HETEROGENEOUS BATTERY SYSTEMS IN BATTERY EQUIPPED PASSENGER TRAINS

(1)

Västerås, Sweden

Thesis for the Degree of Master of Science in Engineering - Dependable

systems and Robotics 30.0 credits

HETEROGENEOUS BATTERY

SYSTEMS IN BATTERY EQUIPPED

PASSENGER TRAINS

Johan Bergelin

Dependable systems

Mälardalen University, Västerås, Sweden

Emil Lundin

Robotics

Examiner: Nikola Petrovic

Supervisor(s): Per-Olof Risman, Luciana Provenzano

Company Supervisor(s): Christer Holmberg

Alstom, Västerås

(2)

Abstract

The rise of batteries in the industry, especially Li-ion, is increasing rapidly. Li-ion battery systems are traditionally composed of a particular type of cell chemistry fit to the system needs. Due to the significant differences between chemistries, different cells have different attributes. The thesis explores the potential of a heterogeneous solution to include different cells to find a suitable compromise between different attributes. An electrified passenger train using a homogenous solution was evaluated against a heterogeneous solution consisting of two cell types, NMC and LTO, which have significant differences in attributes. Simulation with models covering the train kinematics, track characteristics, and battery behaviour generates the thesis results. Validation of simulation results includes comparing previous simulations and the new effects of the heterogeneous solution, which indicate a good fit. Verification of the results encompasses a small-scale experiment with a custom-made physical circuit to observe the proposed solution’s actual behaviour and verify model validity, which implies the correctness of models and implementation. The results indicate that a heterogeneous solution is possible within the scope of electrified trains. Furthermore, several trade-offs exist between NMC and LTO cells, especially regarding rate capability, safety and capacity, which confirms the potential of heterogeneous battery systems.

(3)

Acknowledgements

We want to thank our supervisors for their continuous guidance within their respective fields, which creates the basis for the work performed in the thesis.

For providing us with the opportunity for the thesis and their collaboration, we would like to thank Alstom.

We want to thank Würth Elektronik for sponsoring us with electrical components for our thesis. Additionally, we would like to thank the different faculty members we shared our laboratory facil-ities with for their company.

Lastly, we would like to thank our examiner for providing us with access to lab facilities and assisting us with logistical issues.

(4)

2.12 Dependability . . . 20 2.13 Control theory . . . 21 2.14 Cooling . . . 22 2.15 Related work . . . 24 3. Problem formulation 33 3.1 Hypothesis . . . 33 3.2 Research questions . . . 33 3.3 Limitations . . . 33 4. Method 35 5. Ethical and Societal Considerations 37 6. Implementation 38 6.1 Models . . . 38 6.2 Simulation design . . . 45 6.3 Preliminary simulation . . . 46 6.4 Trade-off simulations . . . 47 6.4.1 Trains . . . 47

6.4.2 Tracks and use cases . . . 48

6.4.3 SoC . . . 48

6.4.4 Fast charging . . . 50

6.4.5 Safety and reliability . . . 50

6.4.6 Cost . . . 51

6.4.7 Losses and temperature . . . 51

6.4.8 Ageing . . . 52 6.4.9 Policies . . . 53 6.5 Hardware . . . 53 6.5.1 Discharge controller . . . 55 6.5.2 Charging controller . . . 57 6.5.3 Fuel gauge . . . 58

6.5.4 Platforms and setup . . . 59

6.6 LabVIEW . . . 60

6.6.1 track profile program . . . 60

6.7 Control theory . . . 61

6.7.1 System identification . . . 62

6.7.2 Controller . . . 63

(5)

6.8 Physical circuit verification of simulation . . . 64

6.8.1 Verification of battery dynamics . . . 65

6.8.2 Verification of nominal discharge . . . 65

6.8.3 Verification of track profile discharge . . . 65

7. Results 66 7.1 Preliminary Simulation . . . 66

7.2 Heterogeneous battery-equipped train simulation . . . 69

7.2.1 SoC . . . 69 7.2.2 Ageing . . . 71 7.2.3 Fast charging . . . 75 7.2.4 Safety . . . 77 7.2.5 Reliability . . . 80 7.2.6 Cost . . . 81 7.2.7 Battery losses . . . 82

7.3 Battery model verification . . . 83

7.4 Model verification summary . . . 87

8. Discussion 88 8.1 Project management & planning . . . 88

8.2 Result interpretations . . . 88

8.3 Research question 1 - How do different heterogeneous battery architectures affect operational SoC? . . . 89

8.4 Research question 2 - What beneficial trade-offs exist in heterogeneous battery sys-tems compared to traditional homogeneous battery syssys-tems? . . . 89

8.5 Relevance of results . . . 89

8.6 Limitations and their effects . . . 89

8.7 Technical implementation . . . 90

8.8 Battery model verification . . . 90

9. Conclusion 92 9.1 Thesis contribution . . . 92

9.2 Potential architectures and trade-offs . . . 92

9.3 Heterogeneous battery systems . . . 94

9.4 Future work and directions . . . 94

References 102 Appendix A Preliminary simulation 103 1.1 Tracks . . . 103

1.2 Interaction effects and regression models . . . 105

1.3 State of charge comparisons . . . 117

Appendix B Physical circuit 121 2.1 Sensor calibrating software . . . 121

2.2 Track profile software . . . 122

2.3 LTSpice . . . 123

2.4 Computer-assisted design . . . 125

Appendix C Project management 129 3.1 Agile methods . . . 129

3.2 Agile project management . . . 129

3.3 Tools and processes . . . 129

3.3.1 Rolling-wave planning . . . 129

3.3.2 Kanban board . . . 129

(6)

Appendix E Modelled tracks 137

Appendix F SoC simulation graphs 146

(7)

List of Figures

1 Figure 1a displays a basic electrical train block diagram while figure 1b displays a block diagram of an electrical train with a secondary propulsion system. . . 4 2 A nominal track profile curve. The distance and relative altitude of the track is

shown in the blue line. The yellow circles represent station stops. The speed limit and availability of external power is placed and the corresponding distance intervals. 5 3 Spider diagram comparing three fictional cells. 8 parameters are drawn on separate

axes and each cell is placed on a scale for each parameter. As such a simple overview can be created between the different attributes. . . 5 4 A primary battery cell with zinc and copper electrodes. A redox reaction occurs in

the system were the anode goes through oxidation while the cathode goes through reduction. The positively charged cathode is supplied by electrons created through the oxidation process at the negative anode. . . 8 5 This figure describes a basic schematic over a typical Li-ion cell chemistry. The

Cathode and Anode in a Li-ion cell create a potential difference, creating a current flow if connected. While electrons travel the external circuit, positive Li-ions traverse the cell internally and combine with the electrons. Due to the Intercalation effect created by the Cathode and Anode structure, discharge can be reversed and repeated. 9 6 Spider diagram comparing LTO and NMC to common attributes. . . 10 7 Electrical schematic of a buck converter. . . 11 8 A typical charge curve for a Li-ion battery cell. Trickle charge is necessary if the

cell voltage is low not to damage the cell, which is the interval between Vmin and

