Applied Thermal Engineering 197 (2021) 117260

Contents lists available at ScienceDirect

Applied Thermal Engineering

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

Incremental thermo-electric CFD modeling of a high-energy

Lithium-Titanate Oxide battery cell in different temperatures: A
comparative study
Eero Immonen, Jussi Hurri
Computational Engineering and Analysis Research Group, Turku University of Applied Sciences, Joukahaisenkatu 3, 20520 Turku, Finland

A R T I C L E I N F O A B S T R A C T

Keywords: In this article, we address the following question: Which of the many CFD models should a designer use for
Battery battery thermal analysis? We present a comparative study of four thermo-electric CFD models for a novel high-
Thermal management energy Lithium-Titanate Oxide (LTO) cell to be used in an electric rallycross car prototype. In contrast to many
CFD
related articles, the data required and the process of parameter identification are described in detail for each of
Equivalent circuit
Internal resistance
the models, and we also address different operating temperatures. Moreover, the models’ relative performance in
Mathematical model predicting cell behavior in constant current and varying current discharge situations is addressed by comparisons
Lithium-ion to cycling measurements and thermal camera imaging. The one key insight of this analysis is that, for purely
LTO thermal considerations, the accuracy of a simple constant internal resistance based heat generation model is
remarkably close to a dynamic equivalent circuit model, which is considerably more tedious to identify. A
limitation of this work is that we focus only on discharging, but the methods and models can also be used in
charging situations in the future.

1. Introduction representations (e.g. [12]). For CFD-based BCS design, researchers have
reported success by using both uniform (e.g. [13,14]) and non-uniform
1.1. Background (e.g. [15,16]) heat source representations for battery models, as well as
reduced-order linear system models (e.g. [17]).
Understanding the thermal behaviour of lithium-ion (Li-ion) batte­ With so many alternatives to choose from, the prospective thermal
ries in different operating conditions is instrumental for both designing engineer may find himself/herself confused about which numerical
and operating modern battery-powered systems. During the past few modeling approach would be the most suitable, in terms of accuracy and
decades, this understanding is cumulated into a myriad of mathematical simulation time, for his/her specific Li-ion battery application. The
models published in the academic literature, each describing different challenge is exacerbated by the fact that, due to the potential need for
aspects of Li-ion batteries. Today, battery cooling systems (BCS) that disassembly or expensive testing equipment, researchers often cite
attempt to maintain cell temperatures close to room temperature in parameter data for the models from previously published articles [18].
battery packs are typically designed based on such computational For some newer battery types, such information is not available, which
models and optimization [1–4]. Furthermore, on-board battery man­ may render the application of an otherwise suitable modeling approach
agement systems (BMS) may contain such numerical models that predict impossible. Also, perhaps surprisingly, researchers seldom describe in
cell temperatures for ensuring safety and durability of the battery system detail their model parameter identification procedures. In particular,
during runtime [5–7]. non-trivial data preprocessing steps are often omitted for the sake of
Depending on the prediction objective, as well as any practical conciseness. Lack of details about the parameter identification proced­
constraints on computational resources and availability of validation ure can also imply uncertainty about the accuracy of the reported
data, the battery model can be based on fundamental physics and parameter values. This can be problematic, as numerical methods, such
chemistry (e.g. [8]), equivalent circuits models (ECM) (e.g. [9]) or even as CFD are considered to require the exact form and parameters of the
black-box like treatment of measurement data (e.g [10]). These models underlying equations be known [19].
also vary in complexity from a zero-dimensional resistor–capacitor cir­ It is the purpose of this article to present and compare different
cuit representations (e.g. [11]) to full-scale 3D models combining thermo-electric CFD-based Li-ion battery models, in order to help the
Computational Fluid Dynamics (CFD) and electrochemical prospective thermal engineer choose the most suitable modeling

https://doi.org/10.1016/j.applthermaleng.2021.117260
Received 11 January 2021; Received in revised form 17 May 2021; Accepted 17 June 2021
Available online 30 June 2021
1359-4311/© 2021 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
E. Immonen and J. Hurri Applied Thermal Engineering 197 (2021) 117260

approach for his/her application, based on availability of measurement Table 1

data and requirements for accuracy. Throughout this work, we attempt List of models considered in the present article.
to carefully describe the model parameter identification procedures Model Description Data source Details
employed, and discuss the model parameters’ representation in CFD
Constant- Uniform heat source with Manufacturer SubSection
simulation applications. Our focus is on Lithium-Titanate Oxide (LTO) Rs constant internal resistance specification 3.2.1 and
battery cells, which are among the prime cell candidates for heavy off­ (Rs ) SubSection
road electric vehicle (EV) applications [20], but the computational 3.2.2
models, the data requirements and model identification procedures are SoC-Rs Uniform State of Charge CC discharge SubSection
(SoC) dependent heat source curves 3.2.1 and
generic and applicable to any other battery chemistry. based on internal resistance SubSection
3.2.3
NTGK Non-uniform SoC-dependent CC discharge SubSection
1.2. Contribution of the article heat source with algebraic curves 3.3.1 and
current- voltage relationship SubSection 3.3.2
ECM Non-uniform SoC-dependent DPPC test SubSection
In this article, we introduce and compare four thermo-electric CFD heat source with dynamic 3.3.1 and
models for a high-energy Lithium-Titanate Oxide (LTO) cell, namely current–voltage relationship SubSection
Toshiba SCiBTM 23Ah [21] described in Fig. 1, to be used in an electric 3.3.3
rallycross car prototype [22]. The models are listed in Table 1, and we
also address modeling hot (+40 ◦ C) and cold (-10 ◦ C) operation condi­
current in continuous use) notably distinct from each other. Thermal
tions, whereby the cell is also initially set at the given ambient tem­
management for fast charging is, then, an important open question left
perature. We focus on data requirements, the models’ parameter
for future work. The methods and models presented in this article can, in
identification and representation (by introducing Douglas-Peucker
the future, also be applied for describing charging situations, given data
approximation) in simulation models as well as comparison of the
for identification.
model accuracies in transient thermal CFD simulations. The set of
models presented herein is incremental: Starting from models based
only on public domain data (as available in manufacturer specifica­ 1.3. Relation to previous work
tions), we progress in steps towards more complex electric circuit rep­
resentations requiring data from Discharge Pulse Power We conclude the introductory section by relating the present work to
Characterization (DPPC) tests, and thermal chambers for identification. previously published research on CFD-based thermo-electric battery
Moreover, we present methods for data-based identification of the modeling, LTO battery cells and EV applications.
thermodynamic parameters. Although none of the presented simulation
model types is new, they have not been considered for LTO cells nor has 1.3.1. CFD-based thermo-electric battery models
their relative performance been previously compared in battery thermal Thermo-electric battery models with varying complexities have been
simulations, especially for different operating temperatures or non- presented in the academic literature; the reader is referred to [3,2] for
constant current cycles. In addition, to the authors’ knowledge, as recent surveys. Out of those employing CFD methods, Che et al. [23]
pointed out in the previous section, the present focus on parameter used OpenFOAM to model heat generation in a battery cell as a result of
identification methods and representation is new. Finally, the LTO power dissipation in the internal resistance Rs . A constant Rs does not
battery cell studied in the present work is lucrative for many offroad EV take into account transient phenomena, such as voltage sag, that may
applications (an example of which the electric rallycross application is) occur in practical battery use, so the extent of applicability of this
thanks to its high cycle life and ability to withstand rapid charging. To approach has been unclear. Nieto et al. [24] and Kim et al. [25]
the authors’ knowledge, this article is the first to report the parameters addressed the dependence of Rs on SoC, and De Vita et al. [13] also
and accuracies of different CFD-based thermo-electric models for this addressed its dependence on temperature. These models are reasonably
cell. lightweight and simple to implement, but assume a uniform heat source
A key limitation of this work is that we focus only on discharging the within the battery. In this article, we consider such battery thermal
cell, which represents driving in an EV application. The reason for this models, from the point of view of identification and model accuracy in
choice is that the LTO cells considered herein can reportedly withstand comparison to more complex non-uniform heat source models, com­
rapid charging at high current. This feature makes charging conditions plementing the recent study of Paccha-Herrera et al. [26].
(higher current for short period) and discharging conditions (lower Non-uniform heat source models allow for addressing

Fig. 1. Specification data and photograph of the cell in test conditions [21].

