Integration of sampling based battery state of health estimation method in electric vehicles

Integration of sampling based battery state of health estimation method in electric vehicles

Applied Energy 175 (2016) 356–367 Contents lists available at ScienceDirect Applied Energy journal homepage: Integ...

3MB Sizes 0 Downloads 42 Views

Applied Energy 175 (2016) 356–367

Contents lists available at ScienceDirect

Applied Energy journal homepage:

Integration of sampling based battery state of health estimation method in electric vehicles Celil Ozkurt a, Fatih Camci a,⇑, Vepa Atamuradov b, Christopher Odorry a a b

Industrial Engineering Department, Antalya International University, Turkey Computer Engineering Department, Mevlana University, Turkey

h i g h l i g h t s  Presentation of a prototype system with full charge discharge cycling capability.  Presentation of SoH estimation results for systems degraded in the lab.  Discussion of integration alternatives of the presented method in EVs.  Simulation model based on presented SoH estimation for a real EV battery system.  Optimization of number of battery cells to be selected for SoH test.

a r t i c l e

i n f o

Article history: Received 8 February 2016 Received in revised form 18 April 2016 Accepted 5 May 2016

Keywords: State of health estimation Electric vehicles Li-ion batteries System level diagnostics

a b s t r a c t Battery cost is one of the crucial parameters affecting high deployment of Electric Vehicles (EVs) negatively. Accurate State of Health (SoH) estimation plays an important role in reducing the total ownership cost, availability, and safety of the battery avoiding early disposal of the batteries and decreasing unexpected failures. A circuit design for SoH estimation in a battery system that bases on selected battery cells and its integration to EVs are presented in this paper. A prototype microcontroller has been developed and used for accelerated aging tests for a battery system. The data collected in the lab tests have been utilized to simulate a real EV battery system. Results of accelerated aging tests and simulation have been presented in the paper. The paper also discusses identification of the best number of battery cells to be selected for SoH estimation test. In addition, different application options of the presented approach for EV batteries have been discussed in the paper. Ó 2016 Elsevier Ltd. All rights reserved.

1. Introduction Battery cost substantially increases the cost of Electric Vehicles (EVs) and it is one of the major obstacles of widely deployment of EVs [1]. Batteries are consumable products with decreasing energy storage capacity. Energy storage capacity loss (i.e., degradation) occurs due to irreversible chemical reactions, formed passive film, and active material dissolution inside the battery during its usage. State of Charge (SoC) is defined as the ratio of available energy stored in the battery at a given time to its energy storage capacity [2], whereas State of Health (SoH) is defined as the ratio of energy storage capacity at a given time to the energy storage capacity in the beginning of the life [3]. Charge/discharge regime, temperature, and depth of charge/discharge are the main parameters affecting SoH degradation speed [4]. The SoH is defined as the ratio ⇑ Corresponding author. E-mail address: [email protected] (F. Camci). 0306-2619/Ó 2016 Elsevier Ltd. All rights reserved.

of the current energy storage capacity to the initial capacity [5]. SoH identification is a critical process affecting the ownership cost, availability, and safety of the battery. Accurate SoH estimation avoids early disposal of the batteries decreasing the ownership cost and unexpected failures [6,7]. Identification of battery SoH is a challenging problem attracting many studies in recent years. Most of the studies in the literature deal with the cell level SoH estimation methods. However, battery systems used in daily life consist of large number of battery cells. SoH estimation in the system level has not been sufficiently addressed in the literature yet. This paper aims to contribute towards system level battery SoH estimation methodology and its integration within a vehicle. The paper is organized as follows: Literature review is presented in Section 2. Section 3 presents the SoH estimation methodology and discusses its integration alternatives to vehicles. Section 4 reports results of presented methodology applied on accelerated degradation tests performed in the lab environment.

C. Ozkurt et al. / Applied Energy 175 (2016) 356–367

The section also involves simulation results for potential integration of the presented method in a sample vehicle. Finally, Section 5 concludes the paper with future work discussion. 2. Literature review The basic SoH estimation method bases on counting the energy stored and consumed during charge discharge process. This approach is called Coulomb counting (CC) and can be applied in two different ways: CC during test with constant load and CC during daily usage with variable load. The former one requires a discharge process just for the sake of testing. Even though CC during test with controlled constant load may be effective, the time and labor required makes this method unpractical. In the latter one, CC is performed while the system is in use without a distinct test discharge. This approach involves accumulative measurement errors as well as errors due to the dynamic usage profile, even though it is more practical [8,9]. Improvements in the methodology and continuous calibration are needed to identify and correct the errors due to varying usage profile [10–12,8]. SoH estimation studies other than CC can be categorized into two main groups. The first category focuses on identification and modeling of a physical phenomenon that is correlated with SoH and SoC such as Open Current Voltage [13,14], resistance [15], cell skin temperature [16], terminal voltage [17,18], voltage and current curves [3]. In this approach, the relationship between the physical phenomenon and battery SoH is approximated by a model. Equivalent circuit models are examples of modeling the relationship between the internal resistance in a battery and the behavior of the battery in charge and discharge process [19,20]. Ouyang and his colleagues used chemical kinetics to model the battery degradation [21]. The models require measurement of specific parameters to estimate the ones that cannot be measured. Advanced methodologies to measure these parameters may include complex devices such as electrochemical impedance spectroscopy (EIS) equipment. EIS is an advanced technique to identify the effects of chemical reactions in the battery, which may be related to its degradation [22]. The relationship between SoH and EIS measurements at specific frequencies has been proven [23]. The EIS measurement requires special and expensive equipment that increases the cost and complexity of the measurement process [8]. The models used to reflect the relationship between measured parameters and hidden ones (e.g., SoH) are often not satisfactory and cannot clearly reveal SoH effectively. Furthermore, the model may require extensive tuning and calibration leading to practical difficulty. The second category involves development of advanced computational or statistical methods to estimate SoH using the measured parameters. Examples of these methods include Artificial Neural Networks, Fuzzy Logic, Relevance Vector Machines, Particle Filters [24,25], Support Vector Machines [26], Kalman Filters, k-Nearest Neighbor Regression, and Particle Swarm Optimization techniques [1,3,8,26–29]. These methods may integrate the physical models as well as other computational methods to enhance the results. For example, Neural Networks and Kalman Filter have been integrated to remove the dependency to the battery model in [30]. In both categories discussed above, the battery system with multiple battery cells is considered a unique system during charge and discharge. The measurement of physical phenomenon during the SoH test and its processing through computational methods are based on the whole system. A large number of battery cells in a system may lead to discrepancy, inefficiency or impracticality in SoH estimation. Thus, good SoH estimation in the cell level does not guarantee in the system level; a system level analysis for SoH estimation is needed for deployment of SoH estimation methods for large battery systems.