Vstart. As the voltage is between Vstartand Vmax, constant current applies until the

measured OCV reaches a critical point Vmax. At Vmaxthe voltage is kept constant,

and the current decreases until the voltage reach Vmaxwith no current applied. . . 12

9 A nominell discharge curve from a Li-ion cell. The blue line represents the SoC of the battery cell during the discharge process while the green line represents the cell voltage. . . 14 10 The ageing effects on a Li-ion cell. The curve represents a constant discharge at a

fixed C-rate and the typical effects of ageing. . . 16 11 A flow of information in a BMS for typical parameters necessary monitoring a

bat-tery. . . 17 12 The different elements of simulation and the corresponding real system test

ele-ments according to Ören and Zeigler 12. Each component of the simulation process associates with the corresponding real system experiment counterparts. . . 18 13 A round-robin process consisting of three different tasks. Each task is continuously

assigned execution time of the same length cyclically without any internal priorities or complex ruleset. . . 19 14 A 3 x 3 Latin square containing the samples A, B and C. . . 20 15 A dependability tree containing typical attributes. It is not necessarily always the

same attributes, Laprie for example proposed Security as a combination of Confid-entiality Integrity and Availability . . . 20 16 Example of a simple open-loop system, where the controller acts on the reference

input, which is then used to control the process. . . 21 17 Example of a simple closed-loop system, where the output is fedback from the sensor

and subtracted from the reference value. . . 22 18 Model of a basic EC of a battery with the internal resistance incorporated. refers

to the ideal voltage source, r is the internal Resistance, R is the load, I is the load current and V is the terminal voltage. . . 24 19 By adding one or more resistor or capacitor pairs to the model in figure 18 the

dynamic behavior of a battery cell is better modelled. . . 25 20 A comparison on the behavior of the voltage over a discharge event between the

(8)

21 A traction curve displaying the typical behaviour of the available traction force at different speeds. The typical running resistance for a train is drawn based on equation 16. The train’s maximum speed is assumed to be 250 km/h, and the force

does not increase beyond that point. . . 29

22 Preliminary activity timeline of the thesis. Green represents planning and docu-mentation, blue simulation and red hardware implementation. . . 35

23 High level architectural description of the model used for train simulation and bat-tery simulation. . . 38

24 Free body diagram of a train in action. . . 38

25 The speed of the train during a track F simulation. The PID controller works well for the implementation needs. . . 39

26 Representation of the tractive effort behavior of motors from Alstom. No numeric information is given to conform to the NDA. . . 40

27 Figure of the two types of cells used for the physical circuit. Both are also modelled and simulated for the verification of the models. . . 42

28 Basic battery system used for preliminary simulations and baseline simulations. Each battery system consists of 8 TBU which in turn are separated into 2 TBS. Further systems such as cooling and Battery Management System (BMS) are part of the battery systems but not observed in simulation. . . 43

29 High level description of the extended battery architecture for heterogeneous battery simulations. Numerous switches are implemented so that different configurations can be achieved during run-time. The same number of total TBUs are the same. . . . 44

30 Pseudocode for the I2_{t limiter in the simulations.} _{. . . .} ₄₅

31 The different tracks used for the preliminary simulation and the different nuances. Three parameters change between tracks which are: Length, CATO availability and the number of stops. . . 47

32 This figure shows the relationship between different terms used for describing the discharge process. DoD is defined as the difference between the starting and final SoC. . . 48

33 Pseudo-code of the logic behind the effect division in the heterogeneous architecture. In this scenario the temperature is not considered as the cooling is assumed to be sufficient and maximum cooling effect is active. . . 49

34 SysML use case diagram for the main actors of the system and the interactions related to the main safety concerns. . . 51

35 Estimation of the relationship between DoD and the total amount of cycles until EoL. The curve was fitted using the MATLAB curve fitting toolbox. . . 52

36 High-level schematic of the proposed circuit. The system used three different perf-boards, a charging controller seen on the left-hand side consisting of the switches and the external power-supplies, the fuel gauge consisting of sensors and the discharge controller seen to the right, consisting of the single pole double throw switch, DC/DC converter, load and current sensor. . . 54

37 Image of the hardware mounted on the printed boards. . . 54

38 Electric schematic of the complete system. The charging controller can be seen to the left of the batteries, the fuel gauge between transistor 1,2 and 3,4 and the discharging controller to the right of the batteries. . . 55

39 Electric schematic of the discharging controller. Control 3 toggles transistor 5, which saturates transistor 4 and 3. Transistor 3 is off when control 3 is high where as transistor 4 is on and vice versa when control 3 is low. . . 56

40 Image of the implemented discharging controller. . . 56

41 HMC 8042 powersupply used for the thesis [103] . . . 57

42 Image of the fuel gauge with batteries and power-supply connected. . . 57

43 Charging controller electrical schematic . . . 58

44 Electrical schematic of the fuel gauge . . . 58

45 National Instruments roboRIO . . . 59

46 Image of the fan, mounted on a custom-made mount . . . 60

(9)

48 Fit for the relationship between the Current and Duty cycle used for the controller based on measurements from the physical circuit. . . 62 49 Duty cycle as the function of output current and Voltage input . . . 63 50 Figure of the control scheme used to control the current consumed by the circuit . 63 51 The graph displays the behavior of the controller aiming to keep a specific current

over a sweep of the usable area for 160 seconds. The Root square mean error of the controller in this case was 0,0626 where the fit was best at lower currents. . . 64 52 Graph of a test-run conducted to validate the controller at a 2.5 Amperage’s, as it

was used for verification of the battery model. . . 64 53 Response surface plot of track 3 between station 2 and 3. . . 66 54 Response surface plot of track 1 between station 2 and 3. . . 67 55 Comparison of the interaction between mass and incline compared to speed and stop

length. . . 68 56 Graph displaying the SoC(%) versus time(s) for tracks 1, 3, 5 and 6 with the same

random train parameters. . . 69 57 The response surface modelling for track F regarding the difference of remaining

capacity from the heterogeneous system and reference simulation at the end of the track. . . 70 58 The response surface modelling for track F regarding the Remaining SoC from the

heterogeneous system and reference simulation at the end of the track. . . 70 59 The discharge curves corresponding to the configuration of 3 LTO modules, 5 NMC

modules, 0.5 LTO Charge policy and 0 LTO discharge policy on track F. The NMC modules of the heterogeneous carry the entire discharge load throughout the run-time without reaching the 30% SoC limit. The reference signal represents the dis-charge curve of a homogeneous NMC installation. . . 71 60 The response surface for increasing NMC SoC within a heterogeneous system. The

analysis is performed on track F and shows that a minority set of LTO cells with full discharge is optimal. . . 72 61 Discharge curve for track F comparing a homogeneous NMC solution with a

hetero-geneous solutions of 3 LTO cells and 5 NMC cells. The policy used prioritizes the LTO cell in all operation where the effect is larger than when cruising or auxiliary consumption. . . 72 62 Discharge curve for track D comparing a homogeneous NMC solution with a

hetero-geneous solutions of 3 LTO cells and 5 NMC cells. The policy used prioritizes the LTO cell in all operation where the effect is larger than when cruising or auxiliary consumption. . . 73 63 Discharge curve for track D comparing a homogeneous NMC solution with a

