IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 66, NO.
8, AUGUST 2017
A Lithium-Ion Battery Current Estimation Technique
Using an Unknown Input Observer
Daniel C. Cambron, Member, IEEE, and Aaron M. Cramer, Senior Member, IEEE
Abstract—Current consumption measurements are useful in a
wide variety of applications, including power monitoring and fault
detection within a lithium-ion battery management system (BMS).
This measurement is typically taken using either a shunt resis-
tor or a Hall-effect current sensor. Although both methods have
achieved accurate current measurements, shunt resistors have in-
herent power loss and require isolation circuitry, and Hall-effect
sensors are generally expensive. This paper explores a novel alter-
native to sensing battery current by measuring terminal voltages
and cell temperatures and using an unknown input observer to
estimate the battery current. An accurate model of a LiFePO4
cell is created, validated, and then used to characterize a model
of the proposed current estimation technique. Finally, the current
estimation technique is implemented in hardware and tested in an
online BMS environment. Results show that the current estima-
tion technique is sufficiently accurate for a variety of applications,
including fault detection and power profiling.
Index Terms—Battery, battery management system (BMS),
current estimation, Hall-effect, LiFePO4 , lithium-ion, parameter
estimation, shunt, state of charge, unknown input observer (UIO).
I. INTRODUCTION
ITHIUM-ION batteries continue to gain popularity for
L use in a wide range of markets, from small consumer elec-
tronics to full-sized electric vehicles. A battery management
system (BMS) must be employed in every battery pack to pro-
tect lithium-ion cells from abuse that could cause fire, gaseous
discharge, or other dangerous conditions. BMSs protect lithium-
ion battery packs by taking, at a minimum, terminal voltage and
temperature measurements on the cells within a battery pack.
When an unsafe condition is detected, the BMS can limit the
current accordingly in order to prevent permanent damage to the
cells or surroundings [1]. Increasingly, BMSs also take current
measurements that can be used for state-of-charge calculations,
power profiling, and fault detection. These current measure-
ments are typically taken using a shunt resistor or a Hall-effect
sensor, and there are advantages to each. Shunt resistors can
be extremely accurate, but generally require additional isolated
measurement circuitry and inherently consume power due to
the series resistance added to the battery system. Hall-effect
Manuscript received April 13, 2016; revised August 8, 2016; accepted January
15, 2017. Date of publication January 24, 2017; date of current version August
11, 2017. The review of this paper was coordinated by Dr. S. Anwar.
The authors are with the Department of Electrical and Computer Engi-
neering, University of Kentucky, Lexington, KY 40506 USA (e-mail: daniel.
cambron@uky.edu; aaron.cramer@uky.edu).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TVT.2017.2657520
0018-9545 © 2017 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.
Authorized licensed use limited to: VIT University. Downloaded on March 14,2021 at 17:33:38 UTC from IEEE Xplore. Restrictions apply.
6707
sensors provide isolated current measurements but are typically
less accurate and more expensive [2]. Thus, there remains an
opportunity to discover an effective, yet inexpensive, solution
for determining the current consumption of a lithium-ion battery
pack.
In this paper, a novel technique for determining current con-
sumption is explored. In essence, the equivalent series resistance
(ESR) of a battery cell could be used in place of a dedicated
shunt resistor. If the cell ESR, open-circuit voltage, and ter-
minal voltage were known, the current consumption could be
directly determined. However, the open-circuit voltage of an ac-
tive cell cannot be measured directly, and the ESR of a cell is
dependent on the cell state of charge, state of health, and tem-
perature [3]. Thus, a current estimate cannot be achieved until
these derivative quantities are estimated.
There have been significant contributions in lithium-ion bat-
tery modeling as well as detailed work in observer theory that
can be utilized to develop a current estimator. Beginning with
battery modeling, most models can be classified as electrochem-
ical, thermal, adaptive, or circuit models. Electrochemical mod-
els consider the chemical processes occurring inside a cell [4],
[5]. These models are accurate, but require detailed parameteri-
zation that is often difficult to obtain in practice. Thermal mod-
els of the form in [6] can be employed, but more information
is required to completely characterize cell behavior. Adaptive
models can be used to model the cell itself, but are more often
used for parameter estimation in the context of a circuit model.
Various forms of the Kalman filter are used to this effect in [1],
and [7]–[10], whereas a proportional-integral observer is used
in [11]. Further adaptive techniques such as a Lyapunov-based
approach [12] are used. Other methods of battery modeling in-
clude works such as [13], which uses a modified form of the
Peukert equation.
Circuit models generally lend themselves well to developing
control systems for practical uses of lithium-ion batteries due to
their simple parameterization. Most lithium-ion battery circuit
models include a state of charge dependent open-circuit voltage
term and a series resistance, which can be followed by two par-
allel RC elements that model dynamic behavior [3], [14]–[18].
Alternatively, three parallel RC elements have been used in an
attempt to achieve greater model convergence [19]. These mod-
els can be further augmented by considering additional effects
such as performance at low temperatures [20], temperature-
dependent self-discharge [21], parameter dependence on state
of health [22], rate-capacity effects [23], or open-circuit voltage
hysteresis terms [24]–[26].
6708
Fig. 1. Circuit model of a lithium-ion battery cell based on a second-order RC
design.
