232 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 29, NO.

1, MARCH 2014

Online Condition Monitoring of Battery Systems

With a Nonlinear Estimator
Günyaz Ablay, Member, IEEE

Abstract—The performance of batteries as uninterruptable

power sources in any industry cannot be taken for granted. The fail-
ures in battery systems of safety-related electric systems can lead
to performance deterioration, costly replacement, and, more im-
portantly, serious hazards. The possible failures in battery systems
are currently determined through periodic maintenance activities.
However, it is desirable to be able to detect the underlying degra-
dation and to predict the level of unsatisfactory performance by
an online real-time monitoring system to prevent unexpected fail- Fig. 1. General working principle of battery systems.
ures through early fault diagnosis. Such an online fault diagnosis
method can also contribute to better maintenance and optimal bat-
are many factors that cause failures in battery systems, includ-
tery replacement programs. A robust nonlinear estimator-based
online condition monitoring method is proposed to determine the ing design errors, battery component failures, overloading of dc
state of health of the battery systems online in industry. Real-world busses, flaking of cell plates, and aging-related failures [3], [4].
experimental data of a modern battery system are used to assess the For these reasons, monitoring and maintenance activities are
efficiency of the proposed approach in the existence of parameter necessary for battery systems. Many industries (e.g., nuclear
uncertainties.
power plants) perform periodic maintenance (monthly, quar-
Index Terms—Battery management, battery modeling, condi- terly, and yearly) and battery parameter testing (capacity and
tion monitoring, fault diagnosis. internal ohmic tests) for battery systems within the mainte-
nance program. The functionality and reliability of batteries can
be enhanced with online fault diagnosis and health monitoring.
I. INTRODUCTION The diagnostic algorithm can monitor the battery system for any
HE increasing needs for high energy, power, life-cycle, possible faults and malfunctions. A battery monitoring system
T FUEL economy, wide-range operating temperature, and
environmentally acceptable batteries in electric and hybrid elec-
is specifically necessary for sealed batteries, e.g., valve regu-
lated lead-acid and lithium-ion batteries, since it is not possible
tric systems are the driving forces for rapid growth of battery to directly measure their internal parameters, i.e., internal cell
technologies. While there are many battery types, some bat- resistances. It helps in detecting deviations of the battery perfor-
teries are more preferable to others in certain industries. For mance from its normal behavior and also isolates the possible
example, Ni-MH batteries are dominant battery technology for causes for the faults. In practice, output signals (measurements)
hybrid electric vehicle applications by having the best overall of the system under consideration are often directly evaluated
performance in the wide-range requirements set by automobile and compared with a given threshold [5]. However, such an
companies [1]. The nuclear plants benefit from lead-acid batter- approach is insufficient in critical processes due to capability
ies for their wide range of electrical systems as battery-backed to react only after a relatively large change in the measured
supplies (UPSs) to ensure continuity of function during outages variable [6], [7]. If the system is modeled, and the difference
or interruptions [2], [3]. Fig. 1 shows a general working princi- between the measurement and its estimation (residual) is ob-
ple of the battery systems in various industries. The dc load is tained, then it is theoretically possible to detect every fault with
powered from battery chargers (rectifiers) during normal oper- residual generation [8], [9]. The measurements of rechargeable
ation, and it is automatically powered from station batteries in batteries are current and voltage at the poles of the batteries and
case of loss of normal power to the battery chargers. the battery temperature, which are enough to model the battery
Reliability of the battery system is significant for any indus- system for fault diagnosis and health monitoring in order to pre-
try due to the safety-related usage aim of the batteries. There vent serious damage and failures and to have better maintenance
activities. In the literature, most of the fault diagnosis and health
monitoring studies are provided for lithium-ion batteries due to
their usage in wide-range applications. The approaches include
Manuscript received July 5, 2013; revised September 23, 2013 and October extended Kalman filter, autoregressive moving average model,
21, 2013; accepted November 14, 2013. Date of publication December 3, 2013; and fuzzy logic, which are summarized in [10], model-based
date of current version February 14, 2014. Paper no. TEC-00379-2013.
The author is with the Department of Electrical-Electronics Engineering, state-of-charge (SOC) estimation [11], model-based fault diag-
Abdullah Gul University, 38039 Kayseri, Turkey (e-mail: gunyaz.ablay@agu. nosis algorithm [12], and parabolic regression algorithm [13].
edu.tr). However, the robustness and efficiency of the above estima-
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. tion approaches are questionable due to parameter uncertainties,
Digital Object Identifier 10.1109/TEC.2013.2291812 simplifications, and linearization in the models.
0885-8969 © 2013 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.
ABLAY: ONLINE CONDITION MONITORING OF BATTERY SYSTEMS WITH A NONLINEAR ESTIMATOR 233