hetero-geneous solutions of 4 LTO cells and 4 NMC cells. The policy used prioritizes the LTO cell in all operation where the effect is larger than when cruising or auxiliary consumption. . . 73 64 Discharge curve for track H comparing a homogeneous NMC solution with a

hetero-geneous solutions of 3 LTO cells and 5 NMC cells. The policy used prioritizes the LTO cell in all operation where the effect is larger than when cruising or auxiliary consumption. . . 74 65 Discharge curve for track H comparing a homogeneous NMC solution with a

het-erogeneous solutions of 3 LTO cells and 5 NMC cells. All cells are assumed to be at EoL. The policy used prioritizes the LTO cell in all operation where the effect is larger than when cruising or auxiliary consumption. . . 74 66 Comparison of two different heterogeneous implementations at twice the nominal

charging rate. . . 75 67 Discharge curve for track H comparing a homogeneous NMC solution with a

het-erogeneous solution of 4 LTO cells and 4 NMC cells. The policy used prioritises the LTO cell in all operation where the effect is larger than when cruising or auxiliary consumption. Charge at station corresponds to 2.5 of the nominal value for the LTO cells, while the reference NMC cells restrict the charging due to the recommended maximum charging current. . . 76

(10)

68 Discharge curve for track H comparing a homogeneous NMC solution with hetero-geneous solutions of 4 LTO cells and 4 NMC cells. The policy used prioritises the LTO cell in all operation where the effect is larger than when cruising or auxiliary consumption. Charge at station corresponds to 3 of the nominal value for the LTO cells, while the reference NMC cells at the recommended maximum charging current. 76 69 Discharge curve for track G comparing a homogeneous NMC solution with a

het-erogeneous solution of 4 LTO cells and 4 NMC cells. The policy used prioritises the LTO cell in all operation where the effect is larger than when cruising or aux-iliary consumption. Charge at station corresponds to 2 times the nominal value for the LTO cells, while the reference NMC cells restrict the charging due to the

recommended maximum charging current. . . 77

70 Sequence diagram describing cooling during CFO with a failure. . . 78

71 FTA comparison of the case of overheat for the homogeneous and heterogeneous battery systems. Introducing LTO cells adds a logical AND gate to the system, increasing safety. . . 79

72 FTA comparison of the case of overcharge for the homogeneous and heterogeneous battery systems. Introducing LTO cells adds a logical AND gate to the system, increasing safety. . . 80

73 The effect on the losses in the battery systems depending on the use of policies. . . 82

74 The losses (W) of the heterogeneous battery system compared with the homogeneous NMC solution in a worst case scenario as seen in figure 73a. . . 83

75 10-second discharge event at 70% SoC showing the measured values of 2 series NMC cells with the model prediction for the same setup. The model follows the measured value relatively close but is smoother due to rounding in the measurements and the mean over several samples used for the current. . . 84

76 The controller tuning the current for 160 seconds on a set level. . . 85

77 Applying a digital low-pass filter reduces noise on the signal but the signal is still somewhat unstable. A large part is the controller which constantly tunes output current to the varying voltage. Furthermore certain limitation exist in the resolution of the sensors and sample times. . . 85

78 The current over time for the LTO and NMC cells measured on the physical circuit during verification track discharge curve. . . 86

79 The measured total current during the physical discharge on verification track dis-charge curve. . . 86

80 Comparison of the estimated voltage and measured voltage for the LTO and NMC cells during the physical circuit verification of case 3. Due to the different connection when charging the behavior is different. Since the model does not consider the charging and discharging separately the fit is worse for charging as the model is based on discharging. . . 87

81 Spider diagram containing homogeneous LTO, NMC and a 4-4 heterogeneous solu-tion formulated through the thesis. . . 94

82 Regression model for Track 1 section 1 . . . 105

94 Interaction between weight and incline for track 1 section 1 . . . 111

(11)

106 Graph displaying the SoC(%) versus time(s) for tracks 5-8 with the same random train parameters. . . 118

107 Graph displaying the SoC(%) versus time(s) for tracks containing 4 stops with the same random train parameters. . . 119

108 Graph displaying the SoC(%) versus time(s) for tracks containing 6 stops with the same random train parameters. . . 120

109 Image depicting the LabVIEW code . . . 121

110 Image depicting the LabVIEW GUI . . . 122

111 Image of the actual code . . . 123

112 Image depicting the GUI . . . 123

113 Image of the preliminary simulation to ensure the single pole double throw switch worked as intended. . . 124

114 Image from a simulation carried out to confirm the circuit is capable of switching between the 4,2 and 3,7 voltage power source. The voltage output can be seen in blue and the control signal in green. . . 124

115 Preliminary simulation to validate the functionality of the SPDT switch integrated with the DC/DC converter. . . 125

116 SolidWorks sketch of the platform used to mount the charging controller and the fuel gauge . . . 125

117 SolidWorks sketch of the platform designed as a mount for the roboRIO . . . 126

118 SolidWorks sketch of the top mount for the fan . . . 126

119 SolidWorks sketch of the bottom attached to the fan mount . . . 127

120 SolidWorks sketch of the designed plate used to mount the discharging controller . 127 121 SolidWorks sketch of the assembly of the system . . . 128

122 Image of the kanban board used throughout the thesis. . . 130

123 Overview of Simulink code. The code blocks represent the different models in section 6.1. . . 131

124 The input and outputs of the battery system model block. Two batteries are ob-served with four reference measurements from NMC and LTO cells at BoL and EoL respectively. . . 131

125 The policy block controls the batteries under inspection via MATLAB scrips. The reference cells are under no particular policy. . . 132

126 Simscape elements modelled the battery. The battery interface manages the condi-tions for the electrical simulation with realistic behavior. . . 132

127 The inputs and outputs of the speed controller. . . 133

128 The controller works as a PID controller. The transfer function creates a small time delay in the system required to avoid algebraic loops found otherwise due to the output directly affecting the system input. . . 133

129 Any parameters of importance are logged to the MATLAB workspace at each time step. . . 134

130 The inputs and outputs of the kinematic model of the train. . . 134

131 The kinematics apply a force summation on the motor traction, gravitational force and running resistance. The sum of forces acts as the basis for acceleration which in turn creates the velocity and traveled distance. . . 135

132 The inputs and outputs of the propulsion system. . . 135

133 The propulsion system applies the motor traction depending on the speed controller input. Furthermore the energy consumption is calculated for the battery system. . 135

(12)

134 The inputs and outputs of the track profile. . . 136

135 The track profile reads data from a MATLAB script containing the track specifics at different positions. Additionally a state machine controls start and stop behavior necessary for the passenger train. . . 136

