You are on page 1of 6

Modeling, estimation, and control challenges for lithium-ion batteries

Nalin A. Chaturvedi

, Reinhardt Klein
,
,
Jake Christensen

, Jasim Ahmed

and Aleksandar Kojic

AbstractIncreasing demand for hybrid electric vehicles


(HEV), plug-in hybrid electric vehicles (PHEV) and electric
vehicles (EV) has forced battery manufacturers to consider
energy storage systems that are better than contemporary lead-
acid batteries. Currently, lithium-ion (Li-ion) batteries are be-
lieved to be the most promising battery system for HEV, PHEV
and EV applications. However, designing a battery management
system for Li-ion batteries that can guarantee safe and reliable
operation is a challenge, since aging and other performance
degrading mechanisms are not sufciently well understood. As
a rst step to address these problems, we analyze an existing
electrochemical model from the literature. Our aim is to present
this model from a systems & controls perspective, and to bring
forth the research challenges involved in modeling, estimation
and control of Li-ion batteries. Additionally, we present a novel
compact form of this model that can be used to study the Li-
ion battery. We use this reformulated model to derive a simple
approximated model, commonly known as the single particle
model, and also identify the limitations of this approximation.
I. BACKGROUND AND MOTIVATION
Lithium ion (Li-ion) batteries are popular as a reliable
source of power in mobile phones, and portable electronic
devices. Li-ion batteries are favored over other battery
technologies since they provide one of the best energy-to-
weight ratios, have no memory effect, and have a slow self-
discharge. Of late, there is a push to commercialize Li-
ion batteries for use in automotive and other applications
(such as aerospace or defense) due to their high energy
density [1]. In the automotive sector, increasing demand
for hybrid electric vehicles (HEVs), plug-in hybrid electric
vehicles (PHEVs), and electric vehicles (EVs) has forced
consideration of other promising battery technologies such
as Li-ion batteries to replace existing lead-acid batteries.
Unfortunately, this replacement is challenging due to the
large power and energy demands placed on such batteries,
while guaranteeing its safe operation.
A battery typically consists of the battery itself and the
battery management system (BMS). A BMS is composed
of hardware and software system that controls charging and
discharging of the battery while guaranteeing reliable and
safe operation [2]. It also takes care of other functions such as
cell balancing. The design of a sophisticated BMS becomes
critical for Li-ion batteries, since these batteries can ignite
and explode when overcharged or under abuse conditions
[2], [3]. To design and build the BMS for Li-ion batteries,
a model is required that can describe the battery dynamics.
One of the key tasks of a BMS is to observe the states of

Nalin.Chaturvedi@us.bosch.com, Robert Bosch LLC, Res. &


Tech. Center, Palo Alto, CA 94304.