Each of these battery models use cell current as an input
in order to estimate the other cell parameters such as state of
charge. In this paper, it is not presumed that a current mea-
surement is available; thus, none of the models can be applied
directly. To remedy this problem, significant work in observer
theory is applied, which attempts to estimate an output signal
when one or more of the input signals are unavailable. These un-
known input observers (UIOs) have been developed for a broad
range of systems. Linear, continuous systems are considered in
[27]–[30]. Each of those works only consider systems without
a direct feed-through term, but this term is required in all cell
models that include a series Ohmic resistance. Fortunately, this
term is included in [30]; however, it is presented in discrete
time. Further works consider nonlinear systems in order to de-
velop more robust structures at the expense of computational
complexity [31]–[33].
This paper contributes a practical battery model developed
from a literature survey, as well as a novel technique by which
the current consumption in a battery pack can be determined
without requiring any additional current-measuring hardware.
Additionally, the battery model and the current estimation tech-
nique are compared with experimental data that demonstrate
the accuracy of the model and the estimator. Finally, this paper
contributes a demonstration of the current estimation technique
employed in a BMS.
This paper is organized as follows. In Section II, a model
for a lithium-ion battery cell is developed. In Section III, the
mathematical model for the UIO is derived and applied to the
current estimation problem. Experimental design is detailed in
Section IV, and experimental results are given in Section V.
Section VI concludes the paper.
II. LITHIUM-ION BATTERY MODEL
An accurate circuit model of a lithium-ion battery cell can
be achieved using a second-order RC circuit with a series re-
sistance and two dependent sources describing the open-circuit
voltage and the hysteresis voltage. This construction is shown
in Fig 1. The open-circuit voltage vo c is a nonlinear, but mono-
tonically increasing, function of the state of charge soc of the
cell. Moreover, the hysteresis voltage vh is a nonlinear function
of soc and is dynamically dependent on load current il . This
circuit model along with relations governing the open-circuit
Authorized licensed use limited to: VIT University. Downloaded on March 14,2021 at 17:33:38 UTC from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 66, NO. 8, AUGUST 2017
voltage and hysteresis voltage is used to develop a complete set
of equations to represent a lithium-ion battery, accounting for
various physical phenomena.
A. State of Charge
State of charge can be represented as the voltage across a
capacitor that represents the battery capacity Cn along with
a parallel resistance Rn , which represents the self-discharge
effect of the battery. The soc represents the ratio of remaining
charge in the cell and the total charge capacity of the cell and,
thus, must be bounded between zero and unity. Further, the sign
convention is defined such that a positive il term represents a
cell that is charging. These constraints result in the following
dynamic equation:
⎧
⎨ il − 1 soc 0 < soc < 1
sȯc = Cn Rn Cn (1)
⎩
0, otherwise.
A nonlinear relationship exists between soc and open-circuit
voltage, but can approximated in a similar manner to [14] as
a linear relationship between much of the useable capacity of
the cell, about 10%–95% soc. In order to maintain accuracy, a
weakly state of charge dependent scale factor α is used such
that
vo c − voc 0 = α · soc, 0 < soc < 1 (2)
where voc 0 represents the open-circuit voltage that appears on
the cell when the cell is discharged to 0% state of charge.
B. Hysteresis
Hysteresis refers to the difference in open-circuit voltage in
the charging and discharging conditions. This effect is included
by adding a hysteresis term to the discharge open-circuit voltage
in the manner of [26]. Total deviation from the discharge vo c
curve can be represented as a percent of the maximum voltage
deviation for a given soc. The percent deviation Hh changes
dynamically with load current and, like soc, must be bounded
between saturation points zero and unity. Further, the hysteresis
voltage changes with current; thus, a self-discharge term Rh
also appears in the following equation:
⎧
⎨ il − 1 Hh 0 < Hh < 1
Ḣh = Ch Ch Rh (3)
⎩
0, otherwise.
Conveniently, the hysteresis equation takes the same form as
the state-of-charge equation, with Ch representing a capacitor,
which provides inertia between the charge and discharge hys-
teresis states, whose value must be experimentally determined.
Similar to the state-of-charge model, a state of charge dependent
scale factor Hm ax can be used to convert the hysteresis term to
a voltage contribution as shown in the following equation:
vh = Hh Hm ax , 0 < Hh < 1. (4)
The Hm ax parameter itself represents the difference between
the open-circuit voltage during a charge cycle and the open-
circuit voltage during a discharge cycle. It is experimentally
determined as a function of soc.
