Development of a Hardware-In-the-Loop Simulator for Battery Management Systems

Transcription

1 Development of a Hardware-In-the-Loop Simulator for Battery Management Systems A Thesis Presented in Partial Fulfillment of the Requirements for the Degree Master of Sciencein the Graduate School of The Ohio State University By Lingchang Wang, B.S. The Ohio State University 2014 Master s Examination Committee: Dr. Giorgio Rizzoni, Advisor Dr. Shawn Midlam-Mohler

2 c Copyright by Lingchang Wang 2014

3 ABSTRACT Battery technology is evolving rapidly with growing energy and power densities. While this energy storage device becomes more prominent and ubiquitous in the automotive industry, safety and reliability of the battery system are attracting more attention than ever before. The battery management system (BMS) becomes the key to improving vehicle safety, prolonging battery life, and reducing cost. This project focuses on constructing a test bench for BMS with Hardware-Inthe-Loop (HIL) technique. A battery model that is single-cell-capable and real-timecapable simulates a battery pack s behavior as a series of connected cells. It takes into account the manufacturing differences between cells that cause voltage deviation from the ideal reference voltage. Said model is then implemented into the dspace mid-size HIL simulator. Several experiments were performed with the virtual battery model to validate prototyped BMS components and BMS software. These components include a module balancing board, a battery controller module, and a battery data logger. These tests indicate it is feasible to validate BMS with HIL techniques, and the developed battery model can reduce the cost of developing BMS software and hardware. ii

4 This document is dedicated to DXI iii

5 ACKNOWLEDGMENTS I would like to thank my advisor, Prof. Giorgio Rizzoni who gave guidance and insight on various issues throughout my time at OSU and CAR. The support provided by my co-advisor, Dr. Shawn Midlam-Mohler has been phenomenal throughout the entire project. I would like to thank the entire Buckeye Bullet team, the Bullet project has been an inspiration. A big thanks to Gian Luca Storti who managed to accel this project. I would also like to thank the EcoCAR2 team who has helped me with everything MIL and HIL. I would like to acknowledge every staff member at CAR for providing an incredible facility for conducting research. Thank you everyone. iv

6 VITA Sycamore High School B.S. Mechanical Engineering, The Ohio State University M.S. Mechanical Engineering, The Ohio State University 2010-Present Graduate Research Associate, The Ohio State University FIELDS OF STUDY Major Field: Mechanical Engineering Specialization: Automotive v

7 TABLE OF CONTENTS Abstract Dedication Acknowledgments Vita List of Tables ii iii iv v ix List of Figures x CHAPTER PAGE 1 Introduction Motivation Project Objective Organization of Thesis Background and Literature Review Battery Management System Overview State of Charge Determination Current-Based SOC Estimation Voltage-Based SOC Estimation Cell Balancing Thermal Management Battery Modeling Zero Order Model First Order Model Second Order Model dspace Battery Automotive Simulation Model MIL and HIL Testing for BMS vi

8 3 Experimental Equipment dspace Mid-Size Hardware-In-the-Loop Simulator MicroAutoBox Prototyping Unit Module Balancing Board TI Module Balancing Board Wilson-Scott Module Balancing Board Battery Overall Experimental Apparatus Summary Battery Model Structure and Configuration General SOC Balancing Simulation Aged Single Cell Simulation Thermal Management System Simulation Increased Internal Resistance Simulation Uniform Resistance Increase Increased Internal Resistance Variance Decreased Capacity Simulation Summary Experimental Methodology Probability Density Functions Current Request Profile dspace MidSize HIL Setup MicroAutoBox Setup MIL and HIL Testing Three-Level MIL Testing Two-Level MIL Testing HIL Testing Summary Battery Simulation Results Baseline Model Simulations Single Cell First Order Model Simulation Parameter Variance Simulations Charging and Discharging Internal Resistance Normal Distribution Simulation Charging and Discharging Internal Resistance Chi-Square Distribution Simulation Cell Capacity Normal Distribution Initial Temperature Normal Distribution Simulation vii

9 6.2.5 Multiple Parameter Variation, Internal Resistance Chi-Square Distribution, Independent Case Multiple Parameter Variation, Internal Resistance Chi-Square Distribution, Dependent Case Summary BMS HIL Testing and Validation Results Traditional Balancing Algorithm MIL Result Algorithm with Wire Voltage Drop Estimation MIL Result Algorithm with OCV Estimation MIL Result Summary Conclusion and Future Work Improving battery Fixture Improving Simulator Efficiency Adding Voltage Amplifier to MidSize HIL Unit Implementing Advance Control Algorithm Bibliography APPENDICES A Editing and Loading Battery Model into MIL/HIL A.1 Creating and Editing Battery Parameter Probability Density Function 88 A.2 Editing Battery Parameter Matrices A.3 Loading Model in ControlDesk A.3.1 Structuring the Simulink Model B Supplementary Battery Simulation Figures viii

10 LIST OF TABLES TABLE PAGE 6.1 Result Table Nomenclature Charging and Discharging Internal Resistance Normal Distribution Results Charging and Discharging Internal Resistance Chi-Square Distribution Results Cell Capacity Normal Distribution Results Initial Temperature Normal Distribution Results Multiple Parameter Variations, Internal Resistance Chi-Square Distribution, Independent Case Multiple Parameter Variations, Internal Resistance Chi-Square Distribution, Dependent Case PDFs Applied, Balancing Case Study Balancing Time ix

11 LIST OF FIGURES FIGURE PAGE 1.1 Buckeye Bullet Battery Management System Basic BMS Structure Sample Voltage Trajectories of Two Battery Cells in Series (EcoCAR, US06, A123 20Ah) SOC Error Caused by Variance in Cell Capacity or Current Measurement OCV Difference Caused by Difference in Cell Capacity Sample Resistor-Based Balancing Circuit Sample Capacity-Based Balancing Circuit Sample Thermal Management System Zero Order Equivalent Circuit Model First Order Equivalent Circuit Model Second Order Equivalent Circuit Model Battery ASM Schematic BMS Simulation Schematic dspace Mid-Size Hardware-In-the-Loop Simulator MicroAutoBox Prototyping Unit TI Module Balancing Board Wilson-Scott Module Balancing Board A Cylindrical Cell x

12 3.6 A123 20Ah Prismatic Module Experimental Setup MIL MicroAutoBox ControlDesk Interface HIL Simulator ControlDesk Interface Battery Model Structure Normal Distribution Example Chi-Square Distribution Example EcoCAR2 US06 Current Request Profile EcoCAR2 US06 Current Request Profile Limited at 200A MidSize HIL Setup MidSize HIL Interface Function MABX Interface Function SOC Observer GUI Three Level MIL Setup Overview dspace SCALEXIO Two-Level MIL Setup Overview dspace Two-Level MIL Setup Overview HIL Setup Overview Single Cell Voltage Under Load Single Cell State of Charge Voltage Spread, Charging and Discharging Internal Resistance, 3 Percent Standard Deviation Voltage Spread, Charging and Discharging Internal Resistance, 3 Percent Standard Deviation, Cross Section t=25[s] Voltage Spread, Charging and Discharging Internal Resistance, Chi- Square 3 Percent Standard Deviation Voltage Spread, Charging and Discharging Internal Resistance, Chi- Square 3 Percent Standard Deviation, Cross Section t=25[s] xi

13 6.7 Voltage Spread, Cell Capacitance, Normal 10 Percent Standard Deviation Voltage Spread, Cell Capacitance, Normal 10 Percent Standard Deviation at Minimum SOC Voltage Spread, Charging and Discharging Internal Resistance, Normal 10 Percent Standard Deviation, Cross Section t=95[s] Voltage Spread, Charging and Discharging Internal Resistance, Chi- Square 3 Percent Standard Deviation Voltage Spread, Charging and Discharging Internal Resistance, Chi- Square 3 Percent Standard Deviation, Cross Section t=25[s] Voltage Spread, Combined Parameter Variation, Independent, Chi- Square 1 Percent Standard Deviation Internal Resistance Voltage Spread, Combined Parameter Variation, Independent, Chi- Square 1 Percent Standard Deviation Internal Resistance, Cross Section t=25[s] Voltage Spread, Combined Parameter Variation, Independent, Chi- Square 1 Percent Standard Deviation Internal Resistance Voltage Spread, Combined Parameter Variation, Independent, Chi- Square 1 Percent Standard Deviation Internal Resistance, Cross Section t=25[s] Balancing Algorithm Flow Chart, Courtesy of Gian Luca Storti Generated Cell Voltages for Balancing Case Study Basic Voltage Based Passive Balancing Balancing Wire and Electronic Voltage Drop Estimation Proposed Balancing Algorithm with Voltage Drop Estimation Improved HIL Battery Fixture A.1 Simulink Model Structure A.2 dspace I/O in Simulink A.3 dspace CAN I/O A.4 dspace CAN I/O xii

