Вы находитесь на странице: 1из 12

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2020.2966266, IEEE
Transactions on Vehicular Technology

On-Board State Estimation in Electrical Vehicles:


Achieving Accuracy and Computational Efficiency
through an Electrochemical Model
M.K.S.Vermaa,c , Suman Basua,d , Rajkumar S. Patila , Krishnan S. Hariharana , Shashishekar P. Adigaa,∗ ,
Subramanya Mayya Kolakea , Dukjin Ohb Taewon Songb , Younghun Sungb
a SAIT-India, Samsung R&D Institute India-Bangalore, India. b Autonomous Material Development Lab, SAIT,

Samsung Electronics, Republic of Korea.

Abstract—Success of electric mobility and connected future of state of charge (SOC), state of health (SOH) [3], [4],
depends on advanced high capacity Lithium-ion batteries and [5], and on cell balancing [6], [7]. Another important task
a tailored battery management system that keeps track of of the BMS is voltage prediction, because it is needed in
battery health and safety to optimize the performance. The
drive for advanced batteries is being pursued on one front estimating available power that is critical in determining if the
via the development of novel chemistry, for example, blended instantaneous acceleration requirement of the EV can be met
composite cathodes to achieve enhanced performance and life. without triggering battery degradation. Even though accurate
On the other front, the need for optimal battery performance SOC and voltage estimations are a critical task of the BMS [8],
via operational control viz. a robust Battery Management System [9], [10], [11], it remains an unmet challenge due to strong
(BMS) that accurately predicts state-of-charge (SOC), state-of-
power (SOP) and state-of-health (SOH) in a computationally nonlinear electrochemical behavior of the battery that changes
efficient way, has been lacking in many ways due to the with temperature [10], [12], [13], [14] and aging [15]. The
reliance on simple and fast-to-compute models such as equivalent lack of a robust electrochemically informed model on-board
circuit models (ECM) that lack the accuracy needed for large can lead to uncertainty in battery performance that poses a
battery packs. This paper reports an on-board reduced-order hindrance to battery life and safety.
electrochemical thermal model (ROTM) based SOC and voltage
estimation for a 12S1P configuration composite-cathode battery The current estimation methods can be broadly classified
pack and demonstrates its practicality by implementing it on into two categories: direct and indirect methods [16], [17]. The
four different micro-controller units (MCU) (ATmega:2560&328, direct methods use physical measurements of battery signals
Infineon:TC275&TC297). We show that on-board ROTM based such as voltage, current, and temperature that are in turn used
SOC and voltage prediction have better accuracy compared to the in analytical expressions for SOC estimation alone without
conventional ECM based methods under both static and dynamic
load conditions. voltage prediction. The indirect methods use measured battery
signals (voltage, current, and temperature) as input to some
Index Terms—Composite Cathode Lithium-ion batteries, appropriate parameterized battery model that can provide both
Reduced-Order-Electrochemical Modeling, Battery Management
System, State of Charge Estimation, Electric Vehicles SOC estimation and voltage prediction. Coulomb counting
(CC) [18] is the most used direct method for short term SOC
estimation. In CC, the SOC is estimated by time integrals of
I. INTRODUCTION charge and discharge currents and it requires the knowledge of
HE remarkable advancement in the field of battery correct initial SOC0 of the battery. As it is based on the time
T technology has made portable electronics, power tools,
medical devices, and electric vehicles (EVs) a reality. With
integral of current, any uncertainty with respect to initial SOC
as well as small errors associated with the integration term get
the continued expansion of lithium-ion battery technology, the accumulated with time and become the source of significant
deployment of on-board battery management system (BMS) inaccuracy. Several other factors also contribute to inaccuracy
for safe and reliable battery operation at different operat- in the CC method including battery age, charge/discharge rate,
ing conditions has become inevitable. The BMS consists and sensor precision. Another commonly used direct method
of sensors, on-board chip and control hardware as well as of SOC estimation is the one based on open circuit voltage
embedded battery model and related software for monitoring (OCV) [19]. In the OCV method, measured cell voltage is
and estimation of states and control of batteries [1], [2]. used to determine the cell SOC by inverting the cell OCV
The main functions of the BMS are to ensure safe and curve (using a polynomial fit or a look-up table). This method
efficient power utilization that relies on accurate estimates requires the battery to have a long resting time to reach
equilibrium voltage, and hence is not suitable for on-board
c Present address: Department of Chemical Engineering, Indian Institute of
applications. Even small errors in the measured voltage can
Technology, Delhi 110016, India lead to significant inaccuracy. On the other hand, the model
d Present address: Energy Systems Department, Mahindra Electric Mobility
Limited, Bangalore 560068, India based indirect techniques for SOC estimation have become
∗ Corresponding author e-mail address: shashi.adiga@samsung.com more common as they surmount the limitations of direct

Copyright (c) 2015 IEEE. Personal use of this material is permitted. However, permission to use this
0018-9545 (c) 2019 IEEE. Personalmaterial
use is permitted,
for any butother
republication/redistribution
purposes must be requires IEEE
obtained permission.
from the IEEESee http://www.ieee.org/publications_standards/publications/rights/index.html
by sending a request to pubs-permissions@ieee.org. for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2020.2966266, IEEE
Transactions on Vehicular Technology

