European Electric Vehicle Congress Brussels, Belgium, 3rd – 5th December 2014
Battery Pack Modelling from the Perspective of Battery Management Systems Jubin Jacob1, Bogdan Rosca1, Steven Wilkins1,2 1
TNO Mobility and Logistics, Steenovenweg 1, 5708HN Helmond, Netherlands EPE Group, Electrical Engineering Department, Technical University of Eindhoven, Netherlands Jubin Jacob (corresponding author)
[email protected]
2
Abstract Battery Management Systems (BMS) have an essential role in hybrid and electric vehicles, contributing to the safety, performance and efficiency of the vehicle. In the years to come, BMS’s will increasingly rely on state estimation to increase the performance of their existing functionalities (such as State of Charge/State of Health estimation and cell balancing). At the same time, innovative new functionality, such as battery ageing mitigation techniques, are made possible by state estimation techniques. To efficiently develop these control strategies, model based development techniques are employed. The core of these is a battery pack simulation tool, allowing fast and efficient scale up from validated cell models to a system level simulation studying the interaction occurring within battery packs. This tool makes various applications possible, such as battery pack sizing analysis, state estimation development, cell balancing system functional specifications analysis, or BMS functionality development. This paper will present the battery pack simulation tool and illustrate some of its applications. Keywords: Battery model, Battery Management System (BMS), Lithium battery, EV (electric vehicle), HEV (hybrid electric vehicle)
1
Introduction
The optimisation of the design of battery modules and packs is of critical importance to the development of hybrid electric or fully electric vehicles. Various aspects need to be taken into account, ranging from electrical to thermal characteristics. A large amount of research has been completed into the various domains, although it is often difficult to combine the various approaches. There is often a need for a tool to combine cell assessment with module/pack design. Battery packs consist of cells connected in parallel and series to result in a high voltage and capacity pack. When connected in series, the
EEVC European Electric Vehicle Congress
pack voltage increases, while the parallel connection helps achieve higher capacity.
Figure 1. Battery pack topology example [1]
1
Theoretically, batteries can be connected in different ways: 1.
2.
3. 4.
Cells connected in parallel forming modules, which are then connected in series to form the pack (Figure 1) Cells connected in series forming modules which are then connected in parallel to form the pack Cells connected in parallel to form a pack Cells connected in series to form a pack
In automotive applications, series connections are required to ensure that high operational voltages can be attained starting from cells (with nominal cell voltages in the region of 3V to 4 V, depending on battery chemistry). With this in mind, the battery pack simulator presented in this paper has the structure shown in Figure 1, where the number of cells forming a module, as well as the number of modules forming the battery pack, are tuneable parameters. The structure, identification and validation of cell models that form the basis of the pack level simulation tool are described in the next section.
2
Battery Cell Models
A battery cell model that is sufficiently representative of the dynamic behaviour of a battery cell is a key requirement for a battery pack level simulator. The structure of the battery model, the experimental identification of the parameters of this model and the experimental validation of the model is described in the subsequent sections.
2.1
Figure 2. Electric circuit equivalent battery model
The description of these influences can be made using various mathematical functions fitted to measurement data, as well as look-up tables derived from standard battery tests, as described in [5].
2.2
Model Identification
For the results presented in this paper, data from a commercially available 18650 Molicel Li-ion battery is used. It consists of a LiCoO2 cathode and a graphitic carbon anode. Some of the important specifications are listed in Table 1. A series of tests were performed to identify the parameters of the model presented in Figure 2 with 2 RC blocks. Table 1: Battery cell specifications
Typical capacity Minimum capacity Nominal voltage Maximum discharge current Maximum charge current Discharge cut-off voltage
1950 mAh 1850 mAh 3.7 V 4A 2A 3V
Model Structure
Various battery chemistries are currently used or considered for use in the automotive domain, which require different model parameters or even structures. Typically, the structure depicted in Figure 2 is used to represent the battery behaviour in the electrical domain [2] [3]. As such, these validated cell models are at the core of the pack-level simulation approach presented in this paper. The structure depicted above remains quite generic and many subsets of it can be used. Depending on the specific battery chemistry and modelling accuracy desired, a different number of RC blocks can be used. Similarly, influences of State-of-Charge (SoC), temperature, current or even ageing can be included in the model parameters [4].
