Prof. Lei Feng Mikael Hellgren. Alexander Bessman Prof. Oskar Wallmark

. Examensarbete MMK 2017: 165 MDA 605 Uppvärmning av litiumjon batterier med växelström vid temperaturer under noll grader Dieter Henz. Godkänt Examinator Handledare 2017-08-30 Prof. Lei Feng Mikael Hellgren Uppdragsgivare Rúdi Soares Alexander Bessman Prof. Oskar Wallmark Kontaktperson Rúdi Soares Sammanfattning. Att göra trafiken eldriven är ett tydligt mål hos många myndigheter i hela världen. Lokalt sett är elektriska fordon utsläppslösa och kan därför bidra till att förbättra luftkvaliteten i snabbt växande städer. Trots att det finns många anledningar till att stödja elektriska fordon accepterar konsumenterna dem inte på en önskad nivå. En grund till den låga populariteten är det dåliga förhållandet mellan körsträcka och inköpskostnad. En möjlighet till att förbättra förhållandet är att minsta antalet komponenter i batteripaketet, eftersom kostnaderna då sjunker samtidigt som körsträckan ökar. Moderna batterier i elektriska fordon bygger på litiumjon teknologin. Denna typ av batterier har ett begränsd brukstemperaturområde och behöver därför värmas upp vid kalla förhållanden. I många elektriska fordon blir batteriet uppvärmt via ett eget värmeelement som byggs in i batteripaketet. Detta värmeelement orsakar mer komplexitet och behöver energi för att fördela värmen över batteriet. I den här studien undersöks möjligheten att använda en ny metod som kallas växelströmvärmning. Denna metod utnyttjar värmeutvecklingen i batteriets inre impedansen för att värma upp batteriet inifrån. Arbetet innehåller både en praktiskt del och en del med simulationer. En litteraturstudie över moderna uppvärmningsmetoder är presenterad i början av rapporten. Efter det finns en impedansmätning av två battericeller. Därefter testas olika driftpunkter för att bestämma hur växelströms frekvens och amplitud påverkar uppvärmningseffekt i en cell. Resultatet av impedansmätningen används för att bestämma driftpunkter för ett följande experiment och härleda ett ekvivalentskrets-schema för en battericell. Studiens slutsats är att det i allmänhet är möjligt att värma litiumjonbatterier med växelström. Låga frekvenser har bättre uppvärmningseffekt på samma strömamplitud och höga strömamplituder har bättre effekt än låga strömamplituder. På grund av detta visar studien också att det behöver genomföras flera undersökningar som fokuserar på hur frekvensen är kopplad med åldringen av batterikapaciteten vid låga temperaturer. Avslutningsvis presenteras en idé som visar hur det är möjligt att höja strömmen i batteriet för att förbättra uppvärmningseffekten. ii

iii

. Master of Science Thesis MMK 2017: 165 MDA 605 Heating Effect of Alternating Current on Lithium-Ion Batteries at Subzero Temperatures Dieter Henz. Approved Examiner Supervisor 2017-08-30 Prof. Lei Feng Mikael Hellgren Commissioner Rúdi Soares Alexander Bessman Prof. Oskar Wallmark Contact person Rúdi Soares Abstract. Electrification of transport is a distinct goal of many authorities. Electric vehicles are locally emission-free and can help to improve the air quality in cities. Despite several stimuli to promote electric vehicles, customers do not accept them at the desired level. One reason is the poor ratio between driving range and purchase cost. One possibility to tackle this poor ratio is to reduce complexity and weight of a battery pack, which helps to reduce its costs while increasing the driving range at the same time. Recent batteries in electric vehicles use lithium-ion technology. This type of batteries have a limited operating temperature, which requires pre-heating in cold ambient. In many electric vehicles pre-heating is realized with a heating element in the battery pack. This heating element does not only bring along additional weight but requires extra energy for distributing the heat. The idea of this study is to propose a more efficient heating method which is called alternating current (AC) heating. This method uses the internal impedance of a battery to generate dissipative heat, heating it up from inside. This work consists of practical measurements and experiments on a cell level as well as simulations. First, a literature research is conducted to present the latest heating technologies. Then, an impedance measurement is performed. The result of the impedance measurement is used to define the operating points for the following experiment. The experiment shows how AC frequency and amplitude influence the heating effect. The results from the impedance measurement and the experiment are further used to construct a cell simulation model that connects the thermal properties with the electrical properties. After verification of the model, the simulation is expanded to a battery pack level. The conclusion of the study is that it is generally possible to heat lithium-ion batteries with alternating current. Lower frequencies provide better heating effect at the same current amplitude and a higher current amplitude gives better heating effect than lower amplitudes. Therefore, the study also shows that further research is necessary to determine how the AC frequency affects aging of lithium-ion batteries at low temperatures. Finally, a method for increasing the current amplitude is proposed. iv

v

Dedication I dedicate this work to my mother and my father who passed away in 2016. So many beautiful pieces of my life which I can not count and not describe are thanks to them. Acknowledgement This work was partially sponsored by Scania AB. Therefore I thank the company and especially Pontus Svens as representative. I thank Rúdi Soares, Alexander Bessman and Oskar Wallmark for this interesting thesis project, for introducing me to the topic as well as for their positive support throughout the whole thesis. Furthermore, I thank Niclas Johannesson and Jesper Freiberg for their practical support during the laboratory work. Last but not least I thank my family, girlfriend and other loved ones who have supported me from far away during my whole master studies. vi

vii

List of Figures 1.1 Comparison of car models with purely electric and internal combustion engine........................................ 2 2.1 Equivalent circtuits of a battery.......................... 15 3.1 Battery with connections and part of insulation................. 17 3.2 Schematic overview of the experimental setup................. 19 3.3 Insulated batteries inside climate chamber.................... 22 3.4 Applied equivalent circuit model......................... 22 3.5 Model of battery pack with controller...................... 25 4.1 Bode graph of battery impedance......................... 27 4.2 Nyquist graph of battery impedance for cold temperatures.......... 27 4.3 Nyquist graph of battery impedance for higher temperatures......... 28 4.4 Surf graph of real part of battery impedance.................. 29 4.5 Surf graph of imaginary parts of battery impedance.............. 30 4.6 SS18 warmup during experiment at 0.5 C.................... 31 4.7 SS20 warmup during experiment ay 0.5 C.................... 32 4.8 SS18 warmup during experiment at 1 C..................... 32 4.9 SS20 warmup during experiment at 1 C..................... 33 4.10 SS18 warmup during experiment at 1.25 C................... 34 4.11 SS20 warmup during experiment at 1.25 C................... 34 4.12 Series resistance and inductance......................... 35 4.13 First parallel R-C branch.............................. 36 4.14 Second parallel R-C branch............................ 36 4.15 Nyquist graph with both measured data and equivalent circuit I....... 37 4.16 Nyquist graph with both measured data and equivalent circuit II...... 38 4.17 Model comparison................................. 39 4.18 Model comparison................................. 40 4.19 Model comparison................................. 41 4.20 Model comparison................................. 42 4.21 Model comparison................................. 43 4.22 Model comparison................................. 44 4.23 Temperature increase with controller in either battery pack or single battery 45 4.24 Controlled current for temperature increase in either battery pack or single battery........................................ 46 viii

LIST OF FIGURES 6.1 Circuit with power supply, resonance capacitor and battery equivalent circuit 54 6.2 Current from power supply and current in resonance circuit......... 54 ix

List of Tables 2.1 Resulting temperatures using different heating methods with different energy inputs with data from [32].......................... 15 3.1 Specifications of the lithium-ion battery used in this study.......... 17 4.1 Qualitative overview of the temperature behavior of the real (R), imaginary capacitive (X C ) and imaginary inductive (X L ) part of the measured impedance....................................... 30 4.2 Comparison of the four electro-thermal battery simulations.......... 45 4.3 Time and power consumption to heat a battery pack.............. 46 5.1 Overview of battery heating methods...................... 50 6.1 Parameters of the equivalent circuit....................... 55 x

Nomenclature Abbreviations AC Alternating current BMS Battery managment system C rate Current that charges or discharges a battery within one hour CAN Controller area network CRG Current ripple generator CV M Constant voltage method DC EIS EV LIB Direct current Electrochemical impedance spctroscopy Electric vehicle Lithium-ion battery P CM Pulsed current method P CS Phase change slurry SEI Solid electrolyte interface SOC State of charge SOH State of health Equation Variables T q Temperature difference [K] Heat generation rate [J] ω Angular frequency [ 1 s ] C Capacitor [F] C th Heat capacity [ J K ] F (s) Controller transfer function xi

LIST OF TABLES G(s) k i k p L P R System transfer function Integral gain Proportional gain Inductor [H] Power [W] Imaginary reactive impedance [Ω] R th Thermal resistance [ K W ] X Z Real active impedance [Ω] Impedance [Ω] xii

Contents 1 Introduction 1 1.1 Background..................................... 1 1.2 Problem Description................................ 2 1.3 Research Questions................................. 3 1.4 Outline........................................ 3 1.5 Limitations...................................... 3 1.6 Methodology.................................... 3 2 Frame of Reference 5 2.1 Battery Principles.................................. 5 2.1.1 Electrochemical Cell............................ 5 2.1.2 Lithium-Ion Batteries............................ 5 2.1.3 Causes of Aging.............................. 6 2.1.4 Temperature Dependent Aging...................... 7 2.1.5 Causes of Heat............................... 7 2.2 Battery Heating Procedures............................ 7 2.2.1 External Battery Heating.......................... 7 2.2.2 Internal Battery Heating.......................... 9 2.3 Battery Modelling.................................. 15 3 Implementation 16 3.1 Methodology.................................... 16 3.2 Battery Specifications................................ 16 3.3 Electrochemical Impedance Spectroscopy.................... 18 3.4 Experimental Setup................................. 18 3.5 Simulations..................................... 21 3.5.1 Electrical Model............................... 22 3.5.2 Thermal Model............................... 23 3.5.3 Temperature Controller.......................... 24 4 Results 26 4.1 Measurements.................................... 26 4.1.1 Electrochemical Impedance Spectroscopy................ 26 4.1.2 Experiment................................. 30 4.1.3 Simulation Model.............................. 35 4.1.4 Heat Simulation............................... 38 xiii

CONTENTS 4.1.5 Temperature Controller.......................... 45 4.2 Analysis....................................... 47 4.2.1 Electrochemical Impedance Spectroscopy................ 47 4.2.2 Experiment................................. 47 4.2.3 Simulation.................................. 48 4.2.4 Controller.................................. 48 5 Conclusion 49 5.1 Answers to Research Questions.......................... 49 5.1.1 What are the existing methods to heat LIBs?.............. 49 5.1.2 To what extent is it possible to heat LIBs with current?........ 50 5.1.3 What are the limitations of the AC heating method.......... 51 5.2 Ethical Aspects................................... 51 6 Outlook 53 6.1 Usability in Real Applications........................... 53 6.2 Future work..................................... 53 Bibliography 56 A Battery pack with temperature- and current controller 59 xiv

Chapter 1 Introduction 1.1 Background Electric vehicles offer many advantages. High efficiency, little noise disturbance and local zero-emission are only a few of them. Even so, electric vehicles are not nearly as popular as their conventional equivalent. Statistical data from the year 2016, provided by the German Federal Motor Transport Authority Kraftfahrt-Bundesamt [1], supports that statement. Less than 2% of all newly registered vehicles in Germany were equipped with alternative propulsion, which also counts in hybrid vehicles. Of all cars with alternative propulsion only 23% were purely electric, which results in a market share of 0.3% purely electric vehicles. This can have many causes like missing charging infrastructure, too long charging time and little driving range combined with higher purchase cost to name only a few of them. This disadvantage becomes more visible when comparing car models that are available with both purely electric and combustion propulsion like VW e-golf [2] and Golf 1.0 TSI [3], Ford Focus Electric [4] and Focus 1.6 Ti-VCT [5] as well as Smart Fortwo Coupé Electric Drive [6] and Fortwo Coupé 1.0 [7], all in their most basic version. Figure 1.1 shows the ratio: The electric version of a car costs double when bought newly, while offering only about a quarter of the driving range. 1