methods by using parameterized battery models to better improved performance and battery life.
estimate the SOC. Although there are number of different Further, an additional challenge has emerged owing to the
approaches for model based SOC estimation, here we only recent trend in EV batteries to use blended (composite) elec-
cite equivalent circuit and electrochemical based models as trodes for better energy densities and cycle life [36]. In blended
they form the basis for other battery models. electrodes, the physical mixing of multiple electrode materials
The most widely adopted indirect method for state estima- results in excellent electrode performance as its properties
tion is the equivalent circuit model (ECM) based estimation become a blend of the individual material properties depending
[20], [21], [22], [23]. In the ECM method, the battery dynam- on mixing fractions. For example, LMO (Lithium manganese
ics are modeled using resistors, capacitors and voltage sources oxide), with its high open circuit potential, excellent rate
to form an electrical circuit network representing the battery. capability, and relatively low cost [37], is mixed with layered
A typical ECM network consists of SOC dependent nonlinear materials (NCA (nickel cobalt aluminum) [38], and NCM
voltage source, a capacitor to model diffusional effects within (nickel cobalt manganese) [39]) to suppress its degradation
the battery, and resistances to model diffusional and internal due to Mn dissolution. The present SOC estimation methods
resistances [23], [24], [25], [26]. By using simple ECM perform very poorly for these blended electrodes since the
models, such as the Thevenin ECM [23], the SOC can be intercalation dynamics of individual electrode components are
directly estimated by the enumeration of model equations. highly non-linearly dependent on the cell SOC that they cannot
Despite not containing the physical and chemical processes of be expressed analytically or using look-up tables.
battery, ECM can be adjusted to reproduce the dynamic battery In this context, physics-based electrochemical thermal mod-
response such as power, energy and heat generation under a els, with their ability to describe composite electrode behav-
wide range of operating conditions reasonably well. Despite ior correctly and to account for the effect of temperature
having several advantages, the ECM has few drawbacks. more accurately, could solve these pertinent issues. The full
ECM is often designed based on battery dynamics using electrochemical thermal models (ECTs) [21], [22], [40], [41],
low amplitude ac signal and lacks prediction accuracy for [42], [43] account for physicochemical phenomena like charge
high power applications such as EVs due to nonlinear battery transfer, diffusion, intercalation, and electrochemical kinetics
dynamics [27]. This issue can be resolved by making the ECM occurring in a battery, and facilitate improved prediction accu-
parameters a function of SOC, SOH, and temperature [28], racy by providing better insight on the system behaviour and
[29], but it leads to a large number parameters that need temperature-dependent parameters. Additionally, these models
to be fit for a wide range of operational conditions. Since, can potentially be augmented with electrochemically more pre-
these parameters do not represent all nuances of physical or cise degradation models to monitor SOH. However, the main
electrochemical behavior of the battery, they do not provide obstacle in using these full order models on-board is that they
detailed insight into the physical or electrochemical processes are complex and computationally expensive owing to a system
occurring inside the battery. Even though there are ECMs of coupled time-varying partial differential equations [44].
that are derived from electrochemical effects and processes On the other hand, the reduced-order models (ROM) [45],
using electrochemical impedence spectroscopy (EIS) [30], which use the profile approximations and volume averaging to
identification, and parametrization of each individual processes simplify the full electrochemical model, are computationally
are challenging [31], [32]. For example, at higher C rates, fast and includes the detailed physics of the electrochemical
a Li concentration gradient often develops in active material model. While ROM based SOC prediction is very attractive for
particles that leads to a very non-linear overall SOC vs cell deployment on onboard BMS, it has not been tested on-board
voltage behaviour. Such behaviour cannot be accounted for in nor has been evaluated against alternative state estimation
ECM models. This issue can lead to large errors in voltage algorithms. Moreover, in EVs, a single battery pack consists
prediction. It’s also become equally important to take account of more than one batteries and SOC estimation using ROM
of battery aging, which can have a complex influence on has not been developed for battery packs. It is widely believed
the cell performance and hence on model prediction. The that electrochemical model based state estimation algorithms
degradation modeling requires a physical or electrochemical will help mitigate driver anxiety towards the end of discharge
understanding of multiscale effects in the battery to enable in EVs because of the promise of higher accuracy under wide
accurate SOH estimation. Although CC, OCV, and ECM are range of operational temperature.
computationally inexpensive and attractive from the point of To address these challenges, we have developed a computa-
view of real-time on-board estimations but are limited in the tionally efficient reduced-order electrochemical thermal model
depth of physical electrochemical interpretation. The predic- (ROTM) for a blended composite cathode battery pack for
tion accuracy becomes even more of a serious issue in the case on-board applications with very accurate SOC and voltage
of electric vehicles (EVs) [33], [34], where the battery pack prediction and tested it on-board. To the best of our knowledge,
consists of hundreds of cells and prediction inaccuracy with this is the first time that an electrochemical model based
respect to one cell gets accumulated over many cells. Thus, SOC and voltage prediction algorithm has been demonstrated
ensuring that BMSs for EVs are much more robust in terms on a chip and we have successfully extended the single
of SOC prediction accuracy [35] under a range of operating cell model to a battery pack. The On-board ROTM predicts
conditions becomes very critical as compared to those for the voltage and SOC of individual cells and also at pack
portable electronics. Thus, models covering the physical and level using a simplified high-speed pack model with higher
electrochemical processes in the battery are advantageous for accuracy than the widely used ECM. The performance of

0018-9545 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2020.2966266, IEEE
Transactions on Vehicular Technology

ROTM is compared on different MCUs (ATmega:2560&328, the electrolyte potential (Φ2ik ). The solid phase concentration
Infineon:TC275&TC297) for real-time voltage and SOC pre- (c1k ) is estimated using the reaction rate at the particle surface.
diction. The On-board Voltage and SOC predictions are shown  
to have > 98% average accuracy under both static and F
jk = 2 jk0 sinh (Φ1k + Φ2k − Uk ) (2)
dynamic load conditions. Further, the developed model is 2RT
computationally very efficient and only takes 7msec for the where
next 100msec voltage and SOC prediction on the most basic jk0 = krk (c1kmax − c1 k)0.5 c1k
0.5 0.5
c2k (3)
MCU chips.
 
2RT jk
II. MODEL DEVELOPMENT Φ1k = Uk + Φ2k + sinh−1 (4)
F 2 jk0
A. Reduced order thermal model (ROTM)
To calculate the solid phase potential (Φ1k ), the reaction
The ROTM is based on pseudo two dimensional electro- rates h jk i, electrolyte concentrations (c2k ), electrolyte poten-
chemical model that uses the porous electrode theory [40], tials (Φ2ik ) and solid phase concentrations (c1k ) are used
[41]. In porous electrode theory, the battery consists of two as input to Butler-Volmer kinetics (Eqs. 2,3), which is then
porous electrodes separated by a porous separator, and the analytically inverted. The pre-factor reaction rates jk0 for the
liquid electrolyte contained in the electrodes and the separator electrodes are estimated using equation 3. For cell voltage
(fig. 1). For thermal effects, the cell is assumed to rest in an estimation, equation 4 is evaluated at the two current collector
environment with a constant ambient temperature (Tamb ). ends of the cell at (x=0) and (x=L), then the cell voltage is
computed as the difference in solid potential at the end of
these two collectors (Eq. 5).
The model cell voltage (Vcell ) is calculated as the difference
between anode and cathode solid potentials at respective
current collectors.

Vcell = Φsp |x=L − Φsn |x=0 (5)


2) Temperature module: Temperature plays an important
role in the performance, life expectancy, safe operation and
diagnostic of Li-ion batteries [48]. Transport and kinetic
parameters as well as OCP of active materials depend on the
Fig. 1. Schematic showing the porous electrode cell model with multiple temperature and have a significant effect on cell performance.
particle of the anode/cathode with separator. In this work, although we have used temperature reading from
the sensor as an input to the ROTM to update the thermally
1) Model equations: The electrochemical model relies on dependent transport and kinetic parameters, having an on-
solving the coupled partial differential equations (PDEs) to de- board thermal model to predict the temperature of individual
scribed the mass and charge balance in the solid and electrolyte cells would still be very desirable since any discrepancy
phases along with boundary conditions. These coupled PDEs between model prediction and measurement may indicate the
are computationally expensive for use on on-board estimation beginning of a thermal runaway problem. However, a full
and control. On the other hand, ROTM is developed based temperature model is not practically possible to implement on
on the principle of spatial volume averaging (Eq.1) for order MCU due to the huge computational requirements. In lieu of
reduction. The model order reduction is achieved through full model for on-board temperature estimation, a simplified
uniform reaction rate approximation that allows us to use six temperature module is required. Therefore, a lumped model
linear ordinary differential equations (ODE) in the place of capable of predicting the temperature of an individual cell
ten coupled PDEs. in the pack is developed. The lumped temperature model
∫ l describes the thermal activities at the continuum scale, which
1
h f (t)i = f (x, t)Sdx (1) is modeled as
S×l 0
∂Tc hc Ac (Tc − Tamb )
The detailed derivation of ROTM has been discussed by ρc Cpc = + QT ot al (6)
Kumar et al. [45] and the governing equations are summarized ∂t νc
in Table I. where ρc , Cpc ,Tc , hc , Ac , νc denote the density, heat capac-
The battery voltage is governed by the lithium concentration ity, temperature, heat transfer coefficient, surface area and
in the solid and liquid phases. The lithium concentration in volume of the cell. For simplification, the contributions from
the electrolyte (c2k ) and solid phase (c1k ) are solved with a individual sources of heat or the spatial variation of tem-
constant reaction rate h jk i approximation by solving the ODEs perature are neglected. The heat source is approximated by
and algebraic expressions in table I. Here, k = n, p repre- Ohmic source (QT ot al ) [49] and for heat dissipation, a constant
sents negative and positive electrode, respectively. First, the convection coefficient is assumed. Thus, the equation6 can be
electrolyte phase concentration (c2k ) is estimated followed by written in a simplified form as

0018-9545 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2020.2966266, IEEE
Transactions on Vehicular Technology

TABLE I
C OMPILATION OF GOVERNING EQUATIONS FOR THE REDUCED ORDER MODEL .