2
E. Immonen and J. Hurri Applied Thermal Engineering 197 (2021) 117260

inhomogeneity of the cells. In a series of papers [27–31] Kwon et al. and reportedly facilitate rapid charging (typically 80% of SoC in 6 min) with
Kim et al. introduced the NTGK model relating a volumetric current little degradation [51,49]. A disadvantage of LTO batteries is their low
transfer rate to the potential field by an Ohm’s law type static specifi­ inherent voltage, 2.3 V, which results in a relatively low energy density.
cation to be identified by polarization tests. This non-uniform current Many researchers have studied LTO cells in the recent years. Nemeth
density generates heat in the cell by overpotential and entropic heating. et al. [20] presented an extensive review of the properties of LTO cells,
The static NTGK model is considered suitable for conditions where along with characterization tests and aging considerations. Others have
changes in the electric load are slow, although no quantitative results of reported lumped-parameter models [52,43,53–55], material properties
such analyses seem to have been reported. In the present article, we [56], uniform heat source models [57], experimental cell characteriza­
study using the NTGK model also in a dynamic DPPC cycle with jump tion [58], aging [20], cold temperature modeling [59].
discontinuities in the discharge current profile. Of the models that However, only few authors seem to have proposed CFD-based
inherently contain mechanisms for describing the effect of dynamic and models for LTO cells. Murashko et al. [60] proposed a first-order ECM
rapid changes in the discharge current, perhaps the most widely used are model for an LTO pouch cell and hybrid mobile working machines,
ECMs. They attempt at mimicking the cell’s transient electric behavior focusing on parameter identification from pulse tests. However, they
by an electrical circuit (see e.g. [9,32,33]). Then, for example, load employed a constant-phase element which is not directly supported by
changes are not instantaneously seen in battery voltage. In the present many commercial solvers, such as ANSYS Fluent. Behi et al. [61] pro­
article, we describe both NTGK and ECM model parameters for a high- posed a battery thermal management system based on a heat pipe design
energy LTO battery cell, and focus on model parameter identification for prismatic LTO cells similar to the ones considered herein. They
and representation. Moreover, we compare the performance of these carried out CFD analysis using a compartment model in which the bat­
non-uniform heat source models to the above simpler uniform heat tery cell was divided into 9 homogeneous zones. Madani et al. [62]
source models, for the LTO cell in question. developed a CFD simulation model for an LTO pouch cell and studied the
As for battery model parameters, identification of the open circuit influence of temperature on its parameters. We complement these re­
voltage (OCV) from different measurement data streams has been sults by carrying out incremental thermo-electric CFD modeling of
studied by many authors. Dubarry et al. [34] considered the pseudo- prismatic LTO cells. Proceeding stepwise from simple uniform heat
OCV as the midpoint of charge and discharge voltage curves. Pattipati source models towards second-order ECMs, we carefully list the data
et al. [35] addressed dependence of OCV on temperature and age. Gurjer requirements, describe the model identification procedure and report
et al. [36] suggested that the variable OCV should be determined the models’ relative accuracy in battery thermal simulations.
through dynamic pulsing, such as the DPPC test considered herein, and
the same requirement has been specified for Rs , too [37,38]. However, 1.4. Electric vehicle applications
the OCV and Rs can, in fact, be identified from constant current
discharge curves. Curiously, though apparently well-known in practice, In this article, we consider battery cells for an electric rallycross car
this technique has been only sparsely documented in the academic prototype [22]. Rallycross is a form of sprint style automobile racing,
literature. The authors are only aware of the original work of Tiedemann with typical race duration lasting only minutes. The application can thus
and Newman [39] who did not consider coupling to CFD simulations, be seen as an example of non-road mobile machinery (NRMM) with
and those of Allu et al. [40] (as part a more complex model develop­ transient duty cycle characteristics [63].
ment), Chacko et al. [41] (who did not reference it to NTGK model), and Several reviews of battery thermal management for EVs have
Kosch et al. [42] (who described and applied it in a significantly more recently been published [3,2,1,64,65]. Moreover, a number of articles
complex extended polarization model). None of these authors addressed have been published on CFD-based battery thermal modeling for EV
the thermal aspects (e.g. imperfect knowledge of battery SoC in cold drive cycles (see e.g. [41,60] and the references therein) along with
conditions) of NTGK model identification, which we do in the present novel ideas for arranging cell cooling [66–68]. The incremental CFD
article. Moreover, here we clarify the identification methodology and based thermo-electric modeling presented in this article attempts to
data requirements, and we discuss accounting for thermal effects. We clarify the data requirements, model identification and relative model
also utilize this methodology for identifying the NTGK model parame­ performance, to help the prospective thermal engineer design and
ters for the LTO cell. Finally, for ECM parameter estimation, Kalogiannis implement better thermal management systems for EVs.
et al. [43] presented a comprehensive review of different approaches,
but did not address state space identification methods nor consider the 1.5. Organization of the article
effect of initial values of the electric subcircuits, which are our concern
in this article. This article is organized as follows. In Section 2, we introduce the
Analysis of two important CFD-based modeling approaches, related studied LTO cells, the equipment for their experimental testing and the
to the ones discussed above, is left for future work. In the present article, test specifications. In Section 3, we present the thermo-electric models
we do not address reduced-order models (ROM; e.g. [17,44–46]) due to and address methods for identifying their parameters. In Section 4, we
abundance of different model reduction techniques. Moreover, since the present a comparison of the models’ accuracies in static and dynamic
models considered herein already yield a reasonable accuracy for BCS current load situations. Finally, Section 5 presents the conclusions of this
design on the LTO cell, we leave analysis of more detailed models, such study and summarizes the proposed directions for future work.
as Newman’s pseudo-2D (P2D) porous electrode model (see e.g.
[47,48]), for future work. Further, we do not address frequency domain 2. Experimental testing
methods for parameter identification [43].
2.1. Cells tested
1.3.2. LTO cells
In this article, we consider thermo-electric modeling of LTO battery Experimental testing was carried out by using two identical Toshiba
cells [49]. Commercialized in 2008, LTO cells can withstand high cur­ SCiBTM LTO cells depicted in Fig. 1b. The cells were new and unused at
rents, repeated cycling and operation in cold temperatures, which make the time of starting the tests. The cell specification is shown in Table 1a,
them suitable for heavy offroad EV applications. Takami et al. [50] with data on the italicized rows obtained from the local supplier Celltech
carried out high-rate full cycling tests at 10C rate, and observed a ca­ Oy.
pacity retention of 95% after 30000 cycles. They also recorded a
discharge capacity retention of 80% at -30 ◦ C relative to room temper­
ature use, indicating decent cold temperature performance. LTO cells

3
E. Immonen and J. Hurri Applied Thermal Engineering 197 (2021) 117260

2.2. Testing equipment

Cell tests were carried out on the Chroma 17011 regenerative

charge/discharge test system. It was used in CC (constant current) and
CV (constant voltage) charge and discharge tests, as well as DPPC tests
with programmed time-dependent discharge current rates. The test
system is capable of collecting data at 10 ms sampling rate. The test
equipment has 24 100A channels and 16 6A channels, supporting par­
allel connection, but the maximum current under investigation was 92A,
i.e. 4C. As usual, here the charge (discharge) rate of a battery is specified
as C-rate, such that NC is the current required for charging (discharging)
the cell from empty (full) to full (empty) in N1 of an hour. Thus, for a 23
Ah battery, 1C is 23A, 2C is 46A etc.
During testing, the cells were either placed on a table at room tem­
perature at 21 ◦ C (Fig. 1b) or in the Espec PL-2KP environmental cabinet
for testing in hot (40 ◦ C) or cold (-10 ◦ C) conditions. Temperature
measurement was carried out by using a thermocouple clamped at the
positive terminal in all tests (the gray wire in Fig. 1b). Thermal imaging,
Fig. 2. The DPPC test sequence for C-rate 1: CC discharge to empty, CC charge
carried out by using Flir E5-XT camera, was used for validating the cell
to full and sequential discharge to empty in 10% SoC steps. Negative C-rate
surface temperature distribution. To minimize the effect of a reflective denotes discharging and positive C-rate denotes charging. For clarity, the figure
surface, one cell was painted and the other covered by duct tape for also shows the preparations preceding the actual DPPC test.
thermal imaging. Results for the painted cell are included in this article.

2.3. Battery tests

2.3.1. Pre-conditioning
To prepare the two new and unused cells for actual cycling tests, to
identify possible manufacturing defects and differences between the
cells, and to determine the actual capacities of the cells, a pre-
conditioning test was first carried out. The pre-test, repeated 5 times
for each cell, follows the procedure suggested by Valmet Automotive
Inc: Rest 1 min; CC charge at 0.2C (4.6A) until voltage above 2.7V; CV
charge at 2.7V until current below 1.15A; Rest 30 min; CC discharge at
0.2C (4.6A) until voltage below 1.5V; CV discharge at 1.5 V until current
below 1.15A; Rest 15 min. The specified cell capacities were also
confirmed at room temperature.