1.2. PROBLEM DESCRIPTION 40 000 35 000 electric combustion 1200 1000 electric combustion 30 000 25 000 20 000 15 000 10 000 5 000 Range in km 800 600 400 200 0 VW Golf Ford Focus smart fortwo 0 VW Golf Ford Focus smart fortwo Figure 1.1: Comparison of car models with purely electric and internal combustion engine In order to increase the popularity of electric vehicles the driving range has to be increased, while the purchasing cost has to be reduced at the same time. Both factors are directly linked to the battery. Storing electric energy in batteries is expensive and one of the main reasons why so little capacity is installed. Currently, the main battery technology used is lithium-ion. 1.2 Problem Description Lithium-Ion Batteries (LIBs) lose capacity much quicker when used at low temperature compared to usage at room temperature. A rule of thumb says, that LIBs are at the end of their lifetime when they have decreased to 80% of their original capacity. Therefore, it would be desirable to avoid operation at cold temperatures. In cold ambients, this would require preconditioning before usage. This could lead to a longer lifetime of the batteries and therefore reduced costs as well as the environmental impact. If in addition a proper operating temperature could be reached without requiring additional heating elements in the battery pack, costs could be reduced at the same time as the driving range would increase due to a lighter battery pack. Since every battery carries an inherent inner ohmic resistance, LIBs generate heat when electric current is flowing into or out of the battery. As under normal operation this heat generation is considered as losses. At cold temperatures this effect can be harvested to heat up the battery from the inside. By applying alternating current (AC), current would flow over the inner resistance, causing ohmic heat but it would not change the state of charge (SOC). If on top of that, the frequency of the alternating current is too high for the chemical reactions in the cell to start, this procedure would not cause aging. 2

CHAPTER 1. INTRODUCTION 1.3 Research Questions In this work, answers are sought for the following questions. What are the existing methods to heat LIBs? To what extent is it possible to heat LIBs with current? What are the limitations of the AC heating method? 1.4 Outline Before the practical activities, a literature study will be presented. The first part of the literature review explains the fundamental principles of LIBs. The second part is dedicated to state of the art research about heating procedures for LIBs. Battery principles will be studied in standard literature books. Heating procedures and current achievements in this discipline will be summarized from various research articles. To achieve a broad and objective overview, studies from multiple research groups with different approaches will be looked into. The practical part is divided in three parts. Firstly, an electrochemical impedance spectroscopy (EIS) is carried out at different temperatures to see how the cell impedance evolves over temperature. The outcome of the EIS will be used to identify the parameters for the following experiment. Secondly, an experiment is conducted. This is to identify the ability to raise a cell s temperature with current. Therefore, the battery cells are insulated to reach a more realistic behaviour. For this experiment, a test bench is set up to test the impact of both AC frequency and amplitude on the heating effect. Finally, a simulation is performed. The goal of this simulation is not only to predict the cell heating under different circumstances but also to draw conclusions about a realistic battery pack for real applications. Therefore, an electric battery model is connected with a thermal battery model. For the electric model, data from the EIS is used and for the thermal model, data from the experiment is used. 1.5 Limitations This project will be conducted on two single lithium-ion cells. A whole battery pack, as it is used in vehicles, is normally composed of many cells and a battery management system (BMS), mounted into a battery box. This changes the temperature behaviour compared to a single cell. Due to practical aspects, such as current output of the supply, this study starts at the cell level. After simulation models for the same case are verified, they can be expanded to a more realistic case by extrapolation of the gathered data. 1.6 Methodology The research fashion of this work is qualitative and will be conducted as critical case study. Generally, in a qualitative study, a phenomenon is investigated to gain more detailed knowledge about it. The idea of a critical case study is to consider unfavorable 3

1.6. METHODOLOGY conditions and proof a certain capability under these unfavorable conditions. In this way a general conclusion can be drawn if the study is successful despite the limited number of samples. The qualitative fashion in this work is a result of the number of samples. There are two cells of the same kind in this study. To ensure data consistency the result of one sample verifies the result of the other sample if equal. The phenomenon in this work is heat generation from internal losses. Naturally, losses are tried to be kept as small as possible. In the batteries case this is done by minimizing the inner impedance. Therefore, these cells are ill suited for being heated by internal losses. Furthermore, the used cells have high thermal capacity, resulting in longer warm up time and large surface area, which leads to more heat loss towards the cold ambient. In this way, it is possible to proof the general feasibility of the AC heating method if it can be proven to work with the described cells. 4

Chapter 2 Frame of Reference 2.1 Battery Principles 2.1.1 Electrochemical Cell In general an electrochemical cell can act either as a galvanic or as an electrolytic cell. Which type it is acting as, depends on the mode of operation. A galvanic cell works by equalizing a thermodynamic imbalance between the two electrodes. A galvanic process is therefore spontaneous and electrons are moving from the negative to the positive electrode. Electrons transport the electric charge outside of the cell. Ions transport charge inside the cell and can move from the positive to the negative electrode or in the opposite direction, depending on their charge. Even both movements can happen at the same time, if ions of both negative and positive charge are present in the cell. During the galvanic process net electric energy can be extracted, as it is done in a fuel cell or while discharging a battery. The opposite is true for an electrolytic cell; it does nothing on its own, but reactions can be forced by inserting electric energy. In a fuel cell this process is described as extracting hydrogen as well as oxygen from water and in a battery as charging. [8] In theory, this process could be repeated unlimitedly. 2.1.2 Lithium-Ion Batteries As Palacín [9] states, the term battery is, even in academia, equivalently used as the term cell. This is common practice, despite battery originally describes a stack of cells. Batteries are divided into primary and secondary batteries, where primary types are nonrechargeable and secondary types are rechargeable [10] [9]. Every LIB consists out of two electrodes, electrolyte and a separator. An electrode itself consists of a chemical structure and a current collector. The chemical structure is metal oxide at the positive electrode and graphite at the negative electrode. These materials are the reaction partner for the lithium ions. The electrodes act as a host for the lithium, which moves in between the two electrodes when the battery is charged or discharged. The electrode materials are electrically conductive, enabling the electrons to move through them to reach the current collectors. The current collectors are of highly conductive metal such as aluminum or copper, which collect the electrons from the chemical structure (or releases them into it) and connects the cell to the externally connected electrical load. As 5

2.1. BATTERY PRINCIPLES electric current is flowing outside of the cell, the same charge has to be transported inside the cell, to keep the balance. Inside the cell this electric charge is transported with Li+ ions, which separate from one electrode, move through the electrolyte and join the other electrode. For this process, the electrolyte acts as an ion conductor, similar as metal acts as an electric conductor for electrons, and is non-conductive for electrons. To prevent the two electrically conductive electrodes from touching, a separator is inserted in between them. This separator lets the lithium ions pass but poses a mechanical, electrically nonconductive barrier to prevent a short-circuit inside the cell. [11] [12] Charging or discharging the battery means to transport charge between the electrodes. In the case of LIBs, lithium is bound at the negative electrode, in the layered graphite when it is charged. When the cell is discharged, the lithium is bound in the tunneled structure of the metal oxide at the positive electrode. When the battery is used as an electric source, lithium at the negative electrode loses one electron and dissolves into the electrolyte. The electron moves through the graphite, into the current collector and goes through the externally connected electrical load. Inside the cell, the lithium ion moves through the electrolyte, passes the separator and intercalates into the tunneled structure of the metal oxide at the positive electrode, where it joins with the electron again. This process can continue as long as there is still lithium bound at the negative electrode. Once all lithium has moved to the positive electrode, the battery is discharged and cannot provide any more electrical current. [12] In order to charge a battery, an external current source is applied to the electrodes. This external electrical source reverses the charging process. All lithium is bound at the positive electrode when the battery is discharged. The lithium atoms lose one electron, become a Li+ ion and dissolve into the electrolyte. The electron moves through the metal oxide and the current collector towards the external source and then to the negative electrode. Inside the cell, the lithium ion again moves through the electrolyte, passes the separator and intercalates into the layered structure of the graphite at the negative electrode, where it obtains the missing electron again. [12] There is several types of LIBs, all with different advantages and disadvantages. Detailed information about the type used in this study can be found in table 3.1 in section 3.1. 2.1.3 Causes of Aging Aging is a general description for the decrease of a battery s performance. The performance can be diminished by either capacity fade or impedance raise. Capacity loss mostly originates in either loss of cyclable lithium or dissolution of electrode material into the electrolyte. [13] [14]. Loss of cyclable lithium is mainly caused by the formation of a solid electrolyte interface (SEI) at the electrode electrolyte interface [15], which happens mostly at the first cycle of a battery. The formation of a SEI layer around the electrodes is actually desired, since it slows down the process of the other two causes of aging (electrode-material dissolution into the electrolyte and impedance increase). [16] [17] This SEI layer can receive cracks and will form a new layer at this crack. This formations cause the SEI layer to grow, which uses up more cyclable lithium as well as increasing the cell impedance. In order to increase the electrical conductivity in the positive electrode, graphite is doted into the metal oxide. At high currents these graphite particles can crack and lose connection with the surrounding metal oxide, causing higher impedance, too. 6

CHAPTER 2. FRAME OF REFERENCE [12] [18] 2.1.4 Temperature Dependent Aging Another process, that causes loss of cyclable lithium is lithium plating. It happens at the negative electrode while charging, when the diffusion rate of the lithium ions is limited, as it is at low temperatures [19] [20] and especially below 0 C [21]. As consequence of the low temperature, the potential of the graphite is decreased. After the lithium ions have passed the SEI layer, they should react with the graphite but cannot due to the decreased potential of the graphite electrode. Anyhow, the lithium ions still receive an electron but since they are not bound to the graphite structure, they depose on top of the graphite as lithium metal. Those lithium ions are now lost for storing electric energy [17]. Deposited lithium ions can even grow a dendrite which can be compared to a spike, which grows from the negative electrode towards the positive electrode. This dendrite can even perforate the separator and cause an electric short circuit inside the battery [12]. 2.1.5 Causes of Heat As already mentioned, LIBs generate heat during any operation. This heat has three sources: charge transfer overpotential, mass transfer overpotential and ohmic losses. Charge transfer overpotential results in reaction heat. It occurs when the Li+ ions perform a chemical reaction at the electrodes. Heat from mass transfer overpotential is caused by friction when the Li+ ions move through the electrolyte. Ohmic heat is generated when electrons move through the electrode structure and the current collectors. [12] As mentioned in section 2.1.4, performing charging actions at low temperatures will cause capacity loss. Therefore the AC frequency must be chosen high enough to prevent the chemical reactions from happening. 2.2 Battery Heating Procedures As described in numerous articles, low temperatures not only affect a LIB s aging and safety during operation but also decreases the efficiency and limits the power output due to an increased inner resistance. Therefore, various research has been conducted on battery warming. Generally spoken, the methods for battery heating can be divided into internal and external heating. In the following section an overview of current heating methods is given. 2.2.1 External Battery Heating Heating with the Air Conditioning System Zou et al. [22] are investigating how a vehicle s air conditioning system can be used to control both the passenger cabin temperature and the battery pack temperature at the same time. A vehicle s air conditioning system is well known from current vehicles. In this case, however, the refrigerant cools/heats both heat exchange media, the cabin s air 7