Governing equations Anode Cathode


I (t )
hjn i (t) = j p (t) = − a pI (t )


Reaction rate a n F ln F lp

d c̄1n (t ) 3h j i(t ) d c̄1p (t ) 3 h j i (t )


Lithium concentration dt = − Rnn dt = − Rpp
d c̄1r n (t ) 30D 1n c̄1r n (t ) 45h j n i(t ) d c̄1r p (t ) 30D 1p c̄1r p (t ) 45 h j p i (t )
Concentration gradient + =− + =−
dt R2 n 2R 2n dt R2p 2R 2p
 
l s l n ε2n l2 ε l l ε
 
= −q2i n (t) + (1 − t + )a n ln hjn i (t)
dq (t ) dq (t )
Interfacial Flux ln ε2n αi n + 2D2s + 3D n 2n 2i n
dt + ln ε2n αi p + s2D n 2n 2i n
dt
2n 2s
 
l 2p ε2p d q2i p (t )
= q2i p (t) + (1 − t + )a p l p j p (t)
d q2i n (t )
l p ε2p αi n + l p ε2p αi n − 3D


dt dt 2p
   
l n l s ε2n l s2 ε2s 2ε
ln l n l s ε2n l s2 ε2s l 2p ε2p
Algebraic expressions αi n = − 2D2s + 6D2s + 3D2n
2n  1  αi p = − 2D2s + 3D2s − 3D2p
 1 
l n ε2n +l s ε2s +l p ε2p l n ε2n +l s ε2s +l p ε2p
q2i k (t )l k R j k av g (t ) 8R c (t )
c2k (t) = c2i k (t) + 2D2k h cs k (t) = ci k (t) + k35D + k 351k r
i h 1k i
c2k (t ) I (t )l k c2i k (t ) (t )l s
Φ2k (t) = Φ2i k (t) + 2Θ ln c2i k (t ) + 2κ2k Φ2i k (t) = 2Θ ln c + I2κ
2mi d (t ) 2s
Θ= RT
F (1 − t+ )

derived for equivalent cell SOC for given component fraction


∂Tc
= θ 0 (Tc − Tamb ) + θ 1 I 2 (7) and SOC [47]. Let us consider an operating voltage range
∂t (Vmax ,Vmin ) that is divided in to ‘n’equal parts (V j , j = 1...n).
where I is the current and θ 1 & θ 0 denote the Ohmic The SOC of i th component at OCP (V j ) is known and denoted
j max )
resistance and convection coefficient. The cell is assumed to by (SOCi ). The maximum solid phase concentrations (c(1,i)
be initially at ambient temperature, T0 = Tamb . Assuming of all the components are also known. For a blended composite
constant current operation, a closed form solution is obtained cathode consisting ‘m’ components with volume fractions
(Eq.8) after expanding and simplifying the exponential series. yi (i = 1...m), the cathode SOC (SOC j ) corresponding to
As the cells were cooled through natural convection, θ 0 is cathode OCP (V j ) is given by the following equation.
expected to be negligibly small. Thus, temperature becomes a Ím max y j
linear function of time and is dependent on only one parameter i=1 SOCi C1,i i
(Eq.8). SOC = j
Ím max (9)
C
i=1 1,i y i
θ1 2 The generated blended composite cathode OCP is vali-
Tc = T0 + I (exp(θ 1 t) − 1) ⇒ T0 + θ 1 I 2 t (8)
θ0 dated against the experimentally measured OCP as shown
This form can be used for dynamic current conditions using in Fig. 2(b). It is observed that the OCP generated with
a piecewise constant current profile. this method is in good agreement with the experimental
measurement as shown in Fig. 2(b). For MCU embedding
purpose, initially, a polynomial fit of generated blended com-
B. Blended composite cathode open circuit potential (OCP)
posite cathode OCP was used for voltage prediction. However,
The blended composite (LMO-NCM-NCA) cathode [46] is fluctuations in the simulated voltage prediction due to the
modelled as a physical mixture of three materials with known limited precision that higher order terms in polynomial fit
fractions, and the composite cathode OCP is generated using carry for single precision, this approach had accuracy issues.
the method developed by Bartlett et al. [47].The blended com- To resolve this issue, the blended composite cathode OCV
posite cathode is assumed to be a physical mixture of LMO lookup table is generated on-board for an operating voltage
(Lithium manganese oxide), NCA (nickel cobalt aluminum) range, and subsequently used to extract the cathode OCV as
and NCM (nickel cobalt manganese) particles of known weight a function of SOC for the on-board voltage prediction.
fractions. Although the open circuit (or equilibrium) potentials
(OCP) of these materials are completely different when plotted
against their respective SOCs (shown in Fig. 2(a)), the average C. High Speed Pack Model
cathode OCP can be modeled as three particles acting in In a battery pack, despite voltage measurement on ecah cell
parallel. Therefore all the three cathode particles are expected is possible, voltage prediction is desirable for the following
to be at same potential at any given point of time. However, reasons. First of all, having the ability to predict individual
the current will split between the each particle and dictated voltage will allow for reconciliation between model and mea-
by the individual components OCP characteristic. As it can surement if needed and correct for associated SOC errors. The
be expected from individual OCP behaviors (Fig. 2(a)), LMO ability to predict voltage will help detecting any sensor errors.
will discharge first, as it reaches close to cathode SOC value Further, being able to predict voltage will allow estimating
of 1 at highest voltage, followed by NCM and then by NCA. available power, which is very crucial towards the end of
Let us describe how the voltage equality based model is discharge, particularly in the electric vehicle application. Given

0018-9545 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2020.2966266, IEEE
Transactions on Vehicular Technology

for cell-to-cell variation due to differences in initial SOC and


electrode loading.
4 (a)

3.8
Cathode OCV

3.6

3.4 LMO
NCA
3.2 NCM
Composite cathode
3 4.2
0.4 0.5 0.6 0.7 0.8 0.9 1
SOC 4
4.2
3.8 Cell 01/R
(b) Cell 02
4
3.6 Cell 03

Voltage
Cell 04
3.8
Cell OCV

3.4 Cell 05
Cell 06 4.09
3.6 Cell 07
3.2
Cell 08 4.08
3.4 Cell 09
3
Experiment Cell 10 4.07
3.2 Cell 11
Composite cathode Cell 2.8
Cell 12 5 10 15 20
3
0 0.2 0.4 0.6 0.8 1 2.6
Cell SOC 0 2000 4000 6000 8000 10000 12000
Time (sec)