In this study, a robust nonlinear estimator-based online con-

dition monitoring methodology is proposed for robust fault di-
agnosis in battery systems in the presence of parameter uncer-
tainties that make the previous studies questionable. In reality,
the periodic maintenance activities are the only way to deter-
mine health of a battery in industries, e.g., nuclear power plants.
With the usage of the proposed online condition monitoring
approach, unexpected failures can be prevented by early fault
detection, and optimization in the replacement programs and Fig. 2. Battery Thevenin equivalent model.
better maintenance activities can be obtained efficiently. TABLE I
The organization of the paper is as follows. Section II de- EXPERIMENTAL BATTERY PARAMETERS
scribes the characterization and modeling of battery systems.
Section III describes the design, synthesis, and analysis of online
condition monitoring strategy. The assessments and conclusions
are given in Sections IV and V, respectively.

II. BATTERY SYSTEMS

There have been significant developments in battery tech-
nologies that offer excellent advantages, including flexible cell thermal, and SOC models. Each of these submodels is described
sizes from 30 mAh to 250 Ah, safe operation at high voltage below.
(+320 V), high volumetric energy and power, maintenance-free
designs, excellent thermal properties, environmentally accept- A. Electrical Model
able and recyclable materials, and simple and inexpensive charg- The voltage (capacity) in the battery cells is represented by
ing and electronic control circuits [1], [14]. The lithium-ion and a voltage source, the electrochemical delay of the battery due
especially nickel–metal hydride (Ni-MH) batteries are the cur- to shifting ion concentrations and plate current densities is de-
rent batteries that can provide the above advantages. However, scribed by a capacitor–resistor model, and the internal resis-
the common use of these new rechargeable battery systems in tance of the battery due to electrolyte diffusion is modeled by
UPSs and industry requires further studies and new technical a resistor [26], [27]. The most commonly used battery mod-
standards. els are ideal model, linear model, and Thevenin equivalent
The electrochemistry of rechargeable batteries is based on model [16], [28], [29]. The Thevenin equivalent model with
charge-discharge mechanism between positive and negative varying parameters due to temperature and SOC variations rep-
electrodes. The electricity generation processes in a battery cell resents batteries with higher accuracy, as experimentally shown
are chemical reactions that either consume or release electrons in many studies [30], and thus, it is considered in this study. This
as the electrode reaction proceeds to completion. Such reac- model is made of electrical values of the open-circuit voltage
tion, depending on battery electrode and electrolyte materials (Eo ), internal resistance (R), capacitance (Co ), and the over-
and mechanisms that affect battery terminal voltage, is similar voltage resistance (Ro ) [16]. The Thevenin equivalent model of
to all types of batteries. Battery terminal voltage changes with a battery is depicted in Fig. 2.
the electrolyte concentration and there are resistive drops in From Fig. 2, the dynamic electric equations of the circuit can
electrodes with current flow [15]. be obtained by applying Kirchhoff’s rules as
To extend the life of a battery system, charge and thermal ⎧
managements play important roles and any problem related ⎨ dVc = − 1 V + 1 I
c
dt Ro Co Co (1)
to overcharge, overdischarge, or control system must be di- ⎩
agnosed with an online condition monitoring system. For de- V = Eo − RI − Vc
velopments of online real-time battery monitoring systems, the where I (in Amperes) is the current at the input, Vc (in Volts) is
large-scale storage batteries have been commonly modeled in the capacitance voltage and V (in Volts) is the output voltage.
three ways: equivalent circuit battery models [16]–[19], electro- All the parameters of (1) are functions of battery temperature (T
chemical battery models [20]–[23], and correlation-based mod- in ◦ C) and SOC (denoted by S), and can be different in charge
els [24]. Since electrochemical and experimental models are not and discharge phases. The battery parameters are determined
well suited to demonstrate battery dynamics accurately [25], the using experimental battery charge and discharge data for a Ni-
equivalent circuit battery model that is based on current-voltage MH battery system, and given in Table I. The meanings of
relations of the batteries is considered in this study. the parameters described by “other parameters” in Table I are
The battery system consists of two main subsystems: battery explained in Section II-C. The data used in this study were
package and temperature controller. Considering battery (load) acquired as part of the activities of the Center for Automotive
current as an input, and battery voltage and temperature as out- Research at the Ohio State University. A general structure of
puts, the battery model consists of three submodels: electrical, the battery test system is illustrated in Fig. 3. The test system
234 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 29, NO. 1, MARCH 2014