14 A.5 Real Time Interface Block Set A.6 Real Time Interface Block Set B.1 Voltage Spread, Charging and Discharging Internal Resistance, 1 Percent Standard Deviation B.2 Voltage Spread, Charging and Discharging Internal Resistance, 1 Percent Standard Deviation, Cross Section t=25[s] B.3 Voltage Spread, Charging and Discharging Internal Resistance, 2 Percent Standard Deviation B.4 Voltage Spread, Charging and Discharging Internal Resistance, 2 Percent Standard Deviation, Cross Section t=25[s] B.5 Voltage Spread, Charging and Discharging Internal Resistance, 1 Percent Chi-Square Deviation B.6 Voltage Spread, Charging and Discharging Internal Resistance, 1 Percent Chi-Square Deviation, Cross Section t=25[s] B.7 Voltage Spread, Charging and Discharging Internal Resistance, 2 Percent Chi-Square Deviation B.8 Voltage Spread, Charging and Discharging Internal Resistance, 2 Percent Chi-Square Deviation, Cross Section t=25[s] B.9 Voltage Spread, Cell Capacitance, 5 Percent Normal Deviation B.10 Voltage Spread, Cell Capacitance, 1 Percent Normal Deviation B.11 Voltage Spread, Initial Temperature, 1 Percent Normal Deviation B.12 Voltage Spread, Initial Temperature, 1 Percent Normal Deviation, Cross Section t=25[s] B.13 Voltage Spread, Initial Temperature, 2 Percent Normal Deviation B.14 Voltage Spread, Initial Temperature, 2 Percent Normal Deviation, Cross Section t=25[s] B.15 Voltage Spread, Combined Parameter Variation, Independent, Chi- Square 2 Percent Standard Deviation Internal Resistance B.16 Voltage Spread, Combined Parameter Variation, Independent, Chi- Square 2 Percent Standard Deviation Internal Resistance, Cross Section t=25[s] xiii

15 B.17 Voltage Spread, Combined Parameter Variation, Independent, Chi- Square 3 Percent Standard Deviation Internal Resistance B.18 Voltage Spread, Combined Parameter Variation, Independent, Chi- Square 3 Percent Standard Deviation Internal Resistance, Cross Section t=25[s] xiv

16 CHAPTER 1 INTRODUCTION 1.1 Motivation Today s increased energy capacity of batteries comes at a cost of increased risk. To maintain the battery pack at a safe and optimal state, the BMS is becoming more complex and more costly. Since the BMS is often application specific, the development cycle of a new BMS often becomes the bottleneck of energy storage. Figure 1.1 shows the BMS used for the Buckeye Bullet, that manages 200 cells. For the Buckeye Bullet, in-vehicle driving test is impossible, and battery testing with large battery packs is cumbersome and time-consuming. One pathway to lower development cost of BMS is the use of HIL techniques, which offers lower cost and higher reproducibility with a virtual test bench. This generates a new set of trade-offs between model fidelity and complexity: the battery model needs to simulate the behaviors of tens or hundreds of individual cells, and it must run in real time with limited processing power. This project takes advantage of the previously collected battery data in CAR and the application of HIL on engine controller units to construct a test bench for BMS. If successful, this test bench can be used to develop BMS for special projects such as the Buckeye Current Motorcycle team. 1

17 Figure 1.1: Buckeye Bullet Battery Management System 1.2 Project Objective The objective of this project is to construct a HIL compatible battery model that is both multi-cell-capable and real-time-capable. This battery model will be used to validate prototype BMS components. The work presented in this thesis is intended as a template for BMS HIL testing at the Ohio State University (OSU) Center for Automotive Research (CAR). 1.3 Organization of Thesis Chapter 1 provides an introduction to the thesis topic, the motivation of the study, and the objective of the project. Chapter 2 contains background information for this project and published literature on the subject matter. This includes the background information on Battery 2

18 Management System (BMS), an overview and application of battery testing and modeling techniques, and Hardware-In-the-Loop (HIL) testing in the context of battery and BMS. Chapter 3 describes the experimental equipments used for this project: the dspace Mid-Size HIL simulator, the MicroAutoBox (MABX) prototyping unit, the TI module Balancing Board, and the test battery cells. A description of the CAN messages and input/output signals is included also. An overall test bench schematic and description is presented at the end of this chapter. Chapter 4 presents the experimental procedures used for system validation. Included is a description of the methodology for evaluating battery model behavior, and MIL/SIL/HIL testing of prototyped module balancing board (MBB) and battery controller module (BCM). A list of faults injected into the BMS is included in this chapter. Also included are the step by step description of the process of each experiment. This chapter serves as an instruction manual on how to set up the existing BMS test bench using the equipments available at CAR. Chapter 5 outlines the results of the battery simulation and validation. Week of Feb 17. Chapter 6 summarizes the results of the BMS validation. Week of Feb 24. Chapter 7 includes the concluding remarks and recommendations for future work. This includes possible applications of the HIL test bench at the Ohio State University s (OSU) Center for Automotive Research (CAR). Appendix A serves as an instruction manual for editing and loading the developed battery model into the HIL simulator. 3

19 CHAPTER 2 BACKGROUND AND LITERATURE REVIEW This chapter introduces the battery management system (BMS), and describes the functions of typical BMS: SOC estimation, cell balancing, and thermal management. Then different battery models are covered, including the zero order, first order, and second order equivalent circuit model. The chapter also compares lower order models with the high order dspace ASM battery model, and makes the argument against the ASM. The typical MIL and HIL validation for a BMS is introduced at the end. 2.1 Battery Management System Overview The BMS serves many purposes in battery applications such as monitoring the state of charge (SOC) or state of health (SOH). It also mitigates several risks associated with high voltage battery packs such as cell over voltage or over heating. Typical BMS is divided into two parts: a cell level module such as a series of balancing board, and a pack or vehicle level module analogous to the engine control unit (ECU). The balancing board is usually programmed to have simple, independent, and passive algorithms. The supervisory controller, on the other hand, handles the overall BMS strategy and communication with other systems such as the vehicle controller. Shown in Figure 2.1 is a simplified example of a BMS with a battery control module (BCM) 4

20 and a module balancing board (MBB). A more detailed BMS model can be found in Section 2.3. Figure 2.1: Basic BMS Structure In a typical vehicle simulator, the BMS is often overlooked or over simplified, because the energy consumption is negligible. The battery pack is often treated as a single cell: the number of series and parallel elements are treated as a multiplication factor for the terminal voltage and capacity respectively. In reality the deviations of capacity, self-discharge rate, internal resistance, operating temperature, and manufacturing quality can create significant differences in cell behavior. Figure 2.2 illustrates the difference in battery voltage trajectories between two cells with different internal resistances. This type of parameter uncertainty often limits the battery operating range, and the BMS plays a central role in maintaining optimal battery performance and prolonging battery life. 5

21 Figure 2.2: Sample Voltage Trajectories of Two Battery Cells in Series (EcoCAR, US06, A123 20Ah) State of Charge Determination The primary functionality of the BMS is state of charge (SOC) determination. In HEV/EV applications, the SOC servers as the fuel gauge, directly affecting the range of the vehicle and energy management strategy Current-Based SOC Estimation The most common type of SOC determination is current-based, also called the Coulomb counting method. The BMS integrates the current with an initial SOC. This method is prone to errors caused by noise in current measurement and variance of battery capacity. Because in battery simulation the Coulomb counting method is 6

22 often used to calculated charge dissipated, the error in current measurement is often neglected. Figure 2.3: SOC Error Caused by Variance in Cell Capacity or Current Measurement Figure 2.3 demonstrates that the effect of the variance in cell capacity is significant when the battery moves through a wide SOC gap. Charge depleting operations are most likely to be affected by this error. The measurement noise in the current sensor, on the other hand, causes a cumulative error in the SOC measurement. The error 7

23 grows linearly over time, implying that the current-based SOC estimation is more accurate during short operations Voltage-Based SOC Estimation Since the open circuit voltage (OCV) of a battery is a function of SOC, it is possible to determine the SOC based on the measured OCV. If the battery is allowed to rest long enough (during balancing or overnight charging conditions, for instance), the OCV can be measured directly. It can also be calculated through the equivalent circuit model, as described in Section 2.2. Because of the simplicity of this model, it is often used in BMS algorithms in conjunction to the current-based method. However, since most Li-ion batteries have a fairly flat OCV curve, there is often a mismatch between estimated OCV and actual OCV. Moreover, the variance in capacity in the battery cells can also effect the result significantly. 8

24 Figure 2.4: OCV Difference Caused by Difference in Cell Capacity Shown in Figure 2.4 is the estimated OCV of two cells with different capacity discharged at 1C. In a battery pack with hundreds of cells, the averaged OCV is not a reliable indication of SOC. Therefore the HIL test bench must be able to simulate the OCV of single cells to test the BMS strategy Cell Balancing Overtime the SOCs of different cells in a battery system will differentiate, which limits the maximum capacity of the battery pack to the weakest cell. Pack charging or discharging shuts off when the battery pack reaches the cut off voltage, effectively limiting the overall capacity. The degradation of cell with the lowest capacity is also faster than other cells, since it is cycled through a wider SOC range at a higher C 9

25 rate than other cells. The BMS can perform cell balancing passively or actively with current shunt balancing circuits or capacitor based balancing circuits. The objective of the balancing circuits is to keep all the cells at a similar SOC, but in implementation most balancing circuits are measuring the OCV instead. The most common balancing method is the current bypass method, shown in Figure 2.5, comprised of a switch and a resistor for each cell unit a series string. During balancing, when the switch is closed, the imbalance in voltage will cause a current to pass through the resistor, effectively lowering the SOC of the cell with higher OCV. Figure 2.5: Sample Resistor-Based Balancing Circuit This method has two disadvantages: the cell voltages are always brought down to the lowest voltage, and the energy lost is not recoverable. A capacitor-based balancing method, on the other hand, can transfer a portion of the excess energy from one cell to another. The trade-off accompanies this method is the increased cost and complexity. The algorithm associated with this balancing method is often more complex and sophisticated thus less passive than the resistor based strategy. An example of the capacitor based balancing circuit is shown in Figure 2.6. The circuit first connects the 10

26 capacitor to the higher voltage cell to store charge, and then connects the capacitor to the lower voltage cell to release charge. It has a non-zero balancing efficiency since a portion of the energy is transferred. Since the current flowing into the capacitor is proportional to the voltage difference, this method becomes less and less effective as the voltage difference decreases. Figure 2.6: Sample Capacity-Based Balancing Circuit To test the switching control algorithm it is typical evaluate the performance of the balancing board on batteries directly. The batteries are often charged or discharged to a certain degree of imbalance first, then the algorithm is will balance the battery OCVs within a certain time window. The traditional testing method is often difficult to reproduce since it is time consuming to charge individual batteries to a certain voltage. Moreover, the resistor or capacitor on the balancing board is often small, and balancing often takes hours. Using battery model, on the other hand, can verify the strategy in a few minutes. The actual verification of the balancing strategy is explained in greater detail in Chapter 4. 11