2.2. BATTERY HEATING PROCEDURES and the water which runs through the battery pack. The system was tested experimentally with a battery pack of lithium-ion cells with a mass of 16kg. It was found that the regular air conditioning system was not capable of heating the battery pack at 20 C on its own. It is explained that the decrease of the refrigerants ability to transport heat and the decrease of the scroll compressor at low temperatures make it necessary to include a positive temperature coefficient (PTC) heating element. This is a simple ohmic heater, which helps the refrigerant to heat up to a suitable temperature under which the known air conditioning system works properly. With this additional PTC element, it was possible for the authors to heat the battery within approx. 800s with a power input of 1.5 kw. It makes sense that it was necessary to add an ohmic heater in the heat exchange medium of the battery. Water freezes below 0 C and cannot be pumped through pipes in that case. This would mean that the air conditioning system first would have to defrost the water before it could be used for heating. Heating using Heat Pipes Wang et al. [23] use heat pipes to transport heat into batteries. In their setup they do not use actual batteries but a metal plate, with a specific thermal capacity that is similar to the one of lithium-ion batteries. Heating block and "batteries" are both placed in a freezer. The heat pipes reach from inside of the heating block into the cell. It is not mentioned what kind of heat transport medium is inside the heat pipes but it freezes at a certain temperature. The authors show that it is possible to heat the batteries with this procedure under different circumstances. The circumstances are different temperatures in the heating block (20 C and 40 C). These circumstances are compared to each other by the time it takes to rise the cell temperature from 15 C or 20 C to 0 C. The rise time to 0 C differs from 300 s (heater at 40 C, start temperature 15 C) to 1500 s (heater at 20 C, start temperature 20 C). The authors state that it was possible to start the system even when the heat transport medium was frozen initially. They further state that heat pipes are a very efficient method to transport heat. They however do not state how much energy is used to maintain a temperature of 20 C or 40 C in the heating block in an environment that has 15 C or 20 C. Thermodynamic Assessment of Different Heating Methods In the study of Zhang et al. [24] three heating methods are compared and evaluated based on the amount of energy they require to heat up a battery pack. The authors call the methods "direct cabin air blow", "PCS cycle" and "refrigerant circulation". PCS stands for phase change slurry and is a mixture of water and micro-encapsulated PCM (phase change material). In all three methods the heat exchange medium inside the battery pack is air, which is circulated with a fan. The battery pack itself is insulated from the ambient air with a box. The difference in the methods is how the air in the battery pack gets heated up. With direct cabin air blow, a channel between the cabin and the battery pack is engaged. Warm air from the cabin is directed into the battery pack and heats the battery. 8

CHAPTER 2. FRAME OF REFERENCE PCS cycle means that there is a pipe system installed, in which the slurry circulates between the passenger cabin and battery pack. In the warm cabin the PCM inside the slurry changes from solid to liquid phase, absorbing heat. In the battery pack the slurry changes phase from liquid to solid releasing heat. The refrigerant method makes use of the air conditioning system. Heat is absorbed from ambient air and released in the condenser. In this case two condensers are included in the closed circuit. One in the passenger cabin and one in the battery pack. In the study a 27.7 kwh battery, consisting of 2400 lithium-ion 18650-type cells, is looked into. The authors conduct a thermodynamic assessment with both the 1st and the 2nd law of thermodynamics to evaluate which method is the most efficient in heating this battery. The authors find that the refrigerant method and the PCS cycle cause the same extra power load at any operating point. For mild heating, blowing air from the cabin causes the least power to transport heat to the battery. However, if the ambient conditions become more extreme and the need for heating increases, the required power for air circulation increases much faster than the required power for refrigerant or PCS circulation. These results are intuitive as it is generally known that heat can be transported more efficiently in liquids as in air, due to higher relative thermal capacity. However, the authors do not state how much energy is required to produce the heat. This is an important factor as the passenger cabin cannot be assumed as an infinite source of heat. 2.2.2 Internal Battery Heating Heating Element Inside a Battery Wang et al. [25] state correctly in their paper that there is no external heat transport medium required if the battery is self-heating. In their study the authors have built a prismatic cell with an additional nickel foil of 20µm thickness inside. The foil is placed behind the negative electrode from the positives electrode perspective and electrically connected to the negative electrode inside the cell. On top of the cell the foil has its own terminal, which the authors call activation terminal. There is also a switch between activation and negative terminal, which can be used to short-circuit those two terminals. The load is always applied between positive and activation terminal. If the cell needs to be heated, the switch is opened and the current flows from the positive terminal through the load to the activation terminal through the foil and then reaches the negative electrode. In this way the foil acts as an ohmic resistance and dissipates heat into the electrolyte. Once the cell has reached a proper temperature, the switch can be closed and the foil is bypassed. The authors say it was possible to heat the cell from 20 C to 0 C within 20 s and from 30 C to 0 C within 30 s while reducing the SOC 3.8% and 5.5% respectively at 1C DC discharge. Heating with Vehicle s Flattening Capacitor, Interter and Electrical Machine Baba and Kawasaki [26] are studying how existing hardware in an electric or hybrid vehicle could be utilized to heat up the battery. The system the authors look into is general motor control circuit. It is composed of the battery as power source in parallel with a 9

2.2. BATTERY HEATING PROCEDURES smoothing capacitor, the three phase inverter with free wheel diodes and a three phase electrical machine in Y-connection. The three phase inverter is connected to the battery at two points, each individually controllable by a relay. The first relay connects the three phases with the positive terminal of the battery. The second relay is special and connects only one phase between the two transistors with the positive terminal of the battery. By disconnecting the first relay and connecting the second relay the inverter can be used to form a buck-boost converter by utilizing the parallel capacitor and the coils of the electric machine. If the transistors in the three phases are called U+ V+ W+ and U- V- W-, then the second relay connects the positive terminal of the battery with the part between U+ and U- of the inverter. The buck-boost procedure then happens in four sequences: Transistors V- and W- are closed, the remaining are open. The inductor is charged. Here, the inductor is comprised as follows: 2 phases are in parallel, the parallel branch is in series with the third phase of the electrical machine. All transistors are open, the inductor builds up a higher voltage in order to maintain the current flow. The voltage in the inductor rises above the capacitor s voltage and current flows over the freewheeling diodes from the inductor into the capacitor. The capacitor has now a higher voltage than the battery. Transistors V+ and W+ are closed, all others are open. The capacitor now acts as a source; battery and inductor are in series, the capacitors voltage discharges into the battery. Transistor V+ and W+ are opened, V- and W- are closed. The voltage in the inductor rises above the battery voltage in order to maintain the current flow. Current continues to flow into the battery until the inductor is discharged. The described charge/discharge currents cause ohmic losses in the battery impedance but the battery is not actually discharged, since the frequency of these currents is too high to oxidize the lithium. Another benefit is that not only the battery heats up, also the motor heats up due to the unavoidable ohmic resistance in the windings. This also causes a temperature rise in the surrounding oil, which is especially beneficial for hybrid vehicles. The authors find that with this method it was possible to heat the system within five minutes from 20 C to 0 C. The procedure was tested on a real setup. In this specific study, the authors used a Toyota Prius to prove their concept. It is worth mentioning that this method presents a very efficient heating method since it only alternating energy between several electric reservoirs, while the occuring losses are acutally intended and therefore of nearly 100% efficiency. Furthermore, the described method can be applied in both cases, with and without a connected charger and as well over a wide battery SOC level. Pulsed Current and Constant Voltage Heating In their study, Mohan, Kim, and Stefanopoulou [27] connect an electrical model with a thermal model of a battery and compare two different heating methods. The authors call them constant voltage method (CVM) and pulse current method (PCM). CVM is DC discharge only with varying current at a constant voltage. PCM cycles energy between the battery and an external energy storage, resulting in alternating current. In the study 10

CHAPTER 2. FRAME OF REFERENCE the authors formulate an optimal control problem with different penalty on energy loss for PCM. The optimal control problem penalizes large energy transport in PCM, since it causes higher losses and requires a larger and therefore more expensive external energy storage. The frequency applied is 10 Hz. The authors find that CVM is faster in any case, even without any penalty in PCM. In other words, this means that PCM can not, even with very high pulse amplitudes, heat a battery as fast as CVM. This can be explained with the lower charge transfer in the battery. With CVM not only dissipative energy from the battery causes heat but also the chemical reactions contribute to the temperature rise, while with PCM almost only ohmic losses cause the temperature rise. It is worth mentioning, that the goal of the authors was to increase the battery s power output. This is achieved by rising the temperature. However, they did not focus on capacity loss and how it would decrease the battery life. Heating with Alternating Current at High Frequency Stuart and Handeb [28] only take AC-heating into consideration and look into the influence of different C-rates, SOC-levels and ambient temperatures. The authors made experiments with lead-acid and 16 series connected 6.6 V nickel metal hybrid (NiMH) batteries. The lead-acid battery was used in a simple setup to prove that the concept of AC heating is actually functional. It was connected to 60 Hz as it is the common grid frequency in American countries and therefore easy to implement. After that the authors built a setup, which could deliver 10-20 khz and connected the NiMH battery and tested the heating effect at different amplitudes, SOC levels and ambient temperatures. In contrast to the previous study, the authors state that it was not possible to achieve a temperature rise with only DC. They also mention that charging at cold temperatures causes permanent capacity loss and even gassing. Gassing bloats the battery, therefore charging a cold battery should be avoided. Even 1 s was time enough to start oxidizing 60 the lead, causing the same effect as pure charging, just at a slower rate. Therefore aging was possibly caused, too. For the NiMH battery, 10-20 khz was applied with current in the range of 60 80A rms (corresponding to 9.23-12.3C). In a first test the authors found a higher current to be more effective. It took 3 min to raise the temperature from 20 C to +20 C with 80A rms and 6 min with 60A rms. This result was confirmed in another test where the ambient temperature was chosen to be 30 C. In this case it also took 3 min to raise the temperature from 30 C to +20 C with 80A rms (12.3C) but 7 min with 60A rms (9.23C). Both tests were done at a SOC-level of 55%. In another test the authors compared different SOC-levels at 60A rms (9.23C). They found that a higher SOC-level is beneficial for heating purposes. While it took 8 min at 25% SOC for the temperature to raise from 30 C to +20 C it took only 6 min at 75% SOC. During the tests the authors as well measured the battery impedance. They found that the impedance in the NiMH batteries decreased from 1.3Ω to 0.4Ω over the mentioned temperature range. Heating with Alternating Current at Mid-Frequency Zhu et al. [29] perform simulations and real measurements with and 18650 2.3 Ah lithiumion battery. Different AC amplitudes and frequencies as well as different AC waveforms (sinusoidal and rectangular) were used. The equivalent circuit is assumed to an ohmic resistance in series with two parallel RC (ohmic resistance in parallel with capacitor)- 11

2.2. BATTERY HEATING PROCEDURES branches. The values for the components were chosen in such way that the equivalent circuit had the same electrical behaviour as the cell. The authors did AC impedance measurements and fitted the data into their equivalent circuit. The figures of the measurement reveal the impedance change over temperature. The real resistance varies between 0.35Ω and 0.02Ω between 25 C and +40 C. In a first step the authors simulated 300 Hz in a range of 1.5 A to 10 A with both sinusoidal and rectangular waveform. In a second step they simulated 8 A and 10 A in a frequency range of 1 Hz to 600 Hz with both waveforms. Result of the first experiment was that, as expected, higher current amplitudes had better heating effect than lower amplitudes and rectangular waveforms also deliver faster temperature rise. In detail Sinusoidal: 24 C to +3 C with 10 A amplitude (3.0C), 24 C to 5 C with 6 A amplitude (1.8C) and no temperature rise with 1.5 A amplitude (0.5C) Rectangular: 24 C to +23 C with 10 A amplitude (4.3C), 24 C to +10 C with 6 A amplitude (2.6C) and no temperature rise with 1.5 A amplitude (0.6C)) This is as expected since power dissipation in an ohmic resistance follows P diss = i 2 rmsr and rectangular waveforms have a higher RMS value at the same peak amplitude. The result of the second experiment was that 30 Hz delivered the highest temperature rise with both waveforms, sinusoidal and rectangular. Continuing from that, neighbouring frequencies (10 Hz and 80 Hz) delivered similar results as 30 Hz while 1 Hz, 300 Hz and 600 Hz gave worse results. Further, 10 A (3.0C for sinusoidal, 4.3C for rectangular) resulted in a higher temperature rise as 8 A (2.5C for sinusoidal, 3.5C for rectangular) and rectangular in a higher temperature than sinusoidal. The best result was achieved with 30 Hz, 10 A rectangular waveform, where the temperature rose from 24 C to +25 C within 600 s. The used equivalent circuit model was able to predict the resulting temperature very well with less than 1 C error. Further findings were that at 1 Hz/10 A/rectangular waveform the capacity faded by 3.7% and by 6.6% after 20 and 40 heating cycles respectively compared to the initial capacity. No capacity fade was observed at higher frequencies. Heating with Alternating Current at Low Frequency Zhang et al. [30] have done a similar study as explained in the preceding study. A model of a 18650 lithium-ion battery was developed according to data gathered from an electrochemical impedance spectroscopy (EIS). The assumed circuit was a series connection of an inductance, an ohmic resistance, a R/Q (ohmic resistance in parallel with a capacitor but with compressed semi-circle in the nyquist diagram) and a Q/R-W (compressed semi circle capacitor in parallel with series connection of ohmic resistance and warborg impedance) branch. The study was carried out with sinusoidal waveform and 3 A to 7 A (1C to 2.25C) at 0.1Hz / 1Hz / 10Hz. The ambient temperature was always assumed to be 20 C and two different heat insulations (R th1 = 21.6 W and R m 2 K th1 = 15.9 W ) were m 2 K applied. As in previous studies, the authors found that higher currents work better for heating purposes compared to lower currents. They further state that 0.1 Hz and 1 Hz achieve similar results, while 10 Hz works less well. As expected, a higher thermal resistance allows a better heating than the insulation with better heat conductivity. During heating the real part of the battery impedance decreased from 0.55Ω at 20 C to 0.15Ω at 12