Otto-von-Guericke University, Institute of Automation Engineering,


Magdeburg 39106, Germany.
the battery and track physically relevant parameters as the
battery ages.
The outline of the paper is as follows. In section II,
we intuitively explain the fundamentals of a Li-ion battery.
In section III, we present equations describing a Li-ion
cells dynamic behavior. The modeling is based on using
electrochemical principles to develop a physics-based model
in contrast to equivalent circuit models [4], [5], [6], [7], [8].
While electrochemical models have been developed earlier,
our goal in this paper is to present this model with a
perspective that appeals to people with diverse backgrounds.
In section IV, we develop a novel compact form of this
model that can be used to study the full Li-ion battery model.
Previous work in this eld [9], [10], [11], [12] avoids such
detailed mathematical constructions since their primary aim
is cell-optimization using numerical simulations. However,
if the intention is to build control or estimation algorithms
for BMS, then a more mathematical and a systems-and-
controls-oriented understanding of the Li-ion battery model
becomes imperative. In this paper, we ll this gap and
additionally derive an approximation of the Li-ion battery
model that is used in the literature [13], [14] in section V.
This approximate model is obtained from the new compact
form derived earlier in section IV, thus demonstrating its use.
Next, in section VI, a comparison between the full model
and the reduced model is presented, identifying the domains
where the approximation holds. In section VII, we present
estimation and control issues for BMS in Li-ion batteries
and present the current status of research in BMS. Finally,
we present the current solutions to estimation and control
problems, and conclude by mentioning future work.
II. INTERCALATION-BASED BATTERIES
The commonly available Li-ion cell is an intercalation-
type cell [15]. The term intercalation-type implies that the
electrodes have a lattice structure and charging (discharging)
the cell causes the Li ions to leave the positive (negative)
electrode and enter the lattice structure of the negative
(positive) electrode. This process of ions being moved in and
out of an interstitial site in the lattice is called intercalation.
A typical Li-ion battery (Figure 1) has four main com-
ponents. A porous negative electrode in a Li-ion cell is the
negative terminal of the cell. It is usually made of graphite.
Similarly, a porous positive electrode is the positive terminal
of the cell. It can have different chemistries, but is usually a
metal oxide or a blend of multiple metal oxides. A separator
is a thin porous medium that physically insulates the negative
from the positive electrode. It is an electrical insulator
that does not allow electrons to ow between the positive
2010 American Control Conference
Marriott Waterfront, Baltimore, MD, USA
June 30-July 02, 2010
WeC12.5
978-1-4244-7427-1/10/$26.00 2010 AACC 1997
and negative electrodes. However, being porous, it allows
ions to pass through it via the electrolyte. The electrolyte
is a concentrated solution that has charged species. These
charged species can move in response to an electrochemical
potential. The key idea behind storing energy in a Li-ion
cell is that lithium has different potentials when placed in
an interstitial site of the lattice of the positive or negative
electrodes solid active material particles (sometimes referred
to as solid particles; see Figure 1). This potential of an
electrode is usually expressed as a function of the lithium
concentration in the electrode, referred to as the open-circuit
potential (OCP) of the electrode [15].
III. MODELING APPROACH
In this section, we present electrochemical equations that
describe the physics of a Li-ion battery. In a Li-ion battery,
lithium can exist either in the dissolved state in the electrolyte
or in the solid phase in the particles (interstitial site), as
shown in Figure 1, at every point along the X-axis.
Fig. 1. Schematic of model of an intercalation cell.
In Figure 1, we assume in our model that spherical solid
particles of radius R
p
(shown by red and green spheres) exist
at every point x along the X-axis, which correspond to the
lattice sites in the electrode. These particles are immersed in
a sea of electrolyte shown by the wavy, dashed line.
The seven variables required to describe this 1D model are
the current i
s
(x, t) in the solid particles (or solid electrode),
the current i
e
(x, t) in the electrolyte, the electric potential

s
(x, t) in solid electrode, the electric potential
e
(x, t)
in electrolyte, the surface molar ux density j
n
(x, t) at the
surface of the spherical particle, the concentration c
e
(x, t) of
the electrolyte, and the concentration c
s
(x, r, t) of lithium in
the solid electrode particles at a distance r from the center
of a spherical particle located at x at time t [15].
The input to the model is the current that is applied to
the battery given by I(t), and the output of the model is the
corresponding output voltage V given by
V (t) =
s
(0
+
, t)
s
(0

, t), (1)
where 0
+
and 0

correspond to the two ends of the battery


as shown in Figure 1.
We next present the equations that describe the behavior of
the seven variables stated above. In the ensuing sections, we
assume, without loss of generality, that the cross-sectional
area of the separator is unity.
The rst two equations follow from Kirchoffs law, i
s
+
i
e
= I, and from Ohms law

s
(x, t)
x
=
i
e
(x, t) I(t)

, (2)
where R
+
is the effective electronic conductivity of
the solid. Equation (2) has no explicit boundary conditions.
However, i
e
= 0 at 0
+
and 0

, and i
e
= I at L
+
and L

.
The third equation relating
e
and i
e
is given by

e
(x, t)
x
=
i
e
(x, t)

+
2RT
F
_
1 t
0
c
_
_
1 +
d ln f

d ln c
e
(x, t)
_
ln c
e
(x, t)
x
, (3)
where F is Faradays constant, R is the universal gas
constant, T is the temperature and f

is the mean molar


activity coefcient in the electrolyte. Also, is the ionic
conductivity of the electrolyte, and t
0
c
is the transference
number with respect to the solvent velocity. The boundary
condition of (3) is
e
(0
+
) = 0.
The fourth equation describing the relationship between
c
e
and i
e
is given by
c
e
(x, t)
t
=