EEVC European Electric Vehicle Congress
Figure 3. Voltage as a function of charge/discharge
The open circuit voltage (OCV), represented as Uoc in Figure 2, is obtained as the average between the voltage measurements during a 0.05C discharge and a 0.05C charge. The procedure is shown in
2
Figure 3, where the voltages are plotted against State of Charge. Determination of OCV is an especially complex process, particularly due to the effect of OCV hysteresis [15]. It is assumed that the voltage during charge and discharge at a low C-rate of 0.05C contains almost negligible contribution of the voltage drop due to the internal resistance. Thus, for the present purpose, it is assumed that the average of the voltage during 0.05C charge and discharge is sufficiently representative of the OCV. The impedance behaviour of the battery was identified based on a test profile derived from Hybrid-Pulse Power Capability (HPPC) test [6]. This test allows the identification of the model parameters describing impedance; namely R0, R1, C1, R2 and C2. Figure 4(a) shows the current profile which was applied to the battery, while the corresponding measured voltage response is visible in Figure 4(b). Figures 4(c) and 4(d) show a closer view of one of the current pulses and the corresponding voltage response respectively. A set of values for the impedance parameters are obtained for each step and the corresponding relaxation period (the zero current phase following the current step). The test is designed such that the impact of charging and discharging, as well as C-rate, are investigated. This is achieved by using current pulses of varying amplitude and direction.
2.3
Model Validation
For validation, a battery current profile corresponding to an New European Driving Cycle (NEDC) electric vehicle test is used [7]. The validation is performed by comparing the measured voltage of the cell undergoing a current load profile and comparing it with the predicted voltage from the model, under the same current load profile. The results are shown in Figure 5. It can be seen that the voltage estimation error is well below 0.2% throughout most of the SoC range, with slightly higher values at the lower end of this range. However, in practice BMS do not allow batteries to reach such low SoC levels. Thus, for the working range of SoC, the model is deemed sufficiently accurate. This model can be used as the basis of the battery pack level simulation tool.
Figure 5. Battery model validation with an automotive cycle
One of the key aspects to consider in modern battery packs, is the inclusion of cell balancing. The following section outlines cell balancing techniques and the algorithms employed. One of these algorithms are then simulated using the battery pack simulation tool, as a demonstration of one of the applications of this tool.
3 Figure 4. Pulsed current profile used for model identification
The steps shown in Figure 4 are repeated at several SoC values, such that SoC dependency is captured in the model. The model with the parameters that are identified with the OCV measurement and HPPC test is then validated, which is described in the next section.
EEVC European Electric Vehicle Congress
Cell Balancing Techniques
The cells of a battery pack can be unbalanced when one or more of the following parameters are different between the cells:
SoC Leakage current (self-discharge) Internal resistance Capacity
These deviations from a balanced condition among the cells can lead to adverse effects on the
3
durability of the battery. For example during a charging process, it can happen that the cell with the least capacity would be fully charged while other higher capacity cells would still be charging. This could cause the lowest capacity cell to get overcharged while other cells reach their full charge. Similarly, during discharging, the cell with least capacity may become over-discharged and fail before the other cells during the discharge process. In order to prolong battery life, it is essential to overcome the aforementioned effects. This can be done via cell balancing. This section describes dissipative and non-dissipative cell balancing algorithms.
3.1
than dissipated as heat. This makes these algorithms more suitable to a wider operational range than their passive counterparts, i.e. both charging and discharging. 3.2.1
Non-dissipative Balancing Algorithm Description When balancing is active, the cell with the minimum SoC receives energy from all the other cells. The balancing current is determined based on the SoC difference between the cells. A lookup table is used to implement this behaviour, as shown in Figure 6.