There are studies in the literature regarding the integration of SoH estimation methodology in the EV system. Test procedure development and architecture design for data processing and transfer as well as user interaction have been presented in [31]. Decision making using SoH information and energy management based on SoH information is presented in [32]. Integration of batteries into an EV battery pack has been investigated based on various factors such as packaging, thermal management, assembly and maintenance in [33]. Sampling based testing idea has been presented for system SoH estimation by performing tests on selected battery cells as representative of the whole battery system [34]. Even though a small prototype has been developed to test the presented approach, the cycling tests could have been applied partially due to the cell balancing problems in the battery management system. Thus, the literature lacks a prototype system performing the presented approach using full cycling tests, sufficient degradation test results and analysis of its potential use in the real systems. The contribution of this paper can be summarized as follows:  Presentation of a prototype system with full charge discharge cycling capability.  Presentation of SoH estimation results using the developed prototype system from four battery systems, each of which includes 32 battery cells degraded in the lab environment.  Discussion of integration alternatives of the presented method in EVs.  Development of a simulation model of a real EV battery system that bases on the presented methodology.  Analysis of simulation results in optimization of number of battery cells to be selected for test. 3. Methodology The methodology is discussed in three subsections: A new circuit design for sampled-based SoH estimation technique is presented in the first subsection. Then, integration alternatives of this technique in EVs are given in the second subsection. The third subsection presents a simulation model for a real battery system of a selected EV architecture. 3.1. New circuit design for battery SoH estimation In the traditional approach all the battery cells are connected to a single output as shown in Fig. 1. The output is obtained with connecting N number of column as parallel, which consists of M number of serially connected battery cells. All the battery cells charge, discharge or be idle together. All estimations regarding SoC (State of Charge) and State of Health (SoH) should be performed based on the single behavior of all battery cells. Presented circuit design includes two outputs, called main and test as shown in Fig. 2. Each battery cell (shown in circle in the figure) can either connect to the main output or test output depending on the need. Two relays for each battery cell create the cell’s ability to be connected with the preferred output. C1 and C2 represent the + and  sides of the battery cell, respectively. C1 is connected to either A1 or B1 and C2 is connected to either A2 or B2. A1 and A2 (B1 and B2) represent the + and  sides of the test (main) output, respectively. The first relay connects C1 (C2) to either A1 or B1 (A2 or B2) through moving the contact. If the contact is pulled by the relay, then C1 (C2) is connected to A1 (A2). If the contact is released, then C1 (C2) is connected to B1 (B2). The battery cells connected to the test output are connected parallel. The battery cells connected to the main output are serially connected in any column. If any cell in a column is connected to the test output, then this cell should be bypassed. The second relay


C. Ozkurt et al. / Applied Energy 175 (2016) 356–367

Fig. 1. Circuit design in traditional approach in connecting battery cells.

is used to achieve this by connecting or disconnecting A and B. In other words, if B2 is connected to C2 (cell is connected to the main output), then A and B should be disconnected to avoid the electricity flow. If C2 is connected to A2 (cell is connected to the test output), then the battery cells connected to the main output should bypass this cell by connecting A and B. The connection of each battery cell to the output is independent from others and is controlled by a micro-controller. The battery system works in two phases: normal use and test. In the normal use phase, main output supplies the required energy by

using the required number of battery cells connected to it, whereas the test output is not connected to any of the battery cells and does not supply any energy. In the test phase, the battery cells are divided into two groups each of which is connected to a distinct output (main or test). Each battery group works as an independent battery. The main output is used for operation of the EV, whereas the test output is used for SoH test. In fact, Fig. 3 demonstrates the normal use and test phases. In the normal use phase, single output is active with all cells connected to it. In the test phase, both outputs may be active, working independently from the other. Fig. 4 displays the battery cell connections during normal use and SoH test phases. All the battery cells are connected in the main output during the normal use phase, whereas dashed battery cells are connected to the test output and the rest are is connected to the main output during the test phase. It is true that the presented approach brings new complexities to the system that may reduce the reliability. Having more components in a system will reduce the reliability with a possibility of their failures. One may question the value of the presented approach compared to the other SoH estimation techniques. Note that most, if not all, of the SoH estimation techniques base on either extraction of physical phenomenon related to battery SoH or development of new computational methods. These techniques focus on the battery cell rather than the whole battery system. The presented method is a system level approach not necessarily competing with the other SoH estimation methods. Thus, existing SoH estimation techniques can be applied in the system level using the presented approach. The presented approach gives the flexibility to the existing techniques to focus only on the selected battery cells. For example, SoH estimation technique based on EIS can be applied to a system with the presented circuit design. The presented circuit

Fig. 2. Microcontroller to control main and test outputs.


