COMSOL Comparison of the simulation results against the experimental
Multiphysics 5.4) and adjusted by fitting the DFN model output to ones, for both constant current discharges, are made in Fig. 5. It can
the experimentally determined GITT curves. GITT consist in be seen that there is close agreement between the experimental and
monitoring the voltage response upon a series of short current simulated total voltages curves, except for right at the end of the
pulses followed by an extended relaxation phase, throughout the discharge this is less than 3%. The error is shown as the relative
entire charge or discharge event. Using the appropriate values of deviation on a voltage basis, calculated as
current pulses and relaxation times, resistances and diffusion (VSimulation − VExperiment)/VExperiment · 100% at each point in time.
coefficients can be estimated as a function of a discrete number of For both the slow and fast discharges we used an identical
SoCs.39,40 To reproduce a GITT experiment in DandeLiion,20 the parameter set, which is given in Table I. For the slow discharge
user has to provide the desired profile of current vs time, following a (Fig. 5, left plot), the relative error does not exceed 1%, except for
square wave pattern of current either constant or zero. This is the last few minutes of discharge when the battery is almost fully
demonstrated in Fig. 6 where the input current is shown as a purple discharged. For the relatively fast discharge (Fig. 5, full agreement is
solid line, and the total voltage obtained numerically (red line) is seen in the initial first hour of test. However, with time, the model
compared with the experimental data (black line). Four zoomed-in increasingly underestimates the voltage, although it is still within an
fragments of the plot are given in Fig. 7. Our results clearly show acceptable error range. As the model considers isothermal conditions
excellent agreement, confirming that the model still captures the throughout the whole experiment (here at 25 °C), these deviations
shape of the entire experimental total voltage profile as well as the almost certainly result from the temperature increase seen in the
shape of each voltage spike throughout the experiment from the middle-to-end of discharge; this is aggravated by increasing current.
beginning of discharge (Fig. 7, top left) till the end (Fig. 7, bottom Self-heating of the batteries is negligible for low current discharges,
right). e.g. C/24, but much more significant for higher rates, such as C/3.2.
This is true even when the experiment is carried out in a thermal
Results chamber,41 as shown in Fig. S7. To account for the variations in
temperature observed experimentally, we display the simulated
After populating our parameters table, we solve the DFN model
discharge profile for C/3.2 and at 30 °C, and significantly improved
to predict the cell voltage from (i) a constant current discharge under
agreement between prediction and experiment (see Fig. 5b, solid
different (slow and fast) discharge rates, and (ii) a drive cycle with a
lines).
highly non-uniform current distribution. In all cases the initial
conditions correspond to the fully pre-charged state (see initial
Drive cycles.—In contrast to laboratory measurements where a
concentrations of lithium in anode and cathode particle in Table I).
constant current, or series of constant current pulses, is typically
The predictions were then compared to experimental results and
applied to a cell in order to characterize its performance, the
used to estimate the absolute error between the cell voltage obtained
operation of batteries in real world applications gives rise to highly
from the fully parametrised model and that obtained from the
non-uniform loads. From an application point of view, it is
equivalent experimental full-cell discharge.
imperative that physics-based models can reproduce battery beha-
vior during practical operation say, for example, in electric vehicles.
