A Comparative Evaluation on State-of-Charge Estimation Methods for Lithium-Ion Batteries in Electric Vehicles

Transcription

1 Paper A Comparative Evaluation on State-of-Charge Estimation Methods for Lithium-Ion Batteries in Electric Vehicles Fuliang HUANG *1 Masashi MUROHOSHI *1 Akira ICHINOSE *1 Tingting SUI *1 Key Words: BMS, LiB, SOC estimation, OCV, EKF 1. Introduction Motorized automobiles such as electric vehicles (EVs) are becoming ever more common worldwide. France and the UK have announced policies to ban the sale of vehicles powered exclusively by internal combustion engines from Subsequently, India has also launched a policy to shift all new cars sales to EVs by China announced that new energy vehicles (NEVs) should account for at least 10% of the sales quantity by 2019 with a target of 7 million annual NEV sales by 2025 and that the future timing for a complete ban on traditional internal combustion engine vehicles powered by only diesel or gasoline is under review. However, there are various challenges for the popularization of EVs. One such challenge is concerned with lithium-ion batteries (LiBs). LiBs with high energy density and high charge and discharge efficiency are commonly used in battery powered EVs. A larger number of batteries are installed to achieve a driving range comparable to current gasoline vehicles. Furthermore, it is difficult to get an accurate understanding of the battery degradation status while driving (1) and this may cause unforeseen consequences. Therefore, to use the battery power safely and reliably in EVs, the battery management system (BMS) needs to estimate the battery capacity accurately at all times using a high performance battery electronic control unit (ECU) and high precision sensors (2), (3). Currently, LiBs used in EVs use different materials (e.g. cathode materials of lithium nickel cobalt aluminum oxide (NCA), lithium nickel cobalt manganese oxide (NMC), lithium manganese oxide (LMO), lithium iron phosphate (LFP), etc.) and cell formats (e.g. cylindrical, prismatic or pouch cell), resulting in different electrochemical and electrophysical properties. LiB capacity estimation, typically represented by SOC, generally uses the open circuit voltage (OCV) method, the current integration method, the modeling method using a Kalman filter as an example, or a combination of these (4), (5), (6). As is well known, the OCV method, which presumes an approximate linear relationship between OCV and SOC, is most widely used but has drawbacks. It is invalid during battery operation, and polarization relaxation takes a long time (2 hours after disconnection from the load). Therefore, OCV tables and maps need to be prepared in Received 29 June 2018 *1 BMS Development Department, R&D Operations -15-

2 A Comparative Evaluation on State-of-Charge Estimation Methods for Lithium-Ion Batteries in Electric Vehicles advance. Another method widely used is the current integration method which integrates charging and discharging currents for estimating SOC. As the charging and discharging current is accumulated continuously over a long period, the accumulative initial error and measurement error also gradually increase, causing the estimated SOC to diverge from the actual SOC. Therefore, correction is required to minimize the error. The modeling method, for instance, using an equivalent electrical circuit model to investigate electrochemical impedance (including adaptive digital filtering algorithm like a Kalman filter), has a trade-off relationship between the accuracy of the model and the accuracy of state estimation. An increment of model parameters improves the accuracy of the model, but impairs accuracy of state estimation. This means that identifying parameters is complicated and difficult in practical applications. In search of a better way to estimate the state of LiB, we worked out theoretical models of the battery, compared estimation methods, and proposed a fast testing procedure to verify the estimation methods in a short time. This paper focuses on the performance evaluation of algorithms for high accuracy SOC estimation based on the existing OCV, current integration, and modeling methods. The two types of cylindrical LiB (i.e., NCA, and NMC) most commonly used for EVs (hereafter referred to as LiB-A and LiB-B respectively) are introduced and pretested in Chapter 2. Chapter 3 describes the model construction for LiB-A. Chapter 4 describes simulation results for existing algorithms for LiB-A and discusses improvement in using these algorithms through comparative analyses. Subsequently, improvements are applied to LiB-B in Chapter 5 to demonstrate the effectiveness. Chapter 6 briefly summarizes this paper, and suggests the direction for future improvement. Table 1 Specifications of LiB-A and LiB-B LiB-A 2. Battery Characteristics LiB-B Type NCA NMC Nominal Voltage (V) Charging Voltage (V) Discharging Voltage (V) Capacity (mah) Temperature Charge ( C) Discharge ( C) 0 to to 60 0 to to 75 Weight (g) Energy density Volumetric (Wh/l) Gravimetric (Wh/kg) Source: data released on battery manufacturer s official website Table 1 shows specifications of the battery cells, LiB-A and LiB-B, which were used in our evaluation. In order to accurately ascertain the battery cell charge and discharge characteristics, we used a test system configured by the KIKUSUI charge and discharge system controller PFX2512, in combination with the DC power supply PWR400L and electronic load PLZ164W, as shown in Figure 1 (7). The exclusive application software BPChecker3000 supports seamless charge and discharge, and high speed data sampling. The main features of this system are: a maximum voltage of V; a maximum current of A; measurement precision for capacity, voltage and current of 0.1mAh, 0.1mV and 0.1mA respectively; a maximum sampling speed of 1ms; pattern charging DC power supply Electronic load Fig. 1 LAN Charge/Discharge system controller LiB-A or LiB-B (Source: KIKUCHI Charge/Discharge System Controller-System configuration) Test environment for LiB cells -16-