C. Ozkurt et al. / Applied Energy 175 (2016) 356–367

Fig. 3. Operational phases of the circuit.

design creates an opportunity to the EIS based methodology to focus on the selected battery cells rather than the whole system, which can reduce the testing time and increase practicality. Thus, the presented approach does not compete with many SoH estimation techniques in the literature but collaborate with them to obtain better SoH estimation results from a battery in the system level. The next subsection discusses the integration alternatives of the presented approach to an EV.

3.2. Integration alternatives The integration of the presented methodology is analyzed in three ways: Test type, system architecture, and test speed. The details are discussed below.

Test Type bases on the active or passive status of the main output when the test output is used for SoH testing. There are two types of test: Parallel and Serial. In the parallel test, the main output is active and continues to supply energy for the EV during the SoH test. In other words, SoH test does not require EV to stop. On the other hand, the main output is passive and energy is not supplied through the main output during the serial test. Thus, the test duration is restricted with the passive time of the EV (time in which the EV is not used, e.g. during park) in the serial test. The advantage of the parallel test is the uninterrupted energy supply during the test process removing the test time and duration restriction. The disadvantage involves the load increase on the battery cells connected to the main battery due to the reduced number of battery cells supplying the energy. The overload to the battery cells may lead to faster degradation. In order to avoid the overload, the number of battery cells may be increased by the number of battery cells to be used in SoH test. For example, if the battery system requires 100 battery cells and the number of cells to be selected for the test is 10; then the total number of battery cells in the system is 110. A hundred of them is randomly selected and used in each cycle of the normal use. During the test, 10 cells are selected randomly for the test. Thus, the excess number of battery cells remains idle during the normal use and get in use during the test process. The battery cells that remain idle will be randomly selected in each cycle to achieve similar degradation among all cells. The consequence of the increasing number of cells includes extra cost and weight. Moreover, system architecture has two options: onboard and off-board testing. In the onboard testing, the test equipment and all related mechanisms should be embedded on the vehicle. In the off-board testing, test equipment or some of the related mechanisms are designed as separate devices from the EV. These devices are plugged into the EV when the SoH testing is needed. The advantage of the onboard system is the convenience and easiness of the test as well as its ‘anytime anywhere’ property. However, the onboard system increases the cost and weight of the system. The off-board system could be applied in maintenance service by maintenance operators or at home by the user. The onboard system can apply serial or parallel test. However, the parallel test in off-board system may be meaningless due to the lack of energy need during the test. Thus, off-board test should be only serial. The test speed is the time it takes to complete a SoH measurement. The proposed approach does not focus on a single physical phenomenon (impedance, EIS, temperature) or computational method for SoH. It rather presents a new circuit design enabling

SoH Test Fig. 4. Battery cell connections in operational phases of the circuit.

Normal use


C. Ozkurt et al. / Applied Energy 175 (2016) 356–367

the selection and separation of some battery cells from the others for testing. Thus, any SoH estimation method could be used in the given circuit. If the speed of the test is fast enough, the SoH of each battery cell can be estimated in a given time period one by one. For example, if a SoH test takes 2 s, then the test of all battery cells can be completed in 200 s for a system with 100 battery cells. This enables individual SoH identification for each battery cell. The fast test enables locating the abnormal battery cells in addition to the full representability of the SoH estimation. Even though the fast test may not be possible in today’s technologies, the presented circuit design opens new opportunities for future’s SoH estimation methods as well as balancing technologies. The slow test is based on the analysis of selected battery cells in a given time period. The main disadvantage of the slow test is the lack of full representability of the selected battery cells to the whole battery system. The localized aging effects due to e.g. temperature differences may lead to errors in SoH estimation. The integration options discussed above involves their own advantage, disadvantage and risk. There also exist uncertainties in the future development of the EVs, which will affect the deployment of these integration options. There are external and usage parameters affecting the SoH estimation results during the test such as external temperature, resting time before the test cycle, and depth of discharge. The best way to minimize the effects of these factors is to minimize their variation in different tests. Thus, a test procedure to determine the depth of discharge, resting time, and temperature should be defined.

below SoC threshold, the discharge process is ended by changing the reset parameter. The reset parameter also triggers the change of the capacity value of the battery for the next cycle. SoC and Voltage parameters at time t are calculated using the SOC and capacity at t  1, as well as current and reset parameters using a battery circuit model. Partnership for Next Generation Vehicles (PNGV) model is used as a battery circuit model in the battery cell [35]. This model calculates the voltage based on the SoC, current and changes in the internal parameters. The details of the PNGV circuit model is displayed in Fig. 6, its interaction with other modules in cell model is represented as a box with subtitle ‘PNGV Model’ in Fig. 5. Voltage formulation of PNGV model is given in (1).