CAMBRON AND CRAMER: LITHIUM-ION BATTERY CURRENT ESTIMATION TECHNIQUE USING AN UNKNOWN INPUT OBSERVER 6709

C. Voltage Relaxation of the cell, which is caused by degradation of the active cell
material and is affected by a number of factors including tem-
The terminal voltage of a battery cell that has been under
discharge load, but is then allowed to rest, will increase asymp- perature, load current, and average state of charge. This effect
is generally captured in the state of health of the cell, which
totically toward the open-circuit voltage. This phenomenon is
will be defined here as a scale factor that represents the ratio of
known as the relaxation effect and is related to active material
diffusion in the electrodes as well as the surface capacitance the actual charge capacity of the cell to the nameplate charge
capacity. Thus, this variable capacitance can be expressed as
of the cell. It exhibits itself as transient behavior at multiple
frequencies and can be represented with two RC circuits placed Cn = f (il ) soh (9)
in series. The RC circuits can be represented using capacitor
voltages as state variables where f (il ) is a nonlinear function that represents the rate-
1 1 capacity effect, and soh is the state of health of the cell.
v̇1 = − v1 + il (5) The state of heath decreases over the lifetime of a cell, but
R1 C1 C1
under normal operating conditions will decrease very slowly
1 1
v̇2 = − v2 + il (6) compared to each of the other effects and is considered a con-
R2 C2 C2 stant in this paper. Indeed, the long-term effects of cell aging,
where the resistance R1 and capacitance C1 represent the fast such as capacity fade and increase in internal resistance, are
dynamics of the relaxation effect, and R2 and C2 represent the not considered, and in a practical setting, this model must go
slower dynamics of the relaxation effect. The value of these through periodic calibration over the lifetime of the cell in order
parameters must be empirically fit to match the response of a for accuracy to be maintained.
particular lithium-ion cell. A relationship between the open-circuit voltage and the state
of charge is found by discharging the cell at a sufficiently small
D. Linear Effects current and approximating the open-circuit voltage with the
terminal voltage. The value of α is directly determined from (2)
The charge path to the terminals, along with physical prop-
by solving for α and expressing both the open-circuit voltage
erties of the electrolyte and the electrode matrix in the cell,
and α itself as a function of state of charge and temperature.
creates a high frequency Ohmic resistance between the open-
Moreover, the relaxation effect has been experimentally deter-
circuit voltage and the terminal voltage
mined to depend on the state of charge of the cell [16], [18], and,
v0 = R0 il (7) as a result, the values of R2 , R1 , C1 , and C2 are experimentally
determined as functions of soc. The internal Ohmic resistance
where R0 is the Ohmic resistance of the cell. The terminal
term R0 exhibits strong temperature dependence, moderate state
voltage can then be computed by summing the individual con-
of heath dependence, and weak state of charge dependence [14].
tributions of all of the cell’s internal effects
It is represented in this model as a function of both temperature
v t = vo c + v h + v 1 + v 2 + v 0 . (8) and state of charge, with state of health appearing as a supplied
constant.
E. Nonlinear Effects Several of the cell parameters are functions of temperature;
however, temperature is not included as an input to a state model
The model presented in this paper is an approximation of the
of this system due to its highly nonlinear relationship with each
battery dynamics about a particular state of charge. The bat-
of the parameters. It will be treated as an input to the functions,
tery parameters have each been presented as constants in the
which update the cell parameters.
battery dynamics. However, in reality, all of these parameters
are dependent on one or more other factors that change very
slowly with respect to the battery dynamics. A suitable approx- F. Linear State Model
imation is made by considering these parameters as constants in The circuit model, less the limiting diodes, can be represented
a saturated, but otherwise linear, system and then periodically as a linear state-variable model with four state variables: soc,
updating the value of these parameters in an outer control loop. Hh , v1 , and v2 . The following state model will act as the basis
In order to avoid interference with the model, the outer control for the development of a current estimator:
loop will be updated at a rate two orders of magnitude slower
than the main model control loop. ẋ (t) =
The energy capacity of the cell will be represented by the ⎡ 1 ⎤ ⎡ 1 ⎤
charge stored in the capacitor Cn . This value is dependent on − 0 0 0
⎢ Rn Cn ⎥ ⎢ Cn ⎥
capacity fade [3] and the rate-capacity effect [16], which are two ⎢ ⎥ ⎢ ⎥
⎢ 1 ⎥ ⎢ 1 ⎥
phenomena that reduce the usable current capacity of a cell. The ⎢ 0 − 0 0 ⎥ ⎥ ⎢ ⎥
⎢ R C ⎢ ⎥
rate-capacity effect does not result in permanent loss of capacity ⎢ h h ⎥x (t) + ⎢ Ch ⎥ w (t) .
⎢ 1 ⎥ ⎢ 1 ⎥
and refers to the fact that a battery will exhibit a smaller effective ⎢ − 0 ⎥ ⎢ ⎥
⎢ 0 0
R1 C1 ⎥ ⎢C ⎥
capacity when discharged or charged at a higher rate of current; ⎢ ⎥ ⎢ 1⎥
⎣ 1 ⎦ ⎣ 1 ⎦
thus, it is dependent on the load current of the cell.
0 0 0 −
Moreover, the capacity fade effect is an irreversible effect R2 C2 C2
that refers to the diminishing effective capacity over the lifetime (10)

Authorized licensed use limited to: VIT University. Downloaded on March 14,2021 at 17:33:38 UTC from IEEE Xplore. Restrictions apply.
6710 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 66, NO. 8, AUGUST 2017

The output equation is given by By inspection, (15) holds if and only if the error equation
  updates such that ė(t) = N e(t), with N Hurwitz. Therefore,
y (t) = α Hm ax 1 1 x (t) + R0 w (t) (11)
from this equation, one can conclude that the observer exists if
where y(t) = vt − voc 0 and w(t) = il (t). The limiting diodes and only if each of state and input coefficients are zero and the
do not appear in this expression but are implemented as saturated error vector coefficient N is Hurwitz. Hence, the four presented
integrators. conditions are necessary and sufficient.