x
_
D
e
c
e
(x, t)
x
_
+
1
F
e
(t
0
a
i
e
(x, t))
x
, (4)
where D
e
is the diffusion coefcient,
e
is the volume
fraction of the electrolyte and t
0
a
is the transference number
for the anion. The boundary conditions for the above equa-
tion are that the uxes of the ions are zero at the current
collectors. Thus,
c
e
x

x=0

=
c
e
x

x=0
+
= 0.
The fth equation relating c
s
and j
n
is described by
diffusion as
c
s
(x, r, t)
t
=
1
r
2

r
_
D
s
r
2
c
s
(x, r, t)
r
_
, (5)
where r is the radial dimension of the particles in the active
material. The boundary and initial conditions for (5) are
given by
c
s
r

r=0
= 0,
c
s
r

r=Rp
=
1
D
s
j
n
, c
s
(x, r, 0) = c
0
s
.
The sixth equation relating i
e
to j
n
is
i
e
(x, t)
x
= aFj
n
(x, t), (6)
where a =
3s
Rp
is a constant. The boundary condition is
i
e
= I at L
+
and L

and i
e
= 0 at 0
+
and 0

.
The seventh and the nal equation relating j
n
to c
e
, c
s
,

s
and
e
is the Butler-Volmer equation [16], [17]
j
n
=
i0
F
_
exp
_
aF
RT

s
_
exp
_
cF
RT

s
_
, (7)
where
a
and
c
are transport coefcients, i
0
is the exchange
current density, and
s
represents the overpotential. The
1998
overpotential
s
represents the change in potential that a
charged species would go through as it passes through the
spherical particle into the electrolyte, and is given as

s
(x) =
s
(x)
e
(x) U(c
ss
(x)) FR
f
j
n
(x), (8)
where for any given t, c
ss
(x) c
s
(x, R
p
). Note that the
equation for the molar ux density j
n
is algebraic. Thus,
all equations need to be solved together to satisfy the
algebraic constraint at every time t. The exchange current
density i
0
in (8) is given by i
0
= r
e
_
c
e
(x, t)(c
s,max

c
ss
(x, t))
_
a
c
ss
(x, t)
c
, where r
e
and c
s,max
are constants.
In summary, given that i
s
+i
e
= I, the six equations (2),
(3), (4), (5), (6) and (7) are solved for with applied current I
as the input and the output given by the voltage V as dened
in (1).
IV. MODEL REDUCTION
In section III, we presented the mathematical equations
describing a Li-ion battery. In this section, we analyze these
equations and present a novel compact structure for the Li-
ion battery model. This reformulation involves changing only
the structure and hence, no information of the model is lost
in the process.
A. Mathematical notation
Let C
r
(a, b) denote the function-space of all real-valued,
r-times continuously differentiable functions with domain
(a, b) R. By abuse of notation, we include 1 value for
r. A function in C
1
(a, b) is discontinuous, but its integral
exists and is continuous. More precisely, this function-space
is the Sobolev space H
1
(R) [18]. Note that C

(R)
C

(R) . . . C
0
(R) C
1
(R), where C

(R) represents
the space of analytic functions.
We next introduce the concept of a function-space map-
ping. In this article, we dene a function-map F as a mapping
that takes an element of C
r
(a, b) to C
q
(a, b) where r and q
are some integers. In other words, F : C
r
(a, b) C
q
(a, b) is
a function-map. We also introduce the notion of restriction of
a function by means of an example. Suppose g(x, t) R is a
function, that is, g : RR R. Then, the t-restriction of the
function g is dened as g
t
(x) g(x, t). Hence, g
t
: R R
denotes the value of g for some xed t.
B. Reduction of PDE system
We begin by reducing the system of ve PDEs and one
algebraic equation to two PDEs with time derivatives and
algebraic constraints. We focus on any one of the positive or
negative electrodes for this purpose. The spatial domain is
assumed to run from 0 to L. Thus, for the positive electrode
and the negative electrode, the corresponding domains are
[0