V L ¼ OCV  OCV 0

Z ðIL dt þ R0 IL þ Rp Ip Þ



where OCV: Open circuit battery voltage. R0 : Ohmic resistance. Rp : Polarization resistance. IL : Load current. Ip : Polarization resistance. 1=OCV 0 : A capacitance that accounts for the variation in open circuit voltage with the time integral of load current IL . s: Polarization time constant (Rp + R0)  C. These internal parameters (s, OCV0 , Rp, R0, OCV) change with the change of SoC and SoH. These parameters have been learned with

3.3. Simulation model Even if it is possible to implement the presented circuit design for a simple prototype in the lab environment, it is not trivial to create its full system version for a real EV. The circuit model has been developed for a simple battery system with 32 battery cells (8  4) and is discussed in the results section. The applicability of the presented methodology for a real EV can be achieved through a simulation model. This section discusses the details of the simulation model used. Simulation involves two levels: cell and system level modeling. In the cell level modeling, a battery cell has been modeled with two inputs (i.e., current and reset) and three outputs (i.e., capacity, SoC, and Voltage) as shown in Fig. 5. Current parameter represents the amount of current discharged in a given cycle, whereas reset is a binary parameter used to control the end of discharge process. When the State of Charge (SoC) of a battery cell in the system falls

Fig. 5. Battery cell level simulation model.

Fig. 6. PNGV battery model.

C. Ozkurt et al. / Applied Energy 175 (2016) 356–367

the degradation tests performed in the lab with the chance of SoC and SoH, which will be discussed in the next section. Sum of current flow is divided by the current capacity to calculate SoC value. The capacity value changes after the end of each cycle based on a given degradation model. The degradation model is obtained using the aging tests in the lab, which is discussed in the next section. The capacity value at cycle i is calculated based on the capacity value at cycle i  1 and the degradation model. SoH is calculated as the ratio of capacity at cycle i to the initial capacity. A battery system consists of battery cells that are connected in series and parallel. In the simulation model, any battery system can be obtained by connecting the battery cells in parallel and series. Example of a battery system with 4 battery cells, (2 parallel and 2 serial connected) is displayed in Fig. 7. Battery cells in a system may degrade differently due to manufacturing variance, difference of environmental factors such as humidity and heat that the battery cell is exposed to as well as variance on the current applied to battery cells in charge and discharge cycles [36]. The current variance occurs due to the variance in distribution of the current among parallel connected battery cells. The degradation variance among battery cells has been modeled as a random variable defined within the model. Thus, when the system is run, the change of SoC and SoH values involve some level of variance controlled by random variables. The variation in the aging tests


obtained in the lab has been used to characterize these random variables leading to variation in SoC and SoH. 4. Results and discussion This section includes following subsections: (a) Specification of an EV battery system, (b) accelerated aging tests, (c) simulation results. 4.1. Specification of an EV battery system The system design of EV batteries shows great variance based on power requirements, preferences of the OEM, various packing and thermal management techniques. A battery system design that bases on a battery cell (LiFePO4 38120) similar to the one (LiFePO4 14505) used in our lab has been selected to simulate the presented method [37]. The battery system consists of 28 serially connected modules. Each module includes 28 parallel connected battery cells. This battery design is used for an EV. If the battery cell LiFePO4 38120 is replaced with the battery cell LiFePO4 14505, then this battery system can be used for an electric bicycle with 1.5 kW h energy storage capability. The comparison of these two battery designs with two different battery cells is displayed in Table 1. Power specifications of the battery system with battery cell LiFePO4 14505 are displayed in Table 2.

Fig. 7. Battery system consisting of multiple battery cells.


C. Ozkurt et al. / Applied Energy 175 (2016) 356–367

Table 1 Comparison of battery designs with LiFePO4 14505 and 38120 battery cells. Parameters

LiFePO4 14505 600 mA h

LiFePO4 38120 8000 mA h

Capacity Nominal voltage Module capacity with 28 cells Module voltage with 28 cells Maximum charge voltage Normal charge voltage Discharge cut off voltage Charge management Energy stored per battery pack Total number of cells Cell connections Weight (g) Battery pack weight (kg)

600 mA h 3.2 V 16.8 A h 89.6 V 3.8 V per cell 3.60 V per cell 2.0 V minimum CC/CV to 3.6 V per cell 1.5 kW h 784 28 Parallel  28 Serial 19.8 15.05

8000 mA h 3.2 V 224 A h 89.6 V 3.8 V per cell 3.60 V per cell 2.0 V minimum CC/CV to 3.6 V per cell 19.5 kW h 784 28 Parallel  28 Serial 355 278.32

Table 4 The internal parameters obtained through the HPPC test for varying SoH with constant 100% SoC.

Table 2 Power specification of battery system with LiFePO4 14505. Voltage 89.6 V

Capacity 16.8 Am h

Energy (V  Am h) 1505 W h

Driving distance (W h/20 W h km)








75.25 km/h

1 21 41 61 81 101 121 141 161 181 201 221 241 261 281 301 321 341 361 381

13 13 13 16 18 19 26 19 25 16 18 25 17 17 19 17 24 14 13 17

0.002 0.002 0.002 0.002 0.003 0.001 0.002 0.003 0.000 0.004 0.003 0.003 0.002 0.004 0.003 0.006 0.003 0.005 0.006 0.005

0.080 0.074 0.076 0.102 0.106 0.137 0.201 0.139 0.206 0.127 0.152 0.245 0.162 0.145 0.173 0.185 0.272 0.196 0.176 0.224

0.101 0.099 0.095 0.135 0.129 0.133 0.144 0.148 0.163 0.171 0.172 0.188 0.202 0.218 0.247 0.255 0.271 0.265 0.263 0.278

3.337 3.340 3.340 3.331 3.365 3.327 3.300 3.336 3.311 3.326 3.351 3.364 3.302 3.310 3.297 3.317 3.301 3.293 3.288 3.298

1 0.96 0.94 0.81 0.73 0.68 0.61 0.54 0.49 0.43 0.37 0.31 0.25 0.20 0.16 0.14 0.12 0.11 0.10 0.10

Table 3 The internal parameters obtained through the HPPC test for varying SoC with constant 100% SoH. SoC






20 30 40 50 60 70 80 90

10,000 12,000 12,000 13,000 14,000 14,000 17,000 18,000

0.002 0.002 0.001 0.001 0.001 0.001 0.001 0.001

0.109 0.111 0.113 0.112 0.104 0.102 0.134 0.127

0.165 0.163 0.156 0.153 0.155 0.154 0.160 0.160

3.238 3.269 3.285 3.285 3.293 3.304 3.324 3.325

The bold values (SoC and SoH) are controlled parameters during the experiment.