27 2.1.3 Thermal Management The performance of a battery pack depends heavily on the operating temperature. In general, temperature affects the following aspects: 1. charge/discharge efficiency. 2. life and life cycle cost. 3. safety and reliability. 4. power and energy capacity. Most BMS monitor the battery temperature through thermal couples and thermistors: by estimating the various thermal resistances and capacitances, the battery core temperature can often be expressed as a function of measured temperatures. The BMS can adjust the cooling level or shut off the battery pack when a certain temperature threshold is exceeded. The design of the thermal management system is based on the trade-offs between performance, volume, mass, and cost. The Buckeye Bullet, for example, does not have any thermal management on board. The battery pack was designed to have a large temperature raise, sacrificing peak power for lower mass and volume. For most automotive applications, liquid cooling/heating and air cooling are more common, such as the one shown in Figure /reffig:thermal Management. Thermal issue is a common cause of failure in HEV battery packs. Validating the thermal management system often involves techniques such as basic heat transfer and fluid flow principles, finite element method (FEM), thermal imaging, and computational fluid dynamics (CFD). Passive thermal management such as constant coolant circulation or air flow can reduce risks of over heating, but only active cooling strategies can maintain the battery at optimal temperature. 12

28 Figure 2.7: Sample Thermal Management System Testing the active thermal management algorithm can be costly: the thermal environment must be controlled and the battery must be exercised to reach a certain temperature. Moreover, since the battery pack is expected to operate in all climates, tests must be conducted over a large temperature range. Chapter 4 suggests several HIL methods to simplify the validation process. 2.2 Battery Modeling This section introduces the equivalent circuit models and the dspace ASM battery model, and explains the advantages/disadvantages of each model. 13

29 2.2.1 Zero Order Model Figure 2.8: Zero Order Equivalent Circuit Model V b = E 0 IR 0 (2.2.1) The zero order equivalent circuit model is the most simplified, consists of an ideal voltage source and a resistor. Both components are dependent of SOC, temperature, and current direction. Because the zero order model is the easiest to identify, it is often used for battery sizing projects or steady state models. It does not capture the transient response of the system, as shown in Equation Chapter 4 suggests using the zero order model to test BMS balancing and thermal management performance, since this model is easy to construct and manipulate First Order Model The first order model contains a capacitor-resistor pair that captures the transient behavior of the battery during charge/discharge, as shown in Figure 2.9, the first order model. The added parameters are both dependent of temperature, SOC, and current direction, making the model more difficult to calibrate. The first order dynamic cannot fully describe the chemical reactions within the battery, but the model is accurate enough for general automotive applications. 14

30 Figure 2.9: First Order Equivalent Circuit Model V = E 0 R 0 V C (2.2.2) dv C dt + 1 R 1 C 1 + V C = 1 C 1 (2.2.3) The first order model is also the mostly widely used model for real time applications, since it strikes a good balance between complexity and accuracy. As shown in Equation 2.2.2, the model is fairly simple to construct and requires low computational power. A multi-cell version of the first order model is used for this project, it is introduced in Chapter Second Order Model Compared to the second order model had an additional resistor-capacitor pair. Since it can capture two sets of first order dynamics, the model is more accurate in terms of battery transient response. The second order model is often used for high fidelity simulations when the necessary parameters are obtained. It is important to note that compared to the first order model, there are two more parameters to identify, both are dependent on temperature, SOC, and current direction, making model identification significantly more costly. 15

31 Figure 2.10: Second Order Equivalent Circuit Model V b = E 0 R 0 I V C1 V C2 (2.2.4) dv C1 dt dv C2 dt + 1 R 1 C 1 + V C1 = 1 C 1 (2.2.5) + 1 R 2 C 2 + V C2 = 1 C 1 (2.2.6) dspace Battery Automotive Simulation Model The dspace Automotive Simulation Model (ASM) takes several step further toward model fidelity by modeling the individual physical effects such as diffusion, double layer capacitance, internal inductance, and internal resistance. The model is more difficult to parametrize than the second order model, making the model identification process difficult to manage. Because of the complexity of the model, it is not single-cell-capable. Linking several ASM models in series requires a large amount of computational power. Instead of single ASM cells, the dspace model employs a voltage deviation model (Delta Model). As shown in Figure 2.11, the delta model represents the SOC imbalance in the cells, adding an additional OCV variance to the reference model. This model structure can significantly reduce the computational requirement and shorten the computation time. However, because the delta model is difficult to manipulate, the 16

32 voltage deviation are not model after aging or manufacturing quality. In this model, the voltage imbalance is often arbitrary. Figure 2.11: Battery ASM Schematic Moreover, the complex ASM model often introduce more modeling error than intended, because of the large number of parameters. The model is intended to be used as a third order model, with three resistor/capacitor pairs. Each parameter is both SOC dependent and temperature dependent, making the reference cell model parameter difficult to extract. 17

33 2.3 MIL and HIL Testing for BMS The purpose of HIL testing is to replace actual batteries with a battery model accurate enough to test the BMS. There are three clear advantages of HIL over conventional testing: 1. Battery parameter manipulation. Most battery parameters cannot be changed easily. For example, if the experiment is designed to test the effect of battery aging on balancing strategy, cells has to be manually aged to match the desired parameter. On the other hand, changing the internal resistance and capacity of a battery model is much easier. 2. Scale of experiment. It is often difficult to perform tests on the BCM, because the BCM is monitoring hundreds or thousands of cells. Setting up a battery pack under load is often costly and time consuming. Using HIL strategy and simulating large quantity of cells can reduce the cost significantly. 3. Fault injection. Battery over-voltage, high temperature operation, or shorting of batteries can often cause catastrophic failures. Even under lab environment, these tests are destructive in nature. Performing fault injection in HIL can reduce the risks and allow a wider range of failure scenarios. Since Simulink can directly compile models into C codes, Model-In-the-Loop simulation is the same as Software-In-the-Loop simulation. This chapter does not cover SIL or programming in C. For BMS application, the general structure of a simulator is shown below in Figure There are three sensors (cell voltage sensor, temperature sensor, and current sensor), and two actuators (discharging balancing circuit, and pr-echarge resistor) in the example system. When the balancing and control strategy becomes more 18

34 complex, it is crucial to update the list of sensors and actuators. The simulator must be structured in a way that the MBB and BCM can be swapped out easily, i.e. the sensor inputs and actuator outputs must be clearly identifiable. The output of the sensors will be simulated by the battery model and read by the BMS. The output actuator signals (balancing current for example) will be fed into the battery to simulate the change in voltage output. Figure 2.12: BMS Simulation Schematic The details of the MIL and HIL setup are described in Chapter 4. 19

35 CHAPTER 3 EXPERIMENTAL EQUIPMENT This chapter covers the available hardware for the MIL and HIL tests, includes the dspace Mid-Size Hardware-In-the-Loop simulator (MidSize HIL), the MicroAutoBox prototyping unit (MABX), the balancing boards (MBB), and the batteries. Two MBBs are available for this project: the TI BQ board, and the Wilson-Scott board. Two types of batteries are available, the A Ah prismatic module, and the cylindrical cells. 3.1 dspace Mid-Size Hardware-In-the-Loop Simulator The dspace HIL simulator shown in Figure 3.1 is designed for ECU function, ideal for power train and vehicle dynamic testing. It features a DS2202 HIL I/O board, which contains signal conditioning for typical signal level of 12V, 24V, and 42V autotmotive systems. The process board runs the real-time model and the communication is handled by the I/O board. It has the following features relevent to BMS testing: 1. A/D Conversion. The ADC unit provides 16 input channels. The unit has 14-bit resolution over a 60V range. 2. D/A Conversion. The DAC unit provides 20 output channels. The unit has 12- bit resolution over a 5-10V range. This resolution can be potentially limiting, since cell balancing requires accurate measurement of OCV. There is an output 20

36 Figure 3.1: dspace Mid-Size Hardware-In-the-Loop Simulator current limit (5mA), therefore the DAC is mostly used to as sensor outputs instead of power supplies. 3. CAN Support. CAN is the main system used for communication between the HIL unit and MABX. 21

37 3.2 MicroAutoBox Prototyping Unit The MicroAutoBox (MABX)l prototyping unit is a stand-alone unit used for realtime ECU simulation. If the MidSize HIL is designed to simulate the plant, the the purpose of the MABX is to simulate the controller. For the purpose of the BMS, this unit can assume the roles of the balancing board and the supervisory controller. Since it is not loaded with resistors or capacitors, it cannot perform high power functions like balancing. It communicates with the MidSize HIL and other MABXs through CAN. The MABX has two CAN channels, making it possible to Figure 3.2: MicroAutoBox Prototyping Unit control two MBBs simultaneously. However, each CAN channel needs to have its own dbc file. In Chapter 4, a MABX interface is introduced that can switch between the MidSize HIL CAN channel and the MBB CAN channel. 22

38 3.3 Module Balancing Board Two module balancing boards (MBB) are used for the HIL portion of the experiment. The balancing functions are not used directly since the balancing current is too high for the MidSize HIL, but the balancing states are observed. In other words, the MBB and BCM will send a set of flag values that indicates whether a cell is being balanced to the MidSize HIL, instead of drawing current from the HIL TI Module Balancing Board Figure 3.3: TI Module Balancing Board The TI MBB, shown in Figure 3.3 is the board used in the Buckeye Current Eletric Motorcycle BMS. It is designed for li-ion battery monitoring and protection, capable 23