Galvanostatic discharge.—We investigated constant current In order to illustrate the use of our parametrised model under
discharges at a low rate, C/24 (200 mA), and at a higher rate, conditions akin to that in the operation of an electric vehicle we
C/3.2 (1.5 A). The higher rate is the standard maximum discharge simulate battery performance when subject to a drive cycle load
rate given in the battery specification, which if exceeded will likely based on a standardized test protocol. Figure 8 shows the current
result in a shortened battery lifetime of below 1,000 charge/ patterns for a single-commute time based on a world harmonized
discharge cycles. In both simulation and experiment, the cell was light-duty vehicles test procedure (WLTP, class 3, version
discharged from full until the cell voltage dropped to the lower acceleration42), which we used to validate our model. Again, the
 Journal of The Electrochemical Society, 2021 168 120522

same set of parameters used in the previous simulations (Table I) are

used to simulate a drive-cycle pattern based on the periodic
(elyte)
⎛
L4 dμ ⎛ ∂c ⎞2
repetition of the current pattern shown in Fig. 8. Simulation results ̇ = ∫L1 ⎝
⎜2 (x ) De (c) e ⎜ ⎟
dc ⎝ ∂x ⎠
for the cell voltage (red) are compared to experimental ones (black)
in Fig. 9. The relative error is shown in yellow. The top plot shows 1 ⎞
the entire discharge cycle, comprised of the periodically repeated + 〈 j〉2 ⎟ dx , [17]
κ (c )  (x ) ⎠
current pattern plotted in Fig. 8), from the fully pre-charged state
until the cell is fully discharged (i.e. when the cell voltage drops to
2.5 V). Figures 9b and c are two 1/2 h fragments of the top plot, (a) b (x ) ⎛ Ra (x )
L2
which correspond to approximately the middle and the end of ̇ part. = −4πF ∫L
1
⎜
4πRa2 (x ) ⎝ 0
∫
Da (ca )
discharge. According to the plots, the model reproduces experi-
mental data with a very high accuracy over the whole discharge ⎛ ∂c ⎞2 dUeq, a 2 ⎞
× ⎜ a⎟ r dr ⎟ dx , [18]
interval. The relative error does not exceed 2% except in the last few ⎝ ∂r ⎠ dca ⎠
minutes of discharge (out of 7.5 hours) when the battery is almost
fully discharged and the total voltage fluctuations increase. Again,
L4 b (x )
the deviation at the end of discharge can be attributed to a non- (c)
negligible increase in battery temperature due to self-heating.
̇ part. = −4πF ∫L 3 4πRc2 (x )
⎛ R c (x ) ⎛ ∂c ⎞2 dUeq, c 2 ⎞
Energy dissipation.—In the results section we demonstrate that
remarkably good agreement can be obtained between a properly
×⎜
⎝
∫0 Dc (cc) ⎜ c ⎟
⎝ ∂r ⎠ dcc
r dr ⎟ dx ,
⎠
[19]
calibrated DFN model and experiment. This should give the user
confidence that the DFN model provides a very good approximation where μe(c) is the chemical potential of the electrolyte and 〈j〉 is the
of the physical and electrochemical processes occurring within the averaged electrolyte current density. The current j̄n in Table III is the
battery. A natural question to ask is how the information gained by component of current density on electrode particle surface in
performing a careful, but relatively costly and time-consuming, direction of outward normal to the particle. The rest of the
parametrisation can be leveraged to improve its design and usage. parameters were introduced in the section above.
Since lithium-ion batteries are energy storage devices a key indicator In Figs. 10 and 11 we show the cumulative energy losses (per
of their efficiency, for a particular task, is the size and location of the unit area), for both C/3.2 constant current discharge and drive cycle,
energy dissipation occurring within the device. Indeed this is a good as a percentage of Etot(t) the total chemical energy consumed by the
way of assessing the fitness of the cell for a particular purpose and of cell (per unit area) throughout the discharge, until time t, as defined
improving its design. A recent rigorous mathematical analysis of the by the relation
DFN model43 has been used to develop an exact energy conservation
law for a device described by the DFN model, which shows that GA, t
the Gibbs free energy of a cell of area A, evolves according to an
energy conservation law of the form
Etot (t ) = ∫0 {IV
A
+ (̇
(elyte) (a)
+ ̇ part.
(a) (a) (c) (c) (c)
+ ̇ ohm + ̇ pol. + ̇ part. + ̇ ohm + ̇ pol.)} dt, [20]
dG (elyte) (a) (a)
−A = IV + A (̇ + ̇ part. + ̇ ohm
dt or equivalently (on using the identity 16) Etot(t) = G(0) − G(t). In
(a) (c) (c) (c)
+ ̇ pol. + ̇ part. + ̇ ohm + ̇ pol.). [16] Fig. 12b we show the cumulative electrical work (per unit area) done
by the cell as a percentage of Etot(t) the total chemical energy
consumed by the cell (per unit area).
Here IV (current times voltage) is the useful power extracted
Notably in both cases, constant current discharge and drive cycle,
from the cell while the terms bracketed on the right-hand side of this
this is a fairly efficient cell which only loses roughly 2.5% and 4% of
equation are the irreversible energy losses (per unit area of cell and
its chemical energy irreversibly. However, it is still interesting to
per unit time). These terms have been divided up into seven
compare where these losses predominantly occur within the cell. For
locations within the cell, where losses can occur, see Table III;
(elyte) (a )
the constant current discharge, the dominant losses are the heat of
thus ̇ is the loss in the electrolyte, ̇ part. is that due to the heat mixing in the cathode particles (a term usually neglected in standard
(a )
of mixing in the anode particles, ̇ ohm is that due to Ohmic losses in treatments of heating in Li-ion batteries) and polarization losses in
(a ) the anode. In contrast, for the drive cycles, polarization losses in the
the carbon binder domain in the anode, ̇ pol. are the polarization anode are by some distance the most pronounced source of unwanted
losses resulting from the overpotential between the anode particles energy dissipation.