Fig. 2. (a) Figure showing the individual component (LMO, NMC and NCA) Fig. 3. Schematic showing the 12S1P configuration battery pack and figure
and blended composite OCP with reference to their corresponding SOC, (b) showing the measured individual cell voltage of the 12S1P battery pack. The
Model prediction of the blended composite cathode cell voltage compared to inset figure is showing variation in the initial SOC between the cells.
the measured cell voltage.
In a typical commercial cell, the capacity is limited by the
cathode capacity. Thus, it’s fair to assume that variation in
a discharge current and the present SOC, an electrochemical cathode active material loading alone will affect the overall
model will allow one to determine voltage at a future SOC cell capacity. The initial SOC of individual cell also has to be
level and thus help predict available power. It is highly taken into account for accurate SOC and voltage estimation.
desirable that both voltage and SOC prediction are as accurate Therefore, SOC (SOCi ) and voltage (Vi ) estimation for ith
as possible to avoid over- or underestimation of available cell can be expressed as multivariable Taylor series expansion
power and charge so that the user experience is not affected (Eq.10,11), where only the first terms are considered for an
nor the battery is operated outside the safe operational window. infinitesimal small variation in initial SOC and electrode active
In a typical battery pack, each cell can be at a different material content wrt. the reference cell.
SOC and voltage when fully charged, until balanced by active ∂SOC
or passive means. Even for two perfectly balanced similar SOCi = SOCr + ∆SOCi0 + ∆εi (10)
∂ε
cells, the capacities can be different due to minor differences
encountered in design parameters, electrode active material ∂V ∂V
Vi = Vr + ∆SOCi0 + ∆εi (11)
content. Thus, solving ROTM for a single cell and using it ∂SOC ∂ε
across all the other cells in the pack, may result in inaccurate where ∆SOCi0 and εi are the variation in SOC and active
SOC and voltage estimation that leads to poor pack perfor- material content of ith cell with respect to reference cell.
mance. On the other hand, solving ROTM for each cell in the SOCr and Vr are the SOC and voltage of the reference cell
pack becomes computationally expensive, and not practically respectively. For the cathode, the total lithium conservation
possible. Thus, there is a need to develop a simplified full can be written as the following equation.
pack model, which can simulate the full pack dynamics while
keeping the physical and electrochemical processes of battery cpmax (SOC − SOC 0 )ε p Ap lp = Q p (12)
intact. Here, we present a hybrid approach, where ROTM is
used to simulate a single cell, which we call reference cell where SOC and SOC 0 are the current and initial state of
(Cell-R), to estimate the SOC (SOCr ) and voltage (Vr ), and cathode respectively. Q p represents the total Lithium interca-
then the SOC and voltage of other cells (SOCi , Vi of the ith lation during the charge/discharge, ε p is the cathode active
cell) are estimated using the simplified model that accounts material volume fraction, Ap is the cathode cross-section area

0018-9545 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2020.2966266, IEEE
Transactions on Vehicular Technology

and lp is the cathode thickness. cpmax is the maximum solid-


phase concentration in the cathode and material constant. It is
assumed that the cell geometry and volume does not change
during cycling and operation. Thus, the equation (Eq.11) can
be differentiated with respect to the ε p and written as

∂SOC SOC 0 − SOC


≈ (13)
∂ε p εp
Fig. 4. Schematic of the ROTM model based BMS set-up for full pack voltage
It is also assumed that the derivative of cell voltage equal to and SOC prediction, and implementation on MCU.
the derivative of cell over-potential with respect to cell SOC as
expressed in equation (Eq.14). Thus, this assumption neglects
the variation in OCPs between cells for the cells with the same (N = 5) points is used for cathode OCP calculation to reduce
chemistry. fluctuations and improve prediction accuracy.
The full pack blended composite cathode ROTM model
∂V ∂V ∂OCP is embedded on the four MCUs (ATmega:2560&328,
≈1⇒ ≈ (14)
∂OCP ∂SOC ∂SOC Infineon:TC275&TC297). The Arduino IDE is used for code
Finally, equation (Eq.10,11) can be rearranged to provide embedding on ATmega2560&328 MCUs. Arduino IDE [53]
SOC and voltage prediction for ith cell in the battery pack as is a freely available tool for Arduino programming. For Infi-
follows. neon TC275&TC297 MCUs, the full pack blended composite
cathode ROTM is embedded using free-entry-toolchain R
SOC 0 − SOC IDE software. The free version of the free-entry-toolchain
R
SOCi = SOCr + ∆SOCi0 + ∆εi (15)
εp IDE was downloaded from the Hightec website [54]. The
simulated voltage data for full pack model is extracted from
∂OCP ∂OCP SOC 0 − SOC the MCUs using serial port and compared against the experi-
Vi = Vr + ∆SOCi0 + · ∆εi (16) mental measurements. A snapshot of the Infineon TC297TFT
∂SOC ∂SOC εp
board running at 0.33C discharge in real time and displaying
The ∆SOCi0 and ∆εi for the ith cell with respect to the the discharge time, SOCs and voltages of individual cells,
reference cell are determined by comparing the experimental including the reference cell in the battery pack is shown in
data (Fig.3) with the electrochemical model results. Once Fig. 5(a). The ROTM is also embedded to the basic MCU
determined, the same values are then used in the simplified (ATmega328) for single cell (Panasonic 18650b) and tested
pack model (equation Eq.15,16) for validation. for real time discharge. The snapshot is showing (Fig. 5(b))
the cell voltage, SOC prediction and the discharge current in
III. EXPERIMENTAL SETUP the real time.
An in-house developed battery pack was used for experi-
ment that consisted of commercially obtained 60Ah prismatic
cells with a blended composite (LMO+NMC+NCA) cathode
and graphite anode. The total voltage of the pack module and
individual cells are measured for a known current load. The
thermocouples are also placed on individual cells for tempera-
ture measurement. The pack module is charged using Constant
Current-Constant Voltage profile while discharge is done for a
range of constant current values at room temperature (298K).
Micro-controller is a key element in the BMS responsible Fig. 5. Snapshots of the on-board SOC and voltage estimation. The embedded
for all the control, data collection, monitoring, processing and ROTM for blended composite cathode battery pack is displaying the on-board
SOC and voltage estimation on Infineon TC297 application board, (b) on-
predictions. In the present study, four different MCUs are used board SOC and voltage estimation on ATmega328 MCU board for single cell
for full pack electrochemical model (ROTM) implementation, (Panasonic 18650b).
testing and validation.The overall scheme of ROTM based
state estimation is shown in Fig. 4.
Initially, the blended composite cathode ROTM Model IV. R ESULTS
for full pack is developed on MATLAB platform using
double precision. However, the MCUs (ATmega:2560&328, A. Validation
Infineon:TC275&TC297) are 8-bit micro-controller chips and The ROTM model for blended composite cathode is first
only support single precision calculations. The single precision validated against full order model (FOM).The FOM is imple-
causes the fluctuations due to the higher order polynomial fit mented in COMSOL Multiphysics framework as described in
used for the blended composite cathode OCP. To address this previous work by our group [51] by appropriately modifying
issue a lookup table is used in the embedded code instead of the existing framework of lithium ion cell model to include
polynomial fit for the blended composite cathode OCP, and composite cathode OCV. Both ROTM and FOM used exper-
the Least Square Regression Line Equation method with five imentally measured values values for model parameters as

0018-9545 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2020.2966266, IEEE
Transactions on Vehicular Technology

given in table II. In Fig. 6 a comparison of discharge profiles respect to a base voltage value of 3.85 V, which correspeonds
obtained from ROTM and FOM for discharge rates of 1C, 3C to the nominal voltage of the cell.
and 6C are provided. As can be seen ROTM agrees very well
with FOM model at 1C and 3C rates and reasonably well at TABLE II
A LIST OF THE PARAMETERS USED IN THE MODEL
6C. The larger deviations at high C-rates are primarily due to
volume averaging of properties and profile approximations for S.no Parameter Anode Cathode Separator
concentration values used in ROTM. The deviations in voltage
1 l 92.0 × 10−6 68.0 × 10−6 16.0 × 10−6
prediction between the two models are given in table III.
2 r 16.9 × 10−6 1.42 × 10−6
3 ε1 0.7535 0.6425
4.2 4 ε2 0.3 0.22 0.4
1C/M 5 D1 3.66 × 10−13 LMO: 6.6 × 10−15
1C/F NCA: 1.44 × 10−13
4 3C/M NCM: 1.44 × 10−13
3C/F 6 D2 2.479 × 10−10
6C/M
Cell Voltage (V)

