Journal of Power Sources 405 (2018) 18–29

Contents lists available at ScienceDirect

Journal of Power Sources

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

Review article

Towards a smarter battery management system: A critical review on battery T

state of health monitoring methods
Rui Xiong∗, Linlin Li, Jinpeng Tian
Department of Vehicle Engineering, School of Mechanical Engineering, Beijing Institute of Technology, Beijing, 100081, China

H I GH L IG H T S

• AThedetailed classification of battery SOH estimation methods was presented.

• Deficiencies
strengths and weaknesses of SOH methods were compared and analyzed.
• SOH estimation of the existing research and the improving directions were pointed out.
• A prospect of future
with ultrasonic is expected to add one-dimensional data to batteries.
• SOH management for battery application has been presented.

A R T I C LE I N FO A B S T R A C T

Keywords: To ensure the driving safety and avoid potential failures for electric vehicles, evaluating the health state of the
Electric vehicles battery properly is of significant importance. This study aims to serve as a useful support for researchers and
Lithium-ion battery practitioners by systematically reviewing the available literature on state of health estimation methods. These
State of health methods can be divided into two types: experimental and model-based estimation methods. Experimental
Degradation
methods are conducted in a laboratory environment to analyze battery aging process and provide theoretical
Capacity
support for model-based methods. Based on a battery model, model-based estimation methods identify the
Impedance
parameters, which have certain relationships with battery aging level, to realize state of health estimation. On
the basis of reading extensive literature, methods for determining the health state of the battery are explained in
a deeper way, while their corresponding strengths and weaknesses of these methods are analyzed in this paper.
At the end of the paper, conclusions for these methods and prospects for the development trend of health state
estimation are made.

1. Introduction The most used SOH indicators are battery capacity and internal
resistance, which reflect the energy capability and power capability
In order to cope with the dual pressures of environmental pollution respectively [5,6]. From the study of internal aging mechanisms of
and energy shortage, electric driven vehicles, which have variety of different battery types, the loss of lithium inventory, active material
energy sources, flexible working modes and the potential of energy- decomposition, structural changes are the common reasons for capacity
saving and environmental protection, have become the key direction loss [7–9]. And the increase of resistance is mainly contributed to solid
for the development of the automobile industry. Lithium-ion batteries electrolyte interface (SEI) layer growth. Battery test procedures man-
(LiBs) are one of the commonly used onboard energy sources of the uals define that when the battery capacity decreases to 80% of the in-
electric vehicles (EVs) [1–3]. The capability to store energy and provide itial rated capacity under a specific test protocol, the battery is deemed
a certain power decreases over the battery lifetime because of aging. So not suitable for vehicle application and need to be replaced [10]. But in
the state of health (SOH) as an important metric to assess the battery some cases, the increase of the internal resistance resulting in power
aging state is critical for safe and pleasant driving experience. Once the decrease obviously will lead to battery failure in advance. So con-
health indicator reaches the predetermined limit, the battery should be sidering these two aspects together is important for EVs to run nor-
removed from the energy storage system and EVs [4]. mally. Battery characteristics which change significantly during the

∗
Corresponding author. Department of Vehicle Engineering, School of Mechanical Engineering, Beijing Institute of Technology, No. 5 South Zhongguancun Street,
Haidian District, Beijing, 100081, China.
E-mail address: rxiong@bit.edu.cn (R. Xiong).

https://doi.org/10.1016/j.jpowsour.2018.10.019
Received 4 June 2018; Received in revised form 18 September 2018; Accepted 8 October 2018
0378-7753/ © 2018 Elsevier B.V. All rights reserved.
R. Xiong et al. Journal of Power Sources 405 (2018) 18–29

aging can also indicate the increasing possibility of battery failure, such 3.1. Direct measurement methods
as charging curves, OCV (open circuit voltage) curves and so on
[11–13]. To reveal the aging rule of these health related indicators, a The direct measurements can be used in lots of specific forms of
large number of experiments have been done in the laboratory. Taking implementation and calculations, like: capacity or energy level, AH
the online and simple implementation requirements for EVs application counting, ohmic resistance, impedance, cycle number counting, de-
into account, much work has concentrated on model-based methods structive methods and so on.
using different kinds of estimation techniques.
As an indispensable technique of BMS (Battery Management 3.1.1. Capacity or energy level
System), SOH estimation methods have been extensively studied using The energy capability, which determines the achievable mileage of
various tools and approaches. Unfortunately, the estimation of health- EVs, is defined by the battery capacity. The battery capacity reflects
related characteristics, and therefore, of the SOH, might be a very how much energy can be stored into a fully charged battery, and thus is
challenging task. Because there are many factors like temperature and widely used as SOH indicator. If the present capacity of a battery can be
charge-discharge rates affecting the battery aging process, studying the measured accurately, the SOH can be determined directly. It is the
aging rule under these factors need long time cycling tests and char- easiest and most precise way. There are mature capacity testers in the
acteristic tests. Although most methods were verified sufficiently in the market which can measure capacity accurately. Unfortunately, for a
abundant experimental data, the feasibility and reliability of these running EV, it is difficult to stop and measure the fully charged capa-
methods on real vehicles still require further discussion. The laboratory, city. Therefore, the method would be only usable at fixed environ-
equipped with advanced equipment and technologies, provides a stable ments, such as the laboratory, to calibrate the capacity. In order to solve
environment to collect high precision experimental data [14]. Yet the the problems that the capacity/energy cannot be measured in real-time,
EVs are not well equipped as laboratory, mainly subject to cost. In a number of online estimation, prediction or identification methods for
addition, there are great differences between the simulated working battery capacity have been proposed with the development of the BMS,
conditions in the laboratory and the actual road conditions, which bring which will be further discussed in Section 4.
difficulty for SOH application. There is thereby an urgent need but it is
a significant challenge to improve these methods for industrial appli- 3.1.2. AH counting method
cation. Hence, a conclusion and analysis of these academic achieve- Through the high precision measurement technique, the charge
ments should be made, which would offer researchers convenience and transferred through the battery during full charge-discharge process is
propose the future research direction. Although all kinds of methods accurately counted, so thus the remaining capacity can be simply cal-
have been proposed, only few literature aimed at a comprehensive re- culated. The computation equation of SOH is shown in Eq. (1):
view for methods on estimating SOH [2,4,15]. Combining analysis of
Qmax
the literature and our practical experience, authors in this paper try to SOH = × 100%
Qn (1)
demonstrate these methods in a clear way, with some suggestions for
improving the existing methods and future SOH development. where Qmax is the maximum available capacity of the present condition,
This paper is organized as follows: the classification for the existing Qn is the nominal capacity of the battery.
battery SOH estimation methods is introduced in Section 2. Different To obtain accurate remaining capacity, long-term monitoring and
SOH monitoring methods and their challenges have been systematically memorizing of the battery current is an indispensable process, which
elaborated in Section 3 and 4. In Section 5, comparison and prospects costs lots of time and energy. And the accurate remaining capacity
for SOH estimation methods are made. Ultimately, Section 6 concludes highly depends on the high-precision current sensor to reduce accu-
this paper. mulated error. What's more, in the laboratory the battery is usually fully
charged and discharged with small constant current at ambient tem-
perature 20–25 °C to measure capacity. Obviously it is difficult to satisfy
2. Classification
such conditions in EVs. Researchers always choose this method to verify
the accuracy of capacity estimation results from other methods [4]. In
Different literature has presented different SOH classifications, and
Ref. [16], the authors used the Ah counting for state of charge (SOC)
each has its own characteristics. In order to track these degradation
and SOH estimation. In the proposed method, however, SOH was re-
behaviors for the online management of batteries in BMS, the battery
evaluated in either fully charged or discharged states, which occur
SOH estimation methods in this paper are divided into two categories:
occasionally in EVs. As will be described in Section 4.1.1, AH counting
the experimental methods and model-based estimation methods. There
method plays an important role in equivalent circuit model (ECM) for
are two branches below each major category and each branch contains
SOC and capacity estimation. By combining it with ECM, there is no
several common methods, which are shown in Fig. 1.
need to fully charge or discharge the battery, which is more applicable
The characteristics and challenges of each approach are analyzed
to on-board usage.
below.
3.1.3. Ohmic resistance or impedance measurement
3. Experimental methods The power ability of the battery is strongly correlated to its internal
resistance. The internal resistance defines the voltage drop over the
Experimental methods require a large number of experiments car- battery when the current is applied. To ensure that the battery is op-
ried out in the laboratory to analyze the battery aging behavior. Owing erated in the safety voltage range, the current rate applied on the
to the need of corresponding experimental equipment and the great battery is limited, therefore, the available power. As a commonly used
difference between actual driving conditions and laboratory environ- parameter to indicate the health state of the battery, different techni-
ments, some experimental methods are difficult to achieve on-board. ques to measure resistance have been investigated extensively. For a
But they can be used to study the aging mechanisms, providing theo- LiB, the internal total resistance includes ohmic resistance and polar-
retical basis for model-based methods. ization resistance. While in normal working condition, the ohmic re-
Experimental methods encompass direct measurement methods and sistance contributes mostly to the voltage drop, which is basically lin-
indirect analysis methods. Direct measurement methods use capacity early dependent on the current magnitude. Due to this reason, the
tests, impedance measurements and other tests to measure the battery ohmic resistance can be calculated following the Ohm's law. Fig. 2(a)
health state indicators directly. Indirect analysis methods require data plots a current and voltage profile in a discharge and charge pulse,
analysis and processing to find the SOH related parameters. where t1 and t2 denote two pulse intervals. Similar pulses can be