136 Track A . . . 137

137 Track A . . . 138

138 Track B . . . 138

139 Track B. Return trip with 10 minute dwell time at end station without charging. . 139

140 Track C . . . 140

141 Track C. Return trip with 10 minute dwell time at end station without charging. . 140

142 Track D . . . 141

143 Track D. Return trip with 10 minute dwell time at end station without charging. . 141

144 Track E . . . 142

145 Track E. Return trip with 10 minute dwell time at end station without charging. . 142

146 Track F . . . 143

147 Track F . . . 143

148 Track G . . . 144

149 Track G. Return trip with 7 minute dwell time at end station with charging. . . . 144

150 Track H . . . 145

151 Track H. Return trip with 7 minute dwell time at end station with charging. . . . 145

152 Response surface on track A regarding capacity . . . 146

153 Response surface on track A regarding SoC . . . 147

154 Response surface on track B regarding capacity . . . 147

155 Response surface on track B regarding SoC . . . 148

156 Response surface on track C regarding capacity . . . 148

157 Response surface on track C regarding SoC . . . 149

158 Response surface on track D regarding capacity . . . 149

159 Response surface on track D regarding SoC . . . 150

160 Response surface on track E regarding capacity . . . 150

161 Response surface on track E regarding SoC . . . 151

162 Response surface on track G regarding capacity . . . 151

163 Response surface on track G regarding SoC . . . 152

164 Response surface on track H regarding capacity . . . 152

165 Response surface on track H regarding SoC . . . 153

166 The SysML block definition diagram for the heterogeneous battery system observed in the thesis. . . 154

167 The internal block definition diagram for the Heterogeneous battery. . . 155

168 Sequence diagram over cooling during CFO containing a monitor function of the BMS and heterogeneous battery system previously described. . . 156

(13)

Acronyms

ANNArtificial Neural Network BMSBattery Management System BoLBeginning of Life

CFOCatenary-Free Operation CATOCatenary Operation

CCMContinuous Conduction Mode DC-DCDirect Current - Direct Current CC-CV Constant Current-Constant Voltage DCMDiscontinuous Conduction Mode DOEDesign of Experiments

DoDDepth of Discharge ECEquivalent Circuit EoLEnd of Life

EEAEuropean Environment Agency EMUElectric Multiple Unit

EVElectrical Vehicle

FPGAField-Programmable Gate Array FTAFault Tree Analysis

KFKalman Filter

LDOLow-Dropout linear regulator Li-ionLithium-ion

LTOLithium-Titanate-Oxide MBDModel-Based Design

MTBFMean Time Between Failures NDANon-Disclosure Agreement NINational Instruments

NMCLithium-Nickel-Manganese-Cobalt-Oxide OCVOpen Circuit Voltage

OSHAOccupational Safety and Health Administration POCProof of Concept

PWMPulse-Width-Modulation

RISEThe Research Institute of Sweden RuLRemaining useful Life

SCBStatistiska Centralbyrån SDBSoftware Defined Batteries SEISolid Electrolyte Interface SoCState of Charge

SoHState of Health

SVMSupport Vector Machine SysMLSystems Modeling Language TBSTraction Battery System TBUTraction Battery Unit

UICInternational Union of Railways UNUnited Nations

(14)

1. Introduction

The usage of batteries is widespread in the modern-day, and their usability is significant in several aspects. With the widespread development of battery technologies, the need for different charac-teristics and attributes of batteries is essential. As a result, specific cells are a better fit for certain operations or environments since cell attributes vary between chemistries [1]. In 2015 a research project by Badam et al. [2] was aimed at creating a battery system including different types of cells with different properties. Combining different cells was a compromise where the trade-offs between cells were hypothesized to improve overall efficiency. According to Badam et al. in their research [2], battery cells have different attributes where an increase in one usually lowers another. Many different cell chemistries aim to maximize one or more of these attributes in the current market. Badam et al. [2] explored the possibility of combining different cells and the corresponding effects. The initial findings indicate that such a combination can garner beneficial trade-offs, and some tests verified this. This thesis will continue on this path and create a system prototype to explore how different topologies involving different cells affect the system’s operation and properties in a real scenario.

The real scenario is that of a small battery-driven passenger train. Trains travelling on tracks mainly use overhead electrical lines to control electrical motors used for propulsion [3]. Many tracks have sections where overhead lines do not exist to power motors or other circumstances that limit overhead line connection [3]. There is also a need for reserve power in emergency and startup procedures of a train. In such cases, the train must operate via alternative means. This thesis will consider a system in the form of a heterogeneous battery system to evaluate the benefits in a dynamic load behaviour found in a Electrical Vehicle (EV) application. Trains especially suit such an evaluation as the tracks are well understood, allowing generalized charge-discharge profiles. Ghaviha [4] performed a simulation with a modelled charge-discharge profile of a battery-driven train, which corresponded well to actual experiments. In combination with varied charge/discharge behaviour, the start-stop behaviour creates well understood dynamic operations for evaluation [5]. The thesis is performed in collaboration with Alstom and Mälardalen University in Västerås, Sweden, aiming to explore a battery system prototype, which utilizes different types of cells used by Alstom. Alstom supplied information regarding electric train operation and characteristics of cells in use by current train implementations. Since some of this information falls under a signed Non-Disclosure Agreement (NDA), the results are generalized and certain specifics withheld.

The scope of the thesis is limited to seven attributes after discussions with the collaborative partner and academic supervisors. The attributes considered are Storage capacity related to the range of a train. More capacity allows for longer travel distances without recharge or external power connection. Charge rate, considered for charging batteries at stations, where a higher charge rate increases the frequency of trips per day. Safety regarding battery failure conditions which have severe risks due to the high energy involved. Reliability, observing the ability of a train to reach the next station in case of battery failure. Cost, in regards to battery technology over its lifetime. Battery losses originating from the heterogeneous system mainly depends on the internal imped-ance. Finally, ageing, which regards the batteries ability to perform the correct function over time, where batteries age differently depending on use case and chemistry. Table1. details the different sections in the thesis where the attributes are addressed:

Table 1.: Analysed attributes

Attribute Background Implementation Results

Storage capacity 2.2,2.6,2.15 6.4.3 7.2.1 Ageing 2.2,2.7,2.15 6.4.8 7.2.2 Charge rate 2.2,2.4,2.15 6.4.4,6.5.2 7.2.3 Safety 2.12 6.4.5 7.2.4 Reliability 2.12 6.4.5 7.2.5 Cost 2.2 6.4.6 7.2.6 Battery losses 2.2,2.7 6.4.7 7.2.7

The evaluation primarily focuses on simulation as the thesis considers a full-scale passenger train. Suitable models for the mechanical properties of the train, track specifics, battery parameters

(15)

and charge-discharge behaviour are used to evaluate the concept of mixing cell types in a real scenario similar to what Ghaviha researched [4].

The rest of the thesis has the following structure: Section2. covers the background necessary to formulate the thesis’s different simulations and the necessary electronics for a prototype im-plementation. In particular, section2.15 covers the works that support the implementation and discussion of this thesis. Section3. describes the thesis problem formulation and formulates the hypothesis, research questions and limitations. Section4. describes the method and methodology used to conduct the thesis and presents the initial time plan. Section 5. discusses the potential ethical and societal considerations. Section6. formulates the experimental setup and tools used to conduct the thesis. Section7. presents the results of the thesis. Section 8. provides a discussion of the thesis results. Section 9. provides the thesis conclusions. Lastly, the appendix contains additional information and documentation regarding the thesis.