, L

] and [0
+
, L
+
], respectively (see Figure 1).
Consider equation (6). For a sufciently regular I(t) R,
we have j
t
n
(x) C
1
[0, L] for each time t R. Note that
I
t
= I(t) is a scalar for each t R. Now, we can solve
(6) as i
e
(x, t) =
_
x
0
aFj
n
(, t)d + i
e0
(I(t)). Dene the
function-map F
ie
: C
1
[0, L] R C
0
[0, L] as
F
ie
(g, ) (x)
_
x
0
aFg()d + i
e0
(). (9)
Then it is clear that a solution of the PDE in (6) for each t
is
i
e
(x, t) = i
t
e
(x) = F
ie
(j
t
n
, I
t
) (x). (10)
Here we have absorbed the constant of integration in F
ie
as the boundary condition of the PDE is known. Indeed, the
constant of integration is obtained by setting i
e
to either zero
or I(t) at one of the boundaries (x = 0 or x = L) of the
domain. Since the value of i
t
e
(x) is known at both x = 0 and
x = L, there are in fact two boundary conditions to satisfy.
Depending on whether it is the positive or negative electrode
or the separator, it is
{i
t
e
(0), i
t
e
(L)} =
_
{0, I(t)} (electrodes),
{I(t), I(t)} (separator).
(11)
For now, we choose any one of the above boundary
conditions (at x = 0 or x = L) to determine i
e0
in (9).
Similarly, we solve PDE in (2) for each t as

s
(x, t) =
t
s
(x) = F
s
(j
t
n
, I
t
) (x) +
s0
(t), (12)
where F
s
: C
1
[0, L] R C
1
[0, L] is dened as
F
s
(g, ) (x)
1

_
x
0
(F
ie
(g, ) (w) ) dw, (13)
w is a dummy variable, and
s0
(t) is an integration constant
that depends on boundary condition. Note that unlike F
ie
, we
do not absorb the integration constant in F
s
. This is because
we do not know a priori what the boundary condition is.
As we show later, the boundary condition appears implicitly
through the ux density j
n
(x, t) and the Butler-Volmer
equation. For now, we write
s0
(t) as a boundary condition
that is unknown.
Similarly, assuming a constant t
0
c
and performing some
manipulations, we solve for the PDE in (3) as

e
(x, t) =
t
e
(x) = F
e
(j
t
n
, c
t
e
, I
t
) (x), (14)
where F
e
: C
1
[0, L] C
0
[0, L] R C
1
[0, L] is
F
e
(g, h, ) (x)
_
x
0
F
ie
(g, ) (w)
(h(w))
dw
+
2RT
F
_
1 t
0
c
_
ln(f

h(x)) +
e0
(g, h, ), (15)
w is a dummy variable and f

is some known function


of h or is set to a constant. Note that in this case we
have absorbed the constant of integration in F
e
since the
boundary condition for this PDE is known. It is given as
F
e
(j
t
n
, c
t
e
, I
t
) (0
+
) = 0.
Summarizing until now, equations (10), (12) and (14)
imply that if j
n
(x, t), c
e
(x, t) and I(t) are given, then we
obtain
s
(x, t),
e
(x, t) and i
e
(x, t) (and hence i
s
(x, t)
since i
s
(x, t) + i
e
(x, t) = I(t) for all (x, t) [0, L] R).
Next, consider equation (7) and (8). Substituting for
s
and
e
from (13) and (15), respectively, in (8) yields

s
(x, t) = F
s
(j
t
n
, I
t
) (x) +
t
s0
F
e
(j
t
n
, c
t
e
, I
t
) (x) U(c
t
ss
(x)) R
f
j
t
n
(x)F. (16)
1999
Equation (16) suggests that we can express
s
as a function
of j
n
, c
e
, c
ss
, I and
s0
for every x and t. Note that

t
s0
=
s0
(t) in (16) is the unknown boundary condition
or constant of integration for the
s
-equation, and that the
exchange current density is also a function of c
e
and c
ss
.
Thus, equation (7) can be expressed as
j
n
(x, t) = j
t
n
(x) = F
jn
(j
t
n
, c
t
e
, c
t
ss
, I
t
,
t
s0
) (x), (17)
where F
jn
: C
1
[0, L] C
0
[0, L] C
0
[0, L] R R
C
1
[0, L] is
F
jn
(f, g, h, , ) (x)
i
0
(x)
F
_
exp
_

a
F
RT

s
(x)
_
exp
_

c
F
RT

s
(x)
__
, (18)
i
0
(x) r
e
_
g(x)
_
a
_
c
s,max
h(x)
_
a
_
h(x)
_
c
, (19)