CHAPTER 2. FRAME OF REFERENCE +5 C The authors also mention that the heat distribution in the cell is very uniform and state that if the battery is discharged, the risk of lithium plating is low. Comparison of Heating Methods Ji and Wang [31] compare three internal and one external heating method. The authors call the internal methods self-internal-, mutual pulse- and AC-heating. The external method is called convective heating. All data is gained from simulations. The authors compared the time that was necessary for heating the battery from 20 C to +20 C. Self-internal heating can be compared to DC discharge, where the battery simply powers an electrical load and heats up from the dissipated power in the battery impedance. For mutual pulse heating the battery is divided into two parts, interconnected with a buck-boost converter. Energy is cycled back and forth between the two parts and heat is dissipated in the internal impedance. For AC-heating an external power source is connected to the battery and injects AC. The battery heats up from the dissipated power of the impedance. During convective heating the battery powers an external ohmic resistance and a fan. In this case the battery heats from in- and outside due to internal losses and heated ambient. For self-internal heating the authors simulate different discharge rates and find that the higher the current, the faster the temperature rise. More precisely they found the rise time from 20 C to +20 C to be 132 s / 228 s / >420 s for 2C / 3C / 4C-rate respectively. In the case of mutual pulse heating the authors only give voltage levels for the discharge pulse and no C-rates. However, the pulses are of 1 s length and it is stated, that the more the discharge voltage level differs from the cell voltage, the higher the current and the faster the temperature increase. For 2.2V / 2.4V / 2.8V discharge voltage it took 80s / 120s / 204s respectively to heat from 20 C to +20 C Considering AC heating the authors assume an AC-voltage source with a 1 V sinusoidal amplitude around the cell voltage. Different frequencies are investigated and, in contrast to other studies, it is found that higher frequency heats the battery faster. It is stated that the dissipated power follows P diss = U RMS 2 and with constant voltage and increasing frequency the real part of the battery impedance decreases, which maximizes R the power. This is in fact not applicable, since in reality the limiting factor usually is the maximum available current. It then follows P diss = i 2 RMSR and with constant current it is obvious that the dissipated power decreases with increasing frequency. With AC-heating the temperature rise time found was 80s / 175s / 270s / 276s / 340s for 1kHz / 60Hz / 1Hz / 0.1Hz / 0.01Hz. As in [30] 1 Hz and 0.1 Hz show barely any difference. Convective heating was carried out at different power levels. The total power was regulated with the ohmic resistor, which produces ohmic heat, while the fan to convect the air was assumed to take 3 W constantly. The different power levels added up to 3.9C / 2.9C / 2.3C, which resulted in a heating time of 80s / 140s / 200s respectively. In a comparison, the required energy for heating is compared and how much the battery-self powered methods have decreased the SOC. Obviously for the AC-heating 13

2.2. BATTERY HEATING PROCEDURES method no discharge rate can be found, since the power comes from an external source. The authors found the following decrease of the SOC level for the three methods: self-internal heating: 14% - 22.5%, depending on the discharge rate (4C - 2C) mutual-pulse heating: 5% convective heating: 7.5% The authors find the mutual-pulse heating to be most suited as a battery self-powered heating method. That is not only due to the fewest power consumption but also due to the more uniform heat distribution. For the AC-heating method the authors suggest 60 Hz frequency. Even though it did not deliver the fastest rise time, they state that it would be most easy to implement, since the grid has the same frequency and therefore only the voltage level would have to be adjusted. Vlahinos and Pesaran [32] compare six different heating methods, which are listed below. The authors created several simulations and then compared the different methods regarding their heat uniformity, energy demand and rise time. The battery pack was constructed of six modules. Each module was encapsulated in a plastic module. All modules were held together by another plastic case. Internal core heating: Heating the battery core with dissipative electric power from constant discharge External jacket heating: Using an ohmic heater, that surrounds each module Internal jacket heating: same as external, but ohmic with an ohmic heater around each cell internal fluid heating: heating with a warmed fluid, that flows around the cells AC-heating: same as internal core heating, but with alternating current (only briefly discussed) As shown in table 2.1 the authors simulated all heating methods with the same magnitude of energy input and compared the temperature rise after 600 s. It can be seen that some methods deliver a more uniform heat distribution among the cells compared with others. The authors state that the core and internal heating methods result in the most uniform heat distribution. The internal core heating showed the fastest temperature rise and was also among the most efficient in terms of energy usage, together with internal jacket heating and internal fluid heating. The authors also found that it is possible to heat with an external AC source. 14

CHAPTER 2. FRAME OF REFERENCE Method 1.09 Wh 2.90 Wh 4.71 Wh 6.53 Wh Internal core h. 35 C 26 C 17 C 8 C min: min: External 37.8 - C 32 - C jacket h. max: max: 37.3 C 24 C Internal jacket h. Internal fluid h. min: 38 C max: 37 C min: 34 C max: 32.5 C min: 30 C max: 28 C min: 27 C max: 23 C 38 C 34.5 C 30.5 C 27 C Table 2.1: Resulting temperatures using different heating methods with different energy inputs with data from [32] 2.3 Battery Modelling As discussed in the beginning, a simulation model is constructed as well in this work. It is possible to capture the electrical dynamics of a battery with an equivalent circuit based on passive components and a voltage source within certain boundaries. Many different equivalent circuits are proposed in literature, all with different advantages. Rahmoun and Biechl [33] and He, Xiong, and Fan [34] have compared several models with respect to accurate prediction of the batteries electrical behaviour and simplicity of the model. They found that the most simple model (Rint-model) is not capable of capturing AC dynamics properly. Further, they both conclude that the double polarization model gives a good balance between simplicity and accuracy within the desired bandwidth. R E R E R C1 R C2 C 1 C 2 U OCV U Bat U OCV U Bat (a) Rint model (b) Double polarity model Figure 2.1: Equivalent circtuits of a battery 15

Chapter 3 Implementation In this chapter, the implementation of the practical work is explained. That covers the electrochemical impedance spectroscopy, the construction of the setup and the realization of the simulation. 3.1 Methodology The goal is to prove that it is possible to internally heat the chosen LIBs by injecting AC. Despite several studies already have proven this concept to work (compare section 2.2.2) it is of interest to investigate further under worse conditions to generalize this procedures validity. Many studies used the very common cylindrical cell type 18650 (r=9.3mm, h=65.2mm) which have an approx. 20 times lower volume than the cell used in this study (see dimensions in table 3.1). Furthermore, Zhu et al. [29] found the real part of their cell impedance varied between 0.02Ω and 0.35Ω and Zhang et al. [30] found it even to be between 0.15Ω and 0.55Ω. The cells in this study show a much lower impedance as can be found in section 4.1.1. While a low battery impedance paired with high specific capacity is desirable for normal operation as it increases both energy density and operating efficiency, it is disadvantageous in terms of heating purposes. In this study two lithium-ion cells of the same kind are tested. 3.2 Battery Specifications The type used in this study is LiNi1 3 the positive electrode is Ni1 3 Mn1 3 Co 1 3 Co 1 3 O 2 (NMC) [18] [35]. The active material at Mn1 3 O 2, the material at the negative electrode is graphite [12]. Unlike the common 18650 type batteries, the batteries in this study are of prismatic shape. The dimensions and further technical data is provided int table 3.1. More detailed information about the impedance is presented in section 4.1.1. Two of these batteries are used and all measurements are done on each battery individually. The batteries are called SS18 and SS20. 16

CHAPTER 3. IMPLEMENTATION Parameter Value Technology LiN i 1 M n 1 Co 1 O2 3 3 3 Nominal voltage 3.67 V (@ 40% SOC) Voltage range 3.45 V - 4.13 V Capacity 28 Ah Max. charge current 104 A (3.7C) Max discharge current 330 A (11.8C) Weight 0.757 kg Length x width x height 148.6 x 27.0 x 91.6 mm3 Operating temperature 40 C to +70 C Table 3.1: Specifications of the lithium-ion battery used in this study Figure 3.1: Battery with connections and part of insulation As the electrodes are swelling when lithium intercalates into them, the batteries need to be compressed. This is done with each battery individually with two pressure plates of steel. Since steel has a high thermal capacity, two wooden press boards are inserted between the battery and the pressure plates to thermally insulate the battery and be able to observe its true temperature response. Inside the battery there are three K-type thermocouples integrated: One beneath the positive terminal, one between the two electrodes in the center of the cell, one in the center of the cell but between one electrode and the outer wall of the battery. A fourth temperature sensor is of Pt100 type and mounted outside of the battery on the negative terminal. The electrical connections at the terminals are, as follows, from bottom to top and can be compared in fig. 3.1. Negative terminal: live ring cable lug, electrically isolating material, ring cable lug with Pt100, ring cable lug for voltage sensing, washer, nut. The 17

3.3. ELECTROCHEMICAL IMPEDANCE SPECTROSCOPY positive terminal is connected similarly but without the cable lug for temperature sensing. There are five 6mm 2 stranded cables connected to the live cable lug to allow high currents. In order to have more realistic temperature dynamics the batteries are insulated in so called expanded poly ethylene (EPE) foam with thermal conductivity of 0.036 W [36]. m K 3.3 Electrochemical Impedance Spectroscopy The electrochemical impedance spectroscopy is a measurement technique that measures the battery impedance as function of frequency. This data can either be presented in magnitude and phase or as real and imaginary part of the impedance. Here, a Zahner IM6 KG [37] was used to do the measurement. To achieve high accuracy, the so called four-terminal sensing method is applied. In this method, there are different cables for carrying current and sensing voltage. Therefore a current induced voltage drop over the two current carrying cables is not influencing the voltage measurement. Additionally, the cable pairs are twisted to minimize electromagnetic interference. The batteries are placed inside a climate chamber with temperature control. Additionally, the chamber temperature is monitored, but not recorded, with a redundant thermometer inside the test chamber. Before the measurement is started, the batteries are mounted into the pressure plates, charged to their nominal voltage of 3.67 V (+/- 0.1 V) and kept in this condition over the whole procedure. The frequency span is chosen from 100 mhz to 10 khz. A lower limit would have prolonged the measurements unnecessarily since several cycles at each frequency are undergone during one measurement. The meaningfulness of a higher upper limit would have been questionable as the battery has a very low inductance which is very difficult to capture, since the cables also show inductive behaviour at high frequencies, even with the taken precautions. There are seven measurements at different temperatures done with each battery individually. Before one measurement is executed, the batteries are kept at the specific temperature for at least two hours to allow the temperature distribution become homogeneous in the whole body. The measurement result is presented in section 4.1.1. 3.4 Experimental Setup The experimental setup is adapted from a more extensive study, done by Bessman et al. [38] and Soares et al. [39]. In fig. 3.2 the alignment of the following units is visualized. Current Ripple Generator (CRG) Debugger for CRG Power Supply (24 Vdc) for CRG Computer with LabView NI cdaq 9174 - chassis for compactdaq I/O modules NI9861 - compactdaq CAN module NI9216 - compactdaq Voltmeter module 18