2.3.2. Constant current charge–discharge cycle test at room temperature

Constant current charge–discharge cycle tests attempt to describe the
battery cell’s overall voltage response to static discharge current, as the
cell is taken from full charge to empty at a constant rate. Several CC
Fig. 3. A 4C DPPC test sequence used for out-of-sample model validation.
charge–discharge cycle tests were carried out at room temperature
conditions for C-rates 0.2C, 0.5C, 1C, 1.5C, 2C, 2.5C, 3C, 3.5C, and 4C.
For C-rate 1, the cycle test is as follows: CC charge at 23A until voltage 2.3.4. Thermal tests
above 2.7V; CV charge at 2.7V until current below 1.15A; Rest 120 min; The effect of hot (+40 ◦ C) or cold (-10 ◦ C) ambient temperature was
CC discharge at 23A until voltage below 1.5V; Rest 15 min. The other C- tested in the environmental cabinet in CC discharge tests for C-rates 1C,
rates were treated analogously. 2C and 4C. The cell, initially charged to full at room temperature, was
first placed in the environmental cabinet overnight for the temperature
2.3.3. Discharge Pulse Power Characterization (DPPC) test to settle to the set ambient conditions. Then, for C-rate 1C, the test
The DPPC tests considered herein attempt to describe the battery involved CC discharge at 23A until voltage dropped below 1.5V. The
cell’s dynamic (i.e. time-dependent or transient) response to changing procedure for the other C-rates was analogous. We emphasize that since
discharge current. The DPPC tests considered in this study were adapted the climate cabinet attempted to maintain a constant temperature in the
from the well-known Hybrid Pulse Power Characterization (HPPC) tests test compartment by fan-based air distribution, the convective heating
(see e.g. [69] and the references therein). They were carried out at room or cooling was potentially slightly different from the room temperature
temperature, for C-rates 1C and 4C. For all DPPC tests, the battery was test conditions (described in SubSection 2.3.2). The cabinet manufac­
initially fully discharged to empty, then charged to full as in the CC turer does not, however, disclose the mechanism nor any associated air
charge–discharge tests, and left to settle. In the actual DPPC test, for C- flow rates that could be used for processing the measurement results.
rate 1C, the following two steps were repeated 10 times or until voltage
was below 1.5V (cf. Fig. 2): CC discharge at 23A for 6 min; Rest 60 min. 3. Battery modeling
For C-rate 4C, the following four steps were repeated 10 times or until
voltage was below 1.5V (cf. Fig. 3): CC discharge at 92A for 10 s; Rest 3.1. Basic principles
290 s; CC discharge at 23A for 320 s; Rest 60 min. The sampling interval
in pulse tests was 50 ms and 1 s during rest periods. To extend the lifetime of a battery cell and to ensure maximal power
output, the cell should be maintained at close to room temperature in all
use conditions [70]. Battery cells generate heat during charge and
discharge due to enthalpy changes, electrochemical polarization and

4
E. Immonen and J. Hurri Applied Thermal Engineering 197 (2021) 117260

resistive heating [71], but in cold conditions, they may also require 3.2.3. State of charge dependent Rs (SoC-Rs)
preheating to be fully operational [72]. In this section, we present an In practice, the cell’s internal resistance Rs (and open circuit voltage
incremental set of models for describing thermal phenomena within V0 ) are not constant but depend on the cell’s State of Charge (SoC) and
battery cells with and without coupling to electrochemistry. We consider temperature. Some authors suggest that SoC-dependent values should be
two groups of models, characterized by whether or not heat generation obtained from the measured immediate voltage drops at current pulses
within the cell is assumed to be uniform, and display the modeling steps during a DPPC test or equivalent; see e.g. [43,73] and Fig. 9. However,
as well as the obtained parameter values for the LTO cells under for modeling cell heating by a single resistance Rs , this approach may not
consideration. be optimal. Indeed, a full equivalent circuit, as in Fig. 4, describing the
dynamical response of the cell, typically contains two or more resistive
components which also contribute to heat generation in the cell.
3.2. Uniform heat source models Consequently, the internal resistance values estimated from the voltage
drop at current pulses may be artificially low (as demonstrated in Fig. 5).
3.2.1. Uniform heat generation within cell Thus, in the present work, we propose to obtain the SoC dependent
According to the Helmholtz-Thévenin theorem, a battery can be values for Rs from linear regression models obtained from at least three
represented as a voltage source V0 in series with an internal resistance Rs distinct CC discharge curves, which are also more widely available from
(as in Fig. 4 without subcircuit V12 ). In this simplistic model, heat gen­ cell manufacturers than DPPC test curves. The same approach can then
eration within a battery cell occurs at rate Q̇ = P = Rs I2 [W], where I is be used for modeling different operating temperatures. The process is
the current through the cell. The battery cell is assumed homogeneous, explained in detail in conjunction with the NTGK model; see SubSection
and thus heat transfer within the cell is governed by the partial differ­ 3.3.2.
ential equation (PDE) for heat conduction: Fig. 5 displays the resulting internal resistance Rs as a function of
∂T SoC. In room temperature conditions, the average of Rs across different
ρc p − ∇⋅(k∇T) = Q̇ (1) SoC values is approximately 1.5 mΩ. This is close to the value (1.3 mΩ)
∂t
calculated from manufacturer specifications in SubSection 3.2.2.
with convective heat transfer boundary condition: Further, when the cell is operated at high SoC values, its internal
resistance is close to that reported by the manufacturer. Clearly, the
∂T
k = h(Ta − T) (2) internal resistance increases with reduction in temperature, and the
∂n
effect is nonlinear in SoC. This is consistent with observations made in
[ ] the literature [59].
Here ρ denotes the cell density mkg3 , cp denotes its specific heat capacity
[ ]
J
3.3. Nonuniform heat source models
kg⋅ ◦ C , T = T(x, y, z, t) denotes the cell temperature [ C], k denotes the
◦

[ ] In practice, the battery cell is not homogeneous and thus the heat
cell’s thermal conductivity m⋅W◦ C , h denotes the total heat transfer co­ generation within it is not uniform. The Multi-Scale Multi-Domain
[ ] (MSMD) approach to battery cell modeling deals with different physics
efficient on the cell surface m2W⋅ ◦ C and Ta denotes the ambient (free- in different solution domains in order to predict the cell temperature
distribution at the battery length scale [74]. This approach also con­
stream) temperature [◦ C]. Since Q̇ is constant throughout the battery
siders the electrical fields in the positive and negative electrodes sepa­
cell domain, also the other parameters can, in this modeling approach,
rately, assuming that the battery micro layers have the same orientation
be approximated by their average values. The main challenge of estab­
(such as in prismatic cells considered herein).
lishing representative values for Rs and the above physical constants is
addressed in the following sections.
3.3.1. Nonuniform heat generation within cell
MSMD battery models are governed by a system of coupled PDEs for
3.2.2. Constant Rs from manufacturer specification (Constant-Rs)
thermal and electrical fields within the cell. The cell temperature T =
Cell manufacturers typically report reference values for the internal
resistance Rs . Thus, preliminary estimates for the rate of heat generation
can often be obtained without measurements. For the cells considered
herein, the manufacturer only reports a maximum output power Pmax =
1000 W, reportedly calculated based on the internal resistance. The
reported nominal cell voltage V0 = 2.3 V then yields, by the Maximum
V2
Power Transfer Theorem, Rs = 4Pmax
0
≈ 1.3 mΩ.

Fig. 5. Battery internal resistance Rs as a function of SoC for different tem­

Fig. 4. Battery representation as an equivalent circuit: Thevénin model peratures. The squares represent the Rs values obtained from the measured
(voltage source V0 and Rs alone) and full ECM (dashed circuit V12 included). voltage drops V0 − Rs I at current pulses in a DPPC test at C-rate 1.