ber of different modeling approaches in the literature including

discharge test, ampere hour counting, linear model, impedance
spectroscopy, internal resistance, Kalman filter, and open-circuit
voltage model [10], [31]–[33]. One of the most acceptable SOC
calculation methods is the ampere hour counting (current inte-
gration) method [10], [31] defined by
 t
Fig. 3. Battery test system. 1
S = S0 − I(t)dt (2)
Cn 0

where I (in Amperes) is the battery (load) current, Cn (in

Ampere-hours) is the nominal capacity, and the initial SOC
is given by S0 . This approach requires dynamic measurement
of the battery current. It is applicable to all battery systems, easy
to calculate and accurate with good current measurements.
Practically, the operating range of batteries with respect to
SOC is confined with specific numbers. For example, a Ni-MH
battery has an operating range around 40–80% with respect to
SOC. Hence, this method implies that the charging operation is
controlled.

Fig. 4. Battery electrical model validation using voltage measurements at

room temperature.
C. Thermal Model
In many systems, it is critical to achieve performance and ex-
tended life of batteries through thermal management. Recharge-
able batteries are designed to work in room temperature (25 ◦ C),
but a battery cannot be held at a certain temperature due to
daily and seasonal temperature changes. It is well known that
low temperatures slow up the chemical reactions in any bat-
tery, resulting in reduced performance; however, from the other
extreme high temperatures destroy batteries. According to the
Arrhenius equation [34], for every 10 ◦ C increase in battery
temperature, battery life is halved due to faster positive grid
corrosion. Suitable modeling for predicting thermal behavior
Fig. 5. Nominal current discharge characteristics for different types of
batteries. of battery systems in applications can help to improve battery
design and development process. Therefore, thermal models for
batteries have been developed based on thermal energy balance
is appropriate for modeling any battery systems. With the use of batteries, and they are coupled with electrochemical or elec-
of this battery test system, the accuracy of the battery electrical tric models [21], [22]. Based on the electric model and thermal
model is tested and showed in Fig. 4. As seen from Fig. 4, the energy balance of a battery, a simple thermal model can be built
battery model has a good level of accuracy. as
To represent a particular battery type, the parameters of the
electrical model can be changed depending on its discharge dT R + Ro 2 hA
= I − (T − T∞ ) (3)
characteristics as illustrated in Fig. 5. It can be seen from the dt mc mc
figure that the exponential voltage drop sections of the batteries
where mc (in J/◦ C) is the effective heat capacity per cell, hA (in
can have more or less wide areas depending on the battery type.
J/◦ C) is the effective heat transfer per cell, and T∞ (in ◦ C) is the
The parameters gathered from the discharge characteristics are
bulk temperature. The effective heat transfer capacity per cell is
usually assumed to be the same for charging when the electrical
given in terms of temperature controller (fan) settings by
model parameters are determined.
hA = h0 A0 (1 + 0.5fs ) (4)
B. SOC Model
SOC provides information about remaining useful energy where h0 A0 (in J/◦ C) is given in Table I for natural convection,
and the remaining usable time of the battery. Many systems and the controller setting fs depends on the temperature. In this
are sensitive to overcharge and too high or too low SOC can study, the following temperature controller settings will be used:
result in irreversible damage in the battery. It can be difficult 
fs = 0 off mode, fs = 1 if T ≥ 30◦ C
to measure the SOC in the battery systems depending on the (5)
type and application of the battery. There have been a num- fs = 2 if T ≥ 35 o C fs = 3 if T ≥ 40◦ C.
ABLAY: ONLINE CONDITION MONITORING OF BATTERY SYSTEMS WITH A NONLINEAR ESTIMATOR 235