(c )
and the surrounding electrolyte, ̇ part. is that due to the heat of
(c )
Conclusions
mixing in the cathode particles, ̇ ohm is that due to Ohmic losses in
(c ) In this work we have shown that a carefully parametrised DFN
the carbon binder domain in the cathode, ̇ pol. are the polarization model is capable of accurately predicting the behavior of a cell for a
losses resulting from the overpotential between the cathode particles wide variety of cell discharges and, in particular, could accurately
and the surrounding electrolyte. In Ref. 43 it is shown how these loss predict cell voltage profiles during both galvanostatic and drive-
terms should be computed from integrals of the solution to the DFN cycles discharges. We were also able to use the solutions to the DFN
model so that Eq. 16 is satisfied exactly. Notably, the result obtained model to accurately compute the irreversible energy losses occurring
in Ref. 43 differs significantly from the standard (and incorrect) within the cell, for different discharge patterns, and infer their
methods that have been used to compute the irreversible energy location. In performing these tasks, we have shown that such well-
losses from solutions to the DFN model. parametrised DFN models are an extremely useful engineering tool
All the heating terms in Eq. 16 are summarised in Table III, to support control systems or to inform new cell designs. In
where the dissipative effects in the electrolyte ̇
(elyte)
and the heat of particular knowledge of the location of the major energy losses
̇ ( a ) ̇ ( c ) can be used to point to components of the cell that should be
mixing in particles ( part. and  part.) can be expressed via the redesigned. We illustrated a simplified experimental workflow to
following integrals: effectively extract the model key parameters from a physical cell.
 Journal of The Electrochemical Society, 2021 168 120522

This was based upon a careful assessment of the critical parameters 16. K. Uddin, S. Perera, W. D. Widanage, L. Somerville, and J. Marco, “Characterising
that most influence the cell’s behavior; these turn out to be those that lithium-ion battery degradation through the identification and tracking of electro-
chemical battery model parameters.” Batteries, 2, 13 (2016).
are related to its total capacity and its transport properties (e.g. 17. S. Santhanagopalan, Q. Guo, and R. E. White, “Parameter estimation and model
particle radius, electrode thicknesses, active material content and discrimination for a lithium-ion cell.” J. Electrochem. Soc., 154, A198 (2007).
diffusion coefficients). Finally, we remark that DandeLiion solver 18. L. Oca, E. Miguela, E. Agirrezabala, A. Herran, E. Gucciardi, L. Otaegui,
has the potential to be leveraged for on-board battery management E. Bekaert, A. Villaverde, and U. Iraola, “Physico-chemical parameter measure-
ment and model response evaluation for a pseudo-two-dimensional model of a
systems in electric vehicles and used to perform real-time computa- commercial lithium-ion battery.” Electrochimica Acta, 382, 138287 (2021).
tions, given its sub-second timescale resolution. 19. M. Ecker, T. K. D. Tran, P. Dechent, S. Käbitz, A. Warnecke, and D. U. Sauer,
“Parameterization of a physico-chemical model of a lithium-ion battery. I.
Acknowledgments Determination of parameters.” J. Electrochem. Soc., 162, A1836–48 (2015).
20. DandeLiion online simulation engine https://www.dandeliion.com/.
The authors were supported by the Faraday Institution Multi- 21. I. Korotkin, S. Sahu, S. E. J. O’Kane, G. Richardson, and J. M. Foster, “DandeLiion
Scale Modelling (MSM) project Grant number EP/S003053/1. We v1: An extremely fast solver for the Newman model of lithium-ion battery (dis)
are grateful for the help of Dr. Simon O’Kane in fitting data for the charge.” J. Electrochem. Soc., 168, 060544 (2021).
22. J. M. Reniers, G. Mulder, and D. A. Howey, “Unlocking extra value from grid
anode and electrolyte properties and Dr. Ferran Brosa-Planella for batteries using advanced models.” J. Power Sources, 487, 229355 (2021).
the discussions on effective diffusion coefficients. 23. A. Allam, E. Catenaro, and S. Onori, “Pushing the Envelope in Battery Estimation
Algorithms.” iScience, 23, 101847 (2020).