5
E. Immonen and J. Hurri Applied Thermal Engineering 197 (2021) 117260

T(x, y, z, t) is described by the PDE (see e.g. [28]): current Jm = jm W, m = 1, 2, 3, respectively (cf. Fig. 6a). Under CC
jWt
∂T
(
dU
) discharge (starting from fully charged state) we have DoD(t, j) = 3600Q 0
,
ρcp − ∇⋅(k∇T) = σ + |∇ϕ+ |2 + σ − |∇ϕ− |2 + j U − (ϕ+ − ϕ− ) − T so that for every constant discharge current J = jW, there is a 1 − 1
∂t dT
(3) correspondence DoD→tJ (DoD) between discharge time t and DoD. We
can thus define a function ψ : {Jm }3m=1 × [0, 1]→[0, ∞) : (Jm ,
with the convective heat transfer boundary condition (2). The electrical DoD)→Vm (tJm (DoD)), i.e. ψ maps a discharge current and a depth of
field on the electrodes is described by the PDE system: discharge to the corresponding voltage seen in the measurements. For
each DoD ∈ [0, 1], we can then find three unique current–voltage pairs
∇⋅(σ + ∇ϕ+ ) = − j (4a)
(Jm , ψ (Jm , DoD)), m = 1, 2, 3, from the CC discharge curves. The NTGK
∇⋅(σ − ∇ϕ − ) = j (4b) specification (5) is then obtained, at known jW = Jm and V = ψ (Jm ,
DoD), from a least-squares regression line across these points: U(DoD) is
Here σ + and σ− are the effective electrical conductivities [mS ]for the 1
the intersection of the vertical axis and Y(DoD) is the slope of the line (cf.
positive and negative electrodes, ϕ+ and ϕ− are phase potentials [V] for Fig. 6b). Three measurement points are needed to confirm the line fit by
the positive and negative electrodes, and j [mA3 ] is the volumetric current a high enough R2 value (a poor line fit indicates erroneous measurement
transfer rate. In this specification, the material constants cp and k are data or nonlinearity not captured by the NTGK Eq. (5)). It is evident that
volume averages of the cell’s constituent components. Moreover, k is U = V0 and Y = R1s , so the proposed method applies for identification of
typically anisotropic, displaying different values for in-plane and
the SoC-Rs model (SubSection 3.2.3).
through-plane directions. The predicted battery voltage V = ϕ+ − ϕ− .
Using the above formalism, the effect of ambient temperature can be
To close these equations, the unknown inputs U and j, which depend on addressed through apparent capacity Q(T). If the cell capacity Q0 =
the battery type and operation, must be addressed. The following sub­ Q(T0 ) is known at some temperature T0 (usually room temperature as is
sections describe the NTGK and ECM models for this purpose. DJ (T)
the case here), then we can simply scale Q(T) = Q(T0 )DJCC(T ), where
CC 0

3.3.2. Newman-Tiedemann-Gu-Kim model (NTGK) DJCC (T) denotes the duration of a CC discharge test at current J and
The hypothesis of the NTGK model is that the volumetric current temperature T. Fig. 7 displays the U and Y curves obtained this way for
transfer rate j is related to the battery voltage by the following algebraic T = − 10 ◦ C, 21 ◦ C and 40 ◦ C, with T0 = 21 ◦ C and capacity Q0 = 23Ah
“polarization expression” [30]: confirmed in measurements. Here, Q( − 10 ◦ C) = 18.8Ah and Q(40 ◦ C)
= 24Ah. The Y curves in Fig. 7b are reciprocals of the Rs curves in Fig. 5.
λY(V − U)
j= (5) Since the U and Y curves can have local minima or maxima (as in
W
Fig. 7 at DoD ≈ 40%), a polynomial approximation (Eqs. (6)) is not
where W is the volume of the active material and λ = QQ0r is a scaling factor necessarily an accurate way to represent these functions. Instead, we
propose to use piecewise linear approximations obtained by the
to be applied if the cell simulated has a nominal capacity Q0 [Ah]
Douglas-Peucker (DP) algorithm [75]. The DP algorithm can be used in a
different from the reference cell capacity Qr [Ah] used for the mea­
parametric search to find piecewise linear approximations, with at most
surements. Throughout this article, we have λ = 1. The functions U [V]
N points (or N − 1 line segments), to the original curves. Instead of
and Y [Ω1 ] are labelled in the literature to as “fitting parameters” [28]
straight line segments of constant width, the DP algorithm attempts to fit
that are specified as polynomial functions of depth of discharge (DoD
variable-length line segments to optimally describe the underlying
=1– SoC), with coefficients ak , bk to be identified from measurements:
function. Fig. 7 displays 20-point DP approximations for the U and Y
U = U(DoD) = ΣNk=0 ak DoDk (6a) functions. For reference, a degree 5 polynomial approximation is also
shown; the DP approximation fit is clearly better and it can be imported
Y = Y(DoD) = ΣNk=0 bk DoDk (6b) into simulation software (e.g. ANSYS Fluent 2019R3) directly from
tables.
In spite of the apparent simplicity of the NTGK law (5), in practice
there appears to be no single accepted way to apply it, especially for 3.3.3. Equivalent Circuit Model (ECM)
varying operating temperatures. Indeed, instead of volumetric current The NTGK model assumes that changes in discharge current are
rates, as in Eq. (5), many authors consider current densities per unit area immediately seen in the battery voltage, while, in practice, the cell’s
[27], whereas battery tests report current in amperes without reference voltage response to changing currents is dynamic (consider, for
to area or volume. Moreover, by definition, example, the well-known recovery phenomenon [76]). To describe the
∫t
DoD = DoD(t) = DoD(0) +3600Q 1
0 0
Wj(t)dt depends explicitly on time dynamic response of the cell voltage to time-varying changing discharge
and implicitly also on temperature (through Q0 ); Especially for cold currents, different Equivalent Circuit Models have been proposed in the
temperatures, Q0 depends significantly on temperature. Finally, literature (see e.g. [9,43]). In the present work, we focus on second-
although the Weierstrass Approximation Theorem guarantees that order systems, as in Fig. 4, at room temperature. The idea of ECM
polynomials can be used for uniformly approximating any continuous modeling is to regard the cell as a dynamical input–output system, with
function at any accuracy, it does not imply that a low-order polynomial parameters V0 , Rs , R1 , C1 , R2 , C2 , and driven by the discharge current
(as supported by commercial CFD solvers) would be sufficient for this input. The modeling challenge is, of course, in accurate identification of
purpose. these parameters that depend on the cell SoC.
The parameter identification problem for U and Y, described above, Let V12 = V1 +V2 denote the voltage across the subcircuit consisting
has been considered by Allu et al. [40], Chacko et al. [41] and Kosch of R1 , C1 , R2 , C2 , i.e. the dashed region in Fig. 4. Since the measured cell
et al. [42], and we attempt to summarize and clarify it here, also voltage V(t) can be represented as (by convention I < 0 in discharge)
addressing its relation to SoC-dependent V0 and Rs parameters (Sub­ V(t) = V0 (SoC) + Rs (SoC)I(t) − V12 (R1 ,C1 ,R2 ,C2 ,t), the piecewise linear
Section 3.2.3). Moreover, we consider representation of the functions time-invariant (LTI) state space system (A, B, C) to be identified is:
U(DoD) and Y(DoD) in simulation software as well as the effect of
ẋ(t) = Ax(t) + Bu(t), x(0) = x0 (7a)
temperature on reduction of available capacity.
For every operation temperature Ta considered, the parameter
y(t) = Cx(t) (7b)
identification procedure requires a minimum of 3 CC discharge voltage
curves Vm (t),m = 1,2,3, each measured at a different constant discharge

6
E. Immonen and J. Hurri Applied Thermal Engineering 197 (2021) 117260

Fig. 6. Using three CC discharge curves to identify the NTGK model parameters at room temperature: Every DoD point corresponds to a line obtained by regression
on three current–voltage pairs. Selected DoD points are shown in the discharge curves (left) as squares, and the corresponding lines are shown on the IU chart (right).

Fig. 7. Fitting parameters U and Y for the NTGK model: The squares correspond to a 20 point Douglas-Peucker piecewise linear approximation. A degree 5 poly­
nomial fit for Y (at Ta = 21 ◦ C) is shown for comparison.

with step in the DPPC test, we can readily employ well-known state space
( ) methods for identification. Here, we attempt to identify the initial state
V1 (t)
x(t) = , u(t) = I(t), y(t) = V12 (t) (8) x0 at 100% SoC as part of the optimization problem, and proceed
V2 (t)
sequentially towards 0% SoC by treating the final value of x(t) of the
previous step as the initial value of the next step.
and
Using MATLAB and the built-in ssest function targeting prediction
⎛ ⎞ ⎛ ⎞
1 1 ( ) error minimization, a stable disturbance-free modal (i.e. diagonal) sys­
⎜− R C 0 ⎟ ⎜C ⎟
⎜ 1 1 ⎟ ⎜ 1⎟ V 01 tem ( A,
̂ B, ̂ without feedthrough is sought for the 9 SoC steps. Since the
̂ C)
A=⎜ ⎟, B = ⎜ ⎟, C = (1 1 ), x0 =
⎝
0 −
1 ⎠ ⎝ 1 ⎠ V 02 function produces diagonal systems, for each step A
̂ = A and the other
R2 C2 C2 ssest outputs B and C can be easily converted to those in (9) with a
̂ ̂
(9)
pre-specified C: Equating transfer functions implies B(i) ̂
̂ C(i) = B(i)C(i)
where also the initial state x0 depends on SoC, and cannot be guaranteed = B(i), with i = 1,2. The resulting SoC dependent parameters are shown
to be zero, an implicit assumption in some related articles. Here, too, in Fig. 8. The model fit, expressed as 100(1 − NRMSE)% (normalized
1
∫t root mean squared error expressed in percent), varies between
SoC(t) = SoC(0) − 3600Q I(t)dt.
0 0 72% − 95%. The best fit is seen close to 50% SoC, and the fits gradually
For identification, the first step is to construct the output y(t) = worsen moving towards either extremal SoC value, 0% and 100%.
V12 (t) = V(t) − V0 (SoC) +Rs (SoC)I(t) by extracting Rs and V0 at discrete
SoC values from the instantaneous step changes in voltage at DPPC
sequence step edges (see Fig. 2 and Fig. 9). Linear interpolation yields 3.4. Thermodynamic parameters
V0 (SoC) and Rs (SoC) for 0⩽SoC⩽100%, and hence we obtain the target
outputs V12 (t) for the entire DPPC test (see Fig. 9 for an example). Thus far we have only addressed modeling the heat generation rate
Assuming that R1 , R2 , C1 , C2 remain constant, and that V0 and Rs vary within the LTO cell in different thermo-electric specifications. To com­
linearly, during each 10% SoC step in the DPPC test, for each 10% SoC plete them, and solve the corresponding PDEs, a number of material-
specific parameter values as well as the rate of heat dissipation to