19
R. Xiong et al. Journal of Power Sources 405 (2018) 18–29

Fig. 1. Classification of battery SOH estimation methods.

pulse.
Because a battery is a complex electrochemical device, the ohmic
resistance is affected by the operation conditions and has relationship
with many parameters like temperature and aging state. Based on this
recognition, deep studies have been conducted to fund the relationships
between the ohmic resistance and the influence factors. The authors in
Ref. [5] used the current pulses methodology to study the dependence
of the ohmic resistance on some parameters, which include temperature
and SOC. It was concluded that temperature had a great influence on
ohmic resistance, while the ohmic resistance had no significant de-
pendency on SOC. Additionally, it should be noted that battery aging
also has impact on the obtained rules in some degree. For example, in
the experimental results of [5], the temperature dependence of the
ohmic resistance remained in its exponential form over the battery
lifetime and this relationship became increasingly obvious with on-
going battery aging.
Another method to measure ohmic resistance is Electrochemical
Impedance Spectroscopy (EIS). The intersection of the spectra and real
axis in a Nyquist plot is the ohmic resistance as can be seen from
Fig. 2(b). EIS is a strong tool in laboratory to study the electrochemical
process inside the battery and it can be used as a diagnostic tool. The
general principle of this method is to measure the characteristic re-
sponse which depends on the cell impedance when a sinusoidal signal is
applied on the cell. Because EIS can measure battery impedance in a
very wide frequency range, besides the ohmic resistance, parameters
like double layer capacitance, SEI resistance and charge transfer re-
sistance can be obtained [9]. The authors in Ref. [18] noticed that
ohmic resistance increased linearly during the aging process from the
experimental results. This regularity was used to determine the health
Fig. 2. (a) Current and voltage profile in a discharge and charge pulse, (b)
state by defining the diagnostic map between ohmic resistance and the
corresponding relationship of EIS and electric elements.
available capacity. Improved ECM is often used to reproduce the im-
pedance spectra, generally using an inductor, a resistance, two parallel
obtained when the EV is in brake or accelerate condition, or in the ZARC elements (a parallel connection of a resistor and a constant-
specific pulse test. The ohmic resistance is obtained as [17]: phase-element) and a Warburg element, as shown in Fig. 2(b) [9].
ΔUt Compared with the simple ECM, the improved ECM can capture more
Ro =
ΔiL (2) complex dynamics of the battery with the sacrifice of computational
efficiency [19].
where ΔUt denotes the pulse voltage, ΔiL denotes the applied current

20
R. Xiong et al. Journal of Power Sources 405 (2018) 18–29

EIS is still a relatively under-utilized method because there are still 3.2. Indirect analysis methods based on the measurements
challenges to use it practically. As is pointed out in Ref. [4], although
the SOH information obtained by EIS analysis is valid, as the battery With the analysis of entire degradation data of batteries, indirect
type changes, there will be greater fluctuations. So, it is necessary to analysis methods to get SOH information can be performed. The in-
combine all possible types of battery to establish calibration platform direct analysis method is a typical multi-step derivation method using
with the use of battery identification algorithm. However, the cost and the health indexes which is associated with the degradation of battery
complexity are not easy to conquer. Moreover, EIS measurement needs capacity or internal resistance. After obtaining the relationship of the
long test time and a stable test environment, not convenient for on- health indicators and capacity or resistance, the battery SOH can be
board application. Considering the auxiliary devices for EIS measure- obtained.
ment and the online application of impedance parameters, other ways
to get impedance parameters are studied. The authors in Refs. [20,21] 3.2.1. Charging curve method
identified some impedance parameters based on the charging and dis- The charging curve can be used to characterize the battery SOH
charging curves, which is free from the constraint of devices. But the because it will change as the battery degradation process. As long as the
proposed method relied on charging and discharging curves at constant amount of data is complete, the method can be used for calculating SOH
current. So easier and more universal method to obtain online EIS accurately.
parameters still needs additional research. Constant current and followed by a constant voltage with current
limit (CCCV) charging mode is widely employed for batteries. A.
Eddahech developed a SOH estimation method using CV step as health
3.1.4. Cycle number counting indicator. In their research [27], they simulated the battery current
The number of cycles of a battery can be observed as the basis of the behavior during CV step by a simple expression as follows:
life model. This is a simple and direct way because only the counter for
numbering the charge and discharge cycles is necessary. It is commonly I (t , Closs) = A (Closs) e−B (C loss) t + C (Closs) (3)
used by the laptop or small electronic products such as nano-satellite where Closs was the capacity loss, A, B and C were parameters which
[22] and phones for their SOH indication of batteries. If the total life had corresponding relationships with the capacity loss. Through the
cycle number of a battery is given by the manufacturer and the current current curve at the CV step, the parameters of the equation could be
cycle number which the battery has experienced is counted or calcu- determined, so thus the corresponding capacity loss was found. Con-
lated, the battery SOH can be calculated. This method primarily de- stant current charging mode without CV period can also be used to
termines the SOH by recording the number of complete discharges calculate SOH. In Refs. [25,28], quantitative analysis of aging me-
(100% depth of discharge). For the case of not fully charging or dis- chanism was done by reconstructing charge voltage curves. The re-
charging operation, the conversion coefficients are usually used to su- constructed charge voltage model contained five parameters which
perimpose and convert the charges and discharges of different depths could be identified by optimal algorithms. The evolution of these
into full charge and discharge. The conversion coefficients can be cal- parameters reflected aging mechanisms quantitatively. In these pro-
culated by experimental test. posed methods, however, the impact of temperature on the charging
curve was not discussed. The estimation accuracy might decrease when
the battery is charged at different temperatures rather than 25 °C. And
3.1.5. Destructive methods these methods were dependent on the particular charge mode, not
They are classical and direct methods for investigating the SOH and usable on fast charging.
clarifying the aging mechanisms of the cells from the angle of micro-
mechanism. Raman spectroscopy [23], X-ray diffraction [24], scanning
3.2.2. ICA and DVA method
electron microscope [25] and so forth, are most used techniques.
Because very little internal information of the battery can be di-
Through the devices, the changes of micro structure during aging
rectly obtained from voltage curves, Dubarry et al. [29,30] used elec-
process can be observed. In Ref. [26], the authors studied the reasons of
trochemical characterization and analysis techniques, incremental ca-
battery aging at storage condition. They pointed out that the leading
pacity analysis (ICA) to process voltage data and get more sensitive
cause of aging at high SOC storage is structure change in the positive
curves. IC curves can be calculated by integrating capacity corre-
electrode, including decomposition of active materials and the change
sponding to small voltage intervals (dQ/dV) through charging or dis-
of stress. However, to analyze the aging mechanisms using these
charging the battery at very small current rate. This process converts
methods, the batteries have to be disassembled, which will permanently
the voltage plateaus of two-phase transition into recognizable IC peaks.
damage the batteries. That is why they are deemed as post-mortem
Another method to get more aging information by processing voltage
methods. Therefore, these methods are only appropriate for laboratory
curves is differential voltage (dV/dQ) analysis (DVA). The distance
research.
between two peaks of the DV curve represents the amount of electricity
participating in the two-phase transition, so it is easier to analyze

Fig. 3. Evolution of the (a) IC curves and (b) DV curves cycled at 40 °C and 50% DOD.