ORCID 24. M. Ecker, S. Käbitz, I. Laresgoiti, and D. U. Sauer, “Parameterization of a physico-
chemical model of a lithium-ion battery. II, Model validation.” J. Electrochem.
Alana Zülke https://orcid.org/0000-0003-2274-0002 Soc., 162, A1849 (2015).
Ivan Korotkin https://orcid.org/0000-0002-5023-3684 25. K.-J. P. K.-J. Park, H.-H. Hwang, F. Ryu, S.-J. Maglia, P. Kim, C. S. Lamp, and Y.-
Jamie M. Foster https://orcid.org/0000-0002-6120-5734 K. Sun, “Degradation Mechanism of Ni-Enriched NCA Cathode for Lithium
Batteries: Are Microcracks Really Critical?” ACS Energy Lett., 4, 1394 (2019).
Mangayarkarasi Nagarathinam https://orcid.org/0000-0001-5485- 26. F. Ciucci and W. Lai, “Derivation of micro/macro lithium battery models from
4237 homogenization.” Transport in Porous Media, 88, 249 (2011).
Harry Hoster https://orcid.org/0000-0001-6379-5275 27. G. Richardson, G. Denuault, and C. Please, “Multiscale modelling and analysis of
lithium-ion battery charge and discharge.” J. Eng. Maths, 72, 41 (2012).
References 28. Y. Zeng and M. Z. Bazant, “Phase separation dynamics in isotropic ion-
intercalation particles.” SIAM J. Appl. Math, 74, 980 (2014).
1. M. Doyle, T. F. Fuller, and J. Newman, “Modeling of galvanostatic charge and 29. J. W. Cahn and J. E. Hilliard, “Free energy of a nonuniform system. I. Interfacial
discharge of the lithium polymer insertion cell.” J. Electrochem. Soc., 140, 1526 freeenergy.” J. Chem. Phys., 28, 258 (1958).
(1993). 30. J. W. Cahn, “Free energy of a nonuniform system. II, Thermodynamic basis.”
2. M. Doyle, J. Newman, A. S. Gozdz, C. N. Schmutz, and J.-M. Tarascon, J. Chem. Phys., 30, 1121 (1959).
“Comparison of modeling predictions with experimental data from plastic lithium 31. J. W. Cahn and J. E. Hilliard, “Free energy of a nonuniform system. III, Nucleation
ion cells.” J. Electrochem. Soc., 143, 1890 (1996). in a two-component incompressible fluid.” J. Chem. Phys, 31, 688 (1959).
3. T. F. Fuller, M. Doyle, and J. Newman, “Simulation and optimization of the dual 32. M. W. Verbrugge and B. J. Koch, “Modeling lithium intercalation of single fiber
lithium ion insertion cell.” J. Electrochem. Soc., 141, 1 (1994). carbon microelectrodes.” J. Electrochem. Soc., 143, 600 (1996).
4. G. L. Plett, Battery Management Systems, Volume II: Equivalent-Circuit Methods 33. S. A. Krachkovskiy, J. M. Foster, J. D. Bazak, B. J. Balcom, and G. R. Goward,
(Artech House, Norwood, MA) (2015). “Operando mapping of Li concentration profiles and phase transformations in
5. A. X. Hu, S. Li, and H. Peng, “A comparative study of equivalent circuit models for graphite electrodes by magnetic resonance imaging and nuclear magnetic resonance
Li-ion batteries.” J. Power Sources., 198, 359 (2012). spectroscopy.” J. Phys. Chem. C, 122, 21784 (2018).
6. J. Newman and K. E. Thomas-Alyea, Electrochemical Systems (Wiley, New York, 34. W. Li, D. Cao, D. Jöst, F. Ringbeck, M. Kuipers, F. Friea, and D. Uwe Sauer,
NY) 3rd ed. (2014). “Parameter sensitivity analysis of electrochemical model-based battery manage-
7. J. Newman, K. E. Thomas, H. Hafezi, and D. R. Wheeler, “Modeling of lithium-ion ment systems for lithium-ion batteries.” Appl. Energy, 269, 115104 (2020).
batteries.” J. Power Sources., 119–121, 838 (2003). 35. R. M. Burrell, A. A. Zulke, P. Keil, and H. Hoster, “Communication-Identifying
8. A. Latz and J. Zausch, “Multiscale modeling of lithium ion batteries: thermal and Managing Reversible Capacity Losses that Falsify Cycle Ageing Tests of
aspects.” Beilstein Journal of Nanotechnology, 6, 987–1007 (2015). Lithium-Ion Cells.” J. Electrochem. Soc., 167, 130544 (2020).