7
E. Immonen and J. Hurri Applied Thermal Engineering 197 (2021) 117260

Fig. 8. ECM parameters obtained by state space LTI system identification. The figures show spline smoothed values and 20 point DP approximations that can be
directly imported to CFD simulation software.

ambient air must be identified. Authors often extract these parameters 3.4.1. Thermodynamic parameters identified from data
from the literature, but such data may not always be available. We thus In this subsection, we discuss three data-based methods for identi­
aim to demonstrate the extent to which the same measurement data used fying the specific heat capacity cp and convective heat transfer coeffi­
for identifying the models Constant-Rs, SoC-Rs, NTGK and ECM can cient h. The first two utilize a differential equation based approach,
also be used for identification of thermodynamic parameters. whereas the last one employs statistical design of experiments and CFD
simulations.
Let us first consider the battery cell as a lumped-parameter (LP)

8
E. Immonen and J. Hurri Applied Thermal Engineering 197 (2021) 117260

the inputs h = hk and cp = ckp . For a sufficiently large NLHS (see [77] for
details), a polynomial response surface p(h, cp ) ≈ e(h, cp ) can fitted to
this data using MATLAB’s built-in fitlm function. Constrained multi­
variable function minimization, e.g. via the MATLAB function fmin­
con, then yields such estimates for h and cp that minimize the error (13).
Table 2 shows the results of data-based thermodynamic parameter
identification for the three methods described above, at room temper­
ature. We utilized the CC discharge test data described in SubSection
2.3.2 and SubSection 2.3.4, totalling 3 data sets. For LHS, the number of
simulations NLHS = 200. For a similar LTO cell, Behi et al. [61] recently
reported cp = 1150 kg⋅J◦ C at room temperature. Moreover, although
natural convection is highly dependent on the geometry of the surfaces,
it typically occurs at h = 2…25 m2W⋅ ◦ C. Consequently, the values reported
in Table 2 for room temperature conditions appear to be reasonable.
The above approach was also used for identifying parameters for the
low (-10 ◦ C) and high (40 ◦ C) temperature conditions. However, there
were some deviations in the resulting parameter values across the
Fig. 9. The 90%→80% SoC step in the DPPC test at C-rate 1: The cell is dis­ methods. They can be attributed to both differences in battery operation
charged for 6 min and left to recover. Parametric identification, carried out for characteristics in subzero temperatures [78] and also to differences in
V12 (t), yields a close correspondence between measured battery voltage and
flow conditions in the climate chamber. In the absence of thermody­
that predicted by the LTI model (7): The model fit for this SoC step is 87%.
namic measurement data from the literature to validate the results, we
utilized cp = 1100…1700 kg⋅J◦ C, h = 2…12 m2W⋅ ◦ C in the simulations. That
system, which means that the cell temperature (measured at a terminal)
cp varies across computational Li-ion battery models is not uncommon
is assumed to be uniform across the cell. Then an elementary thermo­
[79], but the uncertainty in h should be addressed in a future study:
dynamic consideration shows that heat transfer during CC discharge is
Seemingly minor changes in the test layout (such as covering a surface or
governed by the ordinary differential equation (ODE):
forced cooling air flow in climate chamber) can significantly alter h, and
mcp
dT
= hA(Ta − T) + P = hA(Ta − T) + Rs I 2 (10) this variable appears to be the most significant one for thermal model
dt accuracy.

Integration of Eq. (10) from the start of discharge (at t = 0) to the end of
3.4.2. Thermodynamic parameters from the literature
discharge (at t = te ) yields the averaged (ALP) system:
In the absence of suitable measurement equipment, values for ther­
mcp (T(te ) − T(0)) = hA(Ta − Tave )te + Rs I 2 te (11) mal conductivities (in-plane kx , kz , through-plane ky ) and electrical
conductivity (σ ) of the active material were obtained from literature. For
where Tave ≈ 12 (T(te ) − T(0)) represents an average cell temperature thermal conductivity, we used the values reported by Behi et al. [61] for
during discharge. CC discharge measurements for different C-rates a similar cell: kx = kz = 30 m⋅W◦ C and ky = 0.8 m⋅W◦ C. For electrical con­
provide numerical values for the constants te , T(0) and T(te ). With the ductivity, as suggested by many authors (see e.g. [2,18]), we imple­
constants m, A and Rs available from the manufacturer’s specification mented an approximation by weighted averages over the cell’s interior
documents, Eq. (11) results in a matrix equation Ax = b, with x = structure. In the present study, the cells could not be disassembled, and
( )
h hence thicknesses were approximated and material data from similar
. This can be solved by using the least squares method, such as the
cp cells were employed. In the LTO cell, the current collectors for the
backslash operation A\b in MATLAB. This method is referred to as ALP negative and positive electrodes are made from the aluminum. Conse­
in what follows. quently the value σ = σ+ = σ− = 2.74⋅106 mS was used for the active
On the other hand, the unique solution of the LP ODE (10) is: material in simulations.
⎛ ⎞
1 ⎜ 2 t⎟
4. Results
(12)
− Ah
T(t) = ⎝Rs I + AhT a − Ce mcp
⎠, 0⩽t⩽te
Ah
In this section, we discuss the accuracy of the thermo-electric models
Constant-Rs, SoC-Rs, NTGK and ECM using the parameter values
with C = Ah(Ta − T(0)) + Rs I2 . The second, slightly more complex, described in Section 3. Since the NTGK specification is the simplest one
method we consider for identifying h and cp , called LP, is by a parametric to address non-uniform heat generation in the cell, it is considered the
curve fit of the LP solution T(t) in Eq. (12) to measurement data Tm (t). baseline model to which the others are compared.
The built-in function lsqcurvefit in MATLAB implements a trust-
region algorithm, which minimizes the error:
tk =te
∑
e(h, cp ) = ||T − Tm ||2 = ||T(tk ) − Tm (tk )||2 (13)
tk =0
Table 2
[ ] [ ]
J W
at time instants tk where temperature measurements are available. Table of cp and h values identified from the measure­
kg⋅ C
◦ m ⋅ C
2 ◦
In the last method, called LHS, considered herein for identifying
ments (Ta = 21 ◦ C).
thermodynamic constants from data, the NTGK CFD model is simulated
Method h cp
for different h = hk and ckp , k = 1, …, NLHS values which are obtained
from a Latin Hypercube Design matrix. Each simulation then provides a ALP 1.2 1122
LP 2.7 1050
sequence of predicted average cell temperatures T(t),0⩽t⩽te , and hence
LHS 8 1128
a noisy sample ek = e(hk , ckp ) of the error function (13) corresponding to Average 4.0 1100