Remark 1: For a single-input single-output system, as is the
III. UNKNOWN INPUT OBSERVER case for the battery model, the D matrix is a nonzero scalar and,
thus, has full rank and is invertible. The conditions of Theorem 1
A. Derivation of the UIO can be simplified if D has full rank because the only E that ful-
The derivation of the UIO presented here closely follows fills condition 4) is the zero matrix, and is further simplified if D
the derivation as presented in [30], with the exception that this is invertible. This leads to the following corollary to Theorem 1.
derivation considers a continuous-time system. A continuous, Corollary 1: If the D matrix has full rank and is invertible,
linear, time-invariant system can be expressed, as shown in the the UIO (13) for the system (12) exists if and only if the follow-
following: ing conditions are satisfied.
1) N is Hurwitz stable.
ẋ (t) = Ax (t) + Bw (t) , t ≥ t0 2) N = A − LC.
y (t) = Cx (t) + Dw (t) (12) 3) L = BD −1 .
It is also of interest to find an estimate ŵ(t) of the unknown
where x  Rn is the state of the system, w  Rm is the unknown input vector. An expression for ŵ(t) follows directly from the
input to the system, and y  Rp is the measured output of the sys- system output equation, where the estimated input and state are
tem. The matrices A, B,C, and D are real, constant, matrices used in place of the true input and state. If D is invertible and
of appropriate dimensions. E = 0, then ŵ(t) can be expressed directly, as shown in the
A full-order unknown-input state observer for this system can following:
be described by
ŵ (t) = D −1 (y (t) − C x̂ (t))
ż (t) = N z (t) + Ly (t) , t ≥ t0
= D −1 (y (t) − Cz (t)) . (16)
x̂ (t) = z (t) + Ey (t) (13)
where x̂  Rn is the estimated state of the system, and N , L, B. Presentation of the Current Estimator
and E are constant, real matrices of appropriate dimensions. A full-order UIO of the form presented in (13) exists for
Necessarily, this state observer must estimate the state of the this system if the unknown matrices can be found such that the
system with no knowledge of the input since it is unknown or conditions from Corollary 1 are satisfied. For the battery system
otherwise unmeasureable. This observer is required to suitably  T
estimate the state of the system such that the estimate of the state 1 1 1 1
L= (17)
asymptotically converges to the true state. This requirement can R0 Cn R0 Ch R0 C1 R0 C2
be described with an error vector ⎡ 1 ⎤
− 0 0 0
e (t) = x̂ (t) − x (t) (14) ⎢ Rn Cn ⎥
⎢ ⎥
⎢ 1 ⎥
where the observer exists if and only if the matrices N , L, and ⎢ 0 − 0 0 ⎥
⎢ R C ⎥
E can be found such that N = ⎢ ⎢ h h ⎥
1 ⎥
⎢ − ⎥
lim e (t) = 0. (15) ⎢ 0 0
R1 C1
0 ⎥
t→∞ ⎢ ⎥
⎣ 1 ⎦
Theorem 1: The UIO (13) for the system (12) exists if and 0 0 0 −
only if the following conditions are satisfied. R2 C2
⎡ 1 ⎤
1) N is Hurwitz stable.
2) LC − (I n − EC)A + N − N EC = 0. ⎢ R0 Cn ⎥
⎢ ⎥⎡ ⎤T
3) LD − (I n − EC)B = 0. ⎢ 1 ⎥ α
⎢ ⎥
4) ED = 0. ⎢ R0 Ch ⎥ ⎢Hm ax ⎥
Proof: By substituting the system and the observer model into − ⎢ ⎥⎢
⎢ 1 ⎥⎣ 1 ⎦
⎥ (18)
⎢ ⎥
the error vector equation, it can be shown that ⎢R C ⎥ 1
⎢ 0 1⎥
(t) − ẋ (t) ⎣ 1 ⎦
ė (t) = ẋ
R0 C2
= N e (t) + [LC − (I n − EC) A + N − N EC]
E = 0. (19)
× x (t) + [LD − (In − EC) B − N ED] w (t)
The current estimator itself would take the form presented
+ ED ẇ (t) . in (16). Substituting the battery parameters, the expression for

Authorized licensed use limited to: VIT University. Downloaded on March 14,2021 at 17:33:38 UTC from IEEE Xplore. Restrictions apply.
CAMBRON AND CRAMER: LITHIUM-ION BATTERY CURRENT ESTIMATION TECHNIQUE USING AN UNKNOWN INPUT OBSERVER 6711

TABLE I the output current estimate of the observer model. In particular,

CELL PARAMETERS
the Federal Urban Driving Schedule (FUDS) [34] is used as
an input current waveform, and was chosen because it provides
Symbol Parameter Value
an accurate characteristic of the load current an electric vehicle
vo c 0 Voltage at 0% soc 2.7500 V battery pack might expect. The FUDS schedule is given in per-
v o c 100 Voltage at 100% soc 3.6000 V centage of vehicle output power, and was scaled to conform to
Cn Nominal capacity 72 000 F
Ch Hysteresis capacity 15 120 F
the rated current capacity of LiFePO4 cells used in the test, with
soh State of health 94.9% the assumption made that the vehicle battery pack consists of a
Ta m b Ambient temperature 23◦ C single series string of cells.