s
(x) F
s
(f, ) (x) F
e
(f, g, ) (x)
U(h(x)) R
f
f(x)F + . (20)
Note that (17) is an algebraic equation that has to hold
for every x and time t. Given the electrolyte concentration
c
t
e
(x), the surface concentration of the solid particle c
t
ss
(x) =
c
s
(x, R
p
, t), and the current I
t
, we need to nd j
t
n
(x) and

t
s0
that satisfy (17). However, there are two unknowns
j
n
(x, t) and
s0
(t) and only one equation (17). To solve
for j
n
and
s0
together, we now use the second boundary
condition on i
e
. Suppose to derive F
ie
in (9), we used
i
e
(0, t) = 0. Then, the other boundary condition on i
e
at the
separator-electrode interface is i
e
(L, t) = I(t). Thus, from
(10) it follows that
i
e
(L, t) = i
t
e
(L) = F
ie
(j
t
n
, I
t
) (L) = I(t), (21)
which is an algebraic constraint on j
n
(, t). Then, (17) and
(21) are solved together to obtain the current density j
n
(x, t)
and the boundary condition on
s
given by
s0
(t), for a
given electrolyte concentration c
e
(x, t), surface concentra-
tion of the electrode c
ss
(x, t) and current I(t).
The full Li-ion battery model is given by (2), (3), (4), (5),
(6) and (7). In this section, we have reduced these equations
to solving the dynamical equations (involving time) given by
(4) and (5) while satisfying the algebraic constraints in space
given by (17) and (21) at all times t.
V. SIMPLEST DISCRETIZATION OF THE MODEL
In the last section, we studied the equations for the model
given by PDEs (2), (3), (4), (5), (6) and (7), and reduced
them to a compact form given by (4) and (5), and the
algebraic constraint (17) and (21). In general, the equations
obtained from discretization of (17) and (21) along the
spatial dimension cannot be solved analytically. However,
it is possible to solve these equations analytically for the
case where the coarsest discretization is chosen. This yields
a model related to the single particle model (SPM) [13], [14].
A. Assumptions involving coarse discretization
Suppose we choose the lowest order of discretization for
the X-domain. Then, we have one node for the positive
electrode and one node for the negative electrode (and a
node at the separator; see Figure 2) with quantities at this
node representing their average over the whole electrode. Let
us consider the positive electrode. Assume that
ce
x
0 and
ce
t
0. This approximation holds if we assume that I is
small or is large. Then c
e
(x, t) c
0
e
R
+
is a constant.
Also, (4) yields that i
e
(x, t) = cnst(t), yielding that within a
domain (positive electrode, negative electrode, or separator),
i
e
remains a constant or that it does not vary in x. Then we
can express i
e
for the entire electrode by one value in each
of the electrode in the cell.
Fig. 2. Schematic showing variables for the approximate model.
B. Solution for the coarse discretization
Since only one node exists in the electrode, we express
the corresponding variables as scalar functions of time de-
noted as j
+
n
(t), i
+
e
(t),
+
s
(t),
+
e
(t), c
+
e
(t), c
+
s
(r, t), and
similarly for the negative electrode, as shown in Figure 2.
The function-maps in this case are easily solved as follows.
From (9) and (10), we obtain
0 = i
e
(0
+
, t) = i
+
e
(t) = F
ie
(j
t
n
, I
t
) (0
+
)

_
0
0
a
+
Fj
+,t
n
d + i
e0
(I
t
) = i
e0
(I
t
).
Thus, i
e0
(I
t
) = 0 is obtained from the boundary condition
that i
e
(0
+
, t) = i
+
e
(t) = 0. Next, substituting this boundary
condition in F
ie
and solving (21) implies that
I(t) = i
sep
e
(t) = i
e
(L
+
, t) = F
ie
(j
t
n
, I
t
) (L
+
)
=
_
L
+
0
a
+
Fj
+,t
n
d = j
+,t
n
L
+
a
+
F,
where i
sep
e
(t) is the current in the separator and hence,
j
+
n
(t) = j
+,t
n
=
I(t)
Fa
+
L
+
. (22)
Next, it follows from (12) and (13) that