21
R. Xiong et al. Journal of Power Sources 405 (2018) 18–29

capacity fading in a quantitative way using the DV curves [28]. Besides of battery with different aging states at several temperature points. But
getting the IC and DV curves by definition, polynomials can be fitted the charging process before voltage measurement, such as initial SOC,
with the data and then differentiated for IC and DV curves [31]. would have an influence on the estimation accuracy. The authors in
Fig. 3 shows the evolution of the IC and DV curves cycled at 40 °C Ref. [12] measured the stack stress of the battery at different cycling
and 50% DOD and the change trends of the curves have been marked in numbers and noticed that there was a linear relationship between stack
the figure. It can be found that the IC and DV peaks at different aging stress and SOH. The principle of this relationship was analyzed and
states have unique shapes, amplitudes and positions, which is the key attributed to SEI growth. Measuring internal pressure can reveal the
idea for indicating the SOH through the change of the peaks. The au- gases generation caused by the chemical changes inside the battery. The
thors in Refs. [32–35] adopted the position value of the IC peaks as a authors in Ref. [41] found the correlation between internal pressure
signature of battery aging. The local interval between two peak points and capacity fade which would be a useful diagnostic method for da-
of the DV curve was used as SOH indicator in Ref. [36]. In the same mage and SOH prediction.
work, transformation parameters of DV curve were also used to indicate
the battery capacity loss due to the linear relationship between them.
To get more stable results, ICA and DVA can be combined with machine 4. Model-based estimation methods
learning methods. In Ref. [31], the authors used support vector re-
gression to get IC curves and this method was verified using eight cells Although rich degradation information and accurate SOH estima-
with less than 1% error. tion result can be obtained from experimental methods, online, real-
For onboard application of ICA or DVA, some limitations of the time and reliable acquisition of battery health state is more desirable for
method have to be overcome. Firstly, to evaluate the battery health BMS. Capacity, resistance and other parameters can be estimated based
state and ensure an accurate knowing of the aging mechanisms, IC and on a model with adaptive filtering or data-driven algorithms, and then
DV curves are always excited with a very low current rate, like C/25. be used to quantify the degradation of LiBs. The difference between
However, it is very hard to accurately capture such a small current in these two categories lies on the computation procedure. Adaptive al-
practical application. Bigger currents like C/3 [28,37] and C/2 [33] gorithms are commonly using the electrochemical model (EM) and
have been used to analyze the IC behavior, with the compromise of ECM, and the closed-loop control and feedback are indispensable steps.
accuracy. Secondly, traditional methods to obtain IC or DV curves While for the data-driven methods popular are the black box model and
usually require numerical derivation, which brings high computational advanced classification, machine learning, intelligent optimization al-
effort to microcontrollers. Therefore methods with less computation gorithms.
like point counting method [28] and improved center least square It should be noticed that the model-based estimation method is an
method [36] were introduced. Thirdly, to get smoother curves by extension of the indirect method, not simply depending on the real-time
eliminating the noise of the curves, appropriate filtering and smoothing measured data, but to estimate or identify the characteristic parameters
method is of importance. A smoothing method using Gaussian filter was through filtering or intelligent algorithms.
proposed and the noise on the differential curves could be effectively
filtered [33].
4.1. Adaptive filtering methods
3.2.3. Ultrasonic inspection
Ultrasonic inspection is a convenient method to detect all kinds of Adaptive filtering or observer method is popular for battery SOH
minor defects inside the materials. It is an important non-destructive estimation, and is also a progressing kind of method used in the BMS.
technique for quality checks and product maintenance. The application Main implements have been elaborated in the following sections.
of ultrasonic inspection solves some difficulties of internal defect de-
tection without disassembly, such as rail inspection [38] and composite
materials inspection [39]. Recently, the scholars have begun to use 4.1.1. Equivalent circuit model based methods
ultrasonic technique to detect the inner changes during battery aging The electrical model described by a circuit consisting of the basic
[13]. elements, such as resistance, inductor, and capacitor, is called an ECM.
A typical ultrasonic inspection system is composed of pulser, re- Featured with the advantages of less computational burden and easy
ceiver, transducer and a monitor. Receiver can receive the ultrasonic online application, ECM has been widely used for SOC or SOH esti-
waves generated by the pulser. The reflection, refraction and wave mation. Based on a battery model, the adaptive filtering and observer
pattern transformation will be generated at the vital interfaces of the algorithms are good choices to identify the parameters. With the
battery, where degradation processes take place. With these char- identified resistance, OCV and other characterization parameters, the
acteristics, the reflected wave from the defect interfaces can be used for battery SOH can be obtained by lookup table method. The lookup ta-
health diagnosis. As deciphered in Ref. [13], the change of vital inter- bles, obtained through long-term tests in laboratory, define data map-
faces during battery aging can be reflected by the waveform change pings between parameters and health state of the battery [42].
compared with the waveform of new battery. Fig. 4 illustrates a pos- Kalman Filtering is an adaptive method which has received con-
sible combination of ultrasonic inspection and other techniques in a siderable attentions for battery parameter and state estimation. There
BMS. However this method for SOH estimation still needs further study are many derived and improved algorithms based on the KF algorithm,
and improvement. It is hard for end users to know the battery health such as extended KF (EKF) [43], unscented KF (UKF) [44], Particle filter
state through the waveform on the monitor. Comparing the signal (PF) [45,46], adaptive EKF (AEKF) [47], sigma point KF (SPKF) [48],
features with battery health state through machine learning methods etc., to deal with the strong-nonlinear and high computational model.
may be a feasible method, which has been studied in other fields Fig. 5 presents a generic operation framework for identifying and es-
[38,39]. timating the battery parameters through the adaptive filtering or ob-
server algorithms [49]. There are two steps to implement a KF algo-
3.2.4. Other health index methods rithm. In the first step, a prediction state is required, where the filter
There are also many other indirect methods depending on the ex- estimates the current output variable. In the second step, the estimation
perimental degradation data, such as voltage change [40], stack stress is updated in order to obtain a more accurate result and give the esti-
[12], internal pressure [41], etc. In Ref. [40], a linear dependence be- mation a higher certainty. Plett [6] used EKF algorithm to identify cell
tween the OCV obtained after full charging and 30-min rest and re- resistance and capacity by formulating simple models. Taking the cell
maining capacity was found. This method was verified using three types resistance estimation as example, a simple model is formulated:

22
R. Xiong et al. Journal of Power Sources 405 (2018) 18–29

Fig. 4. Illustration of a BMS that uses the ultrasonic-transducer-based SOH assessment technique.

⎧ Rk + 1 = Rk + rk highly used in parameter identification because of their simple im-

⎨ yk = OCV(z k ) − ik R k+ek plementation and less computation. Using offline nonlinear least
⎩ (4)
squares algorithm to identify model parameters is easy to achieve based
where Rk is the cell resistance, rk and ek are process noise and mea- on mathematics software but not feasible for real-time parameter ac-
surement noise, respectively, yk denotes estimated terminal voltage, ik quisition. For online estimation of model parameters with recursive LS
is the load current, OCV is the open circuit voltage, zk represents SOC method, getting a discrete form of the dynamic equations is necessary.
which can be estimated from another estimator. Based on the definition In Ref. [53], the model output equation is rewritten as the discretized
of the state equation, the resistance can be identified using EKF algo- form yk = Φk θk with bilinear transformation method, where yk is the
rithm in real-time. Rather than using a formulated model, in Ref. [50] terminal voltage, Φk is the information matrix and θk is the unknown
parameters of the ECM can be arranged to appear linearly so that an parameter matrix. Compared with other discretized forms in Refs.
adaptive filter method can be applied. The convergence of the proposed [54,55], OCV of the cell can be identified at the same time.
filter was proved by invoking the Lyapunov stability criteria. Considering that the parameters change constantly with the de-
The least squares (LS) [51,52] and its improved algorithm are also gradation and operating conditions, parameters of the model are

Fig. 5. Generic operation framework for identifying or estimating the battery parameters.