9
E. Immonen and J. Hurri Applied Thermal Engineering 197 (2021) 117260

4.1. Simulation environment 4C, 2C, and 1C discharge rates were used for model identification,
whereas the other curves are presented for out-of-sample model vali­
Throughout this article, the ANSYS Fluent 2019 R3 CFD code was dation. Clearly, the NTGK model can accurately capture the voltage
used for thermo-electric simulations. A geometry model was first created response in CC discharge, across the operating temperatures considered.
in ANSYS SpaceClaim according to the cell’s physical dimensions (Fig. 1 Also, considering the entire discharge process, the temperature predic­
hexahedral computational mesh (shown in Fig. 10) of 9312 cells, tar­ tion by the NTGK model is accurate even on unseen data. However, the
geting typical edge length of 3 mm, was subsequently created in ANSYS model fails to capture the temporary temperature peak that is prominent
Meshing for CFD simulations. The active material and the terminal tabs especially on low discharge currents at about 60% SoC (cf. Fig. 11c).
were included as separate cell zones. Table 3 summarizes the average accuracy of the thermal predictions
The Constant-Rs model was implemented in Fluent as a constant of the Constant-Rs, SoC-Rs and NTGK models for the CC discharge
volumetric heat source in the active zone. The varying internal resis­ cases. Here, average is taken across the entire discharge cycle time, and
tance in the SoC-Rs model was implemented in Fluent as a user-defined for the NTGK model, the corresponding temperature curves are shown
function (UDF) producing a volumetric heat source in the active zone. in Figure 11a, Figure 11c and Figure 11e. Although the accuracy of the
The built-in MSMD model implementations in Fluent were used for the NTGK model is the best one, even the computationally simpler Con­
NTGK and ECM models. Here, the user specifies the SoC-dependent stant-Rs and SoC-Rs perform almost equally well for purely thermal
model parameters directly through a graphical user interface in considerations.
Fluent, either as polynomials or tables: For the NTGK model, the input Fig. 12 shows a comparison between the cell surface temperature
parameters were the fitting parameters U and Y in Fig. 7, and for the predicted by the NTGK model and the profile obtained from thermal
ECM model, they were the LTI circuit parameters V0 , Rs , R1 , R2 , C1 and imaging, at the end of the CC discharge test at 1C. Even though the
C2 given in Fig. 8. For a better curve representation accuracy (see predicted temperatures are approximately 0.5 ◦ C-1 ◦ C lower than those
Fig. 7b), in the simulations, the NTGK and ECM electric model param­ seen in the measurements, the temperature distributions have similar
eters were represented as tables using the 20 point Douglas-Peucker characteristics. In particular, the highest temperature is recorded near
approximation described in SubSection 3.3.2. In both of these MSMD the negative terminal in both simulation and measurement. We
representations, the heat source specified for the active material is that emphasize that the simpler Constant-Rs and SoC-Rs models would
given in Eq. (3), with the electrical variables obtained from the predict uniform temperature generation within the cell, and hence, a
respective submodel. uniform surface temperature under the given boundary conditions. On
All simulations were carried out by using double-precision solver the other hand, by Fig. 12a, the thermal image shows an approximately
arithmetic on a laptop computer with 1.9 GHz CPU running on 4 cores 1 ◦ C deviation in the cell surface temperature, the trend of which is
and 16 GB memory. The time step size was 30 s for all CC discharge reproduced by the NTGK model in Fig. 12b.
simulations, whereas for dynamic loading simulations, the time step size
was reduced to 1 s in the vicinity of step changes in the discharge 4.3. Simulation of a dynamic discharge profile
current.
The ECM is a dynamical model facilitating prediction and simulation
4.2. Simulation of CC discharge of transient discharge situations with rapidly changing currents. The
NTGK model, on the other hand, is a static specification without tran­
Fig. 11 displays the NTGK simulation model predictions and sients. Fig. 13 displays a comparison of the predicted voltage and tem­
measured values for CC discharge at various ambient temperatures. The perature response of the LTO cell to the previously unseen (out-of-
sample) 4C DPPC discharge profile shown in Fig. 3. Clearly the ECM
model can accurately predict the cell voltage, which the NTGK model
cannot do (Fig. 13a). However, both of these models are approximately
equally good at predicting the cell temperature (Fig. 13b). This obser­
vation, along with the data in Table 3, clearly suggest that simple in­
ternal resistance based models may well suffice even for practical
battery thermal simulations with dynamic loading profiles.

5. Conclusions

In this article, we have presented a comparative study of four

thermo-electric CFD models, called Constant-Rs, SoC-Rs, NTGK and
ECM, for predicting the behavior of a Lithium-Titanate Oxide battery
cell that is to be used in an electric rallycross car prototype. We have
specifically focused on describing the data requirements and steps for
model parameter identification, while also addressing the effect of high
and low operating temperatures. Model validation was carried out by
comparison to constant current discharge measurements and thermal
imaging. Model performance was also evaluated on previously unseen
test data. All of the presented experimental tests were carried out as a
part of this study, and the purpose of the whole article is to help the
thermal engineer help choose the most suitable CFD model for battery
thermal analysis. The presented results are new for LTO cells, and the
methods and models are applicable to any battery chemistry.
Fig. 10. The computational mesh for the Toshiba SCiBTM cell. The negative The ECM model demonstrated superior performance over the NTGK
terminal (shown in blue) and positive terminal (shown in red) are made of model in predicting the cell voltage on unseen transient current cycle
aluminum. All source terms are implemented in the active material continuum data, and both of the simpler internal resistance based models Constant-
(shown in gray). All surfaces are subjected to the convective thermal boundary Rs and SoC-Rs would perform even worse. The reason is twofold. First,
condition, cf. Eq. (2). the ECM model is a dynamical system employing differential equations

10
E. Immonen and J. Hurri Applied Thermal Engineering 197 (2021) 117260

Fig. 11. NTGK simulation model predictions [s] and measured [m] values for CC discharge at various ambient temperatures.

for describing the electrical phenomena, while the others are algebraic that the Rs -based models require less (and perhaps more standard) data
electrical models. Second, the Constant-Rs and SoC-Rs models do not than the ECM model, it would thus seem sufficient to use them for
address dependence of the open-circuit voltage on the cell state of practical battery thermal design and analysis. Indeed, reasonable accu­
charge. The NTGK model does address it, but a step change in current is racy in thermal predictions was achievable by using the Constant-Rs
immediately seen in the voltage, which neglects, among others, the re­ model and data from manufacturer specifications alone.
covery effect in the DPPC test. However, for purely thermal considerations, An important aspect for future work is addressing battery cell
the accuracy of all models was similar across the test data. Given the fact charging, as in this article we have focused on discharge scenarios under

11
E. Immonen and J. Hurri Applied Thermal Engineering 197 (2021) 117260

Table 3
Mean errors (◦ C) in temperature prediction for CC discharge tests
C-rate 0.5 1 1.5 2 2.5 3 3.5 4 Average

Constant-Rs (21 C) ◦
0.25 0.62 0.24 0.85 0.77 0.89 0.67 2.02 0.79
Constant-Rs (40 ◦ C) - 0.41 - 0.4 - - - 1.95 0.92
Constant-Rs (-10 ◦ C) - 0.18 - 2.7 - - - 6.14 3.0
SoC-Rs (21 ◦ C) 0.3 0.48 0.27 0.57 0.4 0.43 0.27 1.44 0.52
SoC-Rs (40 ◦ C) - 0.51 - 0.32 - - - 1.8 0.88
SoC-Rs (-10 ◦ C) - 0.79 - 0.35 - - - 0.14 0.43
NTGK (21 ◦ C) 0.45 0.54 0.25 0.38 0.29 0.34 0.6 0.64 0.43
NTGK (40 ◦ C) - 0.41 0.41 - - - 1.2 0.67
NTGK (-10 ◦ C) - 0.17 - 0.34 - - - 0.79 0.43

Fig. 12. Cell surface temperature [◦ C] in CC discharge: Thermal image and NTGK model prediction corresponding to the end of 1C discharge (as in Fig. 11c). The
highest temperature is seen near the negative terminal (top left) in both CFD simulation and measurement.

Fig. 13. Comparison of the NTGK (static system) and ECM (dynamical system) predictions and the measured values for the 4C DPPC test sequence shown in Fig. 3.

static and dynamic current profiles and different temperatures. The Declaration of Competing Interest
Lithium-Titanate Oxide cells considered herein can reportedly with­
stand rapid charging at high currents, and thermal management during The authors declare that they have no known competing financial
charging may then have to address cooling both the battery and the interests or personal relationships that could have appeared to influence
charging equipment simultaneously. The methods and models presented the work reported in this paper.
in this article can also be applied for describing charging situations, but
more experimental data is needed for model identification and valida­ Acknowledgements
tion. Another important topic for future research is comparing battery
thermal models for cells with a different chemistry, or a considerably The authors would like to thank Dr. Kirill Murashko, Mr. Juha Nik­
smaller mass, such as the 18650 type cells. The thermal effects for such kanen and Mr. Kjell Björkroth for their help in arranging experimental
cells may be faster than for the relatively large cells considered herein. testing; Celltech Oy for providing the battery cells for this research;