7 Cma x 28200 LMO: 23334 1200


3.8 6C/F NCA: 49195
NCA: 49176
8 kr 7.77 × 10−9 LMO: 5.0 × 10−10
3.6 NCA: 1.0 × 10−10
NCA: 1.0 × 10−10
9 (E M D /R) 6.0136 × 103 6.0136 × 103
3.4 1
10 (E M k r /R) 3.60 × 103 2.4 × 103
11 θ0 0.41
3.2

TABLE III
3 P ERFORMANCE COMPARISON OFROTM WITH COMSOL F ULL O RDER
0 20 40 60 80 M ODEL
Capacity (Ah)
C rate Maximum Average
Fig. 6. Validation of ROTM(C/M) model with COMSOL Full Order Model Voltage deviation Voltage deviation
(C/F). 1 0.59% 0.23%
3 0.79% 0.3%
The ROTM model is then validated aginst experimental data 6 2.18% 1.02%
for a range of C-rates. The experimental data was obtained
from an individual blended composite cathode cell operating
at C-rates 0.1C, 1.0C, 3.0C and 6.0C at room temperature
(298K) in a controlled laboratory setup. The end of discharge
was defined when the cell voltage falls below 3.0V. On the
other hand, the ECM voltage is predicted using the resistance-
capacitance (RC) circuit structure and RC values as shown in
Fig. 7. The ECM consist of two pair RC and Warburg ele-
ment [50], where OCV(z(t)) denotes the SOC(z(t)) dependent
open circuit voltage of the battery, R0 denotes the ohmic resis-
tance, and describes the electrolyte and connection resistance
of the battery, R1,2 and C1,2 denotes charge transfer resistance,
Fig. 7. Schematic of the ECM along with resistance (R) values in Ohm and
and the double layer capacitance, and the Warburg element capacitance (C) values in Farads.
denotes the diffusional impedance for semi-infinite diffusion.
The ROTM voltage prediction is compared with both ECM
prediction and experimental data as shown in Fig. 8(a). The TABLE IV
ROTM perditions are better than ECM and agrees well with P ERFORMANCE COMPARISON OF THE THREE MODELS
the experimental data with more than 98% average accuracy
Model Type Maximum Average Execution
for all C-rates as shown in Fig. 8(a). The predicted temperature Voltage error Voltage accuracy time on Atmega2560
rise for these conditions are also validated against experimental OCV ∼ 13% 90% 0.0049 sec
results (Fig. 8(b)). The average accuracy of temperature ECM ∼ 6% 96% 0.0051 sec
prediction is greater than 90% for all C-rates.The models ROTM ∼ 3% 98% 0.0070 sec
(OCV, ECM, and ROTM) are also compared for voltage
prediction accuracy and on–board computational efficiency. It
is found that even though the ROTM is marginally slower than B. SOC and voltage estimation for a 12S1P configuration
OCV and ECM based models (table IV), it provides better battery pack
voltage accuracy in comparison to ECM and OCV models as The SOC and voltage prediction of the embedded model
shown in table IV. The error calculation is performed with for a 12S1P configuration pack on MCUs (ATmega2560,

0018-9545 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2020.2966266, IEEE
Transactions on Vehicular Technology

4.2 1
0.1C/E
0.1C/M 0.49
4 1.0C/E
1.0C/M 0.8 0.485
3.0C/E 0.48
3.8 3.0C/M
6.0C/E
0.475
Voltage

6.0C/M 0.6
0.1C/C 5400 5500 5600

SOC
3.6 1.0C/C
Cell-R/E
3.0C/C
6.0C/C Cell-R/M
3.4 0.4 Cell-3/E
Cell-3/M
Cell-4/E
3.2 0.2 Cell-4/M
(a) Cell-12/E
3 Cell-12/M
0 20 40 60 80 0
Capacity (Ah) 0 2000 4000 6000 8000 10000 12000
16
0.1C/E Time (sec)
0.1C/M
14 1.0C/E
1.0C/M
Fig. 9. Validation of SOC model estimation (M) against experimental (E)
Temperature rise (K)

12 3.0C/E measurement for reference cell-R, cell-3, cell-4, and cell-12 at 0.33C-rate
3.0C/M discharge.
10 6.0C/E
6.0C/M
8
4
6 Cell−R
Voltage
Experiment
4 ATmega256
3.5
2 TC275
(b) TC297
0 3
0 20 40 60 80 0 2000 4000 6000 8000 10000 12000
Capacity (Ah) 4
Cell−3
Voltage

Fig. 8. Validation of single cell ROTM(C/M) model with experiments(C/E) Experiment


and ECM (C/C) model. (a) Voltage validation at different C-rates, (b) 3.5 ATmega256
Temperature validation. TC275
TC297
3
InfineonTC275&TC297) are compared with the experimental 0 2000 4000 6000 8000 10000 12000
4 Time(sec)
data for a range of C-rates. In the 12S1P configuration pack, Cell−4
the 1st cell is taken as a reference to predict the SOC and
Voltage

Experiment
voltage of other cells. As individual cell have different charge- ATmega256
3.5
discharge characteristic with respect to the reference cell (R), TC275
the variation in parameters like ∆SOC0 and ∆ε with respect TC297
to the reference cell (R) are estimated using ‘Battery and Fuel 3
Cell Module’ of commercially available finite element package 0 2000 4000 6000 8000 10000 12000
Time(sec)
COMSOL Multiphysics R
4.4 [49], [52] for each individual 4
Cell−12
cell. It is found that only cell number 3 of the 11 (#3, 4 and
Voltage

Experiment
12) have a significant variation in terms of initial SOC (SOC 0 )
3.5 ATmega256
and active material loading (ε) concerning to the reference cell TC275
(R). These variations have been tabulated in Table V. TC297
3
TABLE V 0 2000 4000 6000 8000 10000 12000
VARIATION IN INITIAL SOC AND MATERIAL FRACTION BETWEEN CELLS Time(sec)
W. R . T REFERENCE CELL IN THE PACK .
Fig. 10. Validation of on-board voltage model estimation (Atmega256,TC275
Cell number ∆SOC0 ∆ε and TC297 MCU) against experimental measurement for reference cell-R,
cell-3, cell-4, and cell-12 at 0.33C-rate discharge.
Cell-3 0.0117 0.0003656
Cell-4 0.0070 0.0009624
Cell-12 0.0070 0.0015010
First, the SOC and voltage of the reference cell SOCr ,Vr
The SOC and voltage estimation for the reference cell and is estimated, then the estimations are extended to the other
these 3 cells with variations are shown in the Fig. 9,10. cells using a simplified pack model equations (Eq.15,16).