39 of managing 9-18 cells. The ADC has 14-bit resolution and 3mV typical accuracy. Out of the 9 ADC channels, 6 are used for cell voltages, 2 are used for thermistors or thermal couples, and 1 is reserved for general purpose. It is intended to be used with a supervisory controller to maximize the functionalities. Cell balancing is handled by a 47ohm resistor-based balancing circuit. Balancing switches are handled by the supervisory controller Wilson-Scott Module Balancing Board Figure 3.4: Wilson-Scott Module Balancing Board 24

40 The Wilson-Scott balancing board (WSMBB), shown in Figure 3.4 is similar to the TI board in terms of structure and data acquisition. The main difference between the two is that the WSMBB is CAN capable. It has 13ohm balancing resistors, which allows for faster balancing rate. Balancing switches are controlled by the supervisory controller, just like the TI board. The WSMBB is installed into a Deutsch PCB project box, ready for vehicle integration and implementation. 3.4 Battery Two battery chemistries are considered for this experiment: the A cylindrical cell, and the 20Ah A123 prismatic module. The cylindrical cell is easy to assembly into a custom module, making the installation of BMS hardware a simple process (Shown in Figure 3.5). The has also been characterized as a first order model, the various parameters have been identified and are readily avaiable. The anode and cathode have threaded terminals laser-welded for easy of installation. Figure 3.5: A Cylindrical Cell 25

41 The prismatic module, on the other hand, is already assembled with a predetermined number of series element, making hardware testing less flexible. It is designed to have A123 module balancing boards mounted on each end of the module, but it is possible to reconfigure the sense leads to support custom boards. The prismatic module is chosen because it is often used as an automotive energy chemistry (Shown in Figure 3.6. To test the battery model structure, the prismatic module parameters are used, because the simulation result can be compared with the actual behavior of the EcoCAR pack. Figure 3.6: A123 20Ah Prismatic Module 3.5 Overall Experimental Apparatus The HIL station is set up in the EcoCAR room in the Center for Automotive Research (CAR). There is one dedicated desktop controlling the HIL. This control panel 26

42 is responsible for setting up the battery model, loading MBB algorithm, and fault injection. One laptop is used to program the MABX. Figure 3.7: Experimental Setup MIL The MABX ControlDesk (CD) interface performs the following functions: 1. Enable passive and active balancing. Balancing is normally enabled by the supervisory controller, but the CD is able to force the balancing on and off. 2. Indicate fault flags. The CD is able to compose a fault code based on the error flags available. It can track basic faults such as cell over/under voltage and over temperature. 27

43 3. Compare estimated and measured voltages. A reference cell model is included in the control algorithm, and the CD can compare the difference between the reference model and the actual voltages. 4. Switch between the MidSize HIL and actual hardware. Figure 3.8: MicroAutoBox ControlDesk Interface The detailed description of the interface can be found in Chapter 4. It is important to note that the experimental setup is configured to perform specific tasks, and the ControlDesk GUI reflects the intended functions. When setting up validation for SOC estimation, for instance, injecting noise into the current sensor through the GUI becomes essential; validation for balancing, on the other hand, requires voltages of individual cells to be traced. 28

44 Figure 3.9: HIL Simulator ControlDesk Interface The HIL simulator ControlDesk interface, shown in Figure 3.9, serves the following functions: 1. Fault injection. The voltages of the batteries and the output of the thermistors can be changed to simulate over temperature, over voltage, or over temperature. 2. Simulation speed. The battery capacity can be changed through the interface to increase or decrease the C rate without changing the current profile. Because the battery capacity is modeled as an independent parameter, the battery performance is not affected significantly. 3. Reset battery SOC. The detailed description of the MidSize HIL ControlDesk interface can be found in Chapter 4. 29

45 3.6 Summary The MidSize HIL simulator and the MicroAutoBox prototype control units are the primary equipments used for MIL and HIL. Multiple Li battery chemistry are available, but the A cylindrical cells and the 20Ah prismatic modules are selected to construct battery models. The MidSize HIL and the MABX both require a ControlDesk GUI to function, the two units communicate through CAN. 30

46 CHAPTER 4 BATTERY MODEL STRUCTURE AND CONFIGURATION The battery model for the HIL experiments must meet the following requirements: 1. Real-time-capable. The model must be compact enough to run on the MidSize HIL 2. Multi-cell-capable. To test a generic BMS the number of series element must be flexible. A module balancing board can usually handle uo tp 30 cells, but a the BCM monitors data collected from hundreds or thousands of cells. This model needs to simulate multiple cells effectively. 3. Identifiable. The dspace ASM battery model has the highest fidelity, but since the parameters are based on the physical effects and chemical reactions, it is fairly difficult to characterize the cell. The equivalent circuit model, on the other hand, is much easier to identify. To meet the requirements, the first order battery model takes the structure shown below in Figure 4.1. Multiple cells are being simulated at the same time, and the simulation is based on a modified parameter matrix, where the following parameters are modified for each cell: 1. Standard first order model parameters (R0, R1, C1, charging and discharging) 31

47 2. Open circuit voltage (OCV) at different SOC. 3. Initial SOC. 4. Initial battery temperature. 5. Battery to heat sink lumped thermal resistance. Aforementioned parameters are modified through probability density functions: the normal distribution function and the chi-square distribution function. The normal distribution can modify the mean and the variance of the parameters, while the chisquare distribution can add a skewness. Figure 4.1: Battery Model Structure This battery model structure allows different scenarios to be simulated. The procedures to generate these scenarios with the battery model are listed. 32

48 4.1 General SOC Balancing Simulation This is the most common task for a MBB under normal operation condition. Since there are no defects in the batteries, a normal distribution with low parameter variation is used. 1. Determine number of cells used. Generate normal probability density function (PDF) with average of 1 and low variance. 2. Apply PDF to reference parameters, generate matrix. 3. Apply customized initial SOC, normal distribution recommended. 4. Apply uniform initial temperature. If the objective is only to test the balancing signals and observe the behavior in real time, it is recommended to decrease the battery capacity. Since the simulation is done in HIL, there is no downside of having a high C rate. 4.2 Aged Single Cell Simulation The scenario is similar to the general SOC balancing case, except one cell has faulty parameters (high internal resistance, for example). One set of the parameters in the customized matrix needs to be modified manually. It is also possible to apply a flat spread to the parameter variation matrix, i.e. having a large standard deviation to generate an outlier. 4.3 Thermal Management System Simulation The lumped thermal resistance between the heat sink and the battery is part of the parameter matrix, it is possible to simulate a scenario where a portion of the 33

49 batteries do not get proper heating or cooling. Since the internal resistance of the battery depends on temperature, a large temperature difference across the battery pack can potentially reduce the power available (See Chapter 3). The battery model is constructed so that each cell has its own temperature, which makes it possible to simulate the compounded effect of having high internal resistance variance. To simulate the performance of the thermal management system, one can either apply a distribution to the initial temperature, or a distribution to the thermal resistance. 1. Determine number of cells used. Generate normal PDF with average of 1 and low variance. 2. Apply PDF to reference parameters, generate matrix. 3. Determine number of batteries with ineffective thermal management, create PDF with high variance. 4. Generate lumped thermal resistance matrix with high PDF. 5. Apply uniform SOC. 6. Apply customized initial temperature. It is important to note that the actual BMS often has a limited number of thermistor or thermal couples, and only a few of the battery temperatures can be collected. There is no heat transfer between the battery cells either. It is possible to generate a faulty thermal couple reading. Fault injection is covered in the MIL and HIL testing section. 34

50 4.4 Increased Internal Resistance Simulation The increase in internal resistance can be caused by manufacturing defect or aging. This model structure can simulate two types of power fade: an increase in the average of internal resistances, and an increase in the variance of internal resistances Uniform Resistance Increase A uniform increase in internal resistance is often caused by aging, assuming only the internal resistance is affected. This is the most common type of resistance increase used in State of Health estimation, since the battery pack is often lumped into one large cell. To take advantage of the multi-cell capability, it is recommended to apply a low variance to the internal resistance instead of having zero variance. 1. Determine number of cells used. Generate normal PDF with average of 1 and low variance. 2. Apply PDF to reference parameters, generate matrix. 3. Determine percentage increase in internal resistance, apply new average to PDF, generate new internal resistance matrix. 4. Apply uniform initial SOC and temperature This method applies a uniform raise to the internal resistance matrix, but it does not change the relationship between internal resistance, SOC, and temperature. In other words, the lookup table for the internal resistance is still the same, but a gain is applied at the end. It is possible to use more sophisticated techniques and alter the shape of the resistance-temperature curve by making the parameter variation matrix temperature dependent. 35

51 4.4.2 Increased Internal Resistance Variance An increase in internal resistance variance can be caused by uneven aging, poor thermal management, and manufacturing defects. The average internal resistance of the module or the pack is often normal, but certain cells have much higher internal resistance than the average. The increase in variance is often overlooked in the BMS, mostly because it cannot be simulated by a single cell model. In a multi-cell model, the effect of the high internal resistance variance on pack performance can be simulated and observed by the BMS. 1. Determine number of cells used. Generate normal PDF with average of 1 and low variance. 2. Apply PDF to reference parameters, generate matrix. 3. Determine increase in variance internal resistance, apply high variance PDF, generate new internal resistance matrix. 4. Apply uniform initial SOC and temperature. Changing the shape of the probability density function can affect the simulation result. In Section 4.1, two PDFs are introduced. The normal distribution is the more common distribution with a symmetrical bell curve, and the Chi-square distribution has a skewness to the right. Chi-square distribution is recommended for scenarios with manufacturing defects, since the battery parameters have tendency to lean toward higher performance, i.e. defect cells are outliers, but good cells are expected. Normal distribution is recommended for aging and poor thermal management, mostly because of lack of data, a simpler distribution is often the default setting. 36