The rest of the study explores application of presented SoH estimation method in such a system given above. A prototype micro-controller has been developed. Accelerated aging tests for individual batteries and battery system using the prototype microcontroller have been performed in the lab environment. Simulation model for the battery system has been developed based on the parameters obtained in the accelerated aging test. Results of the simulation and accelerated tests are discussed with optimization of the sample size for SoH tests.

The bold values (SoC and SoH) are controlled parameters during the experiment.

4.2. Accelerated aging tests Lithium iron phosphate battery cells (i.e., LiFePO4 14505 with 600 mA h capacity) have been used in accelerated aging tests. Accelerated aging tests have been used to learn two types of parameters used in the simulation model: internal parameters in a cell and degradation variance in a system. Internal parameters have been learned using accelerated tests of individual battery cells. The degradation variance has been learned using the system level test. The details of these tests have been discussed below. 4.2.1. Individual degradation tests The individual accelerated tests have been performed to identify the internal parameters of the battery circuit. Nine battery cells have been degraded individually discharging with 600 mA h between 2 V and 3.2 V. 15 and 30 min waiting times have been used between cycles to increase variability in the degradation. The increased variability in degradation has been desired to

Fig. 8. SoH vs R0 change.

represent variation in the internal parameters with the change of SoH and SoC. Battery degradation is affected with many internal and external parameters such as number of cycles, depth of charge

C. Ozkurt et al. / Applied Energy 175 (2016) 356–367

and discharge, temperature, over charge–discharge. Depth of charge and discharge as well as over charge–discharge have been set as constant. The test has been performed in the room temperature with daily temperature variation. Hybrid Pulse Power Characterization (HPPC) test has been performed during the accelerated degradation test in the lab [20]. The internal parameters of the batteries vary based on the SoC ve SoH values. Hence, HPPC test should be performed for different SoC and SoH level. During the degradation process, one charge discharge cycle is reserved for HPPC test in every 20 cycle. In this cycle, HPPC test has been performed in every 10% SoC decrease from 90% SoC to 20% SoC leading to 8 test in a cycle (i.e., 90%, 80%, 70%,. . .20%). Due to the time consumed during the test, the number of tests per cycle has been reduced to four in the later part of the tests (in SoC levels of 100%, 75%, 50%, and 25%). Tables 3 and 4 display the examples of internal parameters


obtained through HPPC test for different SoC levels (with constant SoH of 100%), and different SoH levels (with constant SoC of 100%), respectively. Fig. 8 displays the change of SoH over R0 for different SoC levels, as an example of internal parameters. The parameters obtained through these tests are used in the simulation of the battery system. 4.2.2. System degradation tests The purpose of the system degradation test is to identify the SoH degradation model, which will be used in the simulation. SoH degradation model is expected represent the change of SoH with the charge–discharge cycles including SoH variance among the battery cells. A prototype model has been developed to represent the presented circuit design. A microcontroller controls the connections of 32 battery cells (4 parallel groups with 8 serially connected cells in a group) to the main or test outputs. As

Fig. 9. Experimental setup.

Fig. 10. Microcontroller developed based on the presented design.


C. Ozkurt et al. / Applied Energy 175 (2016) 356–367

Fig. 11. SoH estimation results of four battery systems on prototype microcontroller.

Fig. 12. SoH estimations of all battery systems combined.

discussed in the previous sections, all battery cells are connected to the main output in the normal use phase and battery cells are separated into two groups during the test phase. One group is connected to the test output and the other one is connected to the main output. The battery system has been charged with constant current of 1 A h until 28.8 V. The system is discharged until one of the battery cell reaches to 2 V. It is possible that the battery cells may charge and discharge differently causing balancing problems. Balancing is achieved during charging process. If a cell reaches to 3.6 V earlier than others (being charged faster), its voltage level is kept at 3.6 V and incoming energy is consumed using a resistance. This assures that all battery cells reaches to 3.6 V leading to a balanced system. Fig. 9 displays the experimental setup and Fig. 10 shows the microcontroller prototype developed based on the presented circuit design.

Fig. 13. SoH degradation curve fitting.

All battery cells are connected to the main output during the normal charge–discharge cycles. Accelerated aging test is performed around 1500 charge–discharge cycles. In approximately every 300 cycles, SoH test has been performed. In SoH test, one cell is selected among 8 serially connected cells in a group. Since there are four groups and one cell is selected from each group, four cells are connected to the test output. In a normal process, SoH estimation obtained from this test is expected to represent the SoH of whole battery system. In order to measure the variation in the SoH values of battery cells, the test is repeated eight times for all battery cells in a group. Distribution of SoH values are obtained in every test. The degradation process has been performed for four battery systems. Note that a battery system consists of 32 battery cells. Thus, total of 128 (4  4  8) battery cells have been employed in all degradations. Fig. 11 displays the SoH measurements for four battery systems in y-axis and test performed in