B. Dynamical Analysis of the Residual Generator

By considering dynamic equations (1) and (3), the state space
form of the battery model can be written as
    
dVc /dt −k1 0 Vc b1 0
= + I+ T∞
dT /dt 0 −k2 T b2 I k2
(8)
where k1 = 1/Ro Co , k2 = hA/mc, b1 = 1/Co , and b2 =
(R + Ro )/mc. Since I and T∞ are bounded, i.e., sup |I(t)| <
Im for Im < ∞ and sup |T∞ (t)| < Tm for Tm < ∞, and it is
Fig. 6. Fault diagnosis scheme for the battery system. evident from the system matrix that the battery system has two
distinct and negative eigenvalues, λV c = −k1 and λT = −k2 .
Hence, battery systems have stable dynamics. Here, Vc (t) has
III. ONLINE CONDITION MONITORING IN BATTERY SYSTEMS much faster stable dynamics than T (t) since |λV c | >> |λT |, im-
The model of battery systems is complicated due to the feed- plying that the slow state variable T (t) determines the dynamic
back loop and the state-depended and time-varying parameters. behavior of the battery system under bounded input values. It is
The equivalent electrical model parameters of batteries are func- clear from (8) that the dynamics of Vc and T are almost com-
tions of temperature and SOC, which make fault detection dif- pletely independent of each other (uncoupled). Temperature T
ficult. Moreover, the temperature feedback controller can mask is a directly measurable state, but capacitance voltage Vc is an
the temperature sensor faults so that the detection problem can indirectly measurable state through battery voltage V as given
pose some additional challenges. A model-based online condi- in (1). For these reasons, the error dynamics of the estimator
tion monitoring scheme for the battery systems is illustrated in can be analyzed separately.
Fig. 6. The residual generator is composed of a nonlinear esti- To analyze the stability of the nonlinear estimator (6), first
mator, and the residual evaluation subsystem is used to provide we can consider the temperature estimation error eT = T − T̂ .
fault information by comparing the residuals with their nom- The dynamic of the temperature estimation error is
inal values. The residual generator and evaluation algorithms
ėT = −k2 eT + (b2 − b̂2 )I 2 − δ1 sign(eT ). (9)
are designed below to detect and isolate possible system faults
online. A candidate Lyapunov function [35], [36] can be selected as
W = e2T /2; then, its time derivative is
A. Robust Nonlinear Estimator for Residual Generation eT ėT = eT ((b2 − b̂2 )I 2 − δ1 sign(eT )) − k2 e2T
A robust nonlinear estimator-based approach is considered to
≤ (|(b2 − b̂2 )I 2 | − δ1 ) |eT | − k2 e2T . (10)
generate residuals in this study. The main advantages of this ap-
proach are robustness and high precision state estimation in the Here,

the observer
gain δ1 is selected large enough to satisfy
presence of parameter variations. These features of the nonlin-
2