52 4.5 Decreased Capacity Simulation Simulation wise, capacity fade is identical to power fade. Instead of changing the parameter variation matrix for the internal resistance, the customized distribution is applied to cell capacity instead. In this particular model, capacity is not temperature dependent, therefore a change in the capacity parameters does not affect the thermal management system. 4.6 Summary The experiment utilizes a first order battery model that is compatible with Li chemistry. The battery properties are readily available. The battery model is structured as multiple cells running with the same current request, and the difference in the cell parameters are generated by a parameter variability matrix. Said matrix can be constructed by a probability density function, covered in Chapter 5. Since every parameter can be modified by said matrix, different simulation scenarios, such as SOC balancing or thermal management, are described in this chapter. 37

53 CHAPTER 5 EXPERIMENTAL METHODOLOGY This chapter covers the basic probability density function (PDF) used for this experiment first, since these PDFs are used to create voltage differences between the cells. Then it covers the current request profile used for the simulation. It also includes the step necessary to initialize the multi-cell battery model for various BMS tests. At the end it explains the hardware setup and the MIL/HIL hardware configurations. 5.1 Probability Density Functions The probability density function (PDF) used for simulation describes the variations in the battery parameters. Unlike the dspace ASM battery model, where a voltage difference matrix is used to generate the voltage imbalance, the battery model used for HIL validation simulates different cells with different parameters. Two common PDFs are used: the normal distribution PDF and the chi-square distribution PDF. The normal distribution, shown in 5.1 can be used to generate variations in battery cell capacity and open circuit voltage at a given SOC. The normal PDF is characterized by the mean and the standard deviation. Uniform aging capacity fade, for example, decreases the mean of the PDF; non-uniform aging capacity fade, on the other hand, increases the standard deviation of the PDF. 38

54 Figure 5.1: Normal Distribution Example The chi-square distribution, shown in 5.2 is skewed. The chi-square PDF is characterized by the degree of freedom, v, where the mean is v and the variance is 2v. The skewness of the PDF can be used to capture the variance of internal resistance: most cells have internal resistance value close to the mean, a few have high resistance, while no cell has extremely low resistance. 39

55 Figure 5.2: Chi-Square Distribution Example The PDFs are not tested through experimental data. The simulation simply assumes the standard deviation as a percentage, from Current Request Profile The current request profile is generated from the EcoCAR2, US06 driving profile, in charge sustaining mode to simulate driving environment. Shown in Figure 5.3, there are several spikes present during the drive cycle, caused by the vehicle supervisory controller. The maximum current draw is limited on the vehicle, thus the current profile is saturated at 200A, shown in Figure

56 Figure 5.3: EcoCAR2 US06 Current Request Profile Figure 5.4: EcoCAR2 US06 Current Request Profile Limited at 200A 41

57 With the same current profile, the simulated battery output can be compared with the data logged from the EcoCAR2 BMS to validate the battery model. 5.3 dspace MidSize HIL Setup The MidSize HIL simulator (HIL) takes the role of the battery and module balancing board (MBB). Because the HIL is configured for an ECU only, the output voltage limit is 10V and the current draw limit is 3mA. Therefore, the MBB is modeled with the HIL as a passive unit that takes command from the supervisory controller, as shown in Figure 5.5. Figure 5.5: MidSize HIL Setup The procedure utilized for setting up the HIL is shown below. 1. Update battery parameter verification matrix. 2. Initialize battery model and sensor lookup curves. 3. Update dbc file for both HIL and MABX. 42

58 4. Load dbc file with RTI CAN configuration block. 5. Change CAN bit rate to hardware or MABX bit rate. 6. Compile battery model and MBB model with MATLAB Load experiment in ControlDesk. 8. Load variable description file. 9. Change battery capacity to observe change in real time. 10. Turn off injected faults and errors in control desk The detailed ControlDesk GUI description is listed below. Figure 5.6: MidSize HIL Interface Function 1. When simulation is stopped, the SOC is reset to the default value in the Simulink model. 43

59 2. The simulated battery cell voltage. These voltages are sent to the supervisory controller through CAN. 3. The thermistor measurements can be forced through the GUI. It is important to note that the temperature sent to the supervisory controller is a voltage, not a temperature. 4. The plotter traces the voltages of all the cells. 5. In MIL simulation, the faults are injected through the GUI and sent to the supervisory controller. 6. The balancing status of the cells monitored by the balancing board. 7. The balancing commands from the supervisory controller. When the balancing circuit is being controlled by the MABX, the balancing status should be the same as the balancing command. 5.4 MicroAutoBox Setup The MABX setup process is similar to that of the MidSize HIL, as shown below. It is important to note that even though the MABX have two CAN channels, they cannot share the same CAN configuration block. In other words, it is not advised to MIL and HIL for the MBB at the same time. 1. Initialize battery model used for parameter lookup in control algorithm. 2. Initialize sensor lookup curves. 3. Update dbc file for both HIL and MABX. 4. Load dbc file with RTI CAN configuration block. 44

60 5. Change CAN bit rate to hardware or HIL bit rate. 6. Compile control algorithm with MATLAB Load experiment in ControlDesk. 8. Load variable description file. 9. Switch algorithm mode. Because the temperature sensors output voltages, the controller needs to convert them to temperature values before they are displayed on the GUI, shown in Figure 5.7. The detailed ControlDesk GUI description is listed below. Figure 5.7: MABX Interface Function 1. Mode switch. At the moment the controller has an active balancing mode and a passive balancing mode. 2. Mode status returned from the HIL or the board. 45

61 3. The MBB error LED is red when an error occurs, and it opens all the balancing gates and stop the controller. The MBB Error Code translates the errors occurred into a code that can be used to diagnose the errors. 4. Temperature values from sensors. 5. Cell voltages measured by the MBB. 6. Balancing command determined by the controller. 7. Balancing status reported by the MBB 8. Over voltage, under voltage, and over temperature errors. Another GUI is made for SOC observers in the control algorithm, as shown in Figure 5.8. Figure 5.8: SOC Observer GUI 46

62 This GUI includes the estimated SOC display and a plotter to trace estimated SOC. The under voltage, over voltage, and over temperature thresholds can be set through the new interface as well. 5.5 MIL and HIL Testing This section covers the various setup for MIL and HIL testing. Not all of the proposed tests are can be implemented due to hardware limitations. The advantages and drawbacks of each test are outlined as well Three-Level MIL Testing The three-level MIL has the same architecture as a typical BMS: cell level represented by battery cells, module level controlled by the MBB, and pack level supervised by the BCM. It is the most modular approach, and the most straight forward for testing the BMS, as shown in Figure 5.9. Figure 5.9: Three Level MIL Setup Overview However, this test requires the MidSize HIL to output cell voltages connected in 47

63 series, which can be up to 60V for a typical module. Because the voltage limit is 10V, this method requires an independent voltage amplifier to work. A different HIL simulator, SCALEXIO by dspace (shown in Figure??, is galvanically isolated can can produce higher voltage, makes it more suitable for this strategy. Figure 5.10: dspace SCALEXIO Because the limited number of outputs on the HIL unit, a three-level MIL test often has a limited number of cells (less than 20, depends on the amount of sensors). If the MBB level is lumped into the cell level, the test can allow a higher number of cells. 48

64 5.5.2 Two-Level MIL Testing The two-level MIL test assumes the balancing board is passive: it performs simple tasks such as measuring, comparing, and balancing, and it executes commands directly from the BCM. Under this assumptions, the model of the MBB is simple enough to be integrated into the cell level, as shown in Figure Figure 5.11: Two-Level MIL Setup Overview The two-level MIL setup is chosen as the primary MIL validation method for the project for the following reasons: 1. Hardware limitations. Even though the three-level MIL is ideal, the MidSize HIL unit cannot output enough current to simulate voltage draw. Because the DACs are not galvanically isolated, it is not possible to connect the cell voltages in series. Therefore the two-level approach is more applicable. 2. Number of cells. The CAN signal between the virtual MBB and the MABX can simulate a large number of cells, while the numbers of DACs and ADCs are limited by the I/O board. 49

65 3. Less controllers. Because the MBB often performs only menial tasks autonomously, it is more economical to have the MBB algorithm lumped into the HIL, instead of loading it on a separate MABX. Most dspace MIL/HIL battery simulations take the form of a two-level test, as shown in Figure The DS1006 is the MidSize HIL unit, that simulates both the virtual MBB and the battery cells; the DS4121 is the IO board that simulates the relays and high voltages, while the DS2211 is the IO board that outputs battery status through CAN. Figure 5.12: dspace Two-Level MIL Setup Overview The dspace two-level approach is more compatible with the dspace ASM battery model, where the reference cell voltage is used to generate the high voltage signal, and the delta model is used to generate the individual cell voltages. 50

66 5.5.3 HIL Testing The hardware used for the HIL validation portion are described in the Experimental Equipment chapter. Because the MidSize HIL is not suitable for battery cell voltage simulation, both the battery and MBB are replaced by hardware, as shown in Figure 5.13 below. Figure 5.13: HIL Setup Overview The configuration reflects the two-level MIL setup. The MABX is configured so that the CAN channel can be switched between the MBB and MidSize HIL easily, as shown in the MABX configuration section. After the control algorithm is validated through MIL, there can still be problems and potential risks with HIL. For example, because the MBB is drawing current through the balancing circuit, it is important to monitor the temperature of the resistor circuit, to prevent the MBB from overheating. 51

67 5.6 Summary To create variability in the battery packs, the normal distribution PDF and chisquare distribution PDF are used in the parameter variability matrix. The current profile is generated by the EcoCAR2 in a US06 driving cycle under charge sustaining operation. Because the MidSize HIL has limited output capabilities, the MBB is modeled with the battery cells, reducing the complexity of the MIL and HIL setup. Since the battery cells and the MBB are grouped, they are both replaced by hardware during the HIL phase of the project. 52