CAN NI9861 12x Volt. NI9215 4x Temp. NI9217 4 x USB CANH & CANL CHAPTER 3. IMPLEMENTATION NI9217 - compactdaq Thermometer for PT100 module Agilent 34970A - Multimeter for thermocouple temperature measurement Agilent 34902A - 16 channel Multiplexer card Power Supply (230 Vac) for Computer, cdaq and Multimeter First order filter with 14.6µH inductance Lithium-ion battery - 28 Ah prismatic cell PT100, 1x - Temperature sensor, mounted at negative terminal K-type, 3x - thermocouple temperature sensor, mounted inside the cell Lead-acid battery (12 V), 2x - power buffer for DC/AC converter TDK Lambda Z+ (0-160 Vdc, 0-4 A) - power supply for DC/AC converter Power Supply 24 VDC Debugger + T TI µc CRG DC/AC converter L1 L2 L3 Power + Supply 24 VDC (adjustable) T PC Filter 2 x + - Lead-Acid Battery + - Power ~ Supply 230 VAC T NI cdaq 9174 PT 100 Li-Ion Battery 3 x K Agilent 34970A Figure 3.2: Schematic overview of the experimental setup The Current Ripple Generator (CRG) controls the current applied to the LIB. The current can be set to AC, DC or a combination of both, from 0 Hz 1.5 khz with sine, triangle or rectangular waveform. The desired current is sent to the CRG via the controller area network (CAN) protocol, together with the propotional integral (PI) controller values kp and ki. The CRG is connected to the 24 Vdc power supply, debugger, controllable power supply TDK Lambda Z+ and lead-acid battery, filter and the NI cdaq. On its TI microcontroller runs C-code which can be flashed with the debugger. The C-code consists of two loops. The faster loop, the current controller, runs at 70 khz rate and follows the internal current reference by controlling the pulse width modulation (PWM) duty cycle. 19

3.4. EXPERIMENTAL SETUP It acts after the principle read-compute-act, where in this case "reading" means measuring the present current, "compute" means performing the control loop computations and "act" means adjusting the PWM duty cycle accordingly. There is only one PWM signal which all three branches of the three leg inverter follow at the same time. This simultaneous operation is necessary to enable higher currents than in a single leg. The converter transforms the connected constant voltage from TDK Lambda Z+ and lead-acid battery into an alternating current for the LIB. The two lead-acid batteries and the TDK Lambda Z+ power supply are connected in parallel. One reason is to be able to allow backward currents that can flow back into the lead-acid battery and would not be allowed by the power supply. The other reason is to be able to temporarily provide high currents while still maintaining a stable voltage. The power supply is connected to the PC via USB. This supports remote control of current and voltage limitation. The 1st order low-pass filter transforms the rectangular, high-frequency PWM signal into the actually desired sine waveform. It is made of an in-line inductor. Initially, a 2nd order filter with additional capacitor after the inductor was planned to have even better filtering and smoother waveform. That was not possible, since the current in the capacitor acted as additional load for the inverter and caused a current error in the CRG during a test run. The rack NI cdaq exchanges data with the computer during the experiment via USB. Three modules are mounted into the rack: NI9861 for CAN communication with the CRG, NI9216 for voltage measurement and NI9217 for temperature measurement with the PT100 on the negative terminal of the LIB. CAN communication is supplied with the same 24 Vdc as the CRG is. The two wires terminated with a 120Ω resistance at each end. The voltage measurement is done with two channels, one for each battery. In this way it is not necessary to switch the cables when switching between the batteries. The temperature measurement is done with a four terminal sensing in two channels. The current source is internal and supplies 1 ma. The Agilent 34970A is a high precision multimeter and used for temperature measurement in this study. It is necessary to use a high precision multimeter since the signal of the thermocouples is in the range of microvolts which is several magnitudes less than the error of the NI9216 voltmeter unit. Furthermore it delivers a temperature directly instead of a voltage. The 16 channel multiplexer Agilent 34902A allows simultaneous measurement of all six thermocouples from the two LIBs and accounts for the voltage error induced from the cable to the clamp of the multiplexer. The computer (PC) has two purposes in this setup. Software for the CRG can be modified with Code Composer and flashed on the CRG. Additionally, the correct functioning of the CRG can be monitored in the debugger mode. Further it runs LabView which is the interface to the setup. All experimental parameters can be monitored, adjusted and recorded. When started, it initializes the USB communications to power supply TDK 20

CHAPTER 3. IMPLEMENTATION Lambda Z+, rack NI cdaq and multimeter Agilent 34970A. After initialization it runs in an endless loop until stopped. In this loop it performs several functions which can be summarized as follows. When the program is stopped, the loop is exited and the communication ports are closed. Read Check from NI cdaq from Agilent 34970A TDK Lambda Z+ input values from user interface if batteries are inside voltage limitation for CAN-error (error in CRG) Decide if operation is safe to continue Compute necessary values Send to NI cdaq to TDK Lambda Z+ The recorded data is: Date, time CAN data Battery Temperature from thermocouples Battery voltage Battery temperature from Pt100 The LIBs are placed in a climate chamber during the entire experiment. The climate chamber runs during the whole experiment and maintains 20 C. Since the batteries are insulated, at least 10 h passed between end of one experiment and start of the next. The alignment can be seen in fig. 3.3. The temperature has to be set manually and is cross checked with a separate thermometer inside the chamber. 3.5 Simulations It is of general interest to have a simulation model to reduce the effort of building a test setup. In this case the heating effect of the battery shall be predicted under different circumstances. Therefore, an electrical and a thermal model of a LIB need to be designed and connected. In this way the dissipative power of the ohmic elements in the electrical model are used as input for the thermal model. 21

3.5. SIMULATIONS Figure 3.3: Insulated batteries inside climate chamber 3.5.1 R Electrical Model L RC1 RC2 E The modelled impedance Zmodel is constructed of one ohmic resistor RE, one inductor LE and two R/C-branches RC1 /C1 C1 C2 and RC2 /C2 in series as shown in fig. 3.4. U UOCV Bat Further there is an open circuit voltage (OCV) UOCV and a battery voltage UBat. The battery voltage and OCV are equal if Figure 3.4: Applied equivalent circuit model there is no load connected to the battery. If there is a load connected then UOCV > UBat and if the battery is charged then UOCV < UBat. As presented in section 4.1.1 the measurements show an inductive part. That is why an inductor is added to the equivalent circuit from section 2.3. The measured impedance in (3.1) consists of a real and imaginary part R and jx. The imaginary part can be positive or negative if the impedance is partly inductive or partly capacitive. (3.1) ZEIS = R + jx Zmodel = RE + jωle + RC1 RC2 + 1 + jωc1 RC1 1 + jωc2 RC2 (3.2) The electrical model is obtained with help of an optimization function. In contrast to Li et al. [40], in this study the simple absolute difference instead of the squared difference between measured and modelled impedance is taken into account, as it is presented in (3.3). 22

CHAPTER 3. IMPLEMENTATION min f opt (ω) = ω max ω min Z model Z eis (3.3) With f opt (ω) as optimization function to be minimized within the frequency range from the lower frequency ω min to the upper frequency ω max, Z model as impedance of the equivalent circuit and Z eis as measured impedance with real and imaginary part from the EIS. This procedure is implemented in Matlab with the fmincon() function [41] and repeated 14 times, once for each battery at each temperature, with the limitation, that the component values can only be positive. Each of the 14 iterations produces a set of those elements which fit the measured impedance of one cell at a specific temperature. The fit is done in a range between 10 Hz and 10 khz. This is to improve matching of the model with the measured impedance by looking only at a limited frequency range. 3.5.2 Thermal Model In order to model the temperature behaviour, the following thermal model consisting of heat flow (ohmic losses in the battery), thermal capacity (battery body) and thermal resistor (insulation). The heat flow splits into two streams: one to heat the battery body (thermal capacity) and one that flows over the thermal resistor to the cold ambient air (thermal resistor) [31]. q = C th d T dt + 1 R th T (3.4) With q as heat generation rate, C th as total thermal capacity of one battery which is constant over temperature, T as temperature difference between ambient and cell and R th as thermal resistance of the insulation. To determine the thermal resistance, final temperature, applied RMS-current and mean real impedance from the experiment are inserted in (3.4) and the dynamic part neglected. This can be verified by analytically computing the thermal resistance from material properties of the insulation material and the dimensions of the box built from it. For a state space model the temperature difference is assumed as state. The input is heat flow and equal to the dissipated electric losses which originate solely from the real part of an impedance. This real part of the impedance is a function of frequency and temperature. q = P diss = Re(Z) i 2 rms (3.5) With P diss as dissipated power from a circuit, Re(Z) as ohmic part of an impedance and i rms as RMS current. Combining and rearranging (3.4) and (3.5) leads to the state space model as presented in (3.6). d T dt = 1 C th R th T + 1 C th Re(Z) i 2 rms (3.6) With A = 1 C th R th and B = 1 C th. Both measured and modelled impedance are implemented in two different electrothermal models. One approach is to build the circuit in the PLECS blockset in simulink. 23

3.5. SIMULATIONS In this blockset it is possible to include passive components in a thermal environment to obtain the dissipated power. The model automatically transforms these electrical losses into a heat generation rate. Another approach is to create a state space model as done in (3.6) and simulate in simulink. The two different simulation environments under usage of two different circuits lead to four individual simulation models. They are compared with respect to accuracy and computational effort in section 4.1.3. 3.5.3 Temperature Controller The most accurate and efficient simulation model is extended with a simple PI temperature controller. Two types of batteries are compared: a single battery and a battery pack. Data for the battery pack is extrapolated from the single battery. It is assumed that all cells with mean voltage of 3.2 V are connected in series to reach 600 V. This 186 cells would have 16.7 kwh storage capacity which is comparable to the installed capacity in the fully electric e.go Life with 120-170 km range [42] 1. All cells reach a combined thermal capacity of 130.2 10 3 J. The cells are assumed to be aligned such to fit in a car and to reach K the minimum possible surface area as a pack. From (3.4) follows the transfer function G(s) in (3.7) with heat flow which is equal to ohmic losses (compare (3.5)), as input and temperature increase as output. This means that the PI-controller gives power as a control variable. This can again be controlled with an additional current controller which ensures the correct current for the desired power. The entire block diagram can be found in appendix A. Normally the current controller is fast and accurate enough to be neglected in slow dynamic systems. Therefore, only the temperature controller is implemented here. Both, transfer function (3.7) and controller (3.8) are connected in a closed loop system G c as in (3.9). R th G(s) = sc th R th + 1 (3.7) F (s) = k p s + k i (3.8) G c (s) = F (s)g(s) 1 + F (s)g(s) = α s + α (3.9) α is the bandwidth and corresponds directly to the rise time t r : α = ln(9) t r. With α chosen to be 3 min, following relation for the control gains are found. k p = ln(9)c th 180 k i = ln(9) 180R th These of course vary if a single battery or a battery pack is considered. Furthermore, to have a more realistic scenario, the current is limited to 100 A as otherwise cabling would become too bulky in a real application [43]. In order to maintain proper control, even if the current is limited for a long time, anti-windup functionality is included, too. The simplified model without current loop is presented in fig. 3.5. 1 Compact model Life from the German startup E.GO mobile AG, located on the campus of the RWTH Aachen 24

CHAPTER 3. IMPLEMENTATION Battery total ohmic resistance # cells -K- -K- EIS: Re(Z(1kHz)) 1-D T(u) power To Workspace4 Controller -C- 1 s Integrator2 energy To Workspace5 T_amb T_ref T_ref -K- K_p max. 100A ctrl current Product heat power -Kthermal capacity 1 dt/dt s Integrator delta T Battery temp. bat_temp To Workspace -K- K_i 1 s Integrator1 heat flow battery -> ambient -Kthermal insulation 1/K_p anti-windup current To Workspace1 Figure 3.5: Model of battery pack with controller 25

Chapter 4 Results In the first part of this chapter, the measurement results of the EIS and the experiment are presented as well as a comparison between the simulation and the experiment s presented. In the second part, the results are analyzed and discussed. 4.1 Measurements 4.1.1 Electrochemical Impedance Spectroscopy There are several ways of displaying the measured battery impedance. Each one is suited to point out a specific detail of the measurement. Firstly, the Bode graph in fig. 4.1 provides an overview of the absolute impedance s variation with temperature with respect to frequency. Secondly, the Nyquist graphs present the complex impedance at different temperatures: In fig. 4.2 30 C, 20 C, 10 C and in fig. 4.3 0 C, +10 C, +20 C, +30 C. Finally, the surf graphs fig. 4.4 and fig. 4.5 show the isolated real and imaginary parts respectively, with respect to both temperature and frequency. As presented in the Bode graph in fig. 4.1, the total impedance changes differently at certain frequencies with temperature. The largest temperature impact can be observed at frequencies below 100 Hz. Here, the impedance increases by several degrees of magnitude from +30 C to 30 C. With increasing frequency, the impact of temperature on impedance decreases. Above 1 khz barely any difference between low and high temperatures can be observed. From the phase plot it can be seen that temperature impact only takes place for negative phases, which indicate a capacitive impedance. The zero phase happens at lower frequencies for high temperatures ( 300 Hz at +30 C) and at higher frequencies for low temperatures ( 1300 Hz at 30 C). Furthermore, minimum impedance shows a similar temperature behavior as the zero phase, although they do not occur at the same frequency. The minimum impedance at +30 C occurs at 1 khz and the minimum impedance at 30 C occurs at 2 khz. 26