9. Y. Li, K. Liu, A. M. Foley, A. Zülke, M. Berecibar, E. Nanini-Maury, J. Van 36. B. D. Bruggeman, “Calculation of different physical constants of heterogeneous
Mierlo, and H. E. Hoster, “Data-driven health estimation and lifetime prediction of substances. i. dielectric constants and conductivities of mixed bodies of isotropic
lithium-ion batteries: A review.” Renew. Sustain. Energy Rev., 113, 109254 substances.” Ann. Phys. (Berlin), 416, 636 (1935).
(2019). 37. J. Landesfeind, A. Hattendorff, W. A. Ehrl, Wall, and H. A. Gasteiger, “Tortuosity
10. J. C. Forman, S. J. Moura, J. L. Stein, and H. K. Fathy, “Genetic identification and Determination of Battery Electrodes and Separators by Impedance Spectroscopy.”
fisher identifiability analysis of the Doyle-Fuller-Newman model from experimental J. Electrochem. Soc., 167, A1373 (2016).
cycling of a LiFePO4 cell.” J. Power Sources, 210, 263 (2012). 38. M. P. Mercer, M. Otero, M. Ferrer-Huerta, A. Sigal, D. E. Barraco, H. E. Hoster,
11. A. Maheshwari, M. A. Dumitrescu, M. Destro, and M. Santarelli, “Inverse and E. P. M. Leiva, “Transitions of lithium occupation in graphite: A physically
parameter determination in the development of an optimized lithium iron informed model in the dilute lithium occupation limit supported by electrochemical
phosphate—Graphite battery discharge model.” J. Power Sources, 307, 160 (2016). and thermodynamic measurements.” Electrochim. Acta., 324, 134774 (2019).
12. C.-H. Chen, F. Brosa Planella, K. O’Regan, D. Gastol, W. D. Widanage, and 39. Y. Zhu and C. Wang, “Galvanostatic intermittent titration technique for phase-
E. Kendrick, “Development of Experimental Techniques for Parameterization of transformation electrodes.” J. Phys. Chem. C., 114, 2830 (2010).
Multi-scale Lithium-ion Battery Models.” J. Electrochem. Soc., 167, 080534 40. A. Nickol, T. Schied, C. Heubner, M. Schneider, A. Michaelis, M. Bobeth, and
(2020). G. Cunibertid, “GITT Analysis of Lithium Insertion Cathodes for Determining the
13. M. Kim, H. Chun, J. Kim, K. Kim, J. Yu, T. Kim, and S. Han, “Data-efficient Lithium Diffusion Coefficient at Low Temperature: Challenges and Pitfalls.”
parameter identification of electrochemical lithium-ion battery model using deep J. Electrochem. Soc., 167, 090546 (2020).
Bayesian harmony search.” Appl. Energy., 254 (2019). 41. A. Zülke, Y. Li, P. Keil, and H. Hoster, “Communication-Why High-Precision
14. S. Park, D. Kato, Z. Gima, R. Klein, and S. Moura, “Optimal Experimental Design Coulometry and Lithium Plating Studies on Commercial Lithium-Ion Cells Require
for Parameterization of an Electrochemical Lithium-Ion Battery Model.” Thermal Baths.” J. Electrochem. Soc., 166, A2921 (2019).
J. Electrochem. Soc., 165, A1309 (2018). 42. “Proposal for a new global technical regulation on the Worldwide harmonized Light
15. J. Sturm, A. Rheinfeld, I. Zilberman, F. B. Spingler, S. Kosch, F. Frie, and vehicles Test Procedure (WLTP).” World Forum for Harmonization of Vehicle
A. Jossen, “Modeling and simulation of inhomogeneities in a 18 650 nickel-rich, Regulations 162nd 15,, UNECE (2014).
silicon-graphite lithium-ion cell during fast charging.” J. Power Sources, 412, 204 43. G. Richardson and I. Korotkin, “Heat Generation and a Conservation Law for
(2019). Chemical Energy in Li-ion batteries.” Electrochimica Acta, 392, 138909 (2021).