3 and discharging capabilities in steps; and temperature measurement during charging and discharging, thus ensuring accurate testing. The battery cell charge and discharge tests were carried out at 0 C, 25 C and 45 C with new battery cells to confirm OCV-SOC and I-V characteristics. Charging procedure: 1. Set the temperature of the constant temperature chamber and perform constant current (CC) charging at 1C rate from the lower limit of cell reference capacity until the voltage reaches 4.2V. 2. Then, perform constant voltage (CV) charging until the charging current drops to 1/50C rate or less. - SOC at this point of time was considered to have reached full charge capacity (FCC) status and was defined as 100%. Discharging procedure: Set the temperature of the constant temperature chamber and repeat the discharging step of 3 minutes CC discharge and 60 minutes pause at 1C rate. Terminate discharging when the terminal voltage of the battery drops to 2.5V. - The terminal voltage after the 60 minutes pause is regarded as the open circuit voltage. Figures 2 a), b) and c) show the charge characteristics of LiB-A at 0 C, 25 C and 45 C; Figures 2 d), e) and f) show the corresponding discharge characteristics of LiB-A at 0 C, 25 C and 45 C respectively through pretest. Figure 3 shows charge and discharge characteristics of LiB-B at 0 C, 25 C, and 45 C for comparison. The measured capacity of LiB-A using the cathode material NCA was 3.08Ah, compared with a nominal capacity of 3.25Ah. The measured capacity of LiB-B using the cathode material NMC was 2.82Ah, compared with a nominal capacity of 3.00Ah. Therefore, both LiB-A and LiB-B were considered to be qualified for performing a comparative evaluation of SOC estimation methods. Figures 4 a), b) and c) are OCV-SOC lookup tables for LiB-A at 0 C, 25 C and 45 C; Figures 4 d), e) and f) are impedance-soc lookup tables for LiB-A at 0 C, 25 C and 45 C. Figure 5 is OCV- SOC and impedance-soc lookup tables for LiB-B at 0 C, 25 C and 45 C, for comparison. The chargeable capacity of a LiB greatly depends a) b) c) d) e) f) Fig. 2 Charge and discharge characteristic curves of LiB-A at 0 C, 25 C and 45 C -17-

4 A Comparative Evaluation on State-of-Charge Estimation Methods for Lithium-Ion Batteries in Electric Vehicles a) b) c) d) e) f) Fig. 3 Charge and discharge characteristic curves of LiB-B at 0 C, 25 C and 45 C a) b) c) d) e) f) Fig. 4 OCV / impedance and SOC lookup tables for LiB-A at 0 C, 25 C, and 45 C a) b) c) d) e) f) Fig. 5 OCV / impedance and SOC lookup tables for LiB-B at 0 C, 25 C, and 45 C -18-