0018-9545 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2020.2966266, IEEE
Transactions on Vehicular Technology

An average SOC and voltage prediction accuracy > 96%


is achieved against the experimental measurements for the 200 (a) FDD FDD: Fast dynamic discharge
SDD: Slow dynamic discharge

Current (Amp)
reference and other cells. The embedded model predictions FDD
SD: Slow discharge
of cell voltage for different MCUs are also compared. No 100 SDD FC: Fast charge
significant difference is observed between the results obtained SC: Slow Charge
R: Rest
from ATmega2560 and InfineonTC275&TC297 as shown in R
0
Fig. 10. SC SC
R SD R
SC
FC
C. Computational response on different MCUs -100
0 2 4 6 8 10 12 14 16 18
A detailed analysis of computational time of embedded full (b) Experiment
pack ROTM is also done for all the MCUs. The computational 4
ROTM
time required for 0.1sec step prediction on different MCUs ECM

Voltage
are compared together. For processing time, the full pack
3.5
ROTM code is divided into two parts: (a) Blended composite
OCP generation (required for blended composite cathode OCP
lookup table), and (b) voltage prediction for input current. The 3
blended composite OCP generation is one time calculation,
and is not required for further calculation until MCU is 0 2 4 6 8 10 12 14 16 18
1
switched off. The computational time is investigated for a (c) Experiment
range of C-rates with 0.1sec voltage and SOC prediction. 0.8 0.12
ROTM
0.1
Table VI summarizes the computational time incurred by 0.08
ECM
the different MCUs for the full pack voltage prediction. It is 0.6
SOC
0.06

observed that the computational time is independent of C-rate, 0.4


but depends on MCUs speed.
0.2
TABLE VI
C OMPUTATIONAL TIME REQUIRED FOR 0.1 SEC TIME STEP PREDICTION ON 0
0 2 4 6 8 10 12 14 16 18
MCU S .
Time (hour)
MCU Speed (MHz) Time(sec)/step
ATmega:2560&328 16 7.0e-3 Fig. 11. ROTM voltage and SOC prediction for dynamic current load: (a)
Infineon:TC275 200 7.8e-5 current load for dynamic condition, (b) On-board model voltage prediction
Infineon:TC297 300 6.0e-5 (ROTM and ECM) against experimental measurements, (c) SOC prediction
(ROTM and ECM) against the Coulomb counting SOC estimation.

D. Voltage prediction for dynamic operating condition note that the during rest period (R), there is voltage recovery
The same methodology is extended for voltage and SOC due to the relaxation of internal chemical potential gradients
prediction for dynamic current load for the 12S1P pack. The towards attaining the equilibrium potential (OCP). This voltage
dynamic cycle (shown in Fig. 11a) consists of six major recovery is more prominent at lower voltages (∼3.0 volt). This
operating conditions: (1) fast dynamic discharge (FDD), (2) voltage recovery during the rest period is captured (shown in
slow dynamic discharge (SDD), (3) slow discharge (SD), (4) Fig 11b) using following equation (Eq.17).
fast charge (FC), (5) slow charge (SC),and (6) rest (R). The
sequence of these operating conditions considered for dynamic
Vrtec = Vrt−∆t
ec exp(−∆t/tr ec ) ⇒ Vcell = Vcell − Vr ec
t 0 t
(17)
current load testing are as follow: FC → FDD → SC → SDD
→ SC → FDD → R → SC → R → SD → R at the end. where Vr ec is the recovery voltage, ∆t is the time step, tr ec
The experimental voltage data is obtained from the 12S1P is voltage recovery rate in seconds (tr ec = 1200), and Vcell 0

pack for each individual cell under dynamic current load is voltage at beginning of rest period. This recovery voltage
condition as shown in Fig. 11a. The on-board (MCU: Infi- is subtracted from the voltage Vcell0 to capture the voltage
neonTC275) voltage and SOC predictions using both ROTM recovery during rest. When this correction was implemented
and ECM models are shown in Fig. 11(b,c) along with in the ROTM, the voltage recovery is captured accurately
experimental measurements for the reference cell. It has been for the rest sections in the dynamic cycles as shown in the
observed that the cell voltage increases rapidly during FC Fig. 11(b). We can see from Table VII that ROTM out
and reduces in response to FDD. It is worthwhile to point performs ECM model under dynamic load conditions both in
out that the per step computational time for dynamic current terms of maximum and average errors in voltage estimation.
load remains the same as compared to the constant current A better voltage prediction accuracy is certainly desirable to
case as presented previously. The voltage accuracy becomes obtain more accurate estimates of available power, particularly
of great important for available power estimation in EVs towards the end of discharge in EVs. The data points in the
during real drive cycles. Additionally, it is also important to rest period were excluded in this error analysis. Along with

0018-9545 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2020.2966266, IEEE
Transactions on Vehicular Technology

10

voltage, the estimated cell SOC is also compared with SOC Voltage estimation, and have demonstrated its practicality by
from Coulomb counting for both ROTM and ECM and shown importing it onto four different micro-controller (MCU) chips
in Fig. 11(c). For ROTM model, the cell SOC is estimated (ATmega:2560&328, Infineon:TC275&TC297). The on-board
continuously in reference to the blended composite cathode ROTM SOC and voltage estimations compare well against
SOC. The estimated cell SOC using both ROTM is able to experimentally measured cell SOC and voltage for a 12S1P
capture these SOC fluctuations due to variation in load current configuration battery pack for both constant and dynamic
and shows good agreement with the SOC with experiments operating load conditions with > 98% average accuracy. The
with > 98% average accuracy during full dynamic cycle. developed on-board ROTM model is fast, robust, accurate and
Although both ROTM and ECM models are able to capture easy to implement on a range MCUs, including the most
the SOC variation, the ECM SOC prediction is slightly higher basic MCUs (ATmega2560&328) for BMS applications. Our
than the experimentally observed SOC. This error can be due work shows that although computational speed depends on the
to the unknown initial SOC, capacity fading, self-discharge MCU’s specification, the estimation accuracy is independent
and can lead to the significant error in long time. of the type of MCU used. The on-board implementation
of this model will offer significant edge over conventional
TABLE VII (ECM, OCV based methods) in terms of SOC and voltage
C OMPARISON OF ERROR IN VOLTAGE PREDICTION
accuracy, while improving the pack performance with minimal
Model Type Maximum Average
computational requirement. The present work demonstrates
Voltage error Voltage accuracy the potential of a physics based electrochemical model for
ECM ∼ 8% 98.3% on-board SOC and voltage estimation for full battery pack
ROTM ∼ 5% 99.6% without any requirement of expensive hardware, and will pave
the way for electrochemically and thermally informed on-
board fuel-gauges for electric vehicles. As the base model
V. D ISCUSSION in our work is electrochemical, it opens up the possibility to
The results we have presented makes a compelling case incorporate low temperature effect and different degradation
for the use of reduced order electrochemical models for models into it. Such a model can then be used to predict the
on-board state estimation since they can be quite compact, life expectancy of battery tailored for each individual user.
computationally efficient enough to run on a power manage- Along with driver/user profile this model can be developed
ment chip and offer better accuracy compared to ECM based into an onboard prognostic tool. Future work will focus on
models. Additionally, ROTM based models are ameanable to development of on-board SOH and available power estimation
incorporation of degradation models to predict state-of-health by including relevant degradation modes for both anode and
that can help maintain prediction accuracy in the presence of blended composite cathode.
degradation. While we have not included a degradation model
in this work, it presents scope for future work in that direction. N OTATION
The accuracy of voltage and SOC estimation is critical in
estimating available power, that is the power that the battery Constants & variables
can deliver to the vehicle drivetrain. At extreme conditions, R: Gas constant, 8.314 (J mol −1 K −1 ).
such as low temperatures and high C-rates or in the presence F: Faraday’s constant, 96487 (C mol −1 ).
of significant battery degradation, it is believed that the use of T: Temperature, (K).
electrochemistry based models such as the ROTM is expected Tamb : Ambient temperature, (K).
to perform better than ECM models in predicting available I(t): External current density at time t, (A m−2 ).
power since they are able to represent thermally activated t+ : Electrolyte transference number.
electrochemical processes in the battery precisely. ln, ls , lp : Thickness of negative electrode, separator and
Despite the above advantages, there are many challenges positive electrode, (m).
that one may face in practical implementation of ROTM. Rn, Rp : Radius of active material particle in the negative and
For example, in realistic drive load conditions and as the positive electrode, (m).
cells age, it is imperative that an precise estimate of initial ε1k , ε2k : Volume fraction of active material in solid and
SOC of all cells is made for accurate state estimation to be electrolyte phase.
made. This challenge can be addressed by having a robust jk : Local surface reaction rate in k = n, p, (mol m−2 s−1 ).
initilizer that estimates the initial SOC of individual cells by jk0 : Over-potential independent rate pre-factor,
compensating for any errors. This would require estimating (mol m−2 s−1 ).
degradation of each cell using an appropriate degradation krk : Suface reaction rate constant in k = n, p electrodes,
model or by estimating degradation related parameters such (mol −0.5 m2.5 s−1 ).
as SEI resistance or capacity fade on-line. c1k , c2k : Concentration in solid and electrolyte phase respec-
tively, (mol m−3 ).
VI. C ONCLUSION max : The maximum value of surface concentration in solid
c1k
In this work, we developed an on-board reduced order phase, (mol m−3 ).
electrochemical thermal model (ROTM) for a 12S1P configu- c̄1rk : Scaled average radial solid phase concentration in k =
ration composite-cathode battery pack for real time SOC and n, p electrodes.