CHAPTER 4. RESULTS abs Impedance Z ( ) Phase ( ) 1 10-1 10-2 10-3 10-4 10-1 10 0 10 1 10 2 10 3 10 4 Frequency (Hz) 90 60 30 0-30 -60-90 10-1 10 0 10 1 10 2 10 3 10 4 Frequency (Hz) Figure 4.1: Bode graph of battery impedance SS18@30 C SS18@20 C SS18@10 C SS18@0 C SS18@-10 C SS18@-20 C SS18@-30 C SS20@30 C SS20@20 C SS20@10 C SS20@0 C SS20@-10 C SS20@-20 C SS20@-30 C 30 C 20 C 10 C 0 C -10 C -20 C -30 C The impedance increase at lower temperatures becomes more obvious on the linear scale of the Nyquist graph in fig. 4.2. For low temperatures, the bow on the negative imaginary plane is so dominant, that only 30 C, 20 C, 10 C can be shown properly in this figure. The bow is similar to a parallel R/C branch half-circle, but compressed. From 0 C to 10 C, from 10 C to 20 C and from 20 C to 30 C, the bow triples in size with every 10 C temperature decrease. -30 C -20 C -10 C 0 C 10 C 20 C 30 C 0.07 -Im(Z) ( ) 0.06 0.05 0.04 0.03 0.02 0.01 0 SS18 @30 C SS18 @20 C SS18 @10 C SS18 @0 C SS18 @-10 C SS18 @-20 C SS18 @-30 C SS20 @30 C SS20 @20 C SS20 @10 C SS20 @0 C SS20 @-10 C SS20 @-20 C SS20 @-30 C 0 0.03 0.06 0.09 0.12 0.15 Re(Z) ( ) Figure 4.2: Nyquist graph of battery impedance for cold temperatures A similar compressed semi-circle can be observed at higher temperatures in fig. 4.3. 27

4.1. MEASUREMENTS It can be observed that after the imaginary part reaches a minimum on the right end of the bow it increases again. The mismatch of zero phase and minimum real part of the impedance can be observed in this figure as well. At the point, where the imaginary part becomes zero, the real part is not yet at its minimum. The minimum real has its minimum slightly in the positive imaginary plane. This minimum is in the range of 0.7mΩ at +30 C and 1.5mΩ at 30 C. This is remarkably small compared to other studies, as mentioned in section 2.2.2. In the positive imaginary plane, the inductive part shows the same shape but at different offsets. -30 C -20 C -10 C 0 C 10 C 20 C 30 C 3 -Im(Z) (m ) 2 1 0-1 -2 SS18 @30 C SS18 @20 C SS18 @10 C SS18 @0 C SS18 @-10 C SS18 @-20 C SS18 @-30 C SS20 @30 C SS20 @20 C SS20 @10 C SS20 @0 C SS20 @-10 C SS20 @-20 C SS20 @-30 C -3 0 1 2 3 4 5 6 7 Re(Z) (m ) Figure 4.3: Nyquist graph of battery impedance for higher temperatures The real part of the impedance shows partly high temperature dependence and partly low temperature dependence, as shown in fig. 4.4. Similarly to the previous figures, the graph can be divided at around 1 khz. Above this frequency, the real impedance is nearly temperature independent. The minimum is located slightly above 1 khz and rises marginally towards 10 khz. Below 1 khz, the real part shows increasing temperature dependence with lower frequency. From little temperature variance at 1 khz, the real part changes from 1mΩ at 30 C to more than 100mΩ at 30 C at 100 mhz. 28

CHAPTER 4. RESULTS 10-1 Re(Z) ( ) 10-2 10-3 20 10-1 1 10 1 30 10 2 10 3 10 4 Frequency (Hz) Figure 4.4: Surf graph of real part of battery impedance 0 10-30 -20-10 Temperature ( C) In fig. 4.5 capacitive (negative) and inductive (positive) imaginary part are split up to left hand and right hand side on the figure due to the zero-crossing, which cannot be displayed on a logarithmic scale. Above 1 khz the imaginary part becomes inductive and increases with increasing frequency. The inductive part does not show a temperature dependence. However, the capacitive imaginary part on the left hand side shows even more temperature dependence than the real part of the impedance. Similar to the real part, the capacitive imaginary part changes more and more with temperature, the more the frequency is below 1 khz. The most significant change can be observed at 1 Hz, where the absolute capacitive imaginary part rises from magnitude 10 4 at +30 C to to magnitude 10 1 at 30 C. 29

4.1. MEASUREMENTS -10-5 Capacitive 3 10-3 Inductive -10-4 2 10-3 Im(Z) ( ) -10-3 -10-2 Im(Z) ( ) 1 10-3 -10-1 30 20 10 0-10 -20-30 10-1 1 10 1 10 2 10 3 10 4 Temp. ( C) Freq. (Hz) Temp. ( C) 0 30 20 10 0-10 -20-30 10-1 1 10 1 10 2 10 3 10 4 Freq. (Hz) Figure 4.5: Surf graph of imaginary parts of battery impedance For a better overview, the impedance change over temperature and frequency is presented qualitatively in table 4.1. 30 to 10 C 10 to +10 C +10 to +30 C R X C X L R X C X L R X C X L 10 1 to 10 Hz 10 to 10 3 Hz 10 3 to 10 4 Hz Table 4.1: Qualitative overview of the temperature behavior of the real (R), imaginary capacitive (X C ) and imaginary inductive (X L ) part of the measured impedance. 4.1.2 Experiment Both cells are tested at nine operating points, all at 20 C. Three current rates (0.5 C, 1 C and 1.25 C), each one with three different frequencies. That was to identify which factor, frequency or current amplitude, has what kind of impact on battery internal heating. Lithium plating must be avoided, therefore the lowest AC-frequency was chosen to be 100 Hz. Since, this study is designed as a critical case study, the second AC frequency was chosen to be 1 khz. That is, where the total impedance, but also the real part, which creates dissipative losses, have their minimum. For comparison, the batteries are also discharged at the corresponding C-rate to the applied RMS current in the AC-case. Despite the fact, that great care was taken to create the insulation layer for the batteries, they turned out to be quite different. That was observed during cooling, when battery SS20 cooled faster than battery SS18. As a result of that, the temperature development 30

CHAPTER 4. RESULTS is presented individually for each battery. Due to safety reasons, the batteries are not charged and discharged to their absolute limits. Therefore, discharging at 0.5C, 1C and 1.25C happens in slightly less time than the corresponding 2 h, 1 h or 48 min, which would result by definition of the C-rate. Fig. 4.6 and fig. 4.7 show the warmup at 0.5 C rate, which corresponds to 14 A. Neither SS18 nor SS20 heated up with AC, independent if 100 Hz or 1 khz was injected. However, pure discharge at the same current rate increases the temperature in both batteries identically by 12.5 C from 20 C to 7.5 C in approx. 110 min. 5 0 SS18-0.5C 0 Hz 100 Hz 1k Hz Battery Temperature ( C) -5-10 -15-20 -25 0 20 40 60 80 100 120 140 160 Time (min) Figure 4.6: SS18 warmup during experiment at 0.5 C 31

4.1. MEASUREMENTS 5 0 SS20-0.5C 0 Hz 100 Hz 1k Hz Battery Temperature ( C) -5-10 -15-20 -25 0 20 40 60 80 100 120 140 160 Time (min) Figure 4.7: SS20 warmup during experiment ay 0.5 C The heating effect of 1 C current (28 A) is presented in fig. 4.8 for SS18 and in fig. 4.9 for SS20. 1 khz raises the temperature by 3.0 C and 2.3 C in SS18 and SS20 respectively. 100 Hz works slightly better. In SS18 the temperature rises by 4.2 C to 15.8 C and in SS20 by 3.0 C to 17 C. All after approx. 120 min. Discharge heats the batteries slightly above (SS18: 1.0 C) or slightly below (SS20: 0.8 C) 0 C within 50 min. 5 0 SS18-1C 0 Hz 100 Hz 1k Hz Battery Temperature ( C) -5-10 -15-20 -25 0 20 40 60 80 100 120 140 160 Time (min) Figure 4.8: SS18 warmup during experiment at 1 C 32

5 0 SS20-1C CHAPTER 4. RESULTS 0 Hz 100 Hz 1k Hz Battery Temperature ( C) -5-10 -15-20 -25 0 20 40 60 80 100 120 140 160 Time (min) Figure 4.9: SS20 warmup during experiment at 1 C The highest temperature increase can be observed with 1.25 C current (35 A), as shown in fig. 4.10 for SS18 and in fig. 4.11 for SS20. As in the previous cases SS18 warms up better. Its temperature increases by 5.3 C to 14.7 C with 1 khz and by 6.6 C to 13.4 C with 100 Hz within 140 min. In the same time, the temperature in SS20 increases by 3.3 C to 16.7 C with 1 khz and by 4.5 C to 15.5 C with 100 Hz. Discharging rises the temperature within 40 min to +4 C in SS18 and to +2.3 C in SS20. 33

4.1. MEASUREMENTS 5 0 SS18-1.25C 0 Hz 100 Hz 1k Hz Battery Temperature ( C) -5-10 -15-20 -25 0 20 40 60 80 100 120 140 160 Time (min) Figure 4.10: SS18 warmup during experiment at 1.25 C 5 0 SS20-1.25C 0 Hz 100 Hz 1k Hz Battery Temperature ( C) -5-10 -15-20 -25 0 20 40 60 80 100 120 140 160 Time (min) Figure 4.11: SS20 warmup during experiment at 1.25 C 34

CHAPTER 4. RESULTS 4.1.3 Simulation Model Equivalent circuit All components of the equivalent circuit vary greatly with temperature except for the inductive part and the second capacitor. As shown in fig. 4.12 the inductance is within 30 45nH. The series resistor R E increases from 0.9mΩ at +30 C to 2.2mΩ at 30 C. It is worth mentioning that the series resistance increases exponentially, which causes a faster increase below 0 C. 2.5 R E SS18 SS20 mean 50 L 2 40 Resistance (m ) 1.5 1 Inductance (nh) 30 20 SS18 SS20 mean 0.5 10 0-30 -20-10 0 10 20 30 Temperature ( C) Figure 4.12: Series resistance and inductance 0-30 -20-10 0 10 20 30 Temperature ( C) Fig. 4.13 shows the behavior of the second RC-branch. The parallel resistance R C1, similarly to R E, increases with decreasing temperature, the parallel capacity C 1 increases moderately from 30 C to +10 C and dramatically above. At +30 C, R C1 is at 0.3mΩ and at 30 C it is at 0.135Ω. Also similar to R E, R C1 increases exponentially, which means slow resistance change at positive temperatures and fast at negative temperatures. C 1 increases on a much higher scale. At 30 C it is at 2.3F and at +30 C it is at 13.5 10 3 F. The behavior above +10 C might seem irregular but is out of the focus of this study, where the LIBs are studied at low temperatures. 35

4.1. MEASUREMENTS 0.15 R C1 SS18 SS20 mean 10 4 C 1 0.12 10 3 Resistance ( ) 0.09 0.06 Cpacity (F) 10 2 0.03 10 1 0-30 -20-10 0 10 20 30 Temperature ( C) Figure 4.13: First parallel R-C branch 10 0-30 -20-10 0 10 20 30 Temperature ( C) While the second parallel resistance R C2 again shows similar behavior to the other resistances, the second capacitor remains absolutely unchanged over temperature (see fig. 4.14). In addition, an exponential increase in resistance can be observed with decreasing temperature. This means in numbers: 1.3mΩ at +30 C, 58mΩ at 0 C and 1.4Ω at @ 30 C. The capacitance is 67.2 10 3 F. 1.5 R C2 SS18 SS20 mean C 2 1.2 10 5 6.72 10 4 Resistance ( ) 0.9 0.6 Capacity (F) 0.3 10 4 0-30 -20-10 0 10 20 30 Temperature ( C) Figure 4.14: Second parallel R-C branch -30-20 -10 0 10 20 30 Temperature ( C) With the above mentioned components included in the equivalent circuit ((3.2)) it is possible to create a Nyquist graph. In this way the modelled and measured impedance 36