68 CHAPTER 6 BATTERY SIMULATION RESULTS 6.1 Baseline Model Simulations The baseline model used is a first order single cell model for the A123 20Ah prismatic module. It assumed that the operating condition is similar to that of the EcoCAR2, where the cooling system is connected to the vehicle radiator, and the current draw is based on the current demand from the vehicle supervisory controller, as shown in Chapter Single Cell First Order Model Simulation The simulation result for the single cell first order model is highlighted below in Figure 6.1, and Figure 6.2. The battery is maintein within 70 percent SOC and 84 percent SOC range, while the voltage fluctuates between 2.8V to 3.7V. Since the vehicle is operating under charge sustaining mode, the SOC range is narrow. 53

69 Figure 6.1: Single Cell Voltage Under Load Figure 6.2: Single Cell State of Charge 54

70 With the given current profile the SOC stays within the 70 to 80 percent window, while the voltage under load fluctuates between 2.8V to 3.65V. 6.2 Parameter Variance Simulations For all the parameter variation simulations, only the first 50 seconds of the simulation is shown. Most figures take the form of a scatter plot, where each point represent the percentage of battery cells in the pack occupying a certain bin at a point in time. For example, a blue point at time 25 second and voltage 2.902V, represents 2-5 percent of the battery pack operating at voltage, while a red point indicates a much higher percentage. The simulations are performed with a 200 cell battery pack. Because the PDFs generate random samples, the result may differ each time the modification matrix is re-evaluated. Increasing the sample size will also increase the chance of having statistical outlier. Large battery pack will generate wider voltage or SOC spreads, even though the PDFs are the same. The results are tabulated, and the nomenclatures for the result table are shown below: 55

71 Case R 0 OCV Capacity SOC 0 T 0 R t M axd StD N 1 C1 C1d Case Number Internal Resistance Open Circuit Voltage Cell Capacity [Ah] Initial State of Charge Initial Temperature Thermal Resistance Maximum Cell Voltage Devation from Mean Cell Voltage Standard Deviation Normal Distribution PDF with 1Percent Standard Deviation Chi-Square Distribution PDF with 1Percent Standard Deviation Chi-Square Distribution, Has Dependency with Other Parameters Table 6.1: Result Table Nomenclature Since the spectrum of the voltage plot represents the concentration of cell voltage bins, a histogram is included in each section, representing the instantaneous voltage concentration at a point in time Charging and Discharging Internal Resistance Normal Distribution Simulation A normal distribution is applied to the charging and discharging internal resistance only, as shown in the table below. 56

72 Case R 0 OCV Capacity SOC 0 T 0 R t MaxD[mV ] StD[mV ] 1 N N N Table 6.2: Charging and Discharging Internal Resistance Normal Distribution Results The result indicates that even though the standard deviation in internal resistance is small, the deviation in the voltage under load is high enough (17mv maximum) to affect the pack performance. With the traditional modeling approach, only the average voltage, shown in Figure 6.3 as the red trajectory, is used for the battery power calculation. The voltage difference generated by the difference in internal resistance can be used to test the robustness of the cell balancing algorithm. Proper cooling plays a big role in the voltage spread as well. Because the internal resistance is not uniform across the pack, the heat generated is much higher for cells with high internal resistance. Because the standard deviation is small, the effect of temperature variance is not significant. However, if there is no proper thermal management and heat is allowed to be built up over time, the voltage spread will be exacerbated. 57

73 Figure 6.3: Voltage Spread, Charging and Discharging Internal Resistance, 3 Percent Standard Deviation It is evident from Figure 6.3 that the voltage deviation is proportional to the current draw, since the voltage drop across the battery cell is the internal resistance times the current draw. Therefore the standard deviation calculated is unique to this drive cycle. When a new drive cycle is applied, the standard deviation needs to be calculated again. 58

74 Figure 6.4: Voltage Spread, Charging and Discharging Internal Resistance, 3 Percent Standard Deviation, Cross Section t=25[s] Charging and Discharging Internal Resistance Chi-Square Distribution Simulation A Chi-Square distribution is applied to the charging and discharging internal resistance only. Compare to the normal distribution, the chi-square distribution has a similar standard deviation, but much higher maximum deviation. It is a much more accurate representation of the actual battery behavior, where the majority of the cells are in the normal region, and a small amount of cells have extremely inferior parameters. Moreover, because the internal resistance is so high in some of the cells, there is a 59

75 Case R 0 OCV Capacity SOC 0 T 0 R t MaxD[mV ] StD[mV ] 1 C C C Table 6.3: Charging and Discharging Internal Resistance Chi-Square Distribution Results large temperature difference across the battery pack, pushing the bad cells toward even worse operating regions, widening the voltage spread. The probability to have high internal resistance cells increases as the number of the cells increases. Because of the long tail of the chi-square PDF, higher number of cells increases the voltage deviation. While this is true for any PDF, it is more visible in the chi-square PDF. 60

76 Figure 6.5: Voltage Spread, Charging and Discharging Internal Resistance, Chi- Square 3 Percent Standard Deviation Shown in 6.5, a few cells are operating out of the normal voltage envelope. Because of the shape of the PDF, the wide spread has a low concentration near the extreme values. The outliers represent the minimum voltage read by the MBB is most BMS systems. When the deviation of the minimum voltage is higher than the deviation of the maximum voltage when the battery is discharging, the chi-square is more suitable than the normal distribution for internal resistance, which is often the case. 61

77 Figure 6.6: Voltage Spread, Charging and Discharging Internal Resistance, Chi- Square 3 Percent Standard Deviation, Cross Section t=25[s] Shown in 6.6 is the instantaneous histogram of the voltage under load. There are two outliers with high internal resistances and low voltages. These outliers will become more common when the number of cells increases in the chi-square distribution Cell Capacity Normal Distribution A normal distribution is applied to cell capacitance only. As shown in the table, the cell capacity has minimal effect on the voltage under load. This is mostly caused by the OCV curve of the Li-ion chemistry: because the OCV curve is flat near the operating point. Moreover, because the energy chemistry has high capacity to begin with, the variance in cell capacity has to be greater than 10 62

78 Case R 0 OCV Capacity SOC 0 T 0 R t MaxD[mV ] StD[mV ] N N N Table 6.4: Cell Capacity Normal Distribution Results percent to have an observable effect on the voltage under load. The voltage variance caused by variation in cell capacitance is independent from the current draw, as shown in Figure 6.7. The deviation in voltage becomes greater if the pack moves across a larger SOC range, shown below in Figure 6.8when the SOC is at a minimal during the drive cycle. Figure 6.7: Voltage Spread, Cell Capacitance, Normal 10 Percent Standard Deviation 63

79 Figure 6.8: Voltage Spread, Cell Capacitance, Normal 10 Percent Standard Deviation at Minimum SOC Figure 6.9: Voltage Spread, Charging and Discharging Internal Resistance, Normal 10 Percent Standard Deviation, Cross Section t=95[s] 64

80 6.2.4 Initial Temperature Normal Distribution Simulation A temperature distribution is applied to the initial temperature alone. Case R 0 OCV Capacity SOC 0 T 0 R t MaxD[mV ] StD[mV ] N N N Table 6.5: Initial Temperature Normal Distribution Results The temperature variation has a similar effect on the voltage spread. However, considering how unpredictable the thermal environment can be, the voltage spread is extremely sensitive to the temperature variation. A 3 percent temperature standard deviation is 0.6C in the simulation, and it causes a 6mV standard deviation in the voltage. Because the thermal management system in the simulation is effective, the temperature difference decreases over time, and the voltage spread decreases over time. This simulation implies that the thermal system can directly influence the battery pack performance. 65

81 Figure 6.10: Voltage Spread, Charging and Discharging Internal Resistance, Chi- Square 3 Percent Standard Deviation Shown in 6.10, the color gradient is red and blue dominant, indicates that the voltage spread is flatter than the normal distribution applied to the temperature. 66

82 Figure 6.11: Voltage Spread, Charging and Discharging Internal Resistance, Chi- Square 3 Percent Standard Deviation, Cross Section t=25[s] Shown in 6.11, the voltage spread does not fit a normal distribution perfectly. Because the relationship between the internal resistance increase and temperature is not linear, the voltage spread does not share the same PDF as the temperature Multiple Parameter Variation, Internal Resistance Chi-Square Distribution, Independent Case The following parameter variations are applied to the model: 1. Internal resistance: chi-square PDF. 2. Cell capacity: normal PDF. 67

83 3. Initial SOC: normal PDF. 4. Initial temperature: normal PDF. The open circuit voltage distribution translates directly to voltage under load distribution, as shown in Equation The PDFs are independent from each other in the simulation. For example, a cell with high internal resistance does not necessarily have low capacity. Case R 0 OCV Capacity SOC 0 T 0 R t MaxD[mV ] StD[mV ] 1 C1 1 N1 1 N C2 1 N1 1 N C3 1 N1 1 N Table 6.6: Multiple Parameter Variations, Internal Resistance Chi-Square Distribution, Independent Case As indicated by the result the combined parameter variations have a significant impact on the voltage spread. The maximum standard deviation of the voltage under load is increased to up to 50mV. Even when the standard deviations of the parameters are only 1 percent, the minimum voltage of the pack is 60mV lower than the mean. 68

84 Figure 6.12: Voltage Spread, Combined Parameter Variation, Independent, Chi- Square 1 Percent Standard Deviation Internal Resistance Shown in 6.12, the voltage spread fits a Chi-Square distribution. The voltage standard deviation increased from below 10mV to 30mV when multiple parameter variations are applied. The simulation demonstrates how the cell voltage trajectories can deviate from the ideal voltage, even when the difference between the cells is small. 69