5 on the environmental temperature during charging. The chargeable capacity around the lowest or highest chargeable temperature may decrease by tens of percent compared with that at nominal voltage charging temperature. Thus, charging at 25 C, discharging at 0 C, re-charging at 25 C, discharging at 45 C were also measured in consideration of the initial FCC. These data were used in the evaluation and classification of battery cells Extended Kalman Filtering (EKF) Method Since the LiB system is nonlinear, a diagram of nonlinear discrete-time state-space considering noise is introduced to use EKF modeling, as shown in Figure 7. Its mathematical expression is formulae (3) and (4). y (k) = V p (k) V (k) (3) SOC = SOC + G y (k) (4) 3. Modeling Modeling was performed using the simplest systematics of a Foster-type equivalent circuit, referring to G.L. Plett s pioneering work (8) and as shown in Fig. 6. Here, R 0 denotes the resistance to mass transfer of lithium ions in the electrolyte; R 1 and C 1 denote resistance to charge transfer on the electrode surface, i.e., solvation and desolvation resistance, and electric double-layer capacitance respectively. The terminal voltage and current are defined as V(t) and I(t) Current Integration Method The current integration method is based on formulae (1) and (2): Here, SOC and SOC denote the prior SOC estimation and updated SOC estimation; V p and V denote the measured voltage and the calculated voltage; y(k) denotes the error between measured voltage and calculated voltage; and G denotes the Kalman gain. Figure 8 shows EKF implementation for SOC estimation on MATLAB R2017b, with reference to the instructions from Adachi, et al (9). Input (Current, Temperature) SOC estimation calculating Kalman gain Measured Voltage + - Q (t) = I (t) dt Q (k) (1) 0t n k=0 Q (k) = (I k + I k+1 ) t/2 (2) Measured Current Fig. 7 Battery Model Calculated Voltage Diagram of nonlinear discrete-time state-space for EKF R 0 R 1 I(t) + OCV C 1 Battery internal impedance V(t) - Fig. 6 Foster-type equivalent circuit model Fig. 8 EKF implementation for SOC estimation on MATLAB -19-

6 A Comparative Evaluation on State-of-Charge Estimation Methods for Lithium-Ion Batteries in Electric Vehicles 4. Simulation on LiB-A Simulation was performed on LiB-A for different algorithms using the current integration method and EKF modeling method for discharge at 0 C, 25 C and 45 C, respectively, in consideration of simulation results shown in Figure 9. The left side of Figure 9 shows simulation results for the current integration method. These results demonstrate that perpetual initial and measurement errors can be considered to be one of the main characteristics of the current integration method. On the other hand, simulation results for the EKF modeling methods shown on the right side of Figure 9 provide highly accurate SOC estimation values where the maximum absolute error is not only below 2%, but also converges in the decreasing direction. Figure 10 proposes a process for a quick SOC estimation. 0 C 0 C 25 C 25 C 45 C 45 C Fig. 9 Simulation results on LiB-A for current integration method (left) and EKF modeling method (right) - at each temperature, upper: transition of SOC true value and SOC estimation value, lower: SOC error in [%] LiB specification OCV test OCV/impedance - SOC Charging: CC(1C)- relationship CV(1/50C cut-off); Discharging: repeat steps CC(1C) 3min pause 60min at 0 C, 25 C, 45 C Ex) equivalent circuit model / parameter identification SOC estimation Current Integration EFK Simulation / Verification Ex) Fig. 10 A proposed process for SOC estimation -20-

7 5. Verification using LiB-B 6. Summary Verification was performed on LiB-B by executing the process described in Figure 10, with different algorithms from the current integration method and the EKF modeling method, for discharge at 0 C, 25 C and 45 C, respectively. Figure 11 shows the verification results which were quite similar to those of LiB-A. Therefore, the verification results themselves do not require detailed explanation. An insight from this verification is that the proposed process for SOC estimation has been proven to have high precision and high efficiency. Firstly, it took a very short time, normally 1 to 2 weeks, to reach the breakpoint test; secondly, the computational complexity was significantly reduced due to application of only known breakpoints to models; and finally, the estimation was reproducible and estimation error was comparable. This study achieved the following. (1) Appropriate and effective pretest procedures were used to ascertain the characteristics of the battery in a short time. Typical charging and discharging data from the measurements provided bases for creating OCV / impedance and SOC relation tables and the models were subsequently identified without backtracking. As mentioned in Chapter 1, each of the methods used for LiB capacity estimation has inherent merits and demerits. The OCV method, the current integration method, and the EKF modeling method were compared by performing experiments in this study. The modeling method using a Kalman filter shows relatively high accuracy and error convergence within a given time period. 0 C 0 C 25 C 25 C 45 C 45 C Fig. 11 Verification results of LiB-B for current integration method (left) and EKF modeling method (right) - at each temperature, upper: transition of SOC true value and SOC estimation value, lower: SOC error in [%] -21-