δ1 >
(b2 − b̂2 )I
. Hence, the derivative of the Lyapunov func-
ear estimator can enhance detection of internal battery problems
tion turns out to be negative definite, eT ėT < 0, which indicates
through detection of internal cell resistance variations. Based on
error convergence to zero.
the system model given in (1)–(5), a nonlinear estimator can be
Finally, the dynamic of capacitance voltage estimation error
designed as
eV c = Vc − V̂c can be written as
⎧
⎪
⎪ dT̂ R̂ + R̂o 2 ĥA ėV c = −k1 Vc + b1 I + k1 V̂c − b1 I
⎪
⎪ = I − (T̂ − T∞ ) + δ1 sign(T − T̂ )
⎪
⎪ dt mc mc
⎪
⎪ = −k1 eV c . (11)
⎨ dV̂c 1 1
=− V̂c + I (6)
⎪
⎪ dt R̂o Co Co It is clear form (11) that eV c is going to approach to zero
⎪
⎪
⎪
⎪ V̂ = Êo − R̂I − V̂c
with increasing time since k1 > 0. As a result, the estimation
⎪
⎪
⎩ errors on the temperature and capacitance voltage decay to zero,
ĥA = h0 A0 (1 + 0.5fˆs ) i.e. eT → 0, eV c → 0, with some time by choosing the observer
gain δ1 sufficiently large.
where (T̂ , V̂c , V̂ ) are the estimates of (T, Vc , V ), the observer
gain δ1 is a constant to be selected, and the sign(·) function is IV. RESIDUAL EVALUATION
the unit vector defined by
The battery system has three measurements: voltage, tem-
sign(e) = e/ |e| (7) perature as outputs, and a load current input. By using these
measurements, it is possible to detect and isolate single faults in
where eT is the error state defined by eT = T − T̂ . The nonlin- temperature and voltage sensors, and internal resistance devia-
ear estimator gain δ1 must be selected large enough to bring the tions. In addition, a failure in the temperature controller can be
estimation errors to zero as explained in the following section. detected.
236 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 29, NO. 1, MARCH 2014

TABLE II
FAULT DIAGNOSIS CHART

Fig. 7. Probability density estimation for voltage residual (normalized).

The residuals can be acquired with the usage of estimation Fig. 8. Flow diagram of the proposed fault diagnosis approach.
error states:

rT ≡ eT = T − T̂ A flow diagram of the proposed fault diagnosis approach is
(12)
rV ≡ eV = V − V̂ illustrated in Fig. 8. This dynamic method is an online real-time
approach, so measurements must be obtained continuously for
where rT and rV are residuals for temperature and voltage
early fault detections.
measurements, respectively.
The generated residuals can be evaluated based on some
threshold selections. There will always be a tradeoff between V. NUMERICAL RESULTS
the avoidance of false alarms and the detection of small faults in The implementation of the online condition monitoring ap-
the existence of noise and modeling errors. Practically, a fault proach for the battery systems are performed with MAT-
can be detected only if it causes the residual evaluation function LAB/Simulink programs by using the experimental current mea-
to surpass a threshold [8] surements of the Ni-MH battery system. The nonlinear estimator
gain is selected as δ1 = 2 (note that since the nonlinear estimator
|ri (t)| ≥ Ri (13) is robust, this gain should not be selected very large in order to
where ri (t) is the function of the ith residual and Ri is the have high sensitivity to faults). Initial conditions for estimator
selected threshold for ith residual function. are taken as To = 19 ◦ C and Vco = 0. For threshold calculations,
A systematic threshold selection may be based on the statisti- the mean and standard deviation of voltage and temperature
cal evaluation of the residuals under the absence of the fault since residuals, by considering their root mean square (rms) values,
residuals can be degraded by the measurement noise or distur- are found as μV = 0.44, σV = 0.65, μT = 0.0004, and σT =
bance. The thresholds for each residual can be calculated from 0.0004. The rms functions of the residuals are used in the resid-
the mean and standard deviation of the corresponding residuals ual evaluations. For the estimator, the sat(·) function defined
under nonfaulty cases. For example, probability density estima- below is used instead of the sign(·) function in order to ease
tions of the voltage residual under nonfaulty and faulty cases numerical calculations via continuous approximation.