Dissipative (Passive) Balancing
The dissipative method refers to dumping charge of individual cells through resistances or other means with the goal of equalizing the charge levels of the various cells within a battery pack. 3.1.1
Dissipative (Passive) Balancing Algorithm description The passive balancing algorithm assumes a battery setup where a set of cells or modules are placed in series, with each cell or module in parallel with a relay and a dissipative resistance. The relay closes the circuit containing the cell and the resistance, when dissipation of excess charge in the cell is desired. This algorithm is feasible in the cases where low pack currents are present, for example during standard charging. The algorithm can be based on any of the following [16]:
3.2
Figure 6. Balancing current dependence on SoC
While currently the algorithm abstracts from the physical implementation of the active balancing device, physics can be introduced by reshaping the balancing current dependence on SoC difference between cells. The algorithm engages cell balancing using a double hysteresis, so as to avoid high frequency oscillations and switching. As such, two levels of the SoC difference are selected to activate and deactivate cell balancing. These relate to maximum SoC variation within the cells.
Cell voltage: during charging, the algorithm dissipates the cell with the highest charge based on matching its voltage with that of other cells Final voltage: similar to the cell voltage algorithm, except that the balancing is done at the end of charge SoC history: the algorithm balances all the time, with the aim of matching the SoC of the different cells based on previous history of cells
Non-Dissipative (Active) Balancing
The principle behind non-dissipative (active) balancing is to redirect charge within the battery pack such that cells’ charge levels are equalized. While having the same goal as dissipative balancing algorithms, the main difference is that the energy is redistributed between cells rather
EEVC European Electric Vehicle Congress
Figure 7 Active balancing algorithm hysteresis logic
Figure 7 describes the ON-OFF switching logic of the active cell balancing algorithm. The y axis represents the maximum SoC difference between any of the cells in series connection, while the x axis depicts time. Three thresholds are used by the algorithm:
ΔSoC1 – if the maximum SoC deviation within the pack drops below this value, the balancing is turned off
4
ΔSoC2 – if the maximum SoC deviation within the pack is above this value, the balancing algorithm is turned on ΔSoC3 – if the maximum SoC deviation is below this value and a different cell than the previous time sample has the minimum SoC, the balancing algorithm is turned off
The last rule is used to insure there is no high frequency switching between different cells to be equalized. This could appear if two cells have a very close SoC, and each time step of balancing shifts the SoC of one over the other, resulting in the two cells alternatively receiving charge from the rest of the battery. 3.2.2 Simulation Case Study In order to understand the use of balancing within a simple example study, the simulation environment was set up to represent a worst-case study based on the cells. The following assumptions are made:
Modules are represented by a collection of similar cells. Each module has 6 cells in parallel, with four such modules in series Only module voltages need to be considered, and not individual cell voltages, as the cells within a module are connected in parallel and it is considered that they have a similar voltage. As described in [8], this assumption holds as long as the interconnect resistances are similar between the cells of the module, assumption that is made within this work. We consider a worst-case where considerable differences exist between modules. Non-dissipative (active) balancing takes place between modules. The starting point following Constant Current Constant Voltage (CCCV) charging is such that the battery pack is fully balanced. The vehicle model is based on a generic EV powertrain. The active balancing rate is limited by DCDC converters within the BMS. The simulation study considers: Case A: NEDC cycle on pack without balancing Case B: NEDC cycle on pack with active balancing The parameters of the battery pack used in the example of balancing algorithms is shown in Table 2.
Table 2. Battery pack parameters
Module number 1 2 3 4
Initial SoC [%] 100 100 100 100
Capacity offset [%] 0 0 10 -20
Ro offset [%] 26 28 13 28
3.2.3 Simulation Results The NEDC profile was simulated on the battery pack, and the resulting cell voltage and SoC can be seen in Figure 8, without any cell balancing. The maximum inter-module deviation in SoC at the end of the cycle is 17%. The results of applying the active balancing algorithm can be seen in Figure 9, where the maximum inter-module deviation in SoC is reduced to around 1%. The functioning of the above algorithm is illustrated in Figure 10. In this simulation example, the values of ΔSoC1, ΔSoC2, and ΔSoC3 are respectively 0.005, 0.008 and 0.01 respectively. It can be seen that at time 4089s, the max deviation in the SoC is above ΔSoC2, because of which the balancing process is started until the maximum SoC difference is below ΔSoC1. The functioning of the third condition with respect to ΔSoC3 can be understood from time instances 3285s and 3450s. It can be seen that the balancing does not start at 3285s, despite the fact that the maximum SoC difference is above ΔSoC2. This is because the maximum SoC difference is less than ΔSoC3, while at the same time, the Module 4 has the minimum SoC, while in the previous time sample the cell to receive current is not determined.