C. Ozkurt et al. / Applied Energy 175 (2016) 356–367

the degradation process in x-axis. Box plot in every test represents the distribution of SoH estimation values. Upper and lower corners of the rectangle show the 25th and 75th percentiles of the SoH distribution. The line in the middle of the box represent the median of the distribution. The vertical lines above and below the box gives the range of data points out of 25th and 75th percentiles that are not outliers. The outliers are shown as single points. A clear progression between in the accelerated tests can be seen in the figure even though there some exist overlap in SoH measurements of some battery cells in different test times. The SoH estimations of four different battery systems in each test have been combined and displayed in Fig. 12. Several curve fitting models have been applied to the SoH estimation results. A 3rd order polynomial model with coefficients (9.6874e11, 2.3074e07, 2.9379e04, 0.9722) is obtained as shown in Fig. 13. In order to evaluate the goodness of fitness of the polynomial model, R-square and Root Mean Square Error (RMSE) have been calculated as 0.9959 and 0.0071, respectively. The figure also shows the confidence intervals obtained using the distribution of SoH estimations. The model obtained is used in generating SoH degradation and variance in SoH values in the simulation model. 4.3. Simulation results The battery system specified in the previous chapter for an e-bike has been simulated and its degradation has been performed based on the data obtained in accelerated aging tests discussed above. The simulation model has been run through charge and discharge cycles until the SoH reaches to 80% level. The charge discharge cycle has been designed in such a way that load for each battery cell becomes similar to the load for each battery cell in the lab experiments. This is important to be able to represent the SoH degradation model obtained in the lab environment. The presented SoH estimation method has been applied using different sample size for the SoH estimation test. The average of the SoH values in the battery group is used as the SoH of the battery group. The number of battery cells to be selected for SoH test is the critical parameter affecting applicability of the presented methodology. Increasing number of battery cells to be selected for SoH test leads to a better SoH estimation since more cells better represent the whole system. The difference between the mean SoH values of all battery cells and mean SoH of the selected battery cells for the test is used to quantify representation capability. Representation capability is highly correlated with the SoH estimation accuracy for the presented method. Increasing representation capability will increase the SoH estimation accuracy. Note that the paper is not focusing on the estimation accuracy of the physical phenomenon or method used during the SoH calculation. Errors in the nature of the SoH calculation related to the physical phenomenon are ignored. This is an acceptable assumption since the same error will occur if the presented circuit design is used or not. There is no single SoH value for the system, since each battery cell in the system has a distinct SoH value. The accuracy of the presented model can be obtained comparing the SoH estimation and the distribution of SoH values of all battery cells in the system. Thus, the term representation capability is used instead of SoH estimation accuracy. The representation capability of a specific test is formulated as the absolute difference between the SoH values of selected battery cells ðSoHsample Þ and the whole system ðSoHsystem Þ as shown in (2). Both SoH values actually are based on a list of SoH values of individual battery cells. The representation capability increases as RC ss approaches to 0.

  RC ss ¼ SoHsystem  SoHsample 



Fig. 14 displays the change of SoH difference between SoH of whole system and selected cells for test with the change in sample size for the SoH test. The test has been repeated for a huge number of times to obtain a distribution for the SoH difference. The dashed line is the average of the SoH difference. As seen from the figure, the difference reduces as the sample size increases, which indicates the increase in representation capability with the sample size. The outer solid lines above and below the dashed line determine the 90% confidence interval for SoH difference. In other words, the SoH difference will be within the solid line in 90% of the all possible test. Note that when the number of sample reaches to 28, perfect representation is achieved since all the battery cells are selected for the SoH test. Even though the increase in sample size increases the representation capability and SoH estimation accuracy; it involves other consequences. In a parallel test, the battery cells connected to the main output will be reduced with the increase in the sample size (i.e., number of the battery cells used for SoH test). This will increase the load on each battery cell leading to faster degradation. If redundant battery cells with the number of battery cells to be used for SoH test are deployed; then the cost of the battery system and its weight will increase. The benefit and consequence should be evaluated together to get the best sample size for SoH test. The cost of the sample size ðC ss Þ is calculated as the difference between cost of battery system with ðC w Þ and without ðC o Þ extra battery cells as shown in (3). Battery cost calculation bases on the project headed by Paul Nelson at Argonne National Laboratory aiming to identify cost of batteries in 2020 [38,39]. Battery costs consist of three parts: price to the OEM (Original Equipment Manufacturer, e.g., Toyota, Ford), thermal management, and battery pack integration [40]. Research and development, material,

Fig. 14. SoH difference between sampled battery cells and whole battery system.

Fig. 15. Cost and SoH difference change with sample size.


C. Ozkurt et al. / Applied Energy 175 (2016) 356–367

Table 5 Analysis of integration alternatives. Alternative no

Test type


Test speed



1 2 3 4 5 6 7 8

Parallel Serial Parallel Serial Parallel Serial Parallel Serial

Onboard Onboard Offboard Offboard Onboard Onboard Offboard Offboard

Slow Slow Slow Slow Fast Fast Fast Fast

Yes Yes No Yes Yes Yes No Yes

16% overloading or 14% extra battery cells Test during parking – Test in service or at home Fast test will open new opportunities

labor, purchased items, sales and administration, profit, warranty and depreciation costs are considered under ‘price to OEM’ cost. Battery management, module control, and packing related costs are presented under pack integration cost. Cooling and heating system costs are considered under thermal management system cost.

C ss ¼ C w  C o


The cost has been quantified as cost difference (Css) between cost with the presented system (Cw) and cost without the presented system (Co) as shown in Eq. (3). Optimization of the sample size is based on the formula that calculates the total cost ðC total Þ of using the presented system as shown in (4). RC ss is the average sampling error of the presented approach and u is the cost of SoH estimation error per unit. If C total is divided by the cost of the whole battery system, the cost percentage (C percentage ) is obtained as in (5). The sample size that gives the minimum C total or C percentage is defined as the optimum sample size for the presented methodology.