85 Figure 6.13: Voltage Spread, Combined Parameter Variation, Independent, Chi- Square 1 Percent Standard Deviation Internal Resistance, Cross Section t=25[s] The histogram shows a normal distribution. Because multiple parameters are affecting the cell voltage, the maximum deviation is not necessarily the worst case scenario. When the number of cells increase, it is more likely to have cells with low capacity and high internal resistance at the same time. The next subsection will demonstrate the effect on cell voltages when there is a correlation between the parameters Multiple Parameter Variation, Internal Resistance Chi-Square Distribution, Dependent Case The following parameter variations are applied to the model: 70

86 1. Internal resistance: chi-square PDF. 2. Cell capacity: chi-square PDF. 3. Initial SOC: normal PDF. 4. Initial temperature: normal PDF. The internal resistance PDF and the capacity PDF are the same. If a battery has high internal resistance, it has low capacity as well. The result is tabulated below: Case R 0 OCV Capacity SOC 0 T 0 R t MaxD[mV ] StD[mV ] 1 C1d 1 C1d 1 N C2 1 C2d 1 N C3 1 C3d 1 N Table 6.7: Multiple Parameter Variations, Internal Resistance Chi-Square Distribution, Dependent Case Compared to the independent case, voltage spread standard deviation is lower. Because two of the parameters are coupled, the spread is more concentrated. However, when an outlier exists, it deviates more from the mean compared to the independent case. 71

87 Figure 6.14: Voltage Spread, Combined Parameter Variation, Independent, Chi- Square 1 Percent Standard Deviation Internal Resistance Shown in 6.14, the voltage spread fits a Chi-Square distribution. The cluster around the mean voltage is much higher, indicates that most cell are concentrated around the average voltage. This result conforms with actual performance of automotive battery packs, where the standard deviation is low, number of outlier is low, but the voltage of the outlier is very different from the mean. 72

88 Figure 6.15: Voltage Spread, Combined Parameter Variation, Independent, Chi- Square 1 Percent Standard Deviation Internal Resistance, Cross Section t=25[s] The histogram shows a chi-square distribution similar to the independent case. It is important to note that increasing the number of batteries will increase the number of outliers and the significance of the concentration. 6.3 Summary Different parameter variabilities are simulated in this chapter to generate voltage difference. 1. Increased deviation in internal resistance increases the voltage difference when the battery pack is operating under high power condition. 73

89 2. Chi-square distribution PDF can generate outliers in the pack, and results in a more realistic simulation 3. Increased deviation in cell capacity increases the voltage difference when the battery pack is at low SOC. 4. Increased deviation in initial operating temperature has a similar effect as the deviation in internal resistance. The results is sensitive to the performance of the thermal management system. 5. When multiple parameters have variability, the voltage spread is increased, and often resembles a normal distribution. 6. Internal resistance and battery capacity should be related, since cells with high internal resistance often have low capacity. The dependency generates higher chance for outlier and larger voltage variability. The simulation result demonstrates that the proposed battery model structure is suitable for MIL and HIL BMS simulations. 74

90 CHAPTER 7 BMS HIL TESTING AND VALIDATION RESULTS The proposed experimental methodology and battery model are used to validate a balancing algorithm. A description of cell balancing is covered in Chapter 2, under the main functions of the BMS. The MIL portion of the validation has the same setup as described in Chapter 4, while the HIL portion utilizes the Wilson-Scott balancing board shown in Chapter 3. While traditional balancing algorithm compares the cell voltages, and determine the resistor balancing switches, it does not consider the difference between the open circuit voltage and the voltage under load. Because there is a voltage drop across the battery cell during balancing, the balancing often stops when the voltage under load reaches the desired threshold. When the voltage under load is within threshold, but the open circuit voltage is not, the switch changes periodically as the cell voltage oscillates between the open circuit voltage and the voltage under load. The proposed algorithm takes into account the voltage drop across the battery during balancing, i.e. uses the open circuit voltage instead of the voltage under load. Since it is often assumed the open circuit voltage is a good indicator of SOC, the new algorithm performs more consistently and more effectively. The flow chart for the balancing algorithm is shown below in Figure

91 Figure 7.1: Balancing Algorithm Flow Chart, Courtesy of Gian Luca Storti As the flow chart indicates, the battery internal resistance is fixed, even though it should be a function of SOC and temperature. Because the algorithm does not use the SOC of the cells, an average value of the internal resistance at 50 percent SOC and 20C is used. The battery cells are generated with the proposed cell structure, described in Chapter 5. Assuming most of the parameters are randomized, and there is no dependency between them, the PDFs applied are shown below: 76

92 Case R 0 OCV Capacity SOC 0 T 0 R t 1 N1 1 N1 N1 N1 1 Table 7.1: PDFs Applied, Balancing Case Study A total of 200 cells are generated, 9 of them are used for MIL and HIL. The generate initial cell voltages are shown below: Figure 7.2: Generated Cell Voltages for Balancing Case Study 77

93 Because normal distribution PDF is applied to all the parameter variabilities, the voltage result exhibits the same distribution. The standard deviation is large enough to cause cell imbalance, and small enough to be realistic. There is no outlier, thus during balancing, the cells will have reasonable voltages. 7.1 Traditional Balancing Algorithm MIL Result The MIL result of the balancing algorithm is shown in Figure 7.3. In this case, the switch of the resistor balancing circuit is based on the comparison between a battery cell with normal voltage and a cell with the lowest voltage. 1. The cyan line represents the switch of the balancing circuit. A value of 1 means the switch is closed, a value of 0 means the switch is open. 2. The blue line represents the voltage output of the cell being balanced. 3. The green line represents the voltage being used for comparison. 4. The red line represents the target voltage, which is the lowest cell voltage in the battery pack or module. 78

94 Figure 7.3: Basic Voltage Based Passive Balancing As shown in Figure 7.3, as the SOC of the cell decreases, the voltage output approaches the target voltage over time. As the cell voltage is balanced to the threshold, the balancing swtich opens to stop balancing. If the open circuit voltage is higher than the threshold, the algorithm will close the balancing switch again to start balancing. The switch changes multiple times near the end of the balancing process, i.e. the effective balancing current is reduced. 7.2 Algorithm with Wire Voltage Drop Estimation MIL Result In this case, instead of comparing the cell output voltage directly, the algorithm subtracts the voltage drop between the balancing resistor and the battery anode and cathode from the output voltage. Because the voltage used for comparison is higher 79

95 than the voltage output, the algorithm balances more aggressively than the traditional algorithm. The result is shown below in Figure 7.4. The difference between the blue line and the green line represents the voltage drop across the wire. Figure 7.4: Balancing Wire and Electronic Voltage Drop Estimation Because the voltage used for comparison is higher than the voltage under load, the number of switch change is significantly reduced, as shown as the cyan line in Figure 7.4. Since the effective balancing current is higher, the balancing time is significantly shorter than the traditional case. 80

96 7.3 Algorithm with OCV Estimation MIL Result The proposed algorithm subtracts both the voltage drop across the wire and the voltage drop across the battery from the cell voltage. If the voltage drop estimations are accurate enough, the algorithm essentially estimates the open circuit voltage of the cell being balanced. Because the lowest cell does not have a current running across it, comparing the OCVs should yield better result. Figure 7.5: Proposed Balancing Algorithm with Voltage Drop Estimation As shown in Figure 7.5, there is no change of switch once the balancing is turned off. The greenline, which represents the voltage being used for comparison, is fairly smooth, with no sudden change when the balancing starts or stops. The OCV estimation is effective. Because there is no unnessesary switch change, the effective 81

97 balancing current is higher than the traditional case. The estimated balancing time is tabulated below: Case Traditional Wire Voltage Drop Estimation Proposed Algorithm Balancing Time 11 Hours 5-6 Hours 4-5 Hours Table 7.2: Balancing Time Since the cells are randomly generated, the balancing time is not a good representation of the actual balancing time. When the algorithm takes the voltage drops into account, the effective balancing current is significantly higher and more consistent. This allows the MBB designer to reduce the nominal balancing current and reduce the maximum temperature of the balancing resistor. 7.4 Summary The validation of the balancing algorithms demonstrate that the proposed MIL/HIL setup is compatible with simple BMS algorithms. Because the voltage and balancing status of each cell can be monitored in real time, it is possible to compare different balancing strategies and improve performance. The cell model structure was able to generate voltage difference to simulate balancing condition, and since each cell has independent current profile, it is possible to add the balancing current to individual cells. 82

98 CHAPTER 8 CONCLUSION AND FUTURE WORK The work presented in this thesis demonstrates that it is possible to utilize the existing MIL/HIL hardware to construct a BMS test bench. The proposed battery model structure is capable of generating cell voltage difference based on parameter variability, and the said model is compatible with MIL/HIL. A first order battery model is used, based on existing battery parameters. Because the battery structure is scalable and multi-cell capable, it is possible to simulate balancing and advanced BMS control algorithm. Normal distribution and chi-square distribution are used to generate the cell variability, where chi-square PDF is favored to generate outliers within a module or a pack. Different combinations of parameter variations with different standard deviations are simulated to generate different output voltages. Variability of cell capacity can be used to simulate passive balancing; variability of internal resistance can be used to validate state of health estimation and active balancing; variability of battery thermal properties can be used to test the thermal management of the BMS. The MIL/HIL are modeled as a two level system: the module level balancing board is modeled with the battery cells, while the supervisory controller is modeled as a separate unit. The two level are connected through standard CAN, where the sensor outputs are sent to the supervisory controller, and the balancing commands are sent to the balancing board and battery cells. 83