12
E. Immonen and J. Hurri Applied Thermal Engineering 197 (2021) 117260

Valmet Automotive Inc. for suggestions for experimental testing; [28] U.S. Kim, C.B. Shin, C.-S. Kim, Effect of electrode configuration on the thermal
behavior of a lithium-polymer battery, J. Power Sources 180 (2) (2008) 909–916.
EDR&Medeso Oy for guidance in simulations; and the anonymous re­
[29] U.S. Kim, J. Yi, C.B. Shin, T. Han, S. Park, Modeling the dependence of the
viewers for many suggestions for improving the manuscript. Funding discharge behavior of a lithium-ion battery on the environmental temperature,
from Business Finland (e3Power project) is gratefully acknowledged. Journal of The Electrochemical Society 158 (5) (2011) A611.
[30] U.S. Kim, C.B. Shin, C.-S. Kim, Modeling for the scale-up of a lithium-ion polymer
battery, J. Power Sources 189 (1) (2009) 841–846.
References [31] U.S. Kim, J. Yi, C.B. Shin, T. Han, S. Park, Modelling the thermal behaviour of a
lithium-ion battery during charge, J. Power Sources 196 (11) (2011) 5115–5121.
[1] J. Kim, J. Oh, H. Lee, Review on battery thermal management system for electric [32] W. Huang, W. Zhang, A. Chen, Y. Zhang, M. Li, A co-simulation method based on
vehicles, Appl. Therm. Eng. 149 (2019) 192–212, https://doi.org/10.1016/j. coupled thermoelectric model for electrical and thermal behavior of the lithium-
applthermaleng.2018.12.020, http://www.sciencedirect.com/science/article/pii/ ion battery, IEEE Access 7 (2019) 180727–180737.
S135943111835614X. [33] A. Rizzello, S. Scavuzzo, A. Ferraris, A.G. Airale, M. Carello, Dynamic electro-
[2] G. Xia, L. Cao, G. Bi, A review on battery thermal management in electric vehicle thermal li-ion battery model for control algorithms, in: 2020 AEIT International
application, Journal of power sources 367 (2017) 90–105. Annual Conference (AEIT), IEEE, 2020, pp. 1–6.
[3] Q. Wang, B. Jiang, B. Li, Y. Yan, A critical review of thermal management models [34] M. Dubarry, N. Vuillaume, B.Y. Liaw, From single cell model to battery pack
and solutions of lithium-ion batteries for the development of pure electric vehicles, simulation for li-ion batteries, J. Power Sources 186 (2) (2009) 500–507.
Renew. Sustain. Energy Rev. 64 (2016) 106–128. [35] B. Pattipati, B. Balasingam, G. Avvari, K. Pattipati, Y. Bar-Shalom, Open circuit
[4] Y. Ma, H. Mou, H. Zhao, Cooling optimization strategy for lithium-ion batteries voltage characterization of lithium-ion batteries, J. Power Sources 269 (2014)
based on triple-step nonlinear method, Energy 117678 (2020). 317–333.
[5] K. Liu, K. Li, Q. Peng, C. Zhang, A brief review on key technologies in the battery [36] L. Gurjer, P. Chaudhary, H. Verma, Detailed modelling procedure for lithium-ion
management system of electric vehicles, Frontiers of mechanical engineering 14 battery using thevenin equivalent, in: 2019 IEEE International Conference on
(1) (2019) 47–64. Electrical, Computer and Communication Technologies (ICECCT), IEEE, 2019,
[6] M. Shen, Q. Gao, A review on battery management system from the modeling pp. 1–6.
efforts to its multiapplication and integration, Int. J. Energy Res. 43 (10) (2019) [37] H.E. Perez, J.B. Siegel, X. Lin, A.G. Stefanopoulou, Y. Ding, M.P. Castanier,
5042–5075. Parameterization and validation of an integrated electro-thermal cylindrical LFP
[7] S. Nejad, D. Gladwin, D. Stone, A systematic review of lumped-parameter battery model, in: Dynamic Systems and Control Conference, Vol. 45318, American
equivalent circuit models for real-time estimation of lithium-ion battery states, Society of Mechanical Engineers, 2012, pp. 41–50.
J. Power Sources 316 (2016) 183–196. [38] S. Yao, P. Liao, M. Xiao, J. Cheng, K. He, Equivalent circuit modeling and
[8] W. Fang, O.J. Kwon, C.-Y. Wang, Electrochemical–thermal modeling of automotive simulation of the zinc nickel single flow battery, AIP Advances 7 (5) (2017)
li-ion batteries and experimental validation using a three-electrode cell, 055112.
International journal of energy research 34 (2) (2010) 107–115. [39] W. Tiedemann, J. Newman, Current and potential distribution in lead-acid battery
[9] M. Chen, G.A. Rincon-Mora, Accurate electrical battery model capable of plates, in: Proceedings of the Symposium on Battery Design and Optimization, Vol.
predicting runtime and iv performance, IEEE transactions on energy conversion 21 79, Battery Division, Electrochemical Society, 1979, p. 39.
(2) (2006) 504–511. [40] S. Allu, S. Kalnaus, W. Elwasif, S. Simunovic, J.A. Turner, S. Pannala, A new open
[10] W. Junping, C. Quanshi, C. Binggang, Support vector machine based battery model computational framework for highly-resolved coupled three-dimensional
for electric vehicles, Energy conversion and management 47 (7–8) (2006) multiphysics simulations of li-ion cells, J. Power Sources 246 (2014) 876–886.
858–864. [41] S. Chacko, Y.M. Chung, Thermal modelling of li-ion polymer battery for electric
[11] M. Einhorn, F.V. Conte, C. Kral, J. Fleig, Comparison, selection, and vehicle drive cycles, J. Power Sources 213 (2012) 296–303.
parameterization of electrical battery models for automotive applications, IEEE [42] S. Kosch, A. Rheinfeld, S.V. Erhard, A. Jossen, An extended polarization model to
Trans. Power Electron. 28 (3) (2012) 1429–1437. study the influence of current collector geometry of large-format lithium-ion pouch
[12] S. Panchal, M. Mathew, R. Fraser, M. Fowler, Electrochemical thermal modeling cells, J. Power Sources 342 (2017) 666–676.
and experimental measurements of 18650 cylindrical lithium-ion battery during [43] T. Kalogiannis, M.S. Hosen, M.A. Sokkeh, S. Goutam, J. Jaguemont, L. Jin, G. Qiao,
discharge cycle for an ev, Appl. Therm. Eng. 135 (2018) 123–132. M. Berecibar, J. Van Mierlo, Comparative study on parameter identification
[13] A. De Vita, A. Maheshwari, M. Destro, M. Santarelli, M. Carello, Transient thermal methods for dual-polarization lithium-ion equivalent circuit model, Energies 12
analysis of a lithium-ion battery pack comparing different cooling solutions for (21) (2019) 4031.
automotive applications, Applied energy 206 (2017) 101–112. [44] F. He, A.A. Ams, Y. Roosien, W. Tao, B. Geist, K. Singh, Reduced-order thermal
[14] R. Zhao, J. Gu, J. Liu, Optimization of a phase change material based internal modeling of liquid-cooled lithium-ion battery pack for evs and hevs, in: 2017 IEEE
cooling system for cylindrical li-ion battery pack and a hybrid cooling design, Transportation Electrification Conference and Expo (ITEC), IEEE, 2017, pp.
Energy 135 (2017) 811–822. 507–511.
[15] O. Yetik, T.H. Karakoc, A numerical study on the thermal performance of prismatic [45] V. Esfahanian, A.B. Ansari, F. Torabi, Simulation of lead-acid battery using model
li-ion batteries for hibrid electric aircraft, Energy 195 (2020) 117009. order reduction, J. Power Sources 279 (2015) 294–305.
[16] S. Jung, D. Kang, Multi-dimensional modeling of large-scale lithium-ion batteries, [46] A.B. Ansari, V. Esfahanian, F. Torabi, Discharge, rest and charge simulation of lead-
J. Power Sources 248 (2014) 498–509. acid batteries using an efficient reduced order model based on proper orthogonal
[17] X. Hu, L. Chaudhari, S. Lin, S. Stanton, S. Asgari, W. Lian, A state space thermal decomposition, Appl. Energy 173 (2016) 152–167.
model for hev/ev battery using vector fitting, in: 2012 IEEE Transportation [47] M. Doyle, T.F. Fuller, J. Newman, Modeling of galvanostatic charge and discharge
Electrification Conference and Expo (ITEC), IEEE, 2012, pp. 1–8. of the lithium/polymer/insertion cell, Journal of the Electrochemical society 140
[18] M. Akbarzadeh, T. Kalogiannis, J. Jaguemont, J. He, L. Jin, M. Berecibar, J. Van (6) (1993) 1526.
Mierlo, Thermal modeling of a high-energy prismatic lithium-ion battery cell and [48] G. Li, S. Li, Physics-based cfd simulation of lithium-ion battery under the fuds
module based on a new thermal characterization methodology, Journal of Energy driving cycle, ECS Trans. 64 (33) (2015) 1.
Storage 32 (2020) 101707. [49] J. Mei, E.K.W. Cheng, Y.C. Fong, Lithium-titanate battery (lto): A better choice for
[19] B.-C. Wang, H.-X. Li, H.-D. Yang, Spatial correlation-based incremental learning for high current equipment, in: 2016 International Symposium on Electrical
spatiotemporal modeling of battery thermal process, IEEE Trans. Industr. Electron. Engineering (ISEE), 2016, pp. 1–4, https://doi.org/10.1109/EENG.2016.7846362.
67 (4) (2019) 2885–2893. [50] N. Takami, H. Inagaki, Y. Tatebayashi, H. Saruwatari, K. Honda, S. Egusa, High-
[20] T. Nemeth, P. Schröer, M. Kuipers, D.U. Sauer, Lithium titanate oxide battery cells power and long-life lithium-ion batteries using lithium titanium oxide anode for
for high-power automotive applications–electro-thermal properties, aging behavior automotive and stationary power applications, Journal of power sources 244
and cost considerations, Journal of Energy Storage 31 (2020) 101656. (2013) 469–475.
[21] Toshiba scib lithium-titanate oxide battery cells, https://www.scib.jp/en/product/ [51] A. Burke, M. Miller, H. Zhao, Fast charging tests (up to 6c) of lithium titanate cells
cell.htm, accessed: 2020-11-16. and modules: Electrical and thermal response, in: EVS26 International Battery,
[22] Turku university of applied sciences erallycross car project, https://erallycross. Hybrid and Fuel Cell Electric Vehicle Symposium, 2012, pp. 1–10.
turkuamk.fi/en/main-page/, accessed: 2020-11-16. [52] W.Y. Low, M.J.A. Aziz, N.R.N. Idris, Modelling of lithium-titanate battery with
[23] Z.H. Che Daud, D. Chrenko, F. Dos Santos, E.-H. Aglzim, A. Keromnes, L. Le Moyne, ambient temperature effect for charger design, IET Power Electronics 9 (6) (2016)
3d electro-thermal modelling and experimental validation of lithium polymer- 1204–1212.
based batteries for automotive applications, Int. J. Energy Res. 40 (8) (2016) [53] A.-I. Stroe, D.-I. Stroe, M. Swierczynski, R. Teodorescu, S.K. Kær, Lithium-ion
1144–1154. battery dynamic model for wide range of operating conditions, in: 2017
[24] N. Nieto, L. Díaz, J. Gastelurrutia, F. Blanco, J.C. Ramos, A. Rivas, Novel thermal International Conference on Optimization of Electrical and Electronic Equipment
management system design methodology for power lithium-ion battery, J. Power (OPTIM) & 2017 Intl Aegean Conference on Electrical Machines and Power
Sources 272 (2014) 291–302. Electronics (ACEMP), IEEE, 2017, pp. 660–666.
[25] J.-K. Kim, C.-S. Lee, Co-simulation approach for analyzing electric-thermal [54] S. Liu, J. Jiang, W. Shi, Z. Ma, H. Guo, et al., Butler–volmer-equation-based
interaction phenomena in lithium-ion battery, International Journal of Precision electrical model for high-power lithium titanate batteries used in electric vehicles,
Engineering and Manufacturing-Green Technology 2 (3) (2015) 255–262. IEEE Transactions on Industrial electronics 62 (12) (2015) 7557–7568.
[26] E. Paccha-Herrera, W.R. Calderón-Muñoz, M. Orchard, F. Jaramillo, K. Medjaher, [55] A. Chen, W. Zhang, C. Zhang, W. Huang, S. Liu, A temperature and current rate
Thermal modeling approaches for a licoo2 lithium-ion battery–a comparative study adaptive model for high-power lithium titanate batteries used in electric vehicles,
with experimental validation, Batteries 6 (3) (2020) 40. IEEE Transactions on Industrial Electronics.
[27] K.H. Kwon, C.B. Shin, T.H. Kang, C.-S. Kim, A two-dimensional modeling of a
lithium-polymer battery, J. Power Sources 163 (1) (2006) 151–157.