C total ¼ C ss þ u  RC ss


C percentage ¼ ðC ss þ u  RC ss Þ=C o


Fig. 15 displays the two terms of the cost percentage and their progression with the number of battery cells to be used for SoH test. u is defined as 1000 as a rough estimation. The identification of this parameter requires further analysis. As seen from the figure, the intersection point of these two lines gives us the optimum sample size, which is 4.35. The total number of battery cells is 28 in a module. We can conclude that the best sample size ratio for the presented sampled SoH estimation approach is 14% (4/28). This result is discussed for integration alternatives explained in Section 3.2. There are eight alternatives that base on three factors as shown in Table 5. The combination of off-board architecture and parallel test does not create any value, since the vehicle should be stable to be able to plug the test equipment. The vehicle will not need energy when it is stable, which will remove the need of parallel test (indicated as No (Alternatives 3 and 7) in reasonability column in the table). Based on the result obtained alternative 1 requires either 14% extra battery cells or 16% overloading to the battery cells connected to the main output during the test. Note that overloading is calculated as the ratio of extra energy required from the cells. If 86 battery cells are handling the energy that is normally supplied by 100 battery cells, (14 of them are used for the test), then energy loading for each battery cell is 1.16 (100/86) leading to 16% overloading. This overloading should be further investigated. The second alternative leads to testing when the EV is not used (during parking). Thus, overloading and redundant battery cells are not needed. If very fast SoH techniques in battery cell level, the presented methodology will create new opportunities. For example, sampling will not be needed for system SoH test since tests will be completed very quickly for all battery cells.

– Test in service or at home

5. Conclusion State of Health (SoH) estimation is important for the effective usage of the batteries that affects the total ownership cost of the battery and the electric vehicle. There are various ways of estimation battery SoH in the literature. Most, if not all, of these methods focus on the battery cell level SoH estimation. This paper focuses on the battery system level SoH estimation. The paper presents a circuit model that randomly selects battery cells and separate them from the main battery system for SoH estimation. A prototype has been developed and used for accelerated aging of a battery system. The prototype controls 32 battery cells (8  4), which is not full representative of real system such as an EV battery. The applicability of the SoH estimation method for EV batteries is discussed in this paper. A simulation model has been developed for a real battery system for an EV. Accelerated aging tests for individual battery cells and battery system have been used to model the parameters for the simulation model. Different potential applications of the presented approach has been discussed. The application that increases the number of cells by the number of cells to be used for the SoH test is selected for further analysis. The number of battery cell samples to be used for SoH estimation is optimized using the benefit and consequence of the presented approach. The benefit has been quantified as the accuracy of the presented method whereas the consequence is quantified as the cost increase ratio with the presented method. The results recommend that the number of battery cells selected for SoH estimation test should 14% of the total number of cells in the system. Analysis of other potential applications of the presented approach and usage of different cycling profiles has been defined as future research. Acknowledgment This research was supported by The Scientific and Technological Research Council of Turkey (TUBITAK) under project number 113M093. References [1] Nuhic A, Terzimehic T, Soczka-Guth T, Buchholz M, Dietmayer K. Health diagnosis and remaining useful life prognostics of lithium-ion batteries using data-driven methods. J Power Sources 2013;239(Oct.):680–8. [2] Joaquin Klee B, Jiahao L, Clemens G. A comparative study and validation of state estimation algorithms for Li-ion batteries in battery management systems. Appl Energy 2015;155:455–62. [3] Chao H, Gaurav J, Puqiang Z. Data-driven method based on particle swarm optimization and k-nearest neighbor regression for estimating capacity of lithium-ion battery. Appl Energy 2014;129:49–55. [4] Fernández IJ, Calvillo CF, Sánchez-Miralles A, Boal J. Capacity fade and aging models for electric batteries and optimal charging strategy for electric vehicles. Energy 2013;60(Oct.):35–43. [5] Kim J, Cho BH. State-of-charge estimation and state-of-health prediction of a Li-ion degraded battery based on an EKF combined with a per-unit system. IEEE Trans Veh Technol 2011;60(9):4249–60. [6] Sun B, Zeng S, Kang R, Pecht MG. Benefits and challenges of system prognostics. IEEE Trans Reliab 2012;61:323–35.