99 Coupling the MIL/HIL structure with the proposed battery model, the test bench is used to validate a set of balancing algorithms. The balancing algorithms assume there is a set of resistors can be used to reduce the voltage of each individual cell, and the module or pack is balanced to the lowest voltage. Without the MIL/HIL test bench, the balancing algorithm can only be verified in either a Simulink only environment or a hardware only environment; the MIL/HIL setup allows the developer to test the algorithm during development, and implement the algorithm on a balancing board easily. The battery model can be used for different chemistries, as long as the battery can be described as a first order model. The MIL test bench is applicable to C++, MATLAB, and StateFlow based BMS algorithms, and the HIL test bench can be used to validate both module level hardware and pack level hardware. Since the test bench is CAN based, most automotive standard BMS are compatible. 8.1 Improving battery Fixture At the moment the cylindrical cells used for MIL are held by a styrofoam fixture, that has a limited operating temperature range. The batteries cannot be charged or discharged with a high C rate, limiting the testing capabilities. A new set of battery fixtures are designed to allow easy assembly and natural convection cooling, as shown below in Figure

100 Figure 8.1: Improved HIL Battery Fixture The new fixture can group nine cells into a small module. The two halves are identical, and cell terminals are exposed to allow easy assembly or disassembly. With the improved fixture, the thermistors used for the MBB are easier to install, and it is possible to monitor the temperature of the terminals directly. 8.2 Improving Simulator Efficiency The maximum population size for the MIL/HIL is limited by the computational power of the MidSize HIL unit. If the model is more efficient, more cells can be simulated at the same time. One possible solution is to derive the close form solution instead of using the numerical ODE to reduce the computational power required. If the simulation population size is increased, the test bench becomes more versatile and making large scale simulation possible. At the moment the population is limited 85

101 to 200 cells, which is typical for an electric hybrid vehicle. However, a large electric vehicle may use a much larger number of batteries. It is important to make the test bench more scalable and increase the population size. 8.3 Adding Voltage Amplifier to MidSize HIL Unit The MIL/HIL are setup as two-level structures, because the MidSize HIL cannot output high voltages to simulate battery cells in series. If a voltage amplifier can be added to the MidSize HIL, it is then possible to adopt a three-level structure to separate the MBB from the MidSize HIL. 8.4 Implementing Advance Control Algorithm At the moment the test bench is only used to test simple balancing algorithms that use a limited number of the sensors. Since the algorithm does not require high computational power, the MABX can execute the control algorithm with minimal delay or error. Kalman filter observers, for example, can be used to estimate the state of charge of individual cells to improve balancing. However, a Kalman filter observer may be too computationally demanding for the MABX prototyping unit to operate properly. Testing more advance control algorithm can make the test bench more adaptable and more versatile. 86

102 BIBLIOGRAPHY [1] Barbarisi, O.,Vasca, F., Gliemoi, L. State of Charge Kalman Filter Estimator for Automotive Batteries. Control Engineering Practice, [2] Caumont, O. Electric Vehicle Symposium. Brussels, Belgium, [3] Kim, Il-Song. The Novel State of Charge Estimation Method for Lithium Battery Using Sliding Mode Observer. Journal of Power Sources, [4] Meissner, E., and Richter, G. The Challenge to the Automotive Battery Industry: The Battery Has Become and Increasingly Integrated Component with the Vehicle Electric Power System. Journal of Power Sources, [5] Plett, G.L. Extended Kalman Filtering for Battery Management Systems of LiPB-Based HEV battery Packs. Journal of Power Sources, [6] Vetter, J. Aging Mechanisms in Lithium-Ion Batteries. Journal of Power Sources, [7] Wei, Liu. Introduction to Hybrid Vehicle System Modeling and Control. New Jersey, John Wiley and Sons, Inc, [8] Yurkovich, B.J., Guezennec, Yann Dynamic Electrothermal Battery Pack Modeling and Simulation of Pack Imbalance

103 Appendix A EDITING AND LOADING BATTERY MODEL INTO MIL/HIL Appendix A covers loading and running the Simulink/MIL/HIL files. Appendix B include the plot relevant to battery simulation. There are two main file paths: Battery Simulation and Masters. Battery Simulation includes the battery model and the scripts used for battery simulation. Masters include all the working versions of MidSize HIL and MABX ControlDesk files. The MidSize HIL template contains the same battery simulation file as the battery folder. A.1 Creating and Editing Battery Parameter Probability Density Function Under Battery Simulation or Masters/HIL Master, there are two files the initializes the battery pack. BatteryRandomizedParameterInitialization.m is the script that generates the battery parameter variability matrix, and BatteryInitialization.m is the script that generates the actual battery parameters. The battery parameters are stored under two structures in MATLAB workspace: BATT-ELEC-RND includes the capacity, the internal resistance, and the first resistor/capacitor pair; BATT-OCV- RND contains the OCV curve. 88

104 A.2 Editing Battery Parameter Matrices Most battery parameter values have three indices: the temperature, the SOC, and the cell index. The cell index, or the third index of the matrix, is directly related to the PDF applied. To modify all the cells at the same time, the parameter variability matrix need to be modified in the BatteryRandomizedParameterInitialization.m script. To modify one cell, the BATT-ELEC-RND structure of the BATT-OCV-RND structure need to be modified in the workspace or in the BatteryInitialization.m script. A.3 Loading Model in ControlDesk The procedure for loading the model in MidSize HIL and the MABX are quite similar. This section covers the prodcedure of loading the Simulink model into ControlDesk for MIL/HIL. A.3.1 Structuring the Simulink Model The master template has a structure shown in the figure below. The dspace Peripheral I/O block configures the MidSize HIL or MABX I/O, and the dynamic model block is the actual model. It is important to distinguish the difference between variable avaialbe from the model and variable available to the dspace I/O, since there is a virtual sensor block that adds noise, faults, and calibration to the signals available to dspace I/O. In ControlDesk, it is often difficult to find the right variable to link to instruments, thus it is recommended to keep the model as modular as possible. It is also important to note that labeling the name of the signals, blocks, and subsystems will make loading model into ControlDesk significantly easier. 89

105 Figure A.1: Simulink Model Structure Note the master enable constant block can be used to reset the simulation in ControlDesk. It is always considered good practice to add controller variables that affect the simulation outside of the dynamic model block and the I/O block, so they can be accessed directly in ControlDesk. The dspace block should have the form shown below: 90

106 Figure A.2: dspace I/O in Simulink The hardware interface includes the DAC and ADC, while the Protocols1 block includes the different CANs. This portion of the template is unique to the hardware. When a new HIL simulator or MABX is used, it is highly important to check the compatibility beforehand. One of the DAC channel is also used as the power supply for the MABX. When the MidSize HIL powers off, the power supply to the MABX is cut off as well. The CAN I/O is structured as below: 91

107 Figure A.3: dspace CAN I/O There are three important types of blocks in this template for CAN: the CAN setup block, the RTI CAN Demo library block, and the actual subsystems. The CAN Controller block configures the dbc file used for CAN, as shown below. 92

108 Figure A.4: dspace CAN I/O The CAN setup block determines which dbc file to use, thus it controls the variable description, the CAN transmit and receive block, and the type of signals available to the simulator. It is recommended to keep one version of dbc file, and load the same dbc for both MidSize HIL and MABX. The RTI demo block includes all the real time interface blocks used for the experiment: 93

109 Figure A.5: Real Time Interface Block Set Primarily the RTICAN Transmit and RTICAN Receive are used for the I/O. The blocks can be selected directly from the Demo block and dragged into the model. The input or output of the CAN block depends on which CAN message is loaded, which is configured through the CAN setup block, as shown below. 94

110 Figure A.6: Real Time Interface Block Set Once all the CAN blocks are properly configured, and model has been properly loaded, it is then possible to compile the Simulink model and load it into MidSize HIL or MABX. 95

111 Appendix B SUPPLEMENTARY BATTERY SIMULATION FIGURES Figure B.1: Voltage Spread, Charging and Discharging Internal Resistance, 1 Percent Standard Deviation 96

112 Figure B.2: Voltage Spread, Charging and Discharging Internal Resistance, 1 Percent Standard Deviation, Cross Section t=25[s] Figure B.3: Voltage Spread, Charging and Discharging Internal Resistance, 2 Percent Standard Deviation 97

113 Figure B.4: Voltage Spread, Charging and Discharging Internal Resistance, 2 Percent Standard Deviation, Cross Section t=25[s] Figure B.5: Voltage Spread, Charging and Discharging Internal Resistance, 1 Percent Chi-Square Deviation 98

114 Figure B.6: Voltage Spread, Charging and Discharging Internal Resistance, 1 Percent Chi-Square Deviation, Cross Section t=25[s] Figure B.7: Voltage Spread, Charging and Discharging Internal Resistance, 2 Percent Chi-Square Deviation 99

115 Figure B.8: Voltage Spread, Charging and Discharging Internal Resistance, 2 Percent Chi-Square Deviation, Cross Section t=25[s] Figure B.9: Voltage Spread, Cell Capacitance, 5 Percent Normal Deviation 100

116 Figure B.10: Voltage Spread, Cell Capacitance, 1 Percent Normal Deviation Figure B.11: Voltage Spread, Initial Temperature, 1 Percent Normal Deviation 101

117 Figure B.12: Voltage Spread, Initial Temperature, 1 Percent Normal Deviation, Cross Section t=25[s] Figure B.13: Voltage Spread, Initial Temperature, 2 Percent Normal Deviation 102

118 Figure B.14: Voltage Spread, Initial Temperature, 2 Percent Normal Deviation, Cross Section t=25[s] Figure B.15: Voltage Spread, Combined Parameter Variation, Independent, Chi- Square 2 Percent Standard Deviation Internal Resistance 103

119 Figure B.16: Voltage Spread, Combined Parameter Variation, Independent, Chi- Square 2 Percent Standard Deviation Internal Resistance, Cross Section t=25[s] Figure B.17: Voltage Spread, Combined Parameter Variation, Independent, Chi- Square 3 Percent Standard Deviation Internal Resistance 104