are illustrated in Fig. 7, which shows different mean and stan- e/ |e| , |e| > 1
dard deviation values for nonfaulty and faulty cases. An optimal sat(e) = (14)
e, |e| ≤ 1.
threshold value can be the intersection of the probability density
functions of nonfaulty and faulty residual signals (see Fig. 7). Fig. 9 displays the load current and the SOC estimation for
However, in such a case, the overlapped regions result in either an initial value 60% and under normal operating conditions.
false alarms or missed detections. Therefore, more conservative While the SOC is a difficult parameter to predict in the battery
threshold values must be selected to avoid false alarms. To avoid management system, the current integral method is commonly
false alarms, thresholds for residuals rV and rT can be selected used to estimate SOC and provides information about remaining
as RV = μV + 4σV and RT = μT + 4σT , similar to the She- useful energy and the remaining usable time of the battery.
whart control chart [37], where μT and μV are the means, and Different faults on measurements and battery parameters are
σT and σV are the standard deviations of related residuals under considered for fault diagnosis as given in Table II. Several incip-
fault-free cases. ient type faults are injected into the battery system for diagnosis
A residual evaluation chart is given in Table II, which shows since the detection of incipient faults is most difficult. Fig. 10 il-
that single faults can be detected and isolated. The temperature lustrates a voltage sensor fault and its detection using the voltage
sensor fault and unexpected internal resistance change may not residual rV . It is obvious from the figure that once an incipi-
be isolated with the given logic because both affect the same ent voltage sensor fault occurs, it is detected and isolated via
residuals. residual surpassing the threshold RV .
ABLAY: ONLINE CONDITION MONITORING OF BATTERY SYSTEMS WITH A NONLINEAR ESTIMATOR 237

Fig. 9. SOC estimation from the load current. (a) Load current. (b) SOC
estimation.
Fig. 11. Control (fan) failure and its detection. (a) Control signal [unitless,
see (5)]. (b) Residual rv . (c) Residual rT .

Fig. 10. Voltage sensor fault and its monitoring. (a) Time response of voltage.
(b) Residual rv . (c) Residual rT .

In Fig. 11, it is assumed that the temperature controller failed

at the time t = 3000 s. This failure is detected by temperature Fig. 12. Temperature sensor fault and its detection. (a) Time response of
temperature. (b) Residual rv . (c) Residual rT .
residual rT . The figure shows that the controller failure immedi-
ately affects the temperature residual so that the residual exceeds
the threshold RT .
An incipient temperature sensor fault that is injected into the
battery system is depicted in Fig. 12. The temperature and volt- An unexpected deviation in the internal resistance of the bat-
age residuals immediately exceed their threshold values to indi- tery is shown in Fig. 13. A small and drift-like deviation in the
cate the existence of temperature sensor fault. We can observe resistance results in alarms in voltage and temperature residuals
from figures that the nonlinear estimator exhibits robustness by exceeding thresholds. The internal resistance of battery pro-
characteristics, which is important for healthy online condition vides knowledge about several problems, including grid growth,
monitoring. It should be noted that rV may not exceed its thresh- sulfation, dry out, and corrosion. Hence, the diagnosis of inter-
old for small temperature sensor fault values unless the threshold nal resistance fault is an important aspect of the methodology
is decreased, but then there will be some false alarms. in terms of its good performance and efficiency.
238 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 29, NO. 1, MARCH 2014