C. Ozkurt et al. / Applied Energy 175 (2016) 356–367 [7] He W, Williard N, Osterman M, Pecht M. Prognostics of lithium-ion batteries based on Dempster–Shafer theory and the Bayesian Monte Carlo method. J Power Sources 2011;196(23):10314–21. [8] Zhang J, Lee J. A review on prognostics and health monitoring of Li-ion battery. J Power Sources 2011;196(15):6007–14. [9] Micea MV, Ungurean L, Cârstoiu GN, Groza V. Online state-of-health assessment for battery management systems. IEEE Trans Instrum Meas 2011;60(6):1997–2006. [10] Codecà F, Savaresi SM, Rizzoni G, Piazza M. On battery state of charge estimation: a new mixed algorithm. In: IEEE inter conf on contrl appl; 2008. p. 102–7. [11] Codecà F, Savaresi SM, Manzoni V. The mix estimation algorithm for battery state-of-charge estimator – analysis of the sensitivity to measurement errors. In: IEEE inter conf on decision and control; 2009. p. 8083–8. [12] Ng K-S, Huang Y-F, Moo C-S, Hsieh Y-C. An enhanced coulomb counting method for estimating state-of-charge and state-of-health of lead-acid batteries. In: INTELEC 2009 – 31st int telecommun energy conf; Oct. 2009. p. 1–5. [13] Lee SJ, Kim JH, Lee JM, Cho BH. The state and parameter estimation of an Li-ion battery using a new OCV-SOC concept. In: 2007 IEEE power electron spec conf; 2007. p. 2799–803. [14] Dong G, Wei J, Zhang C, Chen Z. Online state of charge estimation and open circuit voltage hysteresis modeling of LiFePO4 battery using invariant imbedding method. Appl Energy 2016;162:163–71. [15] Raman SR, Saritha B, John V. Computationally efficient and accurate modeling of Li-ion battery. In: 2013 IEEE innov smart grid technol (ISGT Asia); Nov. 2013. p. 1–6. [16] Dubarry M, Truchot C, Liaw BY. Synthesize battery degradation modes via a diagnostic and prognostic model. J Power Sources 2012;219(Dec.):204–16. [17] Gholizadeh M, Salmasi FR. Estimation of state of charge, unknown nonlinearities, and state of health of a lithium-ion battery based on a comprehensive unobservable model. IEEE Trans Ind Electron 2014;61 (3):1335–44. [18] Chaoui H, Golbon N, Hmouz I, Souissi R, Tahar S. Lyapunov-based adaptive state of charge and state of health estimation for lithium-ion batteries. IEEE Trans Ind Electron 2015;62(3):1610–8. [19] Jin F, Yongling H, Guofu W. Comparison study of equivalent circuit model of Liion battery for electrical vehicles. Res J Appl Sci Eng Technol 2013;6 (20):3756–9. [20] He H, Xiong R, Fan J. Evaluation of lithium-ion battery equivalent circuit models for state of charge estimation by an experimental approach. Energies 2011;4(12):582–98. [21] Ouyang M, Feng X, Han X, Lu L, Li Z, He X. A dynamic capacity degradation model and its applications considering varying load for a large format Li-ion battery. Appl Energy 2016;165:48–59. [22] Zou Y, Lib SE, Shao B, Wang B. State-space model with non-integer order derivatives for lithium-ion battery. Appl Energy 2016;161:330–6. [23] Blanke H, Bohlen O, Buller S, De Doncker RW, Fricke B, Hammouche A, et al. Impedance measurements on lead–acid batteries for state-of-charge, state-of-

[24] [25]


[27] [28]


[30] [31]


[33] [34]



[37] [38]




health and cranking capability prognosis in electric and hybrid electric vehicles. J Power Sources 2005;144(2):418–25. Claudio BM, Marcos OE, et al. Particle-filtering-based estimation of maximum available power state in lithium-ion batteries. Appl Energy 2016;161:349–63. Hu C, Jain G, Tamirisa P, Gorka T. Method for estimating capacity and predicting remaining useful life of lithium-ion battery. Appl Energy 2014;126:182–9. Patil MA, Tagade P, Hariharan SK, Kolake SM, Song T, Yeo T, et al. A novel multistage support vector machine based approach for Li ion battery remaining useful life estimation. Appl Energy 2015;159:285–97. Shahriari M, Farrokhi M. Online state-of-health estimation of VRLA batteries using state of charge. IEEE Trans Ind Electron 2013;60(1):191–202. Xiao-Sheng S. An adaptive prognostic approach via nonlinear degradation modeling: application to battery data. IEEE Trans Ind Electron 2015;62 (8):5082–96. Hu C, Jain G, Zhang P, Schmidt C, Gomadam P, Gorka T. Data-driven method based on particle swarm optimization and k-nearest neighbor regression for estimating capacity of lithium-ion battery. Appl Energy 2014;129(15):49–55. Bai G, Wang P, Hu C, Pecht M. A generic model-free approach for lithium-ion battery health management. Appl Energy 2014;15(135):247–60. Santos VDN, Trovao JP, Branco TP, Goncalves JMR. Information and communication technology solution for the V2G concept implementation. In: IEEE power and propulsion conference (VPPC); 2014. p. 1–6. Ebbesen S, Elbert P, Guzzella L. Battery state-of-health perceptive energy management for hybrid electric vehicles. IEEE Trans Veh Technol 2012;61 (7):2893–900. Saw LH, Ye Y, Tay AAO. Integration issues of lithium-ion battery into electric vehicles battery pack. J Clean Prod 2016;113:1032–45. Camci F, Ozkurt C, Toker O, Atamuradov V. Sampling based state of health estimation methodology for Li-ion batteries. J Power Sources 2015;278:668–74. Long B, Xian W, Jiang L, Liu Z. An improved autoregressive model by particle swarm optimization for prognostics of lithium-ion batteries. Microelectron Reliab 2013;53(6):821–31. Sarasketa-Zabala E, Martinez-Laserna E, Berecibar M, Gandiaga I, RodriguezMartinez LM, Villarreal I. Realistic lifetime prediction approach for Li-ion batteries. Appl Energy 2016;162:839–52. Saw LH, Ye Y, Tay AAO. Electro-thermal analysis and integration issues of lithium ion battery for electric vehicles. Appl Energy 2014;131:97–107. Nelson PA, Gallagher KG, Bloom I, Dees DW. Modeling the performance and cost of lithium-ion batteries for electric-drive vehicles. Argonne Natl Lab Tech Rep; 2012. BatPaC: a lithium-ion battery performance and cost model for electric-drive vehicles; 2012 [Online]. Available: . Dees D, Kevin PN, Gallagher G. PHEV battery cost assessment. Vehicle technologies program annual merit review and peer evaluation meeting; 2012. p. 8.