+
s
(t) =
1

_
0
0
_
i
+
e
(t) I(t)
_
dw +
+
s0
(t) =
+
s0
(t). (23)
2000
Similarly, it can be shown that
+
e
(t) = 0 since
+
e0
(t) =
0 from the boundary condition that
e
= 0 at the current
collector of the positive electrode.
Finally, applying the last algebraic constraint (17) to (22),
and choosing
a
=
c
=
1
2
, it follows that
I(t)
2a
+
L
+
= r
e
_
c
0
e
c
+
ss
(t)(c
+
s,max
c
+
ss
(t)) sinh(
+
), (24)

+
=
F
2RT
_

+
s0
(t) U
+
(c
+
ss
(t)) +
R
+
f
I(t)
a
+
L
+
_
, (25)
where U
+
() is the OCP of the positive electrode and is
usually known from experiments.
Similarly, for the negative electrode, i
e0
(I(t)) = 0 is
obtained from the boundary condition that i
e
(0

, t) =
i

e
(t) = 0. Substituting this boundary condition in F
ie
and
solving (21) yields
j

n
(t) = j
,t
n
=
I(t)
Fa

. (26)
Equations (12) and (13) imply that

s
(t) =
1

_
0
0
_
i

e
(t) I(t)
_
dw +

s0
(t) =

s0
(t). (27)
Also (15) with the boundary condition that
e
= 0 at
the current collector of the positive electrode yields that

+
e0
(t) = 0 and hence

e
(t) =
1

_
0

0
+
i
e
(x, t)dx 0.
The above follows since i
e
(x, t) = I(t) in the separator and
I(t)/ 1 from the assumption for SPM that I(t) is small
or the conductance is large. Lastly, (17), along with the
assumption that
a
=
c
=
1
2
, yields
I(t)
2a

= r
e
_
c
0
e
c

ss
(t)(c

s,max
c

ss
(t)) sinh(

), (28)

=
F
2RT
_

s0
(t) U

(c

ss
(t))
R

f
I(t)
a

_
, (29)
where U

() is the OCP of the negative electrode and is


known from experiments. We can solve (24) and (25) for

+
s0
(t) and (28) and (29) for

s0
(t) yielding
+
s
(t) from (23)
and

s
(t) from (27). Note that we need to compute c
+
ss
(t) =
c
+
s
(R
p
, t) and c

ss
(t) = c

s
(R
p
, t) by solving the PDE (4)
where j
n
(x, t) = j
+
n
(t) and j
n
(x, t) = j

n
(t), respectively.
Then, the output voltage V (t) =
+
s0
(t)

s0
(t).
VI. SIMULATION RESULTS
In section IV, we presented a compact form of the model
that is subsequently used to derive an approximation of
the electrochemical model (SPM) in section V. Since SPM
is an approximation, its applicability is only valid over
certain regimes. In this section, we illustrate some of these
limitations by means of simulation results. We present a brief
comparison of the full model given by (4) and (5), and the
algebraic equations (17) and (21) with SPM. In particular, we
consider cells that correspond to a high energy conguration
with applications in EVs.
To compare model performance, we compare output volt-
ages and surface concentrations computed from the two
models. The comparison of surface concentration c
ss
can be
used as an indicator of when the approximate model starts to
fail. We present a comparison between SPM and full model
for a high energy cell conguration with a nominal capacity
of 3.5 A-h. The applied currents are a constant discharge at
rates of C/25, C/2, 1 C and 2 C, where 1 C corresponds to
3.5 A.
0 1 2 3 4
2.5
3
3.5
4
4.5
V
o
l
t
a
g
e

[
V
]


C/25
0 1 2 3 4
2.5
3
3.5
4
4.5


C/2
0 1 2 3 4
2.5
3
3.5
4
4.5
V
o
l
t
a
g
e

[
V
]
Discharge Capacity [Ah]


1C
0 1 2 3 4
2.5
3
3.5
4
4.5
Discharge Capacity [Ah]