8 A Comparative Evaluation on State-of-Charge Estimation Methods for Lithium-Ion Batteries in Electric Vehicles (2) A process for SOC estimation was proposed and performed on one type of LiB. The effectiveness of this process was verified on another type of LiB, showing its appropriateness for common use. However, factors that are being studied on an ongoing basis, such as battery degradation, cell balancing that influence SOC estimation and thermal effect, and method such as UKF (Unscented Kalman Filter), an algorithm to simultaneously estimate SOC and parameters are not included in this paper. Future studies will cover the above and focus more on peculiar problems of batteries in EVs, especially with additional comparative analyses using the NEDC (New European Driving Cycle), and the C-WTVC (Chinese version World Transient Vehicle Cycle) test. References (1) Huang, F.L., Sumida, Y., Nomura, A., Matsumura, H., Kamiya, Y., Daisho, Y., Morita. K., Analysis of adverse effects on vehicle performance due to hybrid vehicle battery deterioration, Proceedings of the 26th international electric vehicle symposium (EVS26) (2) Tsuchiya, S., Integrated Battery Management System Combining Cell Voltage Sensor and Leakage Sensor, Keihin Review, Vol. 6, pp , 2017 (3) Tsuchiya, S., Kamata, S., Battery Voltage Detection Device, Japan Patent , November 17, 2016 (Request for Substantive Examination) (4) Hato, Y., Kato, J., Hirota, T., Kamiya, Y., Daisho, Y., Improvement of Open Circuit Voltage Estimation Method for Lithium Ion Battery, Proceedings of JSAE Annual congress (Autumn), 2015 (5) Barreras, J. V., Pinto, C., de Castro, R., Schaltz, E., Swierczynski, M. J., Juhl Andreasen, S., Araujo, R. E., An Improved Parametrization Method for Li-ion Linear Static Equivalent Circuit Battery Models Based on Direct Current Resistance Measurement, International Conference on Sustainable Mobility Applications, Renewables and Technology (SMART), pp. 1-9, IEEE Press. DOI: / SMART (6) Oya, M., Takaba, K., Lin, L., Ishizaki, R., Kawarabayasi, N., Fukui. M., SOC Estimation of Lithium-Ion Batteries Based on Parameter- Dependent State-Space Model, Transactions of the Institute of Systems, Control and Information Engineers, Vol. 29, No. 10, pp , 2016 (7) Charge/Discharge System Controller PFX2512 Catalog (Preliminary), catalog/pdf/files/2012/pfx2512.pdf, accessed Nov (8) Plett, G.L., Extended Kalman filtering for battery management systems of LiPB-based HEV battery packs part 1: Background, part 2: Modeling and identification, part 3: Parameter estimation, Journal of Power Sources 134 (2), pp , 2004 (9) Adachi, S., Hirota, Y. (Eds.), Osiage, K., Baba, A., Maruta, I., Mihara, T., Battery Management Systems Engineering - From Battery Structure to State Estimation, Tokyo Denki University Press,

9 Authors F. HUANG M. MUROHOSHI A. ICHINOSE T. SUI Turning the impossible into a possibility has always been the magic performed by scientists. Day after day, we hope that electric vehicles can accelerate faster, run farther and be used longer. Lithium-ion batteries are particularly important among the core components. Currently it is difficult to grasp correctly and instantly what state it is in. The only way to predict its internal state is to monitor some of its external factors. This process of prediction is not easy but inspires full of imagination and fun. (HUANG) -23-