Rn Self-discharge resistance 10 000 Ω
Rh Self-discharge resistance 10 000 Ω
C1 Fast capacitance at 50% soc 28 000 F C. Hardware Implementation
C2 Slow capacitance at 50% soc 300 000 F
R1 Fast resistance at 50% soc 2.132 m·Ω The observer was implemented in a custom designed BMS
R2 Slow resistance at 50% soc 2.440 m·Ω containing an 8-b, 22.1-MHz, PIC18F4580 processor. For this
R0 Internal resistance at 50% soc 1.513 m·Ω
implementation, a sampling period of 0.2323 s was used, and
90% of the 18 kB in program memory was utilized, including
both the observer functions and the other BMS functions.
current estimation becomes The testing apparatus consisted of the BMS serving as a
1     current estimator in communication over USBto a data-logging
ŵ (t) = y (t) − α Hm ax 1 1 z (t) .
R0
IV. EXPERIMENTAL DESIGN
A. Determination of Model Parameters
The model parameters are experimentally determined by con-
ducting a series of tests on a sample 20 Ah prismatic LiFePO4
cell. A pulsed-discharge test is conducted on the cell start-
ing from a maximum open-circuit voltage, defined in the cell
datasheet as the 100% state of charge point. An alternating load
pattern consisting of 15 min of constant-current load followed by
15 min of rest is used to discharge the cell until 0% state of charge
was reached. A load magnitude of 4 A is used, and 0% state of
charge is defined for a particular open-circuit voltage specified
in the cell datasheet. This test is followed by a pulsed-charge test
consisting of an alternating 15 min of constant-current charge
followed by 15 min of rest until the cell has reached 100% state
of charge. The cell parameters are determined by appropriately
fitting the collected test data. The cell parameters are provided
in Table I. Of note, the state of charge limits along with the nom-
inal cell capacity are given from the cell datasheet, and each of
the other parameters are empirically fit from the collected test
data, reported with appropriate significant figures for the test
conducted.
B. Simulation
Simulink models of the cell and the observer have been imple-
mented. In the case of the cell model, cell current data collected
from battery performance tests serves as the input, and the out-
put terminal voltage of this model is compared with the terminal
voltage measured during the cell performance tests. The sim-
ulation itself is conducted on a complete pulsed-discharge test
at a rate of C/2, where the ODE45 Dormand–Prince variable-
step differential equation solver is used with a relative step
tolerance 10e–3.
The observer model accepts the terminal voltage determined
by the cell model and produces an estimated current. Thus, a cur-
rent waveform applied to the cell model can be compared with
Authorized licensed use limited to: VIT University. Downloaded on March 14,2021 at 17:33:38 UTC from IEEE Xplore. Restrictions apply.
(20) PC. The LiFePO4 cell itself was connected, through a relay
controlled by the BMS, to a bidirectional power supply capable
of producing constant current sources from −10 A to 10 A. A
Hall-effect current sensor with an error of 2.5% ± 150 mA was
used to compare the results of the current estimator.
V. RESULTS
A. Validation of the Cell Model
The cell model simulation ran for 25e3 s of simulated time,
taking a total of 256e3 algebraic steps, producing the results
shown in Fig. 2. A similar test was conducted in [14], which
achieved a maximum terminal voltage modeling error of less
than 30 mV, and a further test conducted in [18] achieved an
error of less than 20 mV, whereas the maximum modeling er-
ror of this test was 30 mV. In all three cases, the error took its
maximum value during periods of transient current, but quickly
settled toward a 0-mV modeling error in steady current condi-
tions. These results indicate that the model is sufficient when
compared to similar models in the literature, with the added
advantage that this model lends itself well to implementation in
hardware due to its simplicity and use of look-up tables.
B. Simulation of the Observer
A test was conducted with a 20% discrepancy in initial state
of charge between the cell model and the observer. The results
of this test appear in Fig. 3. At this point, the current estimation
was examined by filtering the error through a low-pass filter
with a break frequency of 0.1 Hz. This is necessary because the
observer itself acts as a filter by smoothing the current response,
and would otherwise suggest artificially high errors since it can-
not respond instantaneously to instantaneous jumps in current.
Further, this filtered current scheme is typical for power profil-
ing in a BMS due to the need of examining average trends of
power consumption over long periods of time. Admittedly, the
current estimator performs poorly in applications when quick
fault detection is desired, and devices such as fuses should be
used when this type of protection is needed.
6712 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 66, NO. 8, AUGUST 2017
Fig. 2. Cell model validation test results for the 10 A pulsed-discharge test.
(a) Pulsed current waveform. (b) State of charge estimate decreasing in steps.
(c) Comparison of measured and modeled voltage waveforms. (d) Discharge
test showing a modeling error under 30 mA.
The results indicate that the current estimation error was
greatest during periods where the state of charge estimation
error was high, with filtered current estimation errors reaching
as high as 9 A. However, the current error was significantly di-
minished as the state of charge estimate converged with the true
Authorized licensed use limited to: VIT University. Downloaded on March 14,2021 at 17:33:38 UTC from IEEE Xplore. Restrictions apply.