2C
Full Model
SPM
Full Model
SPM
Full Model
SPM
Full Model
SPM
Fig. 3. Comparison of the full model and the single particle model (SPM)
for a high-energy cell conguration.
The corresponding voltages for the full model and the
SPM are compared in Figure 3. As seen in the gure, SPM
is accurate until C, /2 where the discharge curves are almost
indistinguishable.
As seen in Figure 4, the surface concentration in an SPM
is the average surface concentration in the electrode. This
averaging follows from the fact that variables in an SPM
represent spatial average over electrodes. At higher discharge
rates of 1 C and 2 C, the uniformity in the concentration is
lost and SPM is no longer valid since concentrations cannot
be effectively represented by its spatial average due to large
variance. This failure of the approximation is noted in the
corresponding rate plots in Figure 3. In fact, even at C/2,
the variance in the surface concentrations shown in Figure 4
is large suggesting that SPM may have signicant errors in
its prediction of states of the full model.
VII. CURRENT STATUS ON BMS AND FUTURE WORK
Earlier in the paper, we mentioned that a BMS has to
perform certain tasks that are critical to the operation of
the battery. In particular for vehicle electrication, a sample
of these tasks include prediction of maximum available
power and energy, safe charging and discharging to meet
regenerative braking and load bearing requirements, tracking
the state of health of the battery pack as it ages, and updating
the BMS to maintain accuracy of its tasks throughout its life.
The importance of prediction of maximum available power
and energy is self-evident since this knowledge allows the
electronic control unit (ECU) to compute the vehicles all-
battery range in miles and the power it can deliver to
accelerate, if demanded. Though ideally, it is desirable to
have the ability to charge or discharge the battery as quickly
as possible, such processes can dangerously stress the battery
2001
0 1 2 3 4
0
0.5
1
c
s
s

/

c
m
a
x


C/25
0 1 2 3 4
0
0.5
1


C/2
0 1 2 3 4
0
0.5
1
Discharge Capacity [Ah]
c
s
s

/

c
m
a
x


1C 1C
0 1 2 3 4
0
0.5
1
Discharge Capacity [Ah]