CHAPTER 4. RESULTS can be compared. In fig. 4.15, where the focus is on lower temperatures ( 30 C and 10 C) it can be seen that, with respect to the real axis, the imaginary part rises faster into the negative imaginary plane than the actual impedance. Furthermore, the modelled impedance also turns earlier and decreases thereafter. This creates a more roundly shaped bow in the negative imaginary part of the modelled impedance compared to the actual impedance. Anyhow, the half circle of a single RC-branch is compressed. The modelled impedance in the 30 C-case does not follow the dent of the measured impedance. The zero crossing at 1 khz matches well in both cases. Nyquist graphs with focus on higher temperatures (+10 C and +20 C) are shown in 4.16. Here, the bows of the modelled impedance show as well a more round shape than the bows of the actual impedance. Even so, the difference at higher temperature can be neglected, since the focus is on low temperatures in this study. 3 2 -Im(Z) (m ) 1 0-1 SS18 @0 C SS18 @-30 C SS20 @0 C SS20 @-30 C eq. circuit @0 C eq. circuit @-30 C -2-3 0 1 2 3 4 5 6 7 Re(Z) (m ) Figure 4.15: Nyquist graph with both measured data and equivalent circuit I 37

4.1. MEASUREMENTS 1 0.5 -Im(Z) (m ) 0 SS18 @20 C SS18 @10 C SS20 @20 C SS20 @10 C eq. circuit @20 C eq. circuit @10 C -0.5-1 0 1 2 3 Re(Z) (m ) Figure 4.16: Nyquist graph with both measured data and equivalent circuit II 4.1.4 Heat Simulation Computation of the Thermal Resistance The thermal resistance is computed as described in section 3.5. Eq. 4.1 shows the procedure for the purpose of computing the thermal resistance. The data is taken from the case cell SS18-1.25C - 1 khz. Reason for choosing this particular case is that SS18 has better insulation, 1.25C caused the highest temperature increase and at 1 khz the impedance changes fewest over temperature. R th = Ṫ q = T i 2 rmsre(z(t )) (4.1) The experimental data T = 5.3K, Re(Z(T )) = 1.17mΩ (mean ohmic resistance), i RMS = 35A gives the thermal resistance R th = 3.7 W. To verify this value the analytic method is K applied ((4.2)). R th = d λa bat.box = d λ 2(hl + hw + lw) (4.2) With d as thickness of the insulation material, λ as thermal conductivity, A bat.box as surface area of the battery box built from the insulation material with the dimensions l - length, w - width and h - height. The box has the dimensions d = 2.5cm, l = 22cm, w = 12cm, h = 18cm. The material is EPE-foam with the thermal conductivity λ = 0.036 W. This results m K in a thermal resistance of R th = 3.96 W. K 38

CHAPTER 4. RESULTS Simulation Results Below, the results of the different simulations are compared. Fig 4.17 shows the temperature development of the experiment compared to the different simulation methods at 0.5C (14A rms and 100Hz. Below the final temperature of each simulation is presented. experiment: 19.59 C state space equivalent circuit: 18.93 C state space EIS Re(Z): 18.74 C PLECS equivalent circuit: 19.94 C PLECS EIS Re(Z): 18.77 C -10 0.5C @ 100 Hz Temperature ( C) -15-20 -25 experiment stsp eq c stsp eis re(z) plecs eq c plecs eis re(z) 0 20 40 60 80 100 120 140 160 Time (min) Figure 4.17: Model comparison Fig. 4.18 shows the temperature development of the experiment compared to the different simulation methods at 0.5C (14A rms and 1kHz. Below the final temperature of each simulation is presented. experiment: 19.42 C state space equivalent circuit: 18.94 C state space EIS Re(Z): 19.15 C PLECS equivalent circuit: 19.95 C PLECS EIS Re(Z): 19.15 C 39

4.1. MEASUREMENTS -10 0.5C @ 1k Hz Temperature ( C) -15-20 -25 experiment stsp eq c stsp eis re(z) plecs eq c plecs eis re(z) 0 20 40 60 80 100 120 140 160 Time (min) Figure 4.18: Model comparison Fig 4.19 shows the temperature development of the experiment compared to the different simulation methods at 1C (28A rms and 100Hz. Below the final temperature of each simulation is presented. experiment: 15.84 C state space equivalent circuit: 16.0 C state space EIS Re(Z): 15.33 C PLECS equivalent circuit: 19.78 C PLECS EIS Re(Z): 15.23 C 40

CHAPTER 4. RESULTS -10 1C @ 100 Hz Temperature ( C) -15-20 -25 experiment stsp eq c stsp eis re(z) plecs eq c plecs eis re(z) 0 20 40 60 80 100 120 140 160 Time (min) Figure 4.19: Model comparison Fig 4.20 shows the temperature development of the experiment compared to the different simulation methods at 1C (28A rms and 1kHz. Below the final temperature of each simulation is presented. experiment: 16.9 C state space equivalent circuit: 16.03 C state space EIS Re(Z): 16.81 C PLECS equivalent circuit: 19.8 C PLECS EIS Re(Z): 16.74 C 41

4.1. MEASUREMENTS -10 1C @ 1k Hz Temperature ( C) -15-20 -25 experiment stsp eq c stsp eis re(z) plecs eq c plecs eis re(z) 0 20 40 60 80 100 120 140 160 Time (min) Figure 4.20: Model comparison Fig 4.21 shows the temperature development of the experiment compared to the different simulation methods at 1.25C (35A rms and 100Hz. Below the final temperature of each simulation is presented. experiment: 13.43 C state space equivalent circuit: 14.03 C state space EIS Re(Z): 12.99 C PLECS equivalent circuit: 19.66 C PLECS EIS Re(Z): 12.85 C 42

CHAPTER 4. RESULTS -10 1.25C @ 100 Hz Temperature ( C) -15-20 -25 experiment stsp eq c stsp eis re(z) plecs eq c plecs eis re(z) 0 20 40 60 80 100 120 140 160 Time (min) Figure 4.21: Model comparison Fig 4.22 shows the temperature development of the experiment compared to the different simulation methods at 1.25C (35A rms and 1kHz. Below the final temperature of each simulation is presented. experiment: 14.7 C state space equivalent circuit: 14.09 C state space EIS Re(Z): 15.15 C PLECS equivalent circuit: 19.69 C PLECS EIS Re(Z): 15.05 C 43

4.1. MEASUREMENTS -10 1.25C @ 1k Hz Temperature ( C) -15-20 -25 experiment stsp eq c stsp eis re(z) plecs eq c plecs eis re(z) 0 20 40 60 80 100 120 140 160 Time (min) Figure 4.22: Model comparison There are three major trends, which can be observed in every figure. First trend is the equivalent circuit in PLECS goes in the right direction but with mean error of 3.16 C the final values are far from the measured data. Furthermore, it shows the wrong dynamics, since the temperature is only increasing in the beginning and stays constant thereafter. The second trend is the real part from the impedance spectroscopy, which produces the same result, independent if simulated in PLECS or the state space model. Moreover, the impedance spectroscopy data produces the most accurate results compared to the experiment with a mean error of only 0.44 C and 0.47 C for the state space and PLECS simulation respectively. The third trend is the result of the state space model with usage of the equivalent circuit. It results in the same temperature, independent if 100 Hz or 1 khz is used. On one hand this means for 1 khz the state space model with equivalent circuit produces slightly too high temperatures and for 100 Hz it produces slightly too low temperatures. On the other hand the state space model with equivalent circuit produces good results with a mean error of 0.56 C, considering the time frame of more than 2.5 hours. Enormous differences were encountered in terms of the simulation time. The most inaccurate simulation, the PLECS model with equivalent circuit, took between 17-18 hours. With the simple lookup table of the EIS data this time could be reduced to 2 hours. Major achievements in reducing simulation time were made with usage of the state space model. Independent if equivalent circuit or EIS data was used, the simulation took only few seconds. 44

environment PLECS state space impedance equivalent circuit CHAPTER 4. RESULTS EIS Re(Z) sim. time: 17h sim. time: 2h accuracy: Worst accuracy: Best sim. time: <5s sim. time: <5s accuracy: Good accuracy: Best Table 4.2: Comparison of the four electro-thermal battery simulations 4.1.5 Temperature Controller Fig. 4.23 shows the temperature increase of a battery pack compared to a single battery. All data originates from simulations of the differential equation with usage of real impedance from the EIS. The reference temperature is selected to be +5 C and the current limit is set to 100A rms. The reference is chosen such that there is a safety margin to the minimum suggested operating temperature found in the literature (compare section 2.1.4). It can be seen that the battery pack reaches the reference temperature after 40 min, if the ambient temperature is 10 C or after 60 min if the ambient temperature is 30 C. On the other side it is difficult to reach the reference temperature with a single battery. Even after 120 min it is not possible to reach the desired +5 C if the ambient temperature is 30 C. 10 5 0 Temperature ( C) -10-20 -30 Pack -30 C -> +5 C Pack -20 C -> +5 C Pack -10 C -> +5 C Cell -30 C -> +5 C Cell -20 C -> +5 C Cell -10 C -> +5 C 0 20 40 60 80 100 120 Time (min) Figure 4.23: Temperature increase with controller in either battery pack or single battery 45

4.1. MEASUREMENTS 120 100 80 Current (A) 60 40 20 0 Pack -30 C -> +5 C Pack -20 C -> +5 C Pack -10 C -> +5 C Cell -30 C -> +5 C Cell -20 C -> +5 C Cell -10 C -> +5 C 0 20 40 60 80 100 120 Time (min) Figure 4.24: Controlled current for temperature increase in either battery pack or single battery Table 4.3 shows how much time and energy is needed to heat the battery pack from a certain temperature, how much power is necessary to maintain +5 C and up to which time it is more energy efficient to maintain the battery temperature instead of letting it cool down and heating it up again. It is interesting to see that the time threshold, where maintaining the battery temperature at +5 C is more efficient than letting the battery cool down and heating it up again, is between 6.4 h and 8.4 h. Considering that it takes some time for the battery to cool down from its field operating temperature to +5 C, it would be most energy efficient to maintain a battery temperature of +5 C. +5 C +5 C +5 C 30 C 20 C 10 C Energy to heat 1.12 kwh 0.74 kwh 0.36 kwh Time to heat 33 min 24 min 13 min Power to maintain 133 W 95 W 57 W Time threshold 8.4 h 7.8 h 6.4 h Table 4.3: Time and power consumption to heat a battery pack Overall, the state space simulations reproduce the experimental results quite well in all operating points. Therefore, it can be assumed to be correct and it can be used further on. 46