13
E. Immonen and J. Hurri Applied Thermal Engineering 197 (2021) 117260

[56] K.A. Murashko, A.V. Mityakov, J. Pyrhönen, V.Y. Mityakov, S. Sapozhnikov, [67] Z. Rao, S. Wang, M. Wu, Z. Lin, F. Li, Experimental investigation on thermal
Thermal parameters determination of battery cells by local heat flux management of electric vehicle battery with heat pipe, Energy Convers. Manage.
measurements, J. Power Sources 271 (2014) 48–54. 65 (2013) 92–97.
[57] M.R. Khan, S.K. Kær, Multiphysics based thermal modeling of a pouch lithium-ion [68] X. Duan, G. Naterer, Heat transfer in phase change materials for thermal
battery cell for the development of pack level thermal management system, in: management of electric vehicle battery modules, Int. J. Heat Mass Transf. 53
2016 Eleventh International Conference on Ecological Vehicles and Renewable (23–24) (2010) 5176–5182.
Energies (EVER), IEEE, 2016, pp. 1–9. [69] D. Dees, E. Gunen, D. Abraham, A. Jansen, J. Prakash, Electrochemical modeling of
[58] A.-I. Stroe, M. Swierczynski, D.-I. Stroe, R. Teodorescu, Performance model for lithium-ion positive electrodes during hybrid pulse power characterization tests,
high-power lithium titanate oxide batteries based on extended characterization J. Electrochem. Soc. 155 (8) (2008) A603.
tests, in: 2015 IEEE Energy Conversion Congress and Exposition (ECCE), IEEE, [70] G.L. Plett, Battery management systems, Volume I: Battery modeling, Artech
2015, pp. 6191–6198. House, 2015.
[59] J. Jaguemont, L. Boulon, Y. Dubé, A comprehensive review of lithium-ion batteries [71] N.H.F. Ismail, S.F. Toha, N.A.M. Azubir, N.H.M. Ishak, M.K. Hassan, B.S.
used in hybrid and electric vehicles at cold temperatures, Appl. Energy 164 (2016) K. Ibrahim, Simplified heat generation model for lithium ion battery used in
99–114. electric vehicle, in: IOP Conference Series: Materials Science and Engineering, Vol.
[60] K. Murashko, J. Pyrhönen, L. Laurila, Three-dimensional thermal model of a 53, IOP Publishing, 2013, p. 012014.
lithium ion battery for hybrid mobile working machines: Determination of the [72] S. Wu, R. Xiong, H. Li, V. Nian, S. Ma, The state of the art on preheating lithium-ion
model parameters in a pouch cell, IEEE Trans. Energy Convers. 28 (2) (2013) batteries in cold weather, Journal of Energy Storage 27 (2020) 101059.
335–343, https://doi.org/10.1109/TEC.2013.2255291. [73] T. Huria, M. Ceraolo, J. Gazzarri, R. Jackey, High fidelity electrical model with
[61] H. Behi, D. Karimi, M. Behi, J. Jaguemont, M. Ghanbarpour, M. Behnia, thermal dependence for characterization and simulation of high power lithium
M. Berecibar, J. Van Mierlo, Thermal management analysis using heat pipe in the battery cells, in: 2012 IEEE International Electric Vehicle Conference, IEEE, 2012,
high current discharging of lithium-ion battery in electric vehicles, Journal of pp. 1–8.
Energy Storage 32 (2020) 101893. [74] G.-H. Kim, K. Smith, K.-J. Lee, S. Santhanagopalan, A. Pesaran, Multi-domain
[62] S.S. Madani, E. Schaltz, S. Knudsen Kær, Simulation of thermal behaviour of a modeling of lithium-ion batteries encompassing multi-physics in varied length
lithium titanate oxide battery, Energies 12 (4) (2019) 679. scales, Journal of the electrochemical society 158 (8) (2011) A955.
[63] A. Lajunen, P. Sainio, L. Laurila, J. Pippuri-Mäkeläinen, K. Tammi, Overview of [75] D.H. Douglas, T.K. Peucker, Algorithms for the reduction of the number of points
powertrain electrification and future scenarios for non-road mobile machinery, required to represent a digitized line or its caricature, Cartographica: the
Energies 11 (5) (2018) 1184. international journal for geographic information and geovisualization 10 (2)
[64] W. Wu, S. Wang, W. Wu, K. Chen, S. Hong, Y. Lai, A critical review of battery (1973) 112–122.
thermal performance and liquid based battery thermal management, Energy [76] R. Rao, S. Vrudhula, D.N. Rakhmatov, Battery modeling for energy aware system
conversion and management 182 (2019) 262–281. design, Computer 36 (12) (2003) 77–87.
[65] S. Yang, C. Ling, Y. Fan, Y. Yang, X. Tan, H. Dong, A review of lithium-ion battery [77] E. Immonen, 2d shape optimization under proximity constraints by cfd and
thermal management system strategies and the evaluate criteria, Int J Electrochem response surface methodology, Appl. Math. Model. 41 (2017) 508–529.
Sci 14 (2019) 6077–6107. [78] Y. Ji, Y. Zhang, C.-Y. Wang, Li-ion cell operation at low temperatures, Journal of
[66] R. Mahamud, C. Park, Reciprocating air flow for li-ion battery thermal The Electrochemical Society 160 (4) (2013) A636.
management to improve temperature uniformity, J. Power Sources 196 (13) [79] D.H. Vega, S. Kelouwani, L. Boulon, Efficient internal resistance and specific heat
(2011) 5685–5696. identification of li-ion battery at low temperature conditions, in: 2015 IEEE Vehicle
Power and Propulsion Conference (VPPC), IEEE, 2015, pp. 1–6.

14