Figure 8 Charge profile without active cell balancing
EEVC European Electric Vehicle Congress
5
implements an additional hysteresis over the one implemented by the thresholds ΔSoC1 and ΔSoC2. The following section describes another application of the BMS simulator: pack-level SoC estimation.
4
Figure 9 Charge profile with active cell balancing
Pack Level SoC Estimation
One other application of the battery pack simulator is to assist the development of battery state estimation strategies at pack level. This topic is recognized in literature as adding to the challenges brought on by cell level SoC estimation [9] [10] [11] [12] [13] [14]. One concept important for energy management is to provide a range of SoC estimations to characterise predictive functionality of the pack as a whole, based on the states of the individual modules. This range decreases with active balancing (and is therefore dynamic and an intrinsic function of the BMS), but is nonetheless important as part of an integration control strategy. A simulation was conducted to investigate this concept further. The simulation is based on the validated cell models described in Section 2.3. For each module, differences in terms of initial SoC, battery capacity and internal resistance are introduced in the model. The scenario is illustrated in Table 3. Table 3. Battery pack parameters
Module Initial SoC Capacity Ro offset number [%] offset [%] [%] 1 94 -4 23 2 94 -5 20 3 86 4 25 4 94 1 27 5 88 2 30 6 91 -8 23 Figure 11 shows the simulation results which were obtained with a configuration of 6 modules in series. Each module consisted of 8 cells in parallel.
Figure 10 Charge profile with active cell balancing (zoomed plot)
Thus the algorithm will wait till the maximum difference in SoC is greater than ΔSoC3. This
EEVC European Electric Vehicle Congress
Figure 11 Current and estimated voltage for the series-connected 6 module battery configuration [1]
6
The tests described in Figure 11 consist of a reallife automotive profile applied to the battery pack, the current describing this presented in the upper plot. The voltage response of each of the modules is visible in the second plot. Figure 12 presents a State of Charge estimation algorithm previously developed for a battery cell [7] enhanced here to estimate the state of the complete battery pack, including indications about the inherent imbalance from the pack. The simulation scenario is the same one described in Figure 5. The thin lines illustrate the SoC of each module, while the thicker lines represent estimated of the highest and lowest SoC within the pack. It can be seen that the upper and lower SoC values throughout the modules are estimated correctly.
dissipative balancing. Thus the BMS simulator provided a good platform for the testing of the non-dissipative balancing algorithm in the context of a worst-case scenario where considerable differences exist between modules. In a second simulation case study, the modelling of an unbalanced pack is considered from the viewpoint of state estimation. The state estimation range concept is introduced, wherein the BMS is able to estimate the upper and lower limit of the SoC for an unbalanced pack. In this example, it is shown that the SoC upper and lower limit SoC estimations are sufficiently accurate. Thus the pack simulator can also be applied to test the performance of SoC estimation algorithms in the context of a pack with modules with significantly different capacities. This concept is important for energy management topics, considering limits on further discharge or charging powers based on inter-module SoC differences within a battery pack.