CHAPTER 4. RESULTS 4.2 Analysis 4.2.1 Electrochemical Impedance Spectroscopy The EIS shows great impedance increase with reduced temperature. This is counter intuitive if compared with metal conductors, which show better electrical conductivity with reduced temperature. Better electrical conductivity at low temperature is also happening in LIBs. Anyhow, the proportion of the parts with better conductivity at low temperatures becomes irrelevant when looking at the whole impedance. The measurement clearly reveals that the proportion of the ion conductivity and the mass transfer ability to the whole impedance is dominating. This is also the reason why LIBs have a reduced power output at low temperatures. However, for the purpose of heating, high impedance, especially high ohmic impedance, is favorable. 4.2.2 Experiment There is a different heating effect depending on both frequency and current amplitude. The difference between 100 Hz and 1 khz can be explained by looking into the equivalent circuit. The impedance of a capacitor follows x c = 1. That means with higher ωc frequency, the capacitor has lower impedance. Since the capacitor is in parallel with an ohmic resistance with fixed impedance, more current will flow over the capacitor with increasing frequency. In exchange for that, less current will flow over the ohmic resistor, where it would cause dissipative heat. The reason for the much higher heat generation at 0 Hz discharge is the higher charge transfer overpotential and mass transfer over potential. At 1 khz, time is too short for Li+ ions to leave or join an electrode. Therefore, the chemical reaction cannot contribute any heat generation. As a result of high frequency flow of Li+ in the electrolyte, which would cause frictional heat. At 0 Hz discharge, charge and mass transfer overpotential is fully established with the result that the cell heats at a much higher rate than at 100 Hz or 1 khz. Different heating at different current levels can also be explained with the equivalent circuit and the charge and mass transfer over potential. The dissipated power follows P diss = i 2 rmsr, which means that doubled current results in four times increased dissipated power. Higher current also means higher chemical reaction rate at the electrode and thus higher flow rate of Li+ ions through the electrolyte. There is also a different temperature increase between the two batteries. This is simply due to deviating thermal insulation. Despite striving for making the battery insulations as equal as possible, they turned out to have slightly different thermal insulation. In some figures, at some points, unexpected temperature progress can be observed. Independent of time, the curve shows a linear progression instead of a typical step response or even a dent, as i.e. in fig. 4.10. The green curve (100 Hz) shows linear behavior until 60 min. The blue curve (1 khz ) shows a dent around 100 min. This behavior is due to the cooling of the climate chamber. The climate chamber regulates the temperature with a so called bang-bang controller. It heats or cools the air if the temperature is detected to be at the lower or upper boundary, respectively. At all time it maintains high rate of air convection. Since the battery, even if insulated, acts as a heat source, the climate chamber dominantly cools to compensate for the heat generated by the battery. Since the 47

4.2. ANALYSIS cooling happens at a high rate, it also affects the battery s temperature in return. 4.2.3 Simulation In the Nyquist graph the equivalent circuit shows a rounder shape than the actual battery impedance. Any parallel RC branch produces a half circle. A second RC-branch can compress that half circle but only to a certain extent. Even more elements could form a more sophisticated shape. Anyhow, the chosen circuit is in good accordance with current literature and reproduces the electrical behavior of the battery properly for temperatures above 10 C. Below 10 C the dent is not captured. This means that the heat causing real part has the same value at 100 Hz and 1 khz. That is the reason why the different behavior at 100 Hz and 1 khz cannot be reproduced in simulations. There is also a subtle deviation for the value of the thermal resistance computed with experimental data or material properties. This deviation is caused by the cabling, which was not taken into account for the computation from material properties. Here, a flawless surface was assumed, while the real insulation has several openings through which the cables are routed. These cables have a good thermal conductivity and are the reason for the lower thermal insulation. The cooling phenomenon of the climate chamber and how it affects the temperature progress has been discussed in section 4.2.2. This feedback coupling from climate chamber to battery is not taken into account in the simulation. Therefore, some discrepancies can be observed during the battery is heated. However, at steady state, the simulations agree very well with the experiment with only about 0.5 C error. 4.2.4 Controller The battery pack simulation of course employs extensive simplifications. Firstly, it is not known how a battery pack is insulated. Here it was assumed that the encapsulation material would have the same thermal insulating properties as EPE foam. Secondly, there is much more material (BMS circuit board, compression frame, fans) inside the battery pack than just the cells. This increases the thermal capacity. Finally, not all cells heat in the same way. Cells located at the edge of the pack will be cooler than cells entirely surrounded by other cells. Here it is assumed that all cells would unite to a huge body, with high thermal conductivity, so that the whole body has a uniform temperature. Nevertheless, this simplified simulation of a battery pack with a controller showed that even though it is not possible to heat a single lithium-ion cell it is possible to heat a battery pack. In this fictional battery pack all batteries are connected in series, which increases the battery impedance by the number of series connected batteries. As a result of that, the heating power increases at the same rate, even with unchanged current. However, the surface area facing the ambient and therefore the heat flow towards the ambient, only increases approx. 15 times. This relation makes it possible to produce more heat in the battery pack than is lost towards the ambient and therefore increase the temperature sufficiently. 48

Chapter 5 Conclusion In this chapter the results are summarized. The summary is presented such that the research questions are answered. The experiment agrees well with the presented literature, which stated that low frequency and high current are favorable for the AC heating method. On top of that, it was possible to construct a valid simulation model, that combines electric and thermal properties in order to predict heating of the LIBs, which were used in this study. An extended model of a battery pack finally proved that it is possible to properly heat LIBs, even if they show low impedance, high thermal capacity, large surface area and even if the AC frequency is in a range, where the cell impedance is at its minimum. 5.1 Answers to Research Questions 5.1.1 What are the existing methods to heat LIBs? Several heating methods, which are currently discussed are reviewed in detail in section 2.2.2 and section 2.2.1. They can be categorized into four groups: internal and external heating, depending on whether the source of heat is inside or the outside of the battery; each method can be battery self-powered or externally powered. An overview is given in table 5.1. In literature external heating methods are considered as self-powered. It is, however, also possible to power them externally from a charger, if the car is connected to one. 49

5.1. ANSWERS TO RESEARCH QUESTIONS Self powered Externally powered Internal heating Additional foil as ohmic heater [25] Buck-boost converter from electric motor and filter capacitor [26] Mutual pulse heating [31] CVM, PCM [27] AC heating 10-20 khz (NiMH) [28] AC sinusoid & rectangular 1-600 Hz [29] AC sinusoid 0.1-10 Hz [30] External heating Refrigerant heating with air-conditioning system and additional ohmic element [22] Heat pipes with heat block [23] Direct cabin air blow, PCS cycle, refrigerant heating [24] - Table 5.1: Overview of battery heating methods External heating methods have the disadvantage of requiring additional components. They can either be mounted inside the battery pack or somewhere else in an electric vehicle. Yet, they all make the vehicle heavier and more complex. Internal heating methods on the other side do not require additional components, but often depend on external power sources. This would make battery heating only available as pre-conditioning, if the car is connected to a charger. However, hardware in battery chargers is already equipped with a current controller, which might be adjustable for injecting AC, too, by upgrading the software functionality. Only few methods discussed in literature combine internal heating with self powering. Anyhow, since internal heating gives advantages for reducing both complexity and and weight of vehicle batteries, self-powered internal heating is necessary to cover both scenarios: when the car is connected to a charger and when it is not. 5.1.2 To what extent is it possible to heat LIBs with current? Heating is mostly correlates with the current level. Therefore, faster and more effective heating will come with higher current. Furthermore, frequency has also an influence on the heating effect. As the experiments in this study show, LIBs have higher real impedance with lower frequencies and therefore generate more dissipative heat at low frequencies at the same current. However, too low frequencies bear the risk of permanent capacity loss through lithium plating and further even safety hazards. Therefore the lowest frequency has to be identified where risk of lithium plating is low and safety is guaranteed. At this frequency the applied AC will be most effective. For faster response, the current needs to be increased. Nonetheless, not only current frequency and amplitude affect effective heating. Attention has to be payed to compact packaging, proper 50

CHAPTER 5. CONCLUSION thermal insulation as well as little thermal conductivity through cabling. Another aspect is the heating strategy, which compares the energy used for heating up a battery and the energy used for maintaining a certain temperature over a limited time. Simulations show that in many cases, maintaining the battery temperature is more efficient than letting it cool down and heating it up again. On one hand, the experiments in this study showed that it is possible to increase temperature even in LIBs with low impedance and high thermal capacity at a frequency, where the battery s real impedance is at its minimum. On the other hand, it was not possible to increase the temperature sufficiently in the existing setup. However, simulations that take a battery pack and a more powerful source into account show that it is possible to heat LIBs even under the worst conditions. 5.1.3 What are the limitations of the AC heating method As previously described it is possible to heat LIBs with alternating current. However, that is if three conditions are fulfilled: 1. The batteries are connected in series, so that the impedance adds up and the power increases by the number of series connected batteries (see section 4.2.4). 2. The batteries are aligned such that they form a compact body with the least possible surface area, that can fit into a vehicle. 3. The batteries are isolated properly Furthermore, if the whole system is taken into account, more factors need to be considered. Generally, battery chargers are designed for DC. However, since they have to support both constant voltage and constant current charging, they have a built-in current controller. This current controller needs to be able to provide high enough frequency in order to make the AC heating method available in real applications. Depending on the type of battery this frequency can vary. More precisely the frequency needs to be high enough to ensure safe operation. Next to frequency, the charger should also be able to provide high enough current amplitude. This again varies from battery to battery and is dependent on thermal insulation of the pack, connection and impedance of the cells. In addition to the importance of the charger, the functionality of the car s drive train plays an important role, too. Additional heating components can only be left out if battery heating can be ensured also with disconnected charger and missing heating elements. This self-powered internal heating requires again additional components i.e. a DC/DC converter for mutual pulse heating or a relay for building a buck-boost converter. However, the effort for self-powered internal heating is expected to be less than for self-powered external heating. 5.2 Ethical Aspects As discussed, the AC heating method can help to lift EV s popularity. In return, a higher share of electric vehicles make the total traffic less polluting and more silent. However, more EVs demand more LIBs and therefore more raw materials. It is important to pay attention to how these raw materials are extracted. Not only environmental pre-cautions 51

5.2. ETHICAL ASPECTS have to be installed in order to not pollute the environment at the extraction sites. Moreover, safety pre-cautions and fair treatment for the workers need to be established. Environmental and social standards are generally met well in the automobile industry. If those standards can be applied to the whole value chain of the vehicle production, also southern American countries, where many of the raw materials are extracted, can benefit from the electrification of transportation. 52

Chapter 6 Outlook 6.1 Usability in Real Applications Experimental data and further simulation show, that it is possible to pre-heat EV-batteries with AC. Even under unfavorable conditions such as low battery impedance and large thermal capacitance it is possible to heat a single battery from 20 C to 15 C in the experiment and a battery pack to from 20 C to +5 C in simulations. These results show that a battery charger could be used to pre-heat an EV s battery, before it is unplugged for driving. This capability makes additional ohmic heating elements inside the battery pack obsolete, leading to fewer components inside the battery pack. Fewer components in the battery pack give the opportunity to reduce weight, size and complexity of the battery pack. In return, these advantages can help to increase the driving range and cut costs of EVs at the same time. However, there are some limitations to pay attention to in order to be able to install this functionality and remove additional heating elements from the battery pack. Firstly, the battery pack has to be insulated properly. This is necessary, to limit the heat loss if the battery has a high temperature difference compared to the ambient as well as to limit the current draw from the power supply if the battery has a low inner impedance. Secondly, the arrangement of the cells inside the battery pack needs to be such that the resulting impedance is as high as possible and the resulting surface area of the pack is as little as possible. This reduces the heating time as well as the necessary power to heat the battery. Finally, self-powered internal heating needs to be installed, too. If not, additional heating elements are still necessary to heat the battery if the car is not plugged to a charger. Solutions for this scenario are already being discussed in literature. 6.2 Future work Further research should be conducted in order to explore the potential of the AC heating method for real applications. Developing a more detailed thermal model of a battery pack is necessary to see the heat distribution among the cells. This detailed thermal model should take into account, that the cells in the center of the pack are easier to heat than the surrounding cells. Further, the detailed model should also take into account the thermal capacity of the cooling liquid 53

6.2. FUTURE WORK Figure 6.1: Circuit with power supply, resonance capacitor and battery equivalent circuit and the thermal capacity of the electronics of the battery management system. Identifying the lowest AC frequency where the chemical reactions still do not happen gives the opportunity to maximize the heat dissipation at a certain AC amplitude. Since the real part of the battery impedance increases at lower frequencies, it would be desirable to apply the lowest frequency possible, that does not cause aging in the battery. The other way to increase the heating effect is to increase the current. Since the current output of battery chargers is limited, another approach could be to build up a resonance circuit, as shown in fig. 6.1. This resonance circuit, consisting of a charger as power supply, a resonance capacitor and the battery, requires only little current from the power supply but would maximize that current in the battery. 150 120 Supply current Resonance Current 90 60 Current (A) 30 0-30 -60-90 -120-150 0 0.1 0.2 0.3 0.4 0.5 Time (ms) Figure 6.2: Current from power supply and current in resonance circuit This is simulated with the equivalent circuit model, as presented in section 3.5 with the parameter as presented in table 6.1. The result is presented in fig. 6.2. 54