(16)

2. Background

Battery systems have been continuously improved over the last decades and developed alongside EV applications [1]. Modern battery cells’ efficiency has increased significantly, and cell technology is continuously developed for the demanding loads of EVs [6]. Trains have under recent years been explored together with different types of battery technologies to shift away from traditional diesel engines [3]. Due to trains’ operation, batteries have been present for startup and emergency functions for some time [3]. With further developed technology, it is feasible to use batteries for propulsion when external power sources are not available. The evolution of propulsion systems has many similarities with electric cars, which have seen considerable improvements in technologies [1]. Today there exist successful development and integration of battery-driven train around the globe [7].

Due to the well-understood load cycles of the train and the static tracks they traverse, invest-igations are efficient in battery-driven trains compared to other EVs applications with relatively simple models [4]. Ghaviha [4] performed a study that aimed to optimise the use of battery-driven trains. The optimisation regards homogeneous battery implementations of Lithium-ion (Li-ion) cells and uses well-understood tracks to model energy optimisation of battery-driven trains. This thesis has a similar method, but instead of focusing on optimisation of battery operation, it will re-gard heterogeneous battery implementations. Similar to Ghaviha [4] the evaluation mainly regards simulation.

The following sections covers the different aspects necessary for the thesis implementation and provides the necessary background to interpret the results. The subsections are summarised as follows: Section2.1 describes electrical vehicles and relevant nuances of electrical trains. Section

2.2provide the necessary background regarding batteries for the thesis scope. Section 2.3 detail the different electrical converters necessary for the physical implementation. Section2.4similarly discusses the nuances of charging while section 2.5 covers the discharging needs of the physical circuit. Section 2.6 regards State of Charge (SoC) and the methods applicable for the thesis while section2.7 covers State of Health (SoH). Section2.8describes the functionality of a BMS. Section 2.9 describes the necessary background within simulation. Section 2.10 formulates the term policy and its relation with the thesis. Similarly sections 2.11, 2.12 and 2.13 provide the necessary background regarding Design of Experiments (DOE), dependability and control theory respectively. Lastly section2.15examines related work and place the thesis in relation to previous research.

2.1 Electric vehicles

Batteries are heavily application dependant, and some attributes of interest within the field are high energy density, high power density, long cycle life, and low-cost [8]. EVs is an industry that requires continually improving rechargeable batteries [8]. Due to the need for adequately long drive range, acceleration ability, and moderate maintenance [8].

Electric trains

Many train applications utilise electrical motors for propulsion [3]. Which is typically performed by connecting the train to an external power source to power the system continuously. The most common method being overhead lines connected to the electrical system [3], [9], as seen in figure

1a. In such a system, the auxiliary power consists of a battery for startup purposes or emergency operation.

Typically, trains also have a secondary system for propulsion for tracks where power lines do not cover entire tracks [3], mainly in the form of diesel engines or auxiliary batteries. Figure 1b

shows a basic block diagram of a system containing a secondary propulsion method via a battery. A secondary propulsion system that is not dependant on any external power supply allows trains to travel more tracks while providing a better ability to handle main power supply failure cases. Us-ing batteries as a secondary power source for propulsion has the benefit of beUs-ing rechargeable with external power during run-time. Operation with external power is usually classified as Catenary Operation (CATO) and Catenary-Free Operation (CFO) otherwise. [9]

(17)

(a) A basic block diagram over a electrical train lo-comotives electrical system. A pantograph supplies the train system with external power from overhead electrical lines which is then converted so that it is usable by the system.

(b) A extended basic block diagram from figure 1a

over a electrical train locomotive equipped with a battery for propulsion. If no overhead electric line is present the battery is connected to the system and acts as the power source. When overhead power is available the battery can be charged if required. Figure 1: Figure1adisplays a basic electrical train block diagram while figure1bdisplays a block diagram of an electrical train with a secondary propulsion system.

2.1.1 Track profiles

Track profiles in this thesis refer to the different parameters that apply to a train track route. Many passenger trains cyclically operate across certain travel routes between two ends stopping at predetermined station stops along the journey. Different information common for train track profiles is summarised as follows:

• Track length

• Speed limit intervals

• Relative incline or decline of the tracks • Stations and other stops

• Duration of stops

• Overhead line availability • Characteristics of overhead lines

Track profiles are the summary of this information. This model is part of the basis for the load characteristics as well as the charging. Figure 2 displays a typical track profile with the corresponding information. [9]

(18)

Figure 2: A nominal track profile curve. The distance and relative altitude of the track is shown in the blue line. The yellow circles represent station stops. The speed limit and availability of external power is placed and the corresponding distance intervals.

The tracks used for the thesis are modelled in the same fashion as from Alstom to compare simulation results.

2.2 Batteries

Batteries are devices that convert chemical energy contained in active materials into electric energy from electrochemical reactions [10]. Batteries consist of one or multiple cells connected in either series or a parallel configuration. Cells are either primary or secondary, where secondary cells are rechargeable and primary cells are not [1], [10]. Primary batteries have certain advantages such as weight, low cost, good shelf-life, ease to use, and high energy density in cases where the discharge rates are adequately low [1], [10], [11]. Secondary batteries are prevalent in applications where there is a need for storing and supplying energy on demand such as, in consumer electronics and EVs as they can be recharged rather than swapped out [10].

There is a wide variety of different batteries with different properties, where some properties affect others negatively and vice versa. Thus, some batteries are more suited for some applications than others. Some Important properties which must be taken into account when designing cells and batteries can be seen in figure3 [12].

Figure 3: Spider diagram comparing three fictional cells. 8 parameters are drawn on separate axes and each cell is placed on a scale for each parameter. As such a simple overview can be created between the different attributes.

(19)

In diagram 3, different cell attributes can be seen on a diagram to visualise their relative strengths and weaknesses. Depending on the use case of an application, cells can be selected with suitable attributes.

The following sections describe the different attributes that affect cell operation: Storage capacity

Storage capacity refers to the amount of energy stored within a battery cell, commonly measured in Ampere hours (Ah) [10]. The capacity varies significantly with the type of cell chemistry and cell construction. Ageing processes will lower the storage capacity of a cell [13].

Cycle life

The cycle life of a cell refers to several areas. In all cases, it is an estimate of when a cell has been forfeit depending on given SoH parameters [14]. The cycle life of a cell primarily depends on the use and time [13]. Most battery cells have a calendar life cycle as they will deteriorate regardless if they are in use or not since the internal chemical balance eventually becomes unstable [14]. Similarly, cells wear during use. A cell manufacturer often estimates the number of charge-discharge cycles expected from a cell before it has deteriorated a set amount [15]. Additionally, different chemistries are more or less prone to ageing processes. Since the chemical compounds differ, so does the ageing process and effects. The manufacturer also plays a role in ageing as the quality of the material and manufacturing processes affect cell ageing [13].

Rate capability

Rate capability refers to the amount of current a battery cell can be charged or discharged with, usually measured in amperes per second (A/s) [1]. A high rate capability is essential in many application areas such as EVs as it requires high currents, with irregular peaks, for propulsion systems [16]. The cell chemistry limits the rate capability as different reactions are more or less stable at high rates [1]. Discharging a battery cell beyond its rated limit risks the battery reaction going out of control [17]. Rate capability is more commonly known as C-rating, which relates the maximum discharge current per second to the maximum capacity [1]. A discharge at 1 C will discharge a battery from 100% SoC to 0% SoC in one hour, while 2 C will discharge the battery in 30 minutes.