References
Figure 12. State of Charge estimation in a pack [1]
5
Conclusions
The activities within this study focus on the development of a modelling environment to capture battery and BMS operation for automotive applications. The developed tool includes validated cell models, extended with definition for balancing algorithms and SoC estimation. A short simulation study was presented to illustrate the impact of active balancing during NEDC discharging cycle. It is shown for this case study, that the deviation in inter-module SoC at the end of the cycle was reduced from 17% without balancing to just over 1% with non-
EEVC European Electric Vehicle Congress
[1] Rosca, Bogdan; Jacob, Jubin, “Battery Pack Simulation for BMS,” TNO, Helmond, 2014. [2] W. Waag, C. Fleischer and D. U. Sauer, “Critical review of the methods for monitoring of lithium-ion batteries in electric and hybrid vehicles,” Journal of Power Sources, vol. 258, pp. 321-339, 2014. [3] S. Bertoni, M. Ceraolo and T. Huria, “Stateof-Charge evaluation of lithium cells showing voltage hysterezis,” in EEVC, Brussels, 2012. [4] B. Rosca, “Enhanced battery model including temperature effects,” in EVS27, Barcelona, 2013. [5] Y. Sumida, A. Nomura, W.-H. Yang , Y. Kamiya, Y. Daisho and K. Morita, “Analysis of Adverse Effects on Vehicle Performance Due to Battery Deterioration Installed in BEV and HEV,” in EEVC, Brussels, 2012. [6] G. Mulder, N. Omar, S. Pauwels, M. Meeus, F. Leemans, B. Verbrugge, W. De Nijs, P. Van den Bossche, D. Six and J. Van Mierlo, “Comparison of commercial battery cells in relation to material properties,” Electrochimica Arta, vol. 87, pp. 473-488, 2013. [7] B. Rosca, J. Kessels, H. Bergveld and P. van den Bosch, “On-line Parameter, State-of-
7
Charge and Aging Estimation of Li-ion Batteries,” in IEEE Vehicle Power and Propulsion Conference, Seoul, 2012. [8] P. J. Miller, W. D. Drury and P. S. Jootel, “An Approach to Cell Selection to Optimize Battery Pack Size and Mass for Small Battery Electric Vehicles,” in EEVC, Brussels, 2012. [9] G. J. Offer, V. Yufit, D. A. Howey, B. Wu and N. P. Brandon, “Module design and fault diagnosis in electric vehicle batteries,” Journal of Power Sources, vol. 392, no. 206, p. 383, 2012. [10] H. Dai, X. Wei, Z. Zun, J. Wang and W. Gu, “Online cell SoC estimation of Li-ion battery packs using a dual time-scale Kalman filtering for EV applications,” Applied Energy, vol. 95, pp. 227-237, 2012. [11] L. Zhong, C. Zhang, Y. He and Z. Chen, “A method for the estimation of the battery pack state of charge based on in-pack cells uniformity analysis,” Applied Energy, vol. 113, pp. 558-564, 2014. [12] G. L. Plett, “Efficient Battery Pack State Estimation using Bar-Delta Filtering,” in EVS24, Stavanger, 2009. [13] C. Truchot, M. Dubarry and B. Y. Liaw, “State-of-charge estimation and uncertainty for lithium-ion battery strings,” Applied Energy, vol. 119, pp. 218-227, 2014. [14] G. L. Plett, “Extended Kalman filtering for battery management systems of LiPB-based HEV battery packs,” Journal of Power Sources, pp. 262-276, 2004. [15] M. A. Roscher, O. Bohlen and J. Vetter, “OCV hysteresis in Li-ion batteries including two-phase transition materials,” International Journal of Electrochemistry, 2011. [16] D. Andrea, Battery Management Systems for Large Lithium-Ion Battery Packs, Norwood: ARTECH HOUSE , 2010.
EEVC European Electric Vehicle Congress
Authors Jubin Jacob obtained a B.Tech degree in Electrical and Electronics Engineering at VIT University, Vellore, India in 2009 after which he joined the Eindhoven University of Technology, the Netherlands, where he obtained an M.Sc in Automotive Technology in 2011 and a Professional Doctorate in Engineering (PDEng) in Automotive Systems Design in 2013. He joined TNO, the Netherlands Organization for Applied Scientific Research, as a research engineer in the powertrains department. Bogdan Rosca obtained his M.Sc. degree in Automotive Technology from the Eindhoven Technical University, in 2011. He is currently a research engineer at TNO, the Dutch Organization of Applied Research, in the Powertrains Department. His activities include battery modelling and state estimation, as well as powertrain modelling and control. He is involved in several European projects, looking at modelling and state estimation of Li-ion batteries, as well as ageing behaviour.. Steven Wilkins works as a scientific research engineer with a background in hybrid and electric vehicle systems powertrain modelling and simulation, originally based at Imperial College London and is now based in the Netherlands as a senior research scientist within TNO Powertrains, involved in various European Projects. He is an active member of EARPA and EGVIA. He is also a part-time Assistant Professor at Eindhoven University of Technology.
8