Fig. 3. Observer validation. An initial error in state of charge of 20% was used.
(a) State-of-charge comparison between cell model and observer. A 5% settling
time of 3200 s was observed. (b) Filtered current estimation error between cell
model and observer. The error exhibits a 9 A magnitude initially but settles
toward zero as the observer converges on the true states.
value. A 5% settling time of less than 3200 s was shown for the
state of charge estimate. This is indeed slower than desired in
a practical application, but the settling time of the observer is
uniquely determined and unable to be set for this design, which
is in contrast to a conventional Luenburger observer where the
observer dynamics can be arbitrarily specified. Mathematically,
the N and L matrices of this UIO have only one solution, and
the degree of freedom that typically allows for pole placement
has been removed as a result of the observer operating with no
knowledge of the unknown input.
In this case, the sluggishness can be attributed to the extremely
slow self-discharge and the relaxation effect, which takes places
over hours. However, it is important to recognize that in a prac-
tical hardware implementation of the observer, the slow settling
time is not cause for concern. In a microcontroller implemen-
tation, the last known observer states can be retrieved from
nonvolatile memory during initialization such that the observer
will never see a 20% state of charge discrepancy, provided the
“burn-in” time of 3200 s–1 h was observed in order to allow the
model to initially converge.
C. Experimental Validation of Hardware Implementation
Two experiments were conducted with the testing apparatus.
First, a pulsed-discharge cycle was carried out on the cell at a
CAMBRON AND CRAMER: LITHIUM-ION BATTERY CURRENT ESTIMATION TECHNIQUE USING AN UNKNOWN INPUT OBSERVER
Fig. 4. Hardware implementation results for the C/2 pulsed-discharge test.
The test begins with a cell at 95% state of charge and concludes with the cell
discharged to 45% state of charge. (a) Comparison of estimated and measured
current. (b) Filtered and instantaneous current error.
rate of 8 A. The results of the test are shown in Fig. 4. From the
results, it is evident that the current estimator correctly predicts
a pulsed-discharge load pattern. The instantaneous error in cur-
rent estimation peaks nearly at the magnitude of the measured
current; however, this is inconsequential because the peaks are
caused by a slight delay between the current sensor and the
current estimator. On the other hand, the filtered current error
limits itself to a maximum ±3.0 A but is significantly less dur-
ing steady-state conditions. The estimated current itself exhibits
a root mean square error (RMSE) of 0.735 A, or 5.9% when
normalized by the range of the measured current.
The second experiment was conducted on the cell to further
characterize the dynamic estimation of current about a single
state of charge. The cell current was continuously varied be-
tween −10 A and 10 A over the course of a 5 min period.
The results appear in Fig. 5 and show that in an average sense,
the hardware implementation was able to estimate cell current
with an accuracy of ±2.0 A, whereas the instantaneous error in
current estimation peaked at 5.2 A. An RMSE of 1.07 A was
achieved, which yields a 6.4% RMSE when normalized by the
range of the measured current.
Multiple sources of error are inherently present in the test-
ing apparatus, and they contribute to the error of the estimator.
Namely, all measured voltages used in this paper exhibit a max-
imum ±1.0 mV error from the sensor and the analog-to-digital
conversion. The magnitude of the R0 term suggests, from this
Authorized licensed use limited to: VIT University. Downloaded on March 14,2021 at 17:33:38 UTC from IEEE Xplore. Restrictions apply.
6713
Fig. 5. Hardware implementation results for the continuously varying current
test at 95% state of charge. (a) Comparison of estimated and measured current.
(b) Presentation of filtered and instantaneous current error. Note that the filtering
technique causes the filtered error to lag behind the instantaneous error.
fact alone, a current error as great as ±0.5 A, or 6.0% for
this apparatus. In addition, the Hall-effect sensor itself exhibits
a maximum 6.8% error at this current magnitude. This high-
lights the fact that extremely precise voltage measurements are
required in order for this method of current estimation to be fea-
sible; however, it also suggests that the observer current errors
seen in this paper are within reason when put in the context of
the hardware limitations.
VI. CONCLUSION
A practical model of a LiFePO4 cell was developed, which
performed similarly to other models in the literature. Using
this model, a UIO was developed, which attempted to estimate
the cell current. The model of the current estimator was used
in conjunction with the model of cell, and simulations were
conducted, which showed that the current estimator converged
toward the true measured current. A hardware implementation
was then tested, with results validating the operation of the
current estimator.
An error analysis showed that much of the observed error
could be attributed to the testing apparatus itself, indicating that
the feasibility of this technique is largely related to the preci-
sion of the terminal voltage measurements. Thus, this technique
could potentially be used in fault detection scenarios alongside
a conventional current transducer, or as a complete replacement
of a current transducer in power monitoring applications where
the current estimator is adequate for the required accuracy.
6714 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 66, NO. 8, AUGUST 2017
However, there remains an opportunity for further exploration
of this novel technique in order to improve its practicality. In
particular, the effects of parallel-connected cells were not con-
sidered in this paper, and the consideration of cell degradation
could be further expanded. Finally, the choice of observer model
used in this paper was not exhaustive, and further work is nec-
essary to determine which observer model would most suitably
implement this technique.