23
R. Xiong et al. Journal of Power Sources 405 (2018) 18–29

usually jointly estimated with SOC to improve the reliability and ac- can see that the accurate capacity estimation is not only related to the
curacy of SOC estimation. In Refs. [56,57], a joint estimation method accumulation of electricity in a certain time interval, but also depends
was proposed using H-infinity and UKF, where the parameters were on the accurate SOC estimation. In practical applications, the accu-
updated online by H-infinity. In Refs. [51,52,54,58], LS was combined mulation of electricity can use high-precision, high sampling frequency
with other filter algorithms to realize parameter and state estimation. ammeter to improve accuracy, but the accurate SOC estimation is often
Compared with other filter methods, LS is frequently used on micro- difficult to obtain, because it has direct connection with the available
controllers, with the advantages of its matured application in many capacity of the battery in the ECM.
fields and easy improving. However, for very complicated battery To better track the battery aging process, modified ECM was de-
model with strong nonlinearity, LS may bring inaccurate solutions and veloped which can model the battery with more physical meanings. In
can even cause algorithms to become divergent [59]. Besides model Refs. [68,69], a half-cell model was developed by modeling the positive
parameters, capacity is also important health indicator which directly and negative electrodes separately. Although more complex model will
reflects the aging degree of the battery. The authors in Ref. [60] used increase the computational intensity, it can reflect more aging me-
two separate particle filters to estimate capacity and SOC. For PF, the chanisms compared with traditional ECM.
amount of defined samples has an important effect on calculation speed
and result accuracy, so reasonable amount of samples is crucial for 4.1.2. Electrochemical model based methods
practical application. EM was originally developed by Doyle, Fuller, and Newman
Because the SOH indicators are usually slow-paced time-varying [70,71], which comprised a serials of nonlinear and coupled partial
parameters, it is not necessary to calculate SOH in every moment. If the differential equations. This kind of model was derived from the working
system states and parameters are estimated at the same time scale, this principle of the battery, with the capability to describe the internal
will not only cause the system parameters change too frequently and electrochemical dynamics of the cell. However, the solution difficulty
reduce the stability of the estimation algorithms, but also increase the and high computational pressure hinder its use for online application
computation of the estimation algorithm and reduce the reaction rate of and state estimation. In recent years, many research institutes have
estimation algorithm. Based on the above considerations, researchers contributed their effort on model simplification based on the traditional
used different time scales to calculate parameters and state respectively. Pseudo-Two-Dimensional (P2D) model. A variety of simplification
Multi-scale EKF [61,62], multi-scale PF [63], multi-scale nonlinear models such as single particle model (SPM) with less computation and
predictive filter [64] were developed and got good accuracy. The au- acceptable accuracy have been developed and get widely used on state
thors in Ref. [43] updated SOH offline with a trigger time of the SOH estimation, optimal charge and other aspects [72–76].
estimator when the error of SOC estimation was unacceptable. The principle of using adaptive filter method to estimate health
Instead of identifying capacity directly, another feasible way to state based on EM is the same as the ECM based method that is iden-
obtain battery capacity is developed based on AH counting equation as tifying health-related parameters. Adaptive output-injection observer
shown in Eq. (5) [65–67]: [73], PI observer [77], EKF [78], multi-time-scale observer [79] and PF
[78] were used to estimate battery resistance and capacity. Besides
t2 ηi (τ )
Q= ∫t1 3600
dτ /(z (t2) − z (t1))
(5)
these two frequently used SOH indicators, electrochemical parameters
which have a certain change trend as battery aging were estimated to
where z (t2)、z (t1) represent SOC at time t2、t1 respectively, Q re- indicate health state. The amount of cyclable Li-ions in the electrodes
presents available capacity, η represents the coulomb efficiency. The largely determines the battery capacity, often used for SOH estimation.
computation procedure can be drawn by the Fig. 6. From the Eq. (4) we Adaptive PDE observer [80] and EKF [81] were developed for SOH
estimation using resistance and the amount of cyclable Li-ions as SOH
indicators. However, the methods were verified by one set of data ne-
glecting the influence of battery aging or operation conditions, as a
result, the methods were not verified adequately. Solid state diffusion
coefficient was used as SOH indicator in literature [82,83]. In Ref. [83],
diffusion time, a variant of diffusion coefficient, was identified by off-
line linear least squares and on-line recursive parameter identification
method. They identified the parameter at aged and fresh battery and
found that the parameter varied monotonically with battery aging. So
this parameter was excellent candidates for SOH estimation. EM can be
combined with chemical/mechanical degradation physics models
which reflect specific aging process. In Refs. [84,85], SPM was im-
proved by considering the SEI layer formation, expressed by a re-
lationship of SEI growth and cycling number. Integrating electro-
chemical model with different aging mechanism can reflect the specific
aging reasons in a quantitative manner. But it is really hard to combine
all the aging mechanism together.

4.1.3. Combined methods

Experimental methods can be combined with model-based methods.
First, the aging mechanisms found from the experimental methods
provide new ideas for model-based methods. Loss of both recyclable
lithium ions and active material and the increase in resistance are the
common causes in most graphite-based LiBs. One of the main con-
tribution to loss of lithium inventory is Li-consuming SEI layer forma-
tion analyzed using ICA [29], so the increase of SEI resistance can be
regarded as SOH indicator. The authors in Ref. [86] found the mono-
Fig. 6. Flowchart for the method of estimating the available capacity based on tonic relationship between identified SEI resistance and remaining ca-
the SOC definition. pacity through ECM. The method to estimate remaining capacity was

24
R. Xiong et al. Journal of Power Sources 405 (2018) 18–29

effectively validated using aged and new battery with an error less than terminal voltage errors is the optimal set of model parameters. This
3%. Furthermore, the parameters identified from the model-based method relies high on the OCV-SOC-capacity relationship. For batteries
method can be processed by experimental methods to get the battery whose OCV-SOC curves don't change significantly as the battery aging
health state. In Ref. [35], based on the OCV identified from the ECM, or the SOC-OCV curves are flat, the identification result will have a
the authors used ICA to get the relationship of IC peaks and remaining large deviation. In Ref. [96], GA was applied to identify the parameters
capacity. The use of this kind of methods depends on the design re- of the ECM. Then SOH was determined by the estimated diffusion ca-
quirements and the ability of the processor. pacitance, taking into account the temperature to improve robustness
and estimation accuracy. GA got widely used in parameter identifica-
4.2. Data-driven methods tion of EM [81,97,98] due to the model's strong nonlinearity. The au-
thors in Ref. [99] constructed the degradation trajectories of five
Due to the complex internal principles and uncertain working electrochemical parameters over the battery whole life time by GA
conditions, it is difficult to establish a battery model which can exhibit identification method. Then these five parameters were applied to re-
the battery dynamic characteristics accurately. The data-driven present the battery health state and the results were promising. Besides
methods for SOH prediction do not need the knowledge of battery GA, other optimization algorithms like particle swarm optimization
working principles and an explicit battery model, and they are only were also used for identifying health-related parameters [25,68].
dependent on the collected aging data. However, to obtain rich and Because these algorithms need some time to find the optimal solu-
complete date, the experiments usually take several months and are tion sets, the operation speed of the algorithm brings it challenge for
costly, which are the main weaknesses of these methods. practical application. In the light of rapid development in computers,
online data can be stored and transported to a powerful computer to
4.2.1. Empirical and fitting methods realize fast calculation. And Internet of Vehicles is an emerging field
They use the available degradation data to predict the future be- which has gotten much attention, the cloud computing platform of
havior of LiBs without detailed knowledge of the electrochemical cell which has the great potential to solve this difficulty by the mass data
design and material characteristic. Polynomial, exponential, power law storage and large-scale computing capabilities.
are usually used as fitting models. The computational simplicity of
empirical or fitting models generally enables faster computation. The 4.2.3. Machine learning methods
main aging factors considered in the papers are SOC, ΔSOC, end of The goal of machine learning is to program computers by using
charge and discharge voltage, DOD, temperature, aging time and the example data or past experience to solve a given problem. With the
current. extensive applications of artificial intelligence, machine learning
Capacity loss models under different aging factors were constructed methods, as the core technology, have been studied in various fields
based on extensive experimental work [87–91]. In Ref. [87], Wang such as image recognition [100], financial sector [101], healthcare
et al. developed a unified capacity loss model using the results from a sector [102], etc. It is thus unsurprising that many research teams are
large cycle-test matrix which includes three important parameters, applying it for battery state estimation. Support Vector Machines
temperature (−30–60 °C), DOD (90–10%), and discharge rate (C-rate, [103–105], Gaussian process regression [106], neural networks
ranging from C/2 to 10C). The model is shown in Table 1, where R is [107–109], Markov Chain [110,111], fuzzy logic [112,113], Monte
the gas constant and T is the absolute temperature. Besides empirical Carlo [114,115], etc., are the main used machine learning methods for
capacity model, the scholars in Refs. [90,92–94] developed resistance SOH estimation. These methods usually have a satisfactory result, but
models, which reflect the power capability of the battery. Empirical and the access to training data is a real time consuming and costly process.
fitting models highly depend on the experimental results, while it is Fortunately, the advent of big data platform could hopefully solve this
difficult to guarantee absolute single variable. For example, in calendar problem. Considering that the big data technique has become a hot
aging process at different temperatures, the batteries are put in the topic in vehicle field, the usage of machine learning will have enormous
thermotank at the aging temperature, hence the battery SOC might potential in health monitoring with the huge amounts of data. Once a
have a change because of self-discharge. In addition, the aging factors big data platform is established, real-time monitored parameters like
have a coupled influence on battery aging, it is hard to take all the voltage, current and temperature will be stored and processed in the
factors into consideration to get a fitting result. platform, which also decreases the requirements of the microcontroller.
These mass data could be used as the training data of the machine
learning algorithm which brings researchers lots of conveniences. How
4.2.2. Optimization algorithms
to select and filter data as the training data is important for these al-
They use the intelligent optimization algorithms to identify model
gorithms. Bad parameters selection will bring bigger error and the al-
parameters and then use one or more identified parameters to infer the
gorithms can even not converge.
SOH. Genetic algorithm (GA) is one of the mostly used optimization
method which can estimate the parameters in a nonlinear system ef-
4.2.4. Sample Entropy
fectively. In Ref. [95], after obtaining the SOC-OCV-capacity mapping
The Approximate Entropy (ApEn) and Sample Entropy (SampEn)
relationship, capacity and initial SOC were identified using GA based on
are important tools for quantifying the complexity of time series and
ECM. In the process of model parameter identification, the changes of
studying the features of time series, widely used in many fields espe-
the model parameters are reflected by the errors of the terminal vol-
cially in the research of biomedical signals [116,117]. SampEn is an
tages between the estimation and experimental results. Different com-
improved form of ApEn. SampEn not only has all the benefits of ApEn,
binations of model parameters provide different terminal voltage er-
but also eliminates self-matches and speeds up the computation. And
rors. A set of parameters which can minimize the root mean value of the
SampEn is not sensitive to record length and shows improved con-
sistency compared with ApEn [118]. Accordingly, when it is applied to
Table 1
Generalized capacity loss model for all C-rates.
battery data which have variation during battery aging, it could serve
as an indicator for SOH estimation [119]. HPPC voltage sequence is
Q loss = B⋅exp[(−31700 + 370.3 × C−Rate)/ RT ](Ah )0.55 highly sensitive to battery aging and often chosen to calculate SampEn
Ah = cycle_number × DOD × 2 as SOH indicator. In Ref. [120], the correspondence between capacity
and SampEn from voltage sequence under HPPC profile was fitted by a
C-Rate C/2 2C 6C 10C
B values 31630 21681 12934 15512 polynomial equation. In Ref. [119], SampEn features were calculated
from complete discharge voltage data at constant current, however