Self discharge

Self-discharge occurs in cells during storage or other idle periods [18]. It reduces capacity due to unwanted chemical reactions, which is highly dependent on cell chemistry and type [13]. Further-more, it is highly dependant on temperature. Self-discharge occurs during more extended storage periods due to the lengthy processes. Self-discharge describes the amount of SOC lost during a period and is expressed by loss of SOC [18].

Energy density

Different cell chemistries have different energy densities depending on their chemical properties, measured as Wh/Kg [1]. As a consequence, different cells with similar storage capacities will often vary in size. Energy density is therefore important as the sizing of batteries is a key factor in many applications [2]. Most batteries are created following sizing standards, enabling simple comparisons [1].

Internal impedance

The internal impedance restrains the charge-discharge current of a cell. Internal impedance depends mainly on the chemistry of a cell and the manufacturing. Like storage capacity, internal impedance is affected by ageing processes and will increase with time. The internal impedance contributes to the heating of a battery cell. A lower internal impedance will lower the losses generated as heat and enable higher C-rates. The increase of internal impedance over a cell life cycle is mainly

(20)

due to the Solid Electrolyte Interface (SEI) layer increase due to chemical reactions. The internal impedance is expressed in ohms. [1], [19]

Coulombic efficiency

Coulombic efficiency is a measure of the losses during charge processes [1]. All energy delivered to a cell does not convert into cell capacity due to losses in the transfer. Therefore it can be described as a ratio between delivered energy and stored energy [1]. The losses can be relatively small and, in many applications, can be disregarded.

Safety

A cell’s safety regards the risk of dangerous events such as fire or explosion and the designed safety mechanisms, such as pressure release vents [20]. Safety is dependant on cell chemistry and cell manufacturing, and in most industries, safety standards dictate the level of safety necessary for applications [1], [13].

Cost

The cost of cells varies as there are differences in many areas such as manufacturing, materials, safety and life cycle replacements. Depending on the application, the cost of a cell implementation varies on several parameters. [1]

When trying to reduce the costs in battery-driven applications, multiple aspects interplay: the operating environment, rate of charging, longevity, capacity and size. Life cycle costs and W/h costs detail the relationships where different cell technologies will vary in cost over these factors. [21]

Cells

A cell is an electrochemical device providing electrical energy by converting chemical energy. Cells consist of three vital components, the negative electrode(cathode), positive electrode(anode) and electrolyte . [10], [22]

Chemical energy converts into electrical energy during electrochemical oxidation and reduction reactions; those processes occur during the discharge process and occur in the anode and cathode. The electrolyte is the medium for the charge transfer in the form of ions and is usually a type of liquid but can also be in a solid-state. [10], [22]

Equation1describes the amount of energy available to be extracted from the chemicals:

∆G0= −nF E0 (1)

Where ∆G0_{is the change of standard free energy, F the Faraday constant, E}0 _{the standard}

elec-tromotive force and n the number of electrons participating in the process. [22]

The rate capability parameter describes the charge and discharge capability of a cell and is dir-ectly affected by the cell’s internal impedance as it creates a voltage drop, which in turn results in a loss in the form of heat, further explored in section2.2. Such internal impedance is known as ohmic polarisation and is the sum of the resistances from the ionic resistance within the electrolyte, electrodes, active mass, collectors, and the contacts [22]. The potential between the anode and cathode is described by equation2.

E = E0− (ηcta+ ηca) − (ηct+ ηcc) − iRi= iR (2)

Where ηcta is the activation polarization at the anode, ηctc the activation polarization at the

cathode, ηcathe concentration polarization of the anode, ηctthe concentration polarization of the

(21)

Primary cells

Primary cells have become significant in businesses where the shelf-life, particularly, is desired since they can be stored for a relatively long amount of time and operate in applications requiring long resting times [11]. The long shelf-life suit long-term applications such as memory backup and backup systems [11]. It is possible to recharge some primary cells; this practice is, however, avoided since it is impractical as much of the energy used for the charging process is lost [1], [11]. Additionally, it might be dangerous, especially when the batteries are tightly sealed and not designed to release gases from the charging procedure [11]. In such circumstances, the cell could leak, rupture or even explode [11], [17].

Secondary cells

Secondary cells are most prominent in the automotive field, such as electric vehicles and industrial truck equipment and hybrid electric vehicles [11]. Secondary cells are also widely used at a smaller scale in consumer electronics [1], [23]. Secondary cells can be used for temporary energy storage, supply energy on-demand without replacements, and reduce costs and time spent on maintenance [1]. While having worse performance in terms of properties such as shelf-life, discharge rates, energy density, cost and capacity, secondary cells are highly coveted due to the possibility of reversing the procedure of converting chemical energy into electrical energy, allowing recharge. [11].

Cell chemistry

Batteries provide electrical energy by converting chemical energy, as described in section2.2. The process of energy conversion depends on the cell type. In a traditional primary cell, the battery cell acts as a pump that transfers electrons between the negative anode and positive cathode [1]. The anode and cathode’s relative potential originates from the chemical process between the anode materials and cathode materials interacting with the cell electrolyte. Figure4 displays a nominal primary cell consisting of a zinc anode and copper cathode.

Figure 4: A primary battery cell with zinc and copper electrodes. A redox reaction occurs in the system were the anode goes through oxidation while the cathode goes through reduction. The positively charged cathode is supplied by electrons created through the oxidation process at the negative anode.

(22)

In this cell, the redox reaction creates electrons at the cathode by oxidation of the zinc [24]. These electrons travel through the salt bridge to the copper cathode, where reduction takes place. In such a system, the chemical process eventually stops as the redox reaction cannot continue [1], [24]. By connecting the anode and cathode as seen in figure 4 the transfer of electrons creates a current that external components can use.

Li-ion cells

Li-ion based cells are used in a large number of applications and especially in EV applications [25]. A large part of the widespread use of Li-ion cells is the ability to act as secondary cells with high energy density and with a relatively high number of life cycles [1], [26]. Li-ion cells generally undergo the same redox reactions as in the case with primary cells, but the key difference is that they are rechargeable [1]. Similarly to a primary cell, a redox reaction is a mechanism for external electron travel. Figure5represents a basic schematic over a Li-ion cell.

Figure 5: This figure describes a basic schematic over a typical Li-ion cell chemistry. The Cathode and Anode in a Li-ion cell create a potential difference, creating a current flow if connected. While electrons travel the external circuit, positive Li-ions traverse the cell internally and combine with the electrons. Due to the Intercalation effect created by the Cathode and Anode structure, discharge can be reversed and repeated.

Li-ion cells can be recharged once discharged due to the Anode and Cathode structure’s in-tercalation mechanism [27]. Intercalation is a mechanism that allows chemical reactions to occur without changing the original chemical crystal structure. Li-ions are continuously transferred from Cathode to Anode in Li-ion cells and reduced and oxidated continuously without changing the internal chemical crystal structures [6]. A typical reaction that takes place in a Li-ion cell is that between lithium-cobalt oxide LiCoO2and lithium-carbon LiC6 [6] seen in equation3:

LiC6+ CoO2 C6+ LiCoO2 (3)

The reaction in equation 3is reversible and goes from left to right during discharge and from right to left during charge [6]. Different materials can be used for a Li-ion cell to create different chemical reactions, but they all involve lithium [6]. Mainly this is because of lithium’s chemical properties that produce a strong tendency for chemical reactions [13].