0018-9545 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2020.2966266, IEEE
Transactions on Vehicular Technology

11

c2ik , q2ik ,: interfacial electrolyte concentrations and [14] H. Aung, K. S. Low, S. T. Goh, "State-of-charge estimation of lithium-
fluxes,(mol m−3 ),(mol m−2 s−1 ). ion battery using square root spherical unscented kalman filter (sqrt-ukfst)
in nanosatellite," IEEE Trans. Power Electron., 30, 4774–4783, 2015.
D1 : Solid diffusivity of k = n, p electrode particles (material [15] X. Tang, B. Liu, F. Gao, F., "State of charge estimation of lifepo4 battery
property),(m2 s−1 ). based on a gain-classifier observer," Energy Procedia 105, 2071–2076,
D2 : Electrolyte diffusivity (material property),(m2 s−1 ). 2017.
brug [16] W. Waag, C. Fleischer, D. U. Sauer, "Critical review of the methods
D2k = D2 × ε2k : Effective electrolyte diffusivity in porous for monitoring of lithium-ion batteries in electric and hybrid vehicles," J.
k = n, p domains. Power Sources 258, 321–339, 2014.
Φ1k ,Φ2k = Solid and liquid phase potential, V. [17] J. P. Rivera-Barrera, N. Munoz-Galeano, H. O. Sarmiento-Maldonado,
"SoC Estimation for Lithium-ion Batteries: Review and Future Chal-
Uk : Open circuit voltage of k = n, p electrodes. lenges," Electronics 6, 102, 2017.
SOCk = c1k /c1k max : State of charge of k = n, p electrodes.
[18] C. R. Lashway, O. A. Mohammed, "Adaptive battery management and
Vcell : Cell voltage, V. parameter estimation through physics-based modeling and experimental
Vr : Reference cell voltage, V. verification," IEEE Trans. Transp. Electrification 2, 454–464, 2016.
[19] L. Lavigne, J. Sabatier,J. M. Francisco, F. Guillemard, A. Noury,
Vre : Recovery voltage, V. "Lithium-ion open circuit voltage (OCV) curve modelling and its ageing
θ 0, θ 1 : Constants representing the convection coefficient adjustment," J. Power Sources 324, 694–703, 2016.
and Ohmic resistance. [20] N. A. Windarko, J. Choi, G. B. Chung, "SOC Estimation of LiPB Bat-
teries Using Extended Kalman Filter Based on High Accuracy Electrical
Model," Proceedings of Power Electronics and ECCE Asia (ICPE and
ECCE), IEEE 8th International Conference, Jeju, Korea, 2015-2022, 2011.
Subscripts and operators: [21] J. Yan, G. Xu, Y. Xu. B. Xie. "Battery State-of-Charge Estimation Based
on H-Filter for Hybrid Electric Vehicle," Proceedings of 10th International
1, 2: Solid active material, liquid electrolyte phases. Conference on Control, Automation, Robotics and Vision (ICARCV 2008),
k = n, s, p: Negative electrode, separator, and positive Hanoi, Vietnam, 464-469, 2008.
electrode. [22] H. Zhang, M. Y. Chow, "Comprehensive Dynamic Battery Modeling for
i: Cell number in the full pack. PHEV Applications," Proceedings of IEEE in Power and Energy Society
General Meeting, Minneapolis, MN, USA, 1-6 2010.
h·i: Volume average,defined in the three domains, n, s, p. [23] Q. Wang, J. Wang, P. Zhao, J. Kang, F. Yan, C. Du, "Correlation between
the model accuracy and model-based soc estimation," Electrochim. Acta
228, 146–159, 2017.
[24] M. Thele, O. Bohlen, D. U. Sauer, E. Karden "Development of a voltage-
Arrhenius function: behavior model for NiMH batteries using an impedance-based modeling
EM
M(Tk ) = M(Tr e f ) exp[ R k ( T1k − Tr1e f )] concept," J. Power Sources 175, 635–643, 2008.
where M is any variable and E is activation energy. [25] S. Lee, J. Kim, J. Lee, B. H. Cho, "State-of-charge and capacity
estimation of lithium-ion battery using a new open-circuit voltage versus
state-of-charge," J. Power Sources 185, 1367–1373, 2008.
R EFERENCES [26] V. H. Johnson, "Battery performance models in ADVISOR," J. Power
Sources 110, 321–329, 2002.
[1] H. Rahimi-Eichi, U. Ojha, F. Baronti, M. Y. Chow, "Battery management [27] U. Krewer, F. Roder, E. Harinath, R. D. Braatz, B. Bedurftig, and R.
system: An overview of its application in the smart grid and electric Findeisen, "Review—Dynamic Models of Li-Ion Batteries for Diagnosis
vehicles," IEEE Ind. Electron. Mag. 7, 4–16, 2013. and Operation: A Review and Perspective," J. Electrochem. Soc., 165 (16)
[2] L. Lu, X. Han, J. Li, J. Hua, M. Ouyang, "A review on the key issues for A3656-A3673, 2018.
lithium-ion battery management in electric vehicles," J. Power Sources,226, [28] M. W. Verbrugge, R. S. Conell, "Electrochemical Characterization of
272-288, 2013. High-Power Lithium Ion Batteries using Triangular Voltage and Current
[3] R. Spotnitz, "Simulation of capacity fade in lithium-ion batteries," J. Excitation Sources," J. Power Sources, 174(1), 2, 2007.
Power Sources, 113, 72-80, 2003. [29] Y. Hu, S. Yurkovich, Y. Guezennec, B. J. Yurkovich, "Electro-Thermal
[4] A. T. Stamps, C. E. Holland, R. E. White, E. P. Gatzke, "Analysis of Battery Model Identification for Automotive Applications," J. Power
capacity fade in a lithium ion battery," J. Power Sources 2005, 150, 229- Sources, 196(1), 449, 2011.
239, 2005. [30] R. L. Sacci, F. Seland, D. A. Harrington, "Dynamic Electrochemical
[5] Q. Zhang, R. E. White, "Capacity fade analysis of a lithium ion cell," J. Impedance Spectroscopy, for Electrocatalytic Reactions," Electrochimica
Power Sources, 179, 793-798, 2008. Acta, 131, 13, 2014.
[6] C. D. Rahn, C. Y. Wang., "Battery Systems Engineering," John Wiley
[31] K. Dong, J. Kirchev, J. Kowal, Y. Bultel, "DynamicModeling of Li-ion
and Sons Inc, State College, PA, 2013.
Batteries using an Equivalent Electrical Circuit," J. Electrochem. Soc., 158,
[7] J. Cao, N. Schofield, A. Emadi, "Battery balancing methods: A compre-
A326, 2011.
hensive review," in Proc. IEEE Conf. Vehicle Power and Propulsion, 1-6
2008. [32] J. Illig, M. Ender, T. Chrobak, J. P. Schmidt, D. Klotz, E. Ivers-
[8] J. Chen, Q. Ouyang,C. Xu, H. Su, "Neural network-based state of Tiffee, "Separation of Charge Transfer and Contact Resistance in LiFePo4-
charge observer design for lithium-ion batteries," IEEE Trans. Control Syst. cathodes by Impedance Modeling," J. Electrochem. Soc., 159(7), A952,
Technol., 1-9, 2017. 2012.
[9] M. Cacciato, G. Nobile, G. Scarcella, G. Scelba, "Real-time model-based [33] T. Stuart, F. Fang, X. P. Wang, C. Ashtiani, A. Pesaran, "A modular
estimation of SoC and SoH for energy storage systems," IEEE Trans. battery management system for HEVs," Future Car Congress 2002,
Power Electron., 32, 794–803, 2017. doi:10.4271/2002-01-1918, 2002.
[10] H. Chaoui, H. Gualous, "Adaptive state of charge estimation of lithium- [34] M. A. Hannan, M. S. H. Lipu, A. Hussain, A. Mohamed, "A review
ion batteries with parameter and thermal uncertainties," IEEE Trans. of lithium-ion battery state of charge estimation and management system
Control Syst. Technol. 25, 752–759, 2017. in electric vehicle applications: Challenges and recommendations," Renew.
[11] H. He, R. Xiong, J. Fan, "Evaluation of lithium-ion battery equivalent Sustain. Energy Rev. 78, 834–854, 2017.
circuit models for state of charge estimation by an experimental approach," [35] S. Piller, M. Perrin, A. Jossen, "Methods for state-of-charge determina-
Energies 4, 582–598, 2011. tion and their applications," J. Power Sources, vol. 96,1,113-120, 2001.
[12] C. Zhang, K. Li, J. Deng,S. Song, "Improved realtime state-of-charge [36] S. B. Chikkannanavar, D. M. Bernardi, L. Liu, "A review of blended
estimation of liFePO4 battery based on a novel thermoelectric model," cathode materials for use in Li-ion batteries," J. Power Sources, 248, 91-
IEEE Trans. Ind. Electron. 64, 654–663, 2017. 100, 2014.
[13] H. Aung, K. S. Low, "Temperature dependent state-of-charge estimation [37] P. Albertus, J. Christensen J. Newman, "Experiments on and Modeling
of lithium ion battery using dual spherical unscented kalman filter" IET of Positive Electrodes with Multiple Active Materials for Lithium-Ion
Power Electron. 8, 2026–2033, 2015. Batteries," J. Electrochem. Soc., 156(7),A606-A618, 2009.