25
R. Xiong et al. Journal of Power Sources 405 (2018) 18–29

Table 2
Comparison of the SOH estimation methods.
Method Advantages Disadvantages Improving directions

Direct measurement • Most direct and simplest methods; • Hard for online direct measurement; • Online and easy EIS realization;
methods • Easy combination with model-based • EIS and destructive methods require • Combined with suitable battery model;
methods; corresponding apparatuses; • Develop offline maintenance and diagnosis strategies.
• Accurate in laboratory environment; • Destructive methods will damage the
• EIS and destructive methods can be
used to study aging mechanisms.
battery permanently and cannot be
implemented in real-time.
Indirect analysis • Get aging information from the • Not suitable for all battery types; • Methods to obtain online ICA and DVA curves with less
methods measured external characteristics; • ICA/DVA/charging curve methods need computation complexity;
• ICA/DVA and ultrasonic inspection particular working condition, such as • Applied combining with machine learning methods;
can be used to study aging constant current charge; • Considering the impact of temperature.
mechanisms in real time; • Loss of accuracy due to the changing
• Accurate in laboratory environment. temperature;
• Difficult for online application.
Adaptive filtering • Accurate and good robustness; • Algorithm development requires a lot of • Improve ECM with more physical meanings
methods • Easily applied for online battery experimental validation and debugging; • Improve EM with more aging mechanisms
energy storage with different • Need high performance controller; • Develop a multi-model fusion method to improve
chemistries. • The precision highly depends on the
model accuracy.
applicability.

Data driven method • Less pre-test required; • High requirements on the efficiency and • Combining machine learning, big data mining methods
• Precise estimation of slowly portability of the algorithm; with on-board adaptive filtering algorithms to improve
changing parameters such as battery
life and health.
• High dependence on the magnitude,
sampling frequency, completeness, etc.
the accuracy and robustness of onboard algorithms in
real-time applications.
of the transmitted data.

such condition is rarely used in EVs. To get better and more accurate
results, SampEn can be combined with machine learning method.
SampEn is employed as the data input and the SOH (usually capacity) is
the target vector of learning algorithm. SVM [119], relevance vector
machine (RVM) [119] were combined with SampEn and showed very
good performance. Besides voltage consequence, surface temperature of
the battery can also be used as SampEn feature. In Ref. [121], by the
combination of PF and SampEn calculated from the battery surface
temperature, capacity fading was estimated with small errors.

5. Comparison and prospect

5.1. Comparison and analysis

Experimental methods store the whole degradation data and use

them to analyze the change rules of main parameters during the battery
degradation. It contains direct and indirect analysis. Direct measure-
ment methods are the simplest and most direct methods, so offline
maintenance and diagnosis strategies are suitable to be proposed based
Fig. 7. A prospect for future novel model system. on these methods. The impedance tests and capacity tests can be per-
formed regularly as parts of the reference performance test during the
cycle aging test. Although they are difficult for real BMS application
directly, they are indispensable methods in the laboratory to research
various cell aging processes.
Indirect analysis methods need analysis and process of the measured
external characteristics (usually including voltage, current and tem-
perature). The analysis results are greatly influenced by the battery
types, not applicable for all batteries. In other words, the linear re-
lationship of selected SOH indicators and capacity may be failing when
the batteries with different chemistries are studied. For practical ap-
plication on EVs, the generality of the method should not be neglected,
so validating these methods with different batteries is necessary. Now
the measurable values of batteries are only current, voltage and tem-
perature. To realize multi-dimensional monitoring of batteries, ultra-
sonic inspection is a good development direction.
Adaptive filtering methods evaluate the SOH by identifying para-
meters that are sensitive to the battery health state. They can reduce the
dependency of performance on the battery data through the advanced
Fig. 8. A framework for combining BMS with big data platform. filtering and state estimation methods and are easily applied for bat-
teries with different chemistries. These methods are accurate but make
a higher demand on microcontrollers. Some measures can be taken to

26
R. Xiong et al. Journal of Power Sources 405 (2018) 18–29

improve the computational efficiency and release computational 6. Conclusions

burden. For example, how to choose appropriate calculation step is a
valuable research topic to reach a balance between estimation accuracy This paper systematically summarized the state of art of the present
and calculation amount. Compared with EM, ECM needs less calcula- battery SOH estimation, calculation and management methods.
tion but lacks physical meanings. Considering the different character- Although a large number of SOH estimation methods have been stu-
istics of these two models, combining them together to create totally died, each method has its deficiencies and possibilities for improve-
new model or multi-model may be helpful for SOH estimation. ment. Further research and tests combined with engineering is neces-
Data-driven methods are the popular methods, which require less sary. Through this paper, the authors hope that for anyone with an
pre-test work. The drawbacks of these methods are high requirements interest in SOH estimation can benefit from this work. For engineers,
on the efficiency and portability of the algorithm and high dependence appropriate method can be chosen to estimate the battery health state
on the transmitted data. These methods have great potential for future according to the actual demands. For researchers, we hope some in-
health management system. Big data, cloud computing, cloud storage spirations can be obtained to further improve the SOH estimation
and other emerging technologies will solve the difficulty of data ac- method.
quisition and improve the accuracy and robustness of onboard algo- In future, advanced sensing, big data and data mining, and multi-
rithms in real-time applications. agent decision-making techniques and methods will also be used to
Most of the above methods were concentrating on the SOH esti- determining the SOH for batteries.
mation with single cells, while the batteries in EVs are generally con-
nected in series and in parallel to construct a battery pack. Because of Acknowledgements
the uncertain operating conditions of EVs, it is difficult to ensure the
uniformity of single cell. Thus health state monitoring for a battery pack This work was supported by the National Natural Science
with strong time-varying and non-uniform characteristics still need Foundation of China (Grant No. 51507012).
systematic theories and methods.
The comparison of these four principle approaches is shown in References
Table 2.
[1] M. Zackrisson, K. Fransson, J. Hildenbrand, G. Lampic, C. O'Dwyer, Life cycle
5.2. Prospect of future health management system assessment of lithium-air battery cells, J. Clean. Prod. 135 (2016) 299–311.
[2] L. Tao, J. Ma, Y. Cheng, A. Noktehdan, J. Chong, C. Lu, A review of stochastic
battery models and health management, Renew. Sustain. Energy Rev. 80 (2017)
Because of its generality and online applicability, the future looks 716–732.
bright for the model-based methods both on laboratory and industrial [3] R. Xiong, J. Cao, Q. Yu, H. He, Critical review on the battery state of charge es-
timation methods for electric vehicles, IEEE Access 6 (2018) 1832–1843.
scales. Along with the higher demand for accurately tracking the bat- [4] M. Berecibar, I. Gandiaga, I. Villarreal, N. Omar, J. Van Mierlo, P. Van Den
tery behavior, traditional single model cannot address the need of fu- Bossche, Critical review of state of health estimation methods of Li-ion batteries
ture BMS. Combining different models together will achieve com- for real applications, Renew. Sustain. Energy Rev. 56 (2016) 572–587.
[5] W. Waag, S. Käbitz, D.U. Sauer, Experimental investigation of the lithium-ion
plementary and get better results. As shown in Fig. 7, we propose an battery impedance characteristic at various conditions and aging states and its
example of novel multi-model fusion system, which combines electro- influence on the application, Appl. Energy 102 (2013) 885–897.
chemical mechanism, electrical model, frequency domain analysis and [6] G.L. Plett, Dual and joint EKF for simultaneous SOC and SOH estimation, Proc.
21st Electric Vehicle Symposium (EVS21), 2005, pp. 1–12.
environment perception together. Multi-model will increase the model
[7] M. Wohlfahrt-Mehrens, C. Vogler, J. Garche, Aging mechanisms of lithium
adaption to uncertain environment and different aging levels, so more cathode materials, J. Power Sources 127 (2004) 58–64.
accurate and reliable SOC/SOH/SOP/SOF and RUL (remaining useful [8] R. Fu, S.Y. Choe, V. Agubra, J. Fergus, Modeling of degradation effects considering
lifetime) estimation can be achieved under complicated operation side reactions for a pouch type Li-ion polymer battery with carbon anode, J. Power
Sources 261 (2014) 120–135.
conditions. We believe multi-model will become the development trend [9] Y. Zhang, C.-Y. Wang, Cycle-life characterization of automotive lithium-ion bat-
to ensure the safety of EVs. teries with LiNiO2 cathode, J. Electrochem. Soc. 156 (2009) A527.
In recent years, how to make good use of massive data from EVs has [10] T.Q. Duong, USABC and PNGV Test Procedures, J. Power Sources 89 (2000)
244–248.
been put on the agenda. As shown in Fig. 8, a simple framework of [11] R. Mingant, J. Bernard, V. Sauvant Moynot, A. Delaille, S. Mailley, J.-L. Hognon,
combining BMS with big data platform is proposed. In the daily driving et al., EIS measurements for determining the SoC and SoH of Li-ion batteries, ECS
process of EVs, voltage, current, temperature and other information are Trans 33 (2011) 41–53.
[12] J. Cannarella, C.B. Arnold, State of health and charge measurements in lithium-ion
constantly transferred to a big data platform, which is constructed on batteries using mechanical stress, J. Power Sources 269 (2014) 7–14.
the cloud technology. Based on the rich collected data, machine [13] B. Sood, M. Osterman, M. Pecht, Health monitoring of lithium-ion batteries, 2013
learning methods can be trained in practical environment and get better 10th Annual IEEE Symposium on Product Compliance Engineering (ISPCE), 2013,
pp. 35–40.
prediction results. Methods which need large computation and memory
[14] R. Xiong, Y. Duan, J. Cao, Q. Yu, Battery and ultracapacitor in-the-loop approach
can be accomplished by the cloud computing technology. The data to validate a real-time power management method for an all-climate electric ve-
processing results will be transferred to a battery monitoring center. In hicle, Appl. Energy 217 (2018) 153–165.
[15] A. Farmann, W. Waag, A. Marongiu, D. Uwe, Critical review of on-board capacity
this monitoring center, battery state and fault information in the whole
estimation techniques for lithium- ion batteries in electric and hybrid electric
life time can be recorded and stored. Real-time battery state and vehicles, J. Power Sources 281 (2015) 114–130.
readable fault information will be transferred to EVs to realize remote [16] K.S. Ng, C.S. Moo, Y.P. Chen, Y.C. Hsieh, Enhanced coulomb counting method for
diagnosis and maintenance. estimating state-of-charge and state-of-health of lithium-ion batteries, Appl.
Energy 86 (2009) 1506–1511.
In a word, the future development of battery SOH estimation maybe [17] C. Zhang, J. Liu, S.M. Sharkh, C. Zhang, Identification of dynamic model para-
focus on the following three aspects: First, the introduction of new meters for lithium-ion batteries used in hybrid electric vehicles, Int. Symp. Electr.
sensors to increase the capture of battery external and internal char- Veh. 1 (2009) 1–11.
[18] M. Galeotti, L. Cinà, C. Giammanco, S. Cordiner, A. Di Carlo, Performance analysis
acteristics, such as ultrasound information, is a new perspective. and SOH (state of health) evaluation of lithium polymer batteries through elec-
Second, the battery model is updated and developed to be more suitable trochemical impedance spectroscopy, Energy 89 (2015) 678–686.
for parameter characterization of aging process and multi-model fusion [19] D. Andre, M. Meiler, K. Steiner, H. Walz, T. Soczka-Guth, D.U. Sauer,
Characterization of high-power lithium-ion batteries by electrochemical im-
method is a new research direction to improve the model adaptability. pedance spectroscopy. II: Modelling, J. Power Sources 196 (2011) 5349–5356.
Finally, with the development of big data technique, the convergence of [20] Y. Cui, P. Zuo, C. Du, Y. Gao, J. Yang, X. Cheng, et al., State of health diagnosis
models and data, and the complementary coordination of off-line and model for lithium ion batteries based on real-time impedance and open circuit
voltage parameters identification method, Energy 144 (2018) 647–656.
online methods, will also enrich and improve the performance of SOH [21] R. Mingant, J. Bernard, V. Sauvant-Moynot, Towards onboard li-ion battery state-
estimate.