A Li-ion cell will experience ageing effects during its lifetime, which decreases the efficiency of the cell and its general properties [14]. A large part of ageing is due to the SEI layer produced by Li-ions traversing the cell [13]. Additionally, the cell will deteriorate over time due to corrosion,

(23)

Figure 6: Spider diagram comparing LTO and NMC to common attributes.

contamination, and other adverse effects [13]. Ageing worsens the cell characteristics, commonly described and measured as SoH, discussed in section2.7.

2.2.1 LTO and NMC

Taking the previous discussion about cell characteristics along with datasheets from Alstom into ac-count, the following comparison of attributes in figure6and table2.2.1between Lithium-Titanate-Oxide (LTO) and Lithium-Nickel-Manganese-Cobalt-Lithium-Titanate-Oxide (NMC) apply:

Table 2.2.1: NMC and LTO comparison (cells with same volume)

Parameter NMC LTO

Storage capacity (Ah) Good ≈ 50%of NMC

Nominal voltage (V) 3.7 2.2

Cycle life Good Best on market

Rate capability (C) Good Several times better than

NMC, but lower capacity.

Self discharge Similar Similar

Energy density (Wh/l) 100% ≈ 35%

Internal impedance (Ω) Similar to other Li-ion Lower

Coulombic efficiency (%) Good Slightly better than NMC

Safety Good Best on market

Cost 100% ≈ 200%

Robustness Similar to other Li-ion Best on market

Temperature range◦_C _{Similar to other Li-ion} _{Can be operated in lower and}

higher temperatures

Although the comparisons are imprecise, it clarifies that differences between the cells exist in several aspects. The differences are further visualized via a spider diagram in figure6:

2.3 DC-DC converters

Direct Current - Direct Current (DC-DC) converters convert a given DC voltage level to a variable DC output voltage level. There are two main categories of DC-DC converters, switch mode and

(24)

linear converters [28]. Linear converters are less complex and provide a more smooth output and low noise, with low efficiency. Thus, linear converters are suitable in certain limited cases, however not suitable in high power applications. Linear converters operate by reducing the input voltage across several resistors, whereas switch-mode converters operate using semiconductors for switching. Two categories divide switch-mode converters, Pulse-Width-Modulation (PWM) converters and soft-switched converters [28], [29].

PWM converters

PWM use square-wave shaped curves to adjust the output voltage level where the switch’s duty cycle is adjusted such that the average output voltage level changes. Advantages of such converters are simplicity of control; however, some downsides include significant electromagnetic interference and more losses. Controlling the output voltage can be done in several ways; a simple method is, however, to use a constant frequency and alter the duty cycle. [28]

Buck converter

A buck converter is a DC-DC converter that steps down the voltage and uses a diode, switch, and an LC filter. The buck converter may operate in two different states. In cases where the inductor current is always positive during operation, it is classified as continuous Conduction Mode (CCM) whereas the Discontinuous Conduction Mode (DCM) is engaged in cases where the input current is low or the switching frequency low. In such circumstances, the inductor current can reach zero during some duration of the switching period. [30]

The barrier between the modes is determined from the value of the inductor which is described in equation4:

Lb=

(1 − D)R

2f (4)

Where D is the duty cycle, R resistance, f frequency, Lb the inductor value, which determines the

threshold between CCM and DCM. The converter is operated in CCM as long as the inductor value L exceeds the threshold Lb, L > Lb [30].

Since the buck converter is used to step down the voltage, the output voltage will never exceed the input voltage. Equation5 depicts the relationship between the duty cycle, input and output voltage. [30]

D =Vout

Vin (5)

Figure7 display an example of a buck converter:

Figure 7: Electrical schematic of a buck converter.

A wide variety of different DC/DC converters are all constructed from the same components. Examples of such converters would be Boost converters, buck-boost converters, Ćuk converters and synchronous converters. Czarkowski further detail the nuances of different converters. [30]

(25)

2.4 Charging

Beyond the actual charging, stabilisation and terminating the charging procedure is vital. A prevalent issue when it comes to charging is the difficulties of charging fast while not negatively affecting the SoH as the lifetime of batteries is directly affected by the rise of temperature during the charging procedure [31]. Measuring the SoC, seen in section 2.6, is important when it comes to charging. An inaccurate SoC estimate can lead to undercharging or overcharging of batteries, resulting in irreversible damage.

Batteries have different maximum charging rates, and the relative charging rates highly affect a battery’s longevity, seen in section2.2. The charging process’s speed is expressed as a multiple of the C-rating of the given battery. The charging process is either slow, quick or fast [1]. Where slow charging usually occurs at 0.1C and typically takes 14-16 hours, Quick charging occurs at approximately 0.3C and takes 3 to 6 hours. Finally, fast charging takes less than an hour and charges the battery at a rate of 1C or above [32]. As charging rates increase, so does safety concerns. Thermal fuses and monitoring the voltage makes the charging procedure more safe [32]. However, fast charging circuitry is more complex and depends on the cell chemistry of the considered cell.

Generally, three chemical processes occur during the charging procedure; initially, charge trans-fer occurs, which lasts for one minute or less; the diffusion process then takes place and can last for several hours. Finally, the intercalation process occurs in some batteries such as Li-ion [32]. There are many different potential methods to charge batteries [33]. Charging profiles describe the different methods which highly affect the battery’s health and performance. The different charging methods have different advantages and disadvantages to be considered [34]. Some of the popular charging methods are

• Constant Current-Constant Voltage (CC-CV) • Pulse charging

• Trickle charging • Negative pulse charging

Figure8 describes a typical Li-ion charging procedure:

Figure 8: A typical charge curve for a Li-ion battery cell. Trickle charge is necessary if the cell voltage is low not to damage the cell, which is the interval between Vminand Vstart. As the voltage

is between Vstart and Vmax, constant current applies until the measured OCV reaches a critical

point Vmax. At Vmaxthe voltage is kept constant, and the current decreases until the voltage reach

(26)

Constant voltage

Charging batteries using only constant voltage takes a longer time than maintaining a constant current; however, it does not require complex hardware. [35]

Constant Current

Constant current charging also does not require as complex hardware as the pulsing charging procedures but requires voltage monitoring of the battery to prevent overcharging. [35]

Constant current-constant voltage

CC-CV is the most prominent charging method for Li-ion batteries. This method combines the constant voltage and constant current methods and does not require complex hardware than other pulse charging methods.

The Open Circuit Voltage (OCV) at the beginning of the charging determines the procedure. If the voltage level is below threshold Vstart in figure8, a small amount of constant current is used

to charge the battery, commonly around 0.1C. Such charging procedure is also known as a trickling charge. This process proceeds until the OCV reaches the Vstart threshold. A constant current is

then applied until Vmax is reached. The remaining charge is then supplied in a constant voltage

fashion until the current reaches the final value. [34] Trickle charging

Trickle charging uses low currents of around 0.1C to charge at slow rates to limit the chemical reactions. Usually, trickle charge aims to fully charge batteries at high voltages or start charging processes at deep discharges. [36]

Pulse charging

Pulse charging is similar to constant current charging except for pulsing the current such that there are pauses which allows the chemical reactions to settle during charging. Pulses can decrease the charging time while also reducing negative SoH effects. [35]

Negative pulse charging

Brief discharging pulses combined with pulse charging aims to reduce the charging time by improv-ing the Coulombic efficiency [37]. Negative pulse charging is an optimal way of charging since it takes the chemical reactions into account; it is, however, more complex and requires bidirectional current flow [36].