REFERENCES
[1] K. W. E. Cheng, B. P. Divakar, H. Wu, K. Ding, and H. F. Ho, “Battery-
management system (BMS) and SOC development for electrical vehicles,”
IEEE Trans. Veh. Technol., vol. 60, no. 1, pp. 76–88, Jan. 2011.
[2] S. Ziegler, R. C. Woodward, H. H.-C. Iu, and L. J. Borle, “Current sens-
ing techniques: A review,” IEEE Sensors J., vol. 9, no. 4, pp. 354–376,
Apr. 2009.
[3] O. Erdinc, B. Vural, and M. Uzunoglu, “A dynamic lithium-ion battery
model considering the effects of temperature and capacity fading,” in Proc.
Int.Conf. Clean Elect. Power, 2009, pp. 383–386.
[4] S. Dey, B. Ayalew, and P. Pisu, “Nonlinear robust observers for state-of-
charge estimation of lithium-ion cells based on a reduced electrochemical
model,” IEEE Trans. Control Syst. Technol., vol. 23, no. 5, pp. 1935–1942,
Sep. 2015.
[5] J. Marcicki, F. Todeschini, S. Onori, and M. Canova, “Nonlinear parameter
estimation for capacity fade in lithium-ion cells based on a reduced-order
electrochemical model,” in Proc. Amer. Control Conf., 2012, pp. 572–577.
[6] M. Rad, D. Danilov, M. Baghalha, M. Kazemeini, and P. Notten, “Thermal
modeling of cylindrical LiFePO4 batteries,” J. Mod. Phys., vol. 4, no. 7B,
pp. 1–7, 2013.
[7] M. Partovibakhsh and G. Liu, “An adaptive unscented Kalman filtering
approach for online estimation of model parameters and state-of-charge of
lithium-ion batteries for autonomous mobile robots,” IEEE Trans. Control
Syst. Technol., vol. 23, no. 1, pp. 357–363, Jan. 2015.
[8] C. Lin, X. Zhang, R. Xiong, and F. Zhou, “A novel approach to state of
charge estimation using extended Kalman filtering for lithium-ion batteries
in electric vehicles,” in Proc. Asia–Pac. IEEE Transp. Electrification Conf.
Expo., 2014, pp. 1–6.
[9] M. Daboussy, D. Chrenko, E.-H. Aglzim, Z. H. Che Daud, and L. Le
Moyne, “Characterisation of a commercial automotive lithium-ion ion
battery using extended Kalman filter,” in Proc. Asia–Pac. IEEE Transp.
Electrification Conf. Expo., 2013, pp. 1–6.
[10] T. Wang, L. Pei, R. Lu, C. Zhu, and G. Wu, “Online parameter identifi-
cation for lithium-ion cell in battery management system,” in Proc. Veh.
Power Propulsion Conf., 2014, pp. 1–6.
[11] J. Xu, C. C. Mi, B. Cao, J. Deng, Z. Chen, and S. Li, “The state of
charge estimation of lithium-ion batteries based on a proportional-integral
observer,” IEEE Trans. Veh. Technol., vol. 63, no. 4, pp. 1614–1621,
May 2014.
[12] H. Chaoui, N. Golbon, I. Hmouz, R. Souissi, and S. Tahar, “Lyapunov-
based adaptive state of charge and state of health estimation for lithium-
ion batteries,” IEEE Trans. Ind. Electron., vol. 62, no. 3, pp. 1610–1618,
Mar. 2015.
[13] E. Leksono, I. N. Haq, M. Iqbal, F. X. N. Soelami, and I. G. N. Merthayasa,
“State of charge (SoC) estimation on LiFePO4 battery module using
Coulomb counting methods with modified Peukert,” in Proc. Rural Inf.
Comm. Technol. Elect.-Veh. Technol., 2013, pp. 1–4.
[14] M. Gholizadeh and F. R. Salmasi, “Estimation of state of charge, unknown
nonlinearities, and state of health of a lithium-ion battery based on a
comprehensive unobservable model,” IEEE Trans. Ind. Electron., vol. 61,
no. 3, pp. 1335–1344, Mar. 2014.
[15] N. Lin, S. Ci, and H. Li, “An enhanced circuit-based battery model with
considerations of temperature effect,” in Proc. Energy Convers. Congr.
Expo., 2014, pp. 3985–3989.
[16] L. W. Yao, J. A. Aziz, P. Y. Kong, and N. R. N. Idris, “Modeling of
lithium-ion battery using MATLAB/simulink,” in Proc. 39th Annu. Conf.
Ind. Electron. Soc., 2013, pp. 1729–1734.
[17] L. Xiwen, M. Yan, and Y. Zhenhua, “Research of SOC estimation for
lithium-ion battery of electric vehicle based on AMEsim-simulink co-
simulation,” in Proc. Control Conf., 2013, pp. 7680–7685.
[18] L. Lam, P. Bauer, and E. Kelder, “A practical circuit-based model for
Li-ion battery cells in electric vehicle applications,” in Proc. Telecommun.
Energy Conf., 2011, pp. 1–9.
Authorized licensed use limited to: VIT University. Downloaded on March 14,2021 at 17:33:38 UTC from IEEE Xplore. Restrictions apply.