[6] R. Isermann, “Model-based fault-detection and diagnosis—Status and ap-

plications,” Annu. Rev. Control, vol. 29, no. 1, pp. 71–85, 2005.
[7] R. Isermann, Fault-Diagnosis Systems. New York, NY, USA: Springer-
Verlag, 2006.
[8] J. Gertler, Fault Detection and Diagnosis in Engineering Systems. New
York, NY, USA: Marcel Dekker, 1998.
[9] C. Hajiyev and F. Caliskan, Fault Diagnosis and Reconfiguration in Flight
Control Systems. Norwell, MA, USA: Kluwer, 2003.
[10] J. Zhang and J. Lee, “A review on prognostics and health monitoring of
Li-ion battery,” J. Power Sources, vol. 196, pp. 6007–6014, 2011.
[11] X. Tang, X. Zhang, B. Koch, and D. Frisch, “Modeling and estimation of
nickel metal hydride battery hysteresis for SOC estimation,” presented at
the Int. Conf. Prognostics Health Manage., Danver, CO, USA, 2008.
[12] C. Suozzo, S. Onori, and G. Rizzoni, “Model-based fault diagnosis for
NiMH battery,” presented at the 1st Annu. Dyn. Syst. Control Conf., Ann
Arbor, MI, USA, 2009.
[13] M. V. Micea, L. Ungurean, G. N. Cârstoiu, and V. Groza, “Online State-of-
health assessment for battery management systems,” IEEE Trans. Instrum.
Meas., vol. 60, no. 6, pp. 1997–2006, Jun. 2011.
[14] D. Ilic, M. Kilb, K. Holl, H. Praas, and E. Pytlik, “Recent progress in
rechargeable nickel/metal hydride and lithium-ion miniature rechargeable
batteries,” J. Power Sources, vol. 80, no. 1–2, pp. 112–115, 1999.
[15] D. Linden and T. B. Reddy, Handbook of Batteries, 3rd ed. New York,
NY, USA: McGraw-Hill, 2001.
[16] Y. Kim and H. Ha, “Design of interface circuits with electrical battery
models,” IEEE Trans. Ind. Electron., vol. 44, no. 1, pp. 81–86, Feb. 1997.
[17] A. C. Margaret and M. Salameh, “Determination of lead-acid battery via
Fig. 13. Internal resistance deviation and its detection. (a) Time response of mathematical model techniques,” IEEE Trans. Energy Convers., vol. 7,
internal resistance. (b) Residual rv . (c) Residual rT . no. 3, pp. 442–446, Sep. 1992.
[18] J. Zhang, S. Ci, H. Sharif, and M. Alahmad, “Modeling discharge be-
havior of multicell battery,” IEEE Trans. Energy Convers., vol. 25, no. 4,
VI. CONCLUSION pp. 1133–1141, Dec. 2010.
[19] T. Kim and W. Qiao, “A hybrid battery model capable of capturing dy-
This paper proposed a nonlinear estimator-based online con- namic circuit characteristics and nonlinear capacity effects,” IEEE Trans.
dition monitoring system for battery systems used in the in- Energy Convers., vol. 26, no. 4, pp. 1172–1180, Dec. 2011.
[20] B. Paxton and J. Newmann, “Modeling of nickel/metal hydride batteries,”
dustry. The study set out to determine the possibility of online J. Electrochem. Soc., vol. 144, no. 11, pp. 3818–3831, 1997.
battery health monitoring system based on battery modeling and [21] W. B. Gu and C. Y. Wang, “Thermal-electrochemical modeling of battery
detection of internal battery problems. This research has shown systems,” J. Electrochem. Soc., vol. 147, no. 8, pp. 2910–2922, 2000.
[22] V. Srinivasan and C. Y. Wang, “Analysis of electrochemical and thermal
that the nonlinear estimator-based approach provides a robust behavior of Li-ion cells,” J. Electrochem. Soc., vol. 150, no. 1, pp. A98–
and efficient fault diagnosis for internal faults through detection A106, 2003.
of internal cell resistance deviations, measurement faults, and [23] V. Agarwal, K. Uthaichana, R. A. DeCarlo, and L. H. Tsoukalas, “De-
velopment and validation of a battery model useful for discharging and
temperature controller failure in the battery system. The results charging power control and lifetime estimation,” IEEE Trans. Energy
of this study indicate that the proposed online real-time health Convers., vol. 25, no. 3, pp. 821–835, Sep. 2010.
monitoring approach can provide early fault detection, extended [24] C. Zhou, K. Qian, M. Allan, and W. Zhou, “Modeling of the cost of EV