Train charging

In electric trains, there are several aspects to consider during charging. The charging process of a train considers three significant sections: Charge during CATO, charge during regenerative braking and charge during standstill. [9]

Charge during CATO is performed using any excess power available from driving the motors and auxiliary systems on the train [9]. The available power is dependant on the needs of the rest of the system and varies accordingly. The sum of the available power is modelled by the sum of the required power and available power, discussed in section 2.15 [4]. Losses originate from the conversion of the overhead lines to the battery specific voltage and current, depending on the different factors of the train used [5].

Regenerative braking converts the mechanical energy used for braking into electrical energy used for the battery, with conversion efficiencies up to 70% for trains [4], [9]. Regenerative braking converts energy to how the motors convert energy [4]. The energy recovered in CFO charges the battery equal to the total power gained from braking with any auxiliary power demands subtracted [9]. During CATO, regenerative braking delivers energy back into the power grid [38].

(27)

During standstill, trains charge via external connections. This type of charging can either occur during more extended periods of standstill or at stations during travel [9]. Charging at a standstill is more controlled as the power demands will not vary greatly [38].

2.5 Discharging

The discharge process of a battery cell is an important factor in the applicational area of batteries as it affects several properties of battery cells. Figure9 represents a nominal discharge curve for Li-ion cell chemistries.

Figure 9: A nominell discharge curve from a Li-ion cell. The blue line represents the SoC of the battery cell during the discharge process while the green line represents the cell voltage.

During the start and end of the operational range, the battery voltage’s behaviour is exponen-tial, while in between, it behaves linearly. Depending on the C-rate of the discharge, the relative capacity of most battery cell chemistries will vary, often with large differences [39]. There is a lower relative capacity during high C-rate discharge, while a higher relative capacity is experienced dur-ing low C-rate discharge [1]. In many applications of EVs the capacity experienced is greater than its specified value as the discharge process is often not continuous and occurs over several hours [1]. The discharge process depends on the load supplied by manufacturers or gathered through testing. Proper battery constructions act so that the relative capacity does not suffer from the discharge characteristics. Load characteristics can be modelled for applications with great accuracy if certain information is delivered. Discharge behavior is further discussed in section2.15.

2.6 State of charge (SoC)

SoC acts as the fuel gauge for a cell or battery [40]. It refers to the available capacity in relation to the total capacity at a full charge with the relationship seen in equation6:

SoC = Ccur Cbat

∗ 100% (6)

Where Ccur refers to the actual capacity and Cbat the capacity at full charge.

It is common for battery cells capacity to be limited in the charge and discharge interval since certain levels of SoC increases ageing processes [13]. A typical range for a operational SoC interval is between 85% and 15% [14]. Waag, Fleischer and Sauer [40] classify different means of SoC estimation for EVs in the following fashion:

• Ampere-hour counting • OCV-based estimation • Model-based estimation • Impedance based estimation

(28)

• Estimation using fuzzy logic and methods of machine learning • Estimation using special measurement techniques

• Monitoring of each cell in a series connection

Not all of the methods described by Waag, Fleischer and Sauer [40] are applicable for the thesis due to the scope. The following methods apply to the thesis:

Ampere-hour counting

Ampere-hour counting is a simple method of measuring SoC as it directly measures the SoC by measuring the battery output. If a good understanding of the cell exists, a proper ampere-hour counting will directly link the SoC without complex computation. As it is a direct translation of measurement, the measurements must be accurate, or errors can accumulate over time, causing the estimation to become erroneous. Regular calibration is a good practice in more extended operating scenarios, but it is difficult to calibrate correctly as the cell ages, changing the characteristics. As it is a direct measurement, it can be used in conjunction with other estimation methods to give an accurate estimation of the current SoC of a cell or battery. Different adaptive algorithms can, for example, collaborate for error correction of an ampere-hour counting application. [40]

OCV-based estimation

Measuring SoC is relatively straightforward under the appropriate conditions. For physical cells, the simplest form is to measure the OCV since there is a relationship between SoC and OCV [40]. A OCV measurement is not always an accurate measurement of the actual battery voltage due to battery overvoltage phenomena after discharge or charge [1]. Therefore, during discharge or charging, batteries may experience altered OCV values, and measurements are not accurate for the actual voltage. Once charging or discharging is complete, a relaxation period starts where the actual voltage will slowly become the OCV [40]. This period lasts for several days and can not always guarantee good OCV measurements in many battery applications. Exact measurement of OCV is not always necessary for accurate estimation of SoC, and therefore OCV measurements are usually incorporated in the SoC estimation regardless of accuracy. A few hours of relaxation of the OCV reduces the error by a large margin and is used in many applications as a basis for SoC estimation since the relative error is relatively low. There exist several different means of extending a OCV measurement to estimate the correct SoC during discharge or charge and immediately after or during the initial relaxation period [26].

Model-based estimation

Model-based estimation is a mean to predict accurate SoC values during charge or discharge [25]. With a battery behaviour model, actual measurements from a battery can be fed into the model to estimate the actual value [40]. It is also possible to extend a simulation model to alleviate the need for real experiments or tests. Section2.15of the report covers the modelling of battery cells and systems.

2.7 State of health (SoH)

SoH is an attribute used to describe the relative health of a battery cell. Unlike other attributes in battery systems, SoH is subjective and defined by the user needs [14]. Depending on the type of operation and cell, the user might consider different aspects of the essence for the SoH estimation. Degradation of cells due to ageing effects, abuse and usage will affect the characteristics of cells throughout service life [13]. Common things to measure in cells to determine the SoH is capacity and internal impedance as it commonly acts as a benchmark [13], [14]. Both attributes are measurable via different means and directly affect cell characteristics. Considering the initial capacity to be 100%, a deviation of 20% is a typical indicator that a cell has been forfeit [14]. Figure10 is a typical curve displaying the ageing effects in a Li-ion battery cell during constant discharge:

(29)

Figure 10: The ageing effects on a Li-ion cell. The curve represents a constant discharge at a fixed C-rate and the typical effects of ageing.

Malmberg [15] provides a grouping of SoH measuring methods but discern model-based method from experimental methods, which aims to separate experimental methods based on stored oper-ational data instead of estimation. Hossain Lipu et al. [14] provides another grouping of methods for SoH estimation in the following manner with common example methods:

• Direct assessment approach – Coulumb counting – Open circuit voltage – Impedance spectroscopy • Adaptive approach

– Kalman filtering – Particle filtering – Least square • Data driven approach

– Fuzzy logic – Neural network

– Support vector machine • Others

– Sample entropy

– Probability density function

Since SoH requires measurements over an extended period with high-quality measurements, it is not part of any thesis implementation. Instead, different levels of static SoH levels apply for cells during the thesis.