[19] R. C. Kroeze and P. T. Krein, “Electrical battery model for use in dynamic
electric vehicle simulations,” in Proc. Power Electron. Spec. Conf., 2008,
pp. 1336–1342.
[20] J. Jaguemont, L. Boulon, and Y. Dube, “Characterization and modeling
of a hybrid-electric-vehicle lithium-ion battery pack at low temperatures,”
IEEE Trans. Veh. Technol., vol. 65, no. 1, pp. 1–14, Jan. 2016.
[21] M. Swierczynski, D.-I. Stroe, A.-I. Stan, R. Teodorescu, and S. K. Kaer,
“Investigation on the Self-discharge of the LiFePO4/C nanophosphate
battery chemistry at different conditions,” in Proc. Transp. Electrification
Asia–Pac. Conf. Expo., 2014, pp. 1–6.
[22] D. Liu, J. Zhou, H. Liao, Y. Peng, and X. Peng, “A health indicator
extraction and optimization framework for lithium-ion battery degradation
modeling and prognostics,” IEEE Trans. Syst., Man, Cybern., Syst., vol. 45,
no. 6, pp. 915–928, Jun. 2015.
[23] C. Blanco, L. Sanchez, M. Gonzalez, J. C. Anton, V. Garcia, and
J. C. Viera, “An equivalent circuit model with variable effective capac-
ity for LiFePO4 batteries,” IEEE Trans. Veh, Technol., vol. 63, no. 8,
pp. 3592–3599, Oct. 2014.
[24] F. Baronti, N. Femia, R. Saletti, C. Visone, and W. Zamboni, “Hysteresis
modeling in Li-ion batteries,” IEEE Trans. Magn., vol. 50, no. 11, pp. 1–4,
Nov. 2014.
[25] F. Baronti, W. Zamboni, N. Femia, R. Roncella, and R. Saletti, “Ex-
perimental analysis of open-circuit voltage hysteresis in lithium-iron-
phosphate batteries,” in Proc. Ind. Electron. Soc., 2013, pp. 6728–6733.
[26] L. Gagneur, C. Forgez, and A. L. D. Franco, “Lithium-ion state of charge
observer with open circuit voltage hysteresis model,” in Proc. 15th Eur.
Conf. Power Electron. Appl., 2013, pp. 1–7.
[27] M. Hou and P. C. Müller, “Design of observers for linear systems
with unknown inputs,” IEEE Trans. Automat. Control, vol. 37, no. 6,
pp. 871–875, Jun. 1992.
[28] J. Chen, R. J. Patton, and H.-Y. Zhang, “Design of unknown input ob-
servers and robust fault detection filters,” Int. J. Control, vol. 63, no. 1,
pp. 85–105, 1996.
[29] F. Yang and R. W. Wilde, “Observers for linear systems with unknown in-
puts,” IEEE Trans. Automat. Control, vol. 33, no. 7, pp. 677–681, Jul. 1988.
[30] M. E. Valcher, “State observers for discrete-time linear systems with
unknown inputs,” IEEE Trans. Automat. Control, vol. 44, no. 2,
pp. 397–401, Feb. 1999.
[31] B. Walcott and S. H. Żak, “State observation of nonlinear uncertain
dynamical systems,” IEEE Trans. Automat. Control, vol. AC-32, no. 2,
pp. 166–170, Feb. 1987.
[32] M. Corless and J. Tu, “State and input estimation for a class of uncertain
systems,” Automatica, vol. 34, no. 6, pp. 757–764, 1998.
[33] J. Moreno, “Unknown input observers for SISO nonlinear systems,” in
Proc. 39th IEEE Conf. Decis. Control, 2000, vol. 1, pp. 790–801.
[34] C.-T. Chen, Linear System Theory and Design, 3rd ed. New York, NY,
USA: Oxford Univ. Press, 1998.
[35] Electric Vehicle Battery Test Procedures Manual, United States Adv. Bat-
tery Consortium, Southfield, MI, USA, Jan. 1996.
Daniel C. Cambron (S’11–M’17) received the B.S.
degree (summa cum laude) in electrical engineer-
ing and computer engineering and the M.S. de-
gree in electrical engineering from the University of
Kentucky, Lexington, KY, USA, in 2015 and 2016,
respectively.
He is currently an Electrical Engineer with Sen-
triLock, West Chester, PA, USA, and CEO and co-
founder of Lexcelon, Lexington, KY, USA.
Aaron M. Cramer (S’02–M’07–SM’13) received
the B.S. degree (summa cum laude) in electrical engi-
neering from the University of Kentucky, Lexington,
KY, USA, and the Ph.D. degree from Purdue Uni-
versity, West Lafayette, IN, USA, in 2003 and 2007,
respectively.
From 2007 to 2010, he was a Senior Engineer with
PC Krause and Associates, West Lafayette. From
2010 to 2016, he was an Assistant Professor with
the University of Kentucky, where he is currently an
Associate Professor. His research interests include
modeling, simulation, and control of power and energy systems.
Dr. Cramer received the ONR Young Investigator Program award in 2015.
He serves as an Editor of the IEEE TRANSACTIONS ON ENERGY CONVERSION.