27
R. Xiong et al. Journal of Power Sources 405 (2018) 18–29

of-health diagnosis by a virtual sensor, World Electr. Veh. J. 5 (2012) 405–411. 113 (2014) 1421–1433.
[22] B. ERW, B.R. Haverkort, M. Jongerden, A score function for optimizing the cycle- [53] H. He, R. Xiong, H. Guo, S. Li, Comparison study on the battery models used for
life of battery-powered embedded, Systems 1 (2015) 305–320. the energy management of batteries in electric vehicles, Energy Convers. Manag.
[23] L.J. Hardwick, H. Buqa, P. Novák, Graphite surface disorder detection using in situ 64 (2012) 113–121.
Raman microscopy, Solid State Ionics 177 (2006) 2801–2806. [54] X. Guo, L. Kang, Y. Yao, Z. Huang, W. Li, Joint estimation of the electric vehicle
[24] K. Maher, R. Yazami, A study of lithium ion batteries cycle aging by thermo- power battery state of charge based on the least squares method and the Kalman
dynamics techniques, J. Power Sources 247 (2014) 527–533. filter algorithm, Energies 9 (2016) 1–16.
[25] M. Ouyang, Z. Chu, L. Lu, J. Li, X. Han, X. Feng, et al., Low temperature aging [55] R. Yang, R. Xiong, H. He, Z. Chen, A fractional-order model-based battery external
mechanism identification and lithium deposition in a large format lithium iron short circuit fault diagnosis approach for all-climate electric vehicles application,
phosphate battery for different charge profiles, J. Power Sources 286 (2015) J. Clean. Prod. 187 (2018) 950–959.
309–320. [56] Q. Yu, R. Xiong, C. Lin, Online estimation of state-of-charge based on the H infinity
[26] J. Li, J. Zhang, X. Zhang, C. Yang, N. Xu, B. Xia, Study of the storage performance and unscented kalman filters for lithium ion batteries, Energy Procedia 105 (2017)
of a Li-ion cell at elevated temperature, Electrochim. Acta 55 (2010) 927–934. 2791–2796.
[27] A. Eddahech, O. Briat, J.M. Vinassa, Determination of lithium-ion battery state-of- [57] Q. Yu, R. Xiong, C. Lin, W. Shen, J. Deng, Lithium-ion battery parameters and
health based on constant-voltage charge phase, J. Power Sources 258 (2014) state-of-charge joint estimation based on H-infinity and unscented Kalman Filters,
218–227. IEEE T. Veh. Technol. 66 (2017) 8693–8701.
[28] X. Han, M. Ouyang, L. Lu, J. Li, Y. Zheng, Z. Li, A comparative study of commercial [58] J. Gao, Y. Zhang, H. He, A real-time joint estimator for model parameters and state
lithium ion battery cycle life in electrical vehicle : aging mechanism identi fi ca- of charge of lithium-ion batteries in electric vehicles, Energies 8 (2015)
tion, J. Power Sources 251 (2014) 38–54. 8594–8612.
[29] M. Dubarry, B.Y. Liaw, Identify capacity fading mechanism in a commercial [59] R. Xiong, H. Chen, C. Wang, F. Sun, Towards a smarter hybrid energy storage
LiFePO4 cell, J. Power Sources 194 (2009) 541–549. system based on battery and ultracapacitor - A critical review on topology and
[30] M. Dubarry, B.Y. Liaw, M.S. Chen, S.S. Chyan, K.C. Han, W.T. Sie, et al., energy management, J. Clean. Prod. 202 (2018) 1228–1240.
Identifying battery aging mechanisms in large format Li ion cells, J. Power Sources [60] S. Schwunk, N. Armbruster, S. Straub, J. Kehl, M. Vetter, Particle filter for state of
196 (2011) 3420–3425. charge and state of health estimation for lithium-iron phosphate batteries, J.
[31] C. Weng, Y. Cui, J. Sun, H. Peng, On-board state of health monitoring of lithium- Power Sources 239 (2013) 705–710.
ion batteries using incremental capacity analysis with support vector regression, J. [61] R. Xiong, F. Sun, Z. Chen, H. He, A data-driven multi-scale extended Kalman fil-
Power Sources 235 (2013) 36–44. tering based parameter and state estimation approach of lithium-ion polymer
[32] C. Weng, J. Sun, H. Peng, A unified open-circuit-voltage model of lithium-ion battery in electric vehicles, Appl. Energy 113 (2014) 463–476.
batteries for state-of-charge estimation and state-of-health monitoring, J. Power [62] C. Hu, B.D. Youn, J. Chung, A multiscale framework with extended Kalman filter
Sources 258 (2014) 228–237. for lithium-ion battery SOC and capacity estimation, Appl. Energy 92 (2012)
[33] Y. Li, M. Abdel-Monem, R. Gopalakrishnan, M. Berecibar, E. Nanini-Maury, 694–704.
N. Omar, et al., A quick on-line state of health estimation method for Li-ion battery [63] M. Ye, H. Guo, R. Xiong, H. Mu, An online model-based battery parameter and
with incremental capacity curves processed by Gaussian filter, J. Power Sources state estimation method using multi-scale dual adaptive particle filters, Energy
373 (2018) 40–53. Procedia 105 (2017) 4549–4554.
[34] C. Weng, X. Feng, J. Sun, H. Peng, State-of-health monitoring of lithium-ion bat- [64] Y. Hua, A. Cordoba-Arenas, N. Warner, G. Rizzoni, A multi time-scale state-of-
tery modules and packs via incremental capacity peak tracking, Appl. Energy 180 charge and state-of-health estimation framework using nonlinear predictive filter
(2016) 360–368. for lithium-ion battery pack with passive balance control, J. Power Sources 280
[35] J. Tian, R. Xiong, S. Member, Q. Yu, S. Member, Fractional order model based (2015) 293–312.
incremental capacity analysis for degradation state recognition of lithium-ion [65] X. Tang, X. Mao, J. Lin, B. Koch, Capacity estimation for Li-ion batteries, Am.
batteries, IEEE Trans Ind. ELetron. 46 (2018). Control Conf. (ACC), 2011, 2011, pp. 947–952.
[36] L. Wang, C. Pan, L. Liu, Y. Cheng, X. Zhao, On-board state of health estimation of [66] J.T.B.A. Kessels, B. Rosca, H.J. Bergveld, P.P.J. Van Den Bosch, On-line battery
LiFePO4battery pack through differential voltage analysis, Appl. Energy 168 identification for electric driving range prediction, 2011 IEEE Veh Power Propuls
(2016) 465–472. Conf VPPC 2011, 2011, pp. 0–5.
[37] X. Han, M. Ouyang, L. Lu, J. Li, A comparative study of commercial lithium ion [67] M. Einhorn, F.V. Conte, C. Kral, J. Fleig, A method for online capacity estimation