2C 2C
c
ss
(x=0

)
c
ss
(x=L

)
c
ss
(SPM)
c
ss
(x=0

)
c
ss
(x=L

)
c
ss
(SPM)
c
ss
(x=0

)
c
ss
(x=L

)
c
ss
(SPM)
c
ss
(x=0

)
c
ss
(x=L

)
c
ss
(SPM)
Fig. 4. Surface concentrations over the electrode computed from SPM and
the full model for a high-energy cell conguration.
and accelerate aging. Thus, a BMS has to monitor overpoten-
tials and other relevant states that indicate potential damage
to the battery [19]. Finally, as the battery ages, a BMS
needs to track model parameters to maintain accuracy of
power and energy prediction throughout the life of the battery
pack. Each of the above BMS tasks, reects an estimation
or control problem.
We now briey mention some of the contemporary re-
search work on design of a BMS. A large section of the
research work on batteries uses a simple equivalent circuit
model for design of the BMS [2], [6], [3], [5], [7], [8]. This
choice stems from the fact that existing BMS for portable
electronics mostly models the battery as an equivalent circuit
and hence, its use in modeling is naturally extended to Li-ion
batteries for high-energy applications.
In contrast to equivalent circuit approach, [19], [20], [21],
[22], [23] study estimation problems using other models,
including electrochemical-based models. In particular, they
use approximations of electrochemical models and other
physics-based models to improve accuracy of estimation
algorithms for BMS. With this picture in mind, we mention
the future work that needs to be addressed for design of
improved and sophisticated BMS.
Referring to our earlier discussion in this section, the
future challenges are characterization of an approximation or
reduction of the full electrochemical model given by (4) and
(5), and the algebraic equations (17) and (21) such that the
model is accurate over a large range of operation and simple
enough that it is analytically tractable. Retaining the physical
signicance of the parameters is important since it helps in
characterizing aging phenomena in batteries. Next, the design
of simple algorithms for observing states of this model is an
open problem, especially when applied to the whole battery
pack and not just one cell. Finally, the estimation of all
parameters of the model to maintain accuracy of the model
and to identify age of the battery pack by tracking relevant
physical parameters is also an open problem.
REFERENCES
[1] M. Armand and J.-M. Tarascon, Building better batteries, Nature,
vol. 451, pp. 652657, 2008.
[2] T. Stuart, F. Fang, X. Wang, C. Ashtiani, and A. Pesaran, A modular
battery management system for HEVs, SAE Future Car Congress,
vol. 2002-01-1918, 2002.
[3] Y.-S. Lee and M.-W. Cheng, Intelligent control battery equalization
for series connected Lithium-ion battery strings, IEEE Transactions
on Industrial Electronics, vol. 52, no. 5, pp. 12971307, 2005.
[4] I. J. Ong and J. Newman, Double-layer capacitance in a dual lithium
ion insertion cell, Journal of the Electrochemical Society, vol. 146,
no. 12, pp. 43604365, 1999.
[5] M. W. Verbrugge and R. S. Conell, Electrochemical and thermal
characterization of battery modules commensurate with electric vehicle
integration, Journal of the Electrochemical Society, vol. 149, no. 1,
pp. A45A53, 2002.
[6] B. Schweighofer, K. M. Raab, and G. Brasseur, Modeling of high
power automotive batteries by the use of an automated test system,
IEEE Transactions on Instrumentation and Measurement, vol. 52,
no. 4, pp. 10871091, 2003.
[7] M. Chen and G. A. Rinc on-Mora, Accurate electrical battery model
capable of predicting runtime and I-V performance, IEEE Transac-
tions on Energy Conversion, vol. 21, no. 2, pp. 504511, 2006.
[8] M. W. Verbrugge and R. S. Conell, Electrochemical characterization
of high-power lithium ion batteries using triangular voltage and current
excitation sources, Journal of Power Sources, vol. 174, no. 1, pp. 28,
2007.
[9] J. Newman and W. Tiedemann, Porous-electrode theory with battery
applications, AIChE J., vol. 21, no. 1, pp. 2541, 1975.
[10] M. Doyle, T. F. Fuller, and J. Newman, Modeling of galvanostatic
charge and discharge of the lithium/polymer/insertion cell, Journal
of The Electrochemical Society, vol. 140, no. 6, pp. 15261533,
1993. [Online]. Available: http://link.aip.org/link/?JES/140/1526/1
[11] T. F. Fuller, M. Doyle, and J. Newman, Simulation and optimization
of the dual lithium ion insertion cell, Journal of The Electrochemical
Society, vol. 141, pp. 110, 1994.
[12] K. Thomas, J. Newman, and R. Darling, Advances in lithium-ion
batteries: Mathematical Modeling of Lithium Batteries. New York:
Springer US, 2002.
[13] G. Ning and B. N. Popov, Cycle life modeling of Lithium-ion
batteries, Journal of the Electrochemical Society, vol. 151, no. 10,
pp. A1584A1591, 2004.
[14] S. Santhanagopalan, Q. Guo, P. Ramadass, and R. E. White, Review
of models for predicting the cycling performance of lithium ion
batteries, Journal of Power Sources, vol. 156, no. 2, pp. 620628,
2006.
[15] N. Chaturvedi, R. Klein, J. Christensen, J. Ahmed, and A. Kojic,
Algorithms for Advanced Battery Management Systems: Modeling,
estimation, and control challenges for lithium-ion batteries, IEEE
Control Systems Magazine, vol. 30, no. 3, 2010.
[16] A. J. Bard and L. R. Faulkner, Electrochemical Methods: fundamentals
and applications. New York: John Wiley & Sons, Inc., 2001.
[17] J. Newman and K. E. Thomas-Aleya, Electrochemical Systems. New
Jersey: John Wiley & Sons, Inc., 2004.
[18] L. C. Evans, Partial Differential Equations. New Jersey: American
Mathematical Society: Graduate Studies in Mathematics, 1998.
[19] R. Klein, N. Chaturvedi, J. Christensen, J. Ahmed, R. Findeisen, and
A. Kojic, State estimation of a reduced electrochemical model of a
lithium-ion battery, Proceedings of the American Control Conference,
2010.
[20] G. L. Plett, Extended Kalman ltering for battery management
systems of LiPB-based HEV battery packs: Part 2. Modeling and
identication, Journal of Power Sources, vol. 134, no. 2, pp. 262276,
2004.
[21] , Extended Kalman ltering for battery management systems
of LiPB-based HEV battery packs: Part 3. State and parameter
estimation, Journal of Power Sources, vol. 134, no. 2, pp. 277292,
2004.
[22] S. Santhanagopalan and R. E. White, Online estimation of the state
of charge of a lithium ion cell, Journal of Power Sources, vol. 161,
no. 2, pp. 13461355, 2006.
[23] K. Smith, C. Rahn, and C.-Y. Wang, Control oriented 1D elec-
trochemical model of lithium ion battery, Energy Conversion and
Management, vol. 48, no. 9, pp. 25652578, 2007.
2002

You might also like