battery wear due to V2G application in power systems,” IEEE Trans.
useful life, better maintenance, and optimized replacement tim- Energy Convers., vol. 26, no. 4, pp. 1041–1050, Dec. 2011.
ing for battery systems. [25] O. Tremblay and L. A. Dessaint, “Experimental validation of a battery
dynamic model for EV applications,” World Electr. Veh. J., vol. 3, pp. 1–
10, 2009.
ACKNOWLEDGMENT [26] C. R. Gould, C. M. Bingham, D. A. Stone, and P. Bentley, “New battery
model and state-of-health determination through subspace parameter esti-
The author gratefully would like to thank Prof. G. Rizzoni, mation and state-observer techniques,” IEEE Trans. Veh. Technol., vol. 58,
the Director of the Center for Automotive Research, Ohio State no. 8, pp. 3905–3916, Oct. 2009.
University, for his valuable guidance and contributions. [27] J. Marcos, A. Lago, C. M. Penalver, J. Doval, A. Nogueira, C. Castro, and
J. Chamadoira, “An approach to real behaviour modeling for traction lead-
acid batteries,” in Proc. IEEE 32nd Annu. Power Electron. Spec. Conf.,
REFERENCES 2001, vol. 2, pp. 620–624.
[28] A. Emadi, M. Ehsani, and J. M. Miller, Vehicular Electric Power Systems.
[1] M. A. Fetcenko, S. R. Ovshinsky, B. Reichman, K. Young, C. Fierro, New York, NY, USA: Marcel Dekker, 2004.
J. Koch, A. Zallen, W. Mays, and T. Ouchi, “Recent advances in NiMH [29] V. H. Johnson, “Battery performance models in ADVISOR,” J. Power
battery technology,” J. Power Sources, vol. 165, pp. 544–551, 2007. Sources, vol. 110, no. 2, pp. 321–329, Aug. 2002.
[2] D. Kim and H. Cha, “Analysis on lead-acid battery bank for nuclear power [30] H. He, R. Xiong, and J. Fan, “Evaluation of lithium-ion battery equivalent
plants in Korea,” in Proc. Int. Power Electron. Motion Control Conf., 2012, circuit models for state of charge estimation by an experimental approach,”
vol. 3, pp. 2118–2122. Energies, vol. 4, no. 12, pp. 582–598, 2011.
[3] N. K. Trehan, “Impact of failures of direct current systems on nuclear [31] S. Piller, M. Perrin, and A. Jossen, “Methods for state-of-charge determi-
power generating stations,” in Proc. IEEE Nucl. Sci. Symp. Conf. Rec., nation and their applications,” J. Power Sources, vol. 96, no. 1, pp. 113–
2001, vol. 4, pp. 2482–2485. 120, Jun. 2001.
[4] S. Eckroad and B. M. Radimer, “Review of engineering design consid- [32] S. Lee, J. Kim, J. Lee, and B. H. Cho, “The state and parameter estimation
erations for battery energy management systems,” IEEE Trans. Energy of an Li-ion battery using a new OCV-SOC concept,” presented at the
Convers., vol. 6, no. 2, pp. 303–309, Jun. 1991. IEEE Power Electron. Spec. Conf., Orlando, FL, USA, 2007.
[5] P. M. Frank and X. Ding, “Survey of robust residual generation and eval- [33] L. Liu, L. Y. Wang, Z. Chen, C. Wang, F. Lin, and H. Wang, “Integrated
uation methods in observer-based fault detection systems,” J. Process system identification and state-of-charge estimation of battery systems,”
Control, vol. 7, no. 6, pp. 403–424, 1997. IEEE Trans. Energy Convers., vol. 28, no. 1, pp. 12–23, Mar. 2013.
ABLAY: ONLINE CONDITION MONITORING OF BATTERY SYSTEMS WITH A NONLINEAR ESTIMATOR 239

[34] S. R. Logan, “The origin and status of the Arrhenius equation,” J. Chem. [37] M. Basseville and I. V. Nikiforov, Detection of Abrupt Changes: Theory
Educ., vol. 59, no. 4, p. 279, Apr. 1982. and Application. Englewood Cliffs, NJ, USA: Prentice-Hall, 1993.
[35] G. Ablay and T. Aldemir, “Observation of the dynamics of nuclear systems
using sliding mode observers,” Nucl. Technol., vol. 174, no. 1, pp. 64–76,
2011.
[36] G. Ablay and T. Aldemir, “Fault detection in nuclear systems using sliding
mode observers,” Nucl. Sci. Eng., vol. 173, no. 1, pp. 82–98, 2013. Author’s photograph and biography not available at the time of publication.