battery cycle life in electric vehicle : capacity loss estimation, J. Power Sources of lithium ion battery cells using the state of charge and the transferred charge,
268 (2014) 658–669. 2010 IEEE Int Conf Sustain Energy Technol, 2010, pp. 1–6.
[38] F.L. di Scalea, P. Rizzo, S. Coccia, I. Bartoli, M. Fateh, E. Viola, et al., Non-contact [68] Z. Ma, Z. Wang, R. Xiong, J. Jiang, A mechanism identification model based state-
ultrasonic inspection of rails and signal processing for automatic defect detection of-health diagnosis of lithium-ion batteries for energy storage applications, J.
and classification, Insight Non-Destr. Test Cond. Monit. 47 (2005) 346–353. Clean. Prod. 139 (2018) 379–390.
[39] T. D'Orazio, M. Leo, A. Distante, C. Guaragnella, V. Pianese, G. Cavaccini, [69] M. Dubarry, C. Truchot, B.Y. Liaw, Synthesize battery degradation modes via a
Automatic ultrasonic inspection for internal defect detection in composite mate- diagnostic and prognostic model, J. Power Sources 219 (2012) 204–216.
rials, NDT E Int. 41 (2008) 145–154. [70] M. Doyle, T.F. Fuller, J. Newman, Modeling of galvanostatic charge and discharge
[40] I. Baghdadi, O. Briat, P. Gyan, J. Michel, Electrochimica Acta State of health as- of the lithium/polymer/insertion cel, J. Electrochem. Soc. 140 (1993) 1526–1533.
sessment for lithium batteries based on voltage – time relaxation measure, [71] T.F. Fuller, M. Doyle, J. Newman, Simulation and optimization of the dual lithium
Electrochim. Acta 194 (2016) 461–472. ion insertion cell, J. Electrochem. Soc. 141 (1994) 1–10.
[41] A. Matasso, Identification of the Interdepedance of Aging and Internal Pressure [72] S.K. Rahimian, S. Rayman, R.E. White, Extension of physics-based single particle
Evolution of Commercial Li-ion Cells, PhD Dissertations University of Texas, 2014. model for higher charge-discharge rates, J. Power Sources 224 (2013) 180–194.
[42] Z. Chen, Q. Wang, The application of UKF algorithm for 18650-type lithium bat- [73] N. Lotfi, J. Li, R.G. Landers, J. Park, Li-ion battery state of health estimation based
tery SOH estimation, Appl. Mech. Mater. 519–520 (2014) 1079–1084. on an improved single particle model, 2017 Am Control Conf, 2017, pp. 86–91.
[43] Y. Zou, X. Hu, H. Ma, S.E. Li, Combined State of Charge and State of Health es- [74] X. Han, M. Ouyang, L. Lu, J. Li, Simplification of physics-based electrochemical
timation over lithium-ion battery cell cycle lifespan for electric vehicles, J. Power model for lithium ion battery on electric vehicle . Part II : Pseudo-two-dimensional
Sources 273 (2015) 793–803. model simplification and state of charge estimation, J. Power Sources 278 (2015)
[44] Wang Qiuting, Lu Yunhao, JY, State of health estimation for lithium-ion battery 814–825.
based on D-UKF, Int. J. Hybrid Inf. Technol. 8 (2015) 16. [75] K.D. Stetzel, L.L. Aldrich, M.S. Trimboli, G.L. Plett, Electrochemical state and in-
[45] H. Dong, X. Jin, Y. Lou, C. Wang, Lithium-ion battery state of health monitoring ternal variables estimation using a reduced-order physics-based model of a li-
and remaining useful life prediction based on support vector regression-particle thium-ion cell and an extended Kalman filter, J. Power Sources 278 (2015)
filter, J. Power Sources 271 (2014) 114–123. 490–505.
[46] R. Xiong, Y. Zhang, H. He, X. Zhou, M.G. Pecht, A double-scale, particle-filtering, [76] S.K. Rahimian, S. Rayman, R.E. White, Optimal charge rates for a lithium ion cell,
energy state prediction algorithm for lithium-ion batteries, IEEE Trans. Ind. J. Power Sources 196 (2011) 10297–10304.
Electron. 65 (2017) 1526–1538. [77] L. Zheng, L. Zhang, J. Zhu, G. Wang, J. Jiang, Co-estimation of state-of-charge,
[47] H. He, R. Xiong, H. Guo, Online estimation of model parameters and state-of- capacity and resistance for lithium-ion batteries based on a high-fidelity electro-
charge of LiFePO4batteries in electric vehicles, Appl. Energy 89 (2012) 413–420. chemical model, Appl. Energy 180 (2016) 424–434.
[48] G.L. Plett, Sigma-point Kalman filtering for battery management systems of LiPB- [78] A. Bartlett, J. Marcicki, S. Onori, G. Rizzoni, Xiao Guang Yang, T. Miller, Model-
based HEV battery packs Part 2: Simultaneous state and parameter estimation, 161 based state of charge estimation and observability analysis of a composite elec-
(2006) 1369–1384. trode lithium-ion battery, 52nd IEEE Conf Decis Control, 24 2016, pp. 7791–7796.
[49] R. Xiong, Q.Q. Yu, L.Y. Wang, C. Lin, A novel method to obtain the open circuit [79] C. Zou, C. Manzie, D. Ne, Multi-time-scale observer design for state-of-charge and
voltage for the state of charge of lithium ion batteries in electric vehicles by using state-of-health of a lithium-ion battery, J. Power Sources 335 (2016) 121–130.
H infinity filter, Appl. Energy 207 (2017) 346–353. [80] S.J. Moura, M. Krstic, N.A. Chaturvedi, Adaptive PDE observer for battery SOC/
[50] Y.H. Chiang, W.Y. Sean, J.C. Ke, Online estimation of internal resistance and open- SOH estimation, ASME 2012, Dynamic Systems and Control Conference joint with
circuit voltage of lithium-ion batteries in electric vehicles, J. Power Sources 196 the JSME 2012 11th Motion and Vibration Conference, 2012, pp. 101–110.
(2011) 3921–3932. [81] X. Zhou, J.L. Stein, T. Ersal, Control Engineering Practice Battery state of health
[51] J. Remmlinger, M. Buchholz, M. Meiler, P. Bernreuter, K. Dietmayer, State-of- monitoring by estimation of the number of cyclable, Contr. Eng. Pract. 66 (2017)
health monitoring of lithium-ion batteries in electric vehicles by on-board inter- 51–63.
nalresistance estimation, J. Power Sources 196 (2011) 5357–5363. [82] S. Dey, B. Ayalew, P. Pisu, Combined estimation of state-of-charge and state-of-
[52] R. Xiong, F. Sun, X. Gong, C. Gao, A data-driven based adaptive state of charge health of Li-ion battery cells using SMO on electrochemical model, 13th IEEE
estimator of lithium-ion polymer battery used in electric vehicles, Appl. Energy International Workshop on Variable Structure Systems (VSS), 2014, pp. 1–6.