0018-9545 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2020.2966266, IEEE
Transactions on Vehicular Technology

12

[38] S. T. Myung, M. H. Cho, H. T. Hong, T. K. Kang, C. S. Kim,


"Electrochemical evaluation of mixed oxide electrode for Li-ion secondary
batteries: [Li1. 1Mn1. 9O4] and [LiNi0. 8Co0. 15Al0. 05O2]," J. Power
Sources, 146, 222, 2005.
[39] S. K. Jeong, J. S. Shin, K. S. Nahm T. P. Kumar, A. M.
Stephan, "Electrochemical studies on cathode blends of LiMn2O4 and
Li[Li1/15Ni1/5Co2/5Mn1/3O2]," Materials Chemistry and Physics, 111,
213-217, 2008.
[40] T. F. Fuller, M. Doyle, J. Newman, "Simulation and Optimization of the
Dual Lithium Ion Insertion Cell," J. Electrochem. Soc., 141, 1, 1994.
[41] J. Newman, "Optimization of Porosity and Thickness of a Battery
Electrode by Means of a Reaction-Zone Model," J. Electrochem. Soc.,
142, 97, 1995.
[42] N. Lotfi, R. G. Landers, J. Li, J. Park, "Reduced-order electrochemical
model-based soc observer with output model uncertainty estimation," IEEE
Trans. Control Syst. Technol, 25, 4, pp 1217-30, 2017.
[43] S.J. Moura, F.B. Argomedo, R. Klein, A. Mirtabatabaei, M. Krstic,
"Battery state estimation for a single particle model with electrolyte
dynamics," IEEE Trans. Control Syst. Technol. 25, 453–468, 2017.
[44] V. R. Subramanian, D. Tapriyal, R. E. White, "A boundary condition for
porous electrodes," Electrochem. Solid-State Lett., 7, A259–A263, 2004.
[45] V. S. Kumar, "Reduced order model for a lithium ion cell with uniform
reaction rate approximation," J. Power Sources, 222, 426-441, 2013.
[46] Z. Mao, M. Farkhondeh, M. Pritzker, M. Fowler, Z. Chen, M. Sa-
fari "Model-Based Prediction of Composition of an Unknown Blended
Lithium-Ion Battery Cathode," J. Electrochem. Soc., 162(4), A716-721,
2015.
[47] A. Bartlett, J. Marcicki, S. Onori, G. Rizzoni, X. G. Yang, T. Miller,
"Electrochemical Model-Based State of Charge and Capacity Estimation
for a Composite Electrode Lithium-Ion Battery," IEEE Transactions on
Control Systems Technology, 24(2), 384-399, 2016.
[48] A. E. Mejdoubi, A. Oukaour, H. Chaoui, H. Gualous, J. Sabor, Y. Sla-
mani, "State-of-charge and state-of-health lithium-ion batteries’ diagnosis
according to surface temperature variation," IEEE Trans. Ind. Electron. 63,
2391–2402, 2016.
[49] S. Basu, R. S. Patil, S. Ramachandran, K. S. Hariharan, S. M. Kolake, T.
Song. D. Oh, T. Yeo, S. Doo, "Non-isothermal electrochemical model for
lithium-ion cells with composite cathodes," J. Power Sources, 283, 132-50,
2015.
[50] B. Boukamp, "Electrochemical Impedance Spectroscopy," Nano-
Electrolysis, U.Leiden, 24028th Nov, 2008.
[51] R. S. Patil, A. Khandelwal, K. Y. Kim, K. S. Hariharan, S. M. Kolake,
"Model Based Design of Composite Carbonaceous Anode for Li-Ion
Battery Fast Charging Applications," J. Electrochem. Soc., 166 (6), A1185-
A1196, 2019.
[52] COMSOL 4.4 Documentation, COMSOL Inc., 2014.
[53] https://www.arduino.cc/en/Main/Software/.
[54] http://free-entry-toolchain.hightec-rt.com/.

0018-9545 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

Вам также может понравиться