28
R. Xiong et al. Journal of Power Sources 405 (2018) 18–29

[83] G.K. Prasad, C.D. Rahn, Model based identification of aging parameters in lithium Conference (IEEE IEMCON), 2016, pp. 1–6.
ion batteries, J. Power Sources 232 (2013) 79–85. [102] R.C. Deo, Basic Science for Clinicians, Circulation 132 (2015) 1920–1930.
[84] J. Li, K. Adewuyi, N. Lotfi, R.G. Landers, J. Park, A single particle model with [103] D. Andre, C. Appel, T. Soczka-Guth, D. Uwe, Advanced mathematical methods of
chemical/mechanical degradation physics for lithium ion battery State of Health SOC and SOH estimation for lithium-ion batteries, J. Power Sources 224 (2013)
(SOH) estimation, Appl. Energy 212 (2018) 1178–1190. 20–27.
[85] R. Deshpande, M. Verbrugge, Y.-T. Cheng, J. Wang, P. Liu, Battery cycle life [104] V. Klass, M. Behm, G. Lindbergh, A support vector machine-based state-of-health
prediction with coupled chemical degradation and fatigue mechanics, J. estimation method for lithium-ion batteries under electric vehicle operation, J.
Electrochem. Soc. 159 (2012) A1730–A1738. Power Sources 270 (2014) 262–272.
[86] R. Xiong, J. Tian, H. Mu, C. Wang, A systematic model-based degradation behavior [105] A. Nuhic, T. Terzimehic, T. Soczka-Guth, M. Buchholz, K. Dietmayer, Health di-
recognition and health monitoring method for lithium-ion batteries, Appl. Energy agnosis and remaining useful life prognostics of lithium-ion batteries using data-
207 (2017) 372–383. driven methods, J. Power Sources 239 (2013) 680–688.
[87] J. Wang, P. Liu, J. Hicks-Garner, E. Sherman, S. Soukiazian, M. Verbrugge, et al., [106] Z. Wang, J. Ma, L. Zhang, State-of-Health estimation for lithium-ion batteries
Cycle-life model for graphite-LiFePO4 cells, J. Power Sources 196 (2011) based on the multi-island genetic algorithm and the Gaussian process regression,
3942–3948. IEEE Access 5 (2017) 21286–21295.
[88] M. Petit, E. Prada, V. Sauvant-Moynot, Development of an empirical aging model [107] A. Eddahech, O. Briat, N. Bertrand, J.Y. Delétage, J.M. Vinassa, Behavior and
for Li-ion batteries and application to assess the impact of Vehicle-to-Grid stra- state-of-health monitoring of Li-ion batteries using impedance spectroscopy and
tegies on battery lifetime, Appl. Energy 172 (2016) 398–407. recurrent neural networks, Int. J. Electr. Power Energy Syst. 42 (2012) 487–494.
[89] N. Omar, M.A. Monem, Y. Firouz, J. Salminen, J. Smekens, O. Hegazy, et al., [108] G. Bai, P. Wang, C. Hu, M. Pecht, A generic model-free approach for lithium-ion
Lithium iron phosphate based battery - assessment of the aging parameters and battery health management, Appl. Energy 135 (2014) 247–260.
development of cycle life model, Appl. Energy 113 (2014) 1575–1585. [109] You G. won, S. Park, D. Oh, Real-time state-of-health estimation for electric ve-
[90] M. Ecker, J.B. Gerschler, J. Vogel, S. Käbitz, F. Hust, P. Dechent, et al., hicle batteries: a data-driven approach, Appl. Energy 176 (2016) 92–103.
Development of a lifetime prediction model for lithium-ion batteries based on [110] M. Zhu, S. Member, W. Hu, N.C. Kar, S. Member, The SOH estimation of LiFePO4
extended accelerated aging test data, J. Power Sources 215 (2012) 248–257. battery bbased on internal resistance with Grey Markov chain, 2016 IEEE
[91] Z. Li, L. Lu, M. Ouyang, Y. Xiao, Modeling the capacity degradation of LiFePO4/ Transportation Electrification Conference and Expo (ITEC), 2016, pp. 1–6.
graphite batteries based on stress coupling analysis, J. Power Sources 196 (2011) [111] R. Xiong, J. Cao, Q. Yu, Reinforcement learning-based real-time power manage-
9757–9766. ment for hybrid energy storage system in the plug-in hybrid electric vehicle, Appl.
[92] R.B. Wright, C.G. Motloch, J.R. Belt, J.P. Christophersen, C.D. Ho, Energy 211 (2018) 538–548.
R.A. Richardson, et al., Calendar- and cycle-life studies of advanced technology [112] P. Singh, R. Vinjamuri, X. Wang, D. Reisner, Fuzzy logic modeling of EIS mea-
development program generation 1 lithium-ion batteries, J. Power Sources 110 surements on lithium-ion batteries, Electrochim. Acta 51 (2006) 1673–1679.
(2002) 445–470. [113] A. Zenati, P. Desprez, H. Razik, A. Daney, U. Lyon De, U. Lyon, et al., Estimation of
[93] J. Schmalstieg, S. Käbitz, M. Ecker, D.U. Sauer, A holistic aging model for Li the SOC and the SOH of Li-ion Batteries, by Combining Impedance Measurements
(NiMnCo)O2 based 18650 lithium-ion batteries, J. Power Sources 257 (2014) with the Fuzzy Logic Inference, 36th Annual Conference of the IEEE Industrial-
325–334. Electronics-Society (IECON), 2010, pp. 1773–1778.
[94] I. Baghdadi, O. Briat, J.Y. Delétage, P. Gyan, J.M. Vinassa, Lithium battery aging [114] R. Xiong, Y. Zhang, J. Wang, H. He, S. Peng, M. Pecht, Lithium-ion battery health
model based on Dakin's degradation approach, J. Power Sources 325 (2016) prognosis based on a real battery management system used in electric vehicles,
273–285. IEEE T. Veh. Technol. (2018), https://doi.org/10.1109/TVT.2018.2864688.
[95] R. Yang, R. Xiong, H. He, H. Mu, C. Wang, A novel method on estimating the [115] Y. Zhang, R. Xiong, H. He, M. Pecht, Lithium-ion battery remaining useful life
degradation and state of charge of lithium-ion batteries used for electrical ve- prediction with Box-Cox transformation and Monte Carlo simulation, IEEE Trans.
hicles, Appl. Energy 207 (2017) 336–345. Ind. Electron. 46 (2018).
[96] Z. Chen, C.C. Mi, Y. Fu, J. Xu, X. Gong, Online battery state of health estimation [116] W. Chen, J. Zhuang, W. Yu, Z. Wang, Measuring complexity using FuzzyEn , ApEn
based on Genetic Algorithm for electric and hybrid vehicle applications, J. Power , and SampEn, Med. Eng. Phys. 31 (2009) 61–68.
Sources 240 (2013) 184–192. [117] Y. Kumar, M.L. Dewal, Complexity measures for normal and epileptic EEG signals
[97] J.C. Forman, S.J. Moura, J.L. Stein, H.K. Fathy, Genetic identification and Fisher using ApEn, Int. J. Comput. Commun. Technol. 2 (2011) 6–12.
identifiability analysis of the Doyle-Fuller-Newman model from experimental cy- [118] J.S. Richman, J. Randall Moorman, Physiological time-series analysis using ap-
cling of a LiFePO4 cell, J. Power Sources 210 (2012) 263–275. proximate entropy and sample entropy, Am. J. Physiol. Hear Circ. Physiol. 278
[98] L. Zhang, L. Wang, C. Lyu, J. Li, J. Zheng, Non-destructive analysis of degradation (2000) H2039–H2049.
mechanisms in cycle-aged graphite/LiCoO2 batteries, Energies 7 (2014) [119] A. Widodo, M.C. Shim, W. Caesarendra, B.S. Yang, Intelligent prognostics for
6282–6305. battery health monitoring based on sample entropy, Expert Syst. Appl. 38 (2011)
[99] R. Xiong, L. Li, Z. Li, Q. Yu, H. Mu, An electrochemical model based degradation 11763–11769.
state identification method of Lithium-ion battery for all-climate electric vehicles [120] H. Han, H. Xu, Z. Yuan, Y. Shen, A new SOH prediction model for lithium-ion
application, Appl. Energy 219 (2018) 264–275. battery for electric vehicles, 2014 17th Int Conf Electr Mach Syst, 2014, pp.
[100] M. Kallas, P. Honeine, C. Richard, C. Francis, H. Amoud, Non-negative pre-image 997–1002.
in machine learning for pattern recognition, 19th European Signal Processing [121] J. Li, C. Lyu, L. Wang, L. Zhang, C. Li, Remaining capacity estimation of Li-ion
Conference (EUSIPCO 2011), 2011, pp. 931–935. batteries based on temperature sample entropy and particle filter, J. Power
[101] S. Birla, K. Kohli, A. Dutta, Machine learning on imbalanced data in credit risk, 7th Sources 268 (2014) 895–903.
IEEE Annual Information Technology, Electronics and Mobile Communication

29