PROJECT SEMESTER REPORT

on
DEVELOPMENT OF BATTERY MANAGEMENT SYSTEM
MODEL FOR ELECTRIC AND HYBRID VEHICLES

Submitted by
Pranav Mathur
101504089

Under the Guidance of

Rouble Singh Sandhu Dr S. K. Jain
Sr. Engineer Associate Professor
FEV India Pvt. Ltd. TIET

2019

Electrical and Instrumentation Engineering Department

Thapar Institute of Engineering and Technology, Patiala
(Declared as Deemed-to-be-University u/s 3 of the UGC Act., 1956)
Post Bag No. 32, Patiala – 147004
Punjab (India)

1|Page
 TABLE OF CONTENT

Declaration 4
Acknowledgement 5
Abstract 6
List of Tables 7
List of Figures 8-9
List of Abbreviations 10

Chapter 1: Introduction 11-20

1.1 About FEV India
1.2 Capabilities at FEV India
1.3 Categories of Hybrid Vehicles
1.4 Role of Electric Motor
1.5 Cells and Batteries
1.7 Cell Chemistries
1.8 Li-ion Batteries
1.9 Definitions
1.10 Battery Management Systems
1.11 Standards Used

Chapter 2: Equivalent Cell Model 21-25

2.1 Requirement of Cell Model
2.2 Equivalent Circuit of Cell
2.3 Input Data
2.4 Output Data
2.5 Model Overview
2.5.1 Detailed Model
2.5.2 Different Components of Voltage
2.5.3 Calculation of Current through Diffusion Voltage
2.5.4 Calculation of Hysteresis Value

Chapter 3: State-of-Charge Estimation 26-32

3.1 Introduction
3.2 Coulomb Counting Method
3.3 Neural Networks
3.4 Model Based Estimators

2|Page
 3.5 Input Data
3.6 Output Data
3.7 Model Overview
3.7.1 Step 1a: State Estimate Time Update
3.7.2 Step 1b: Error Covariance Time Update
3.7.3 Step 1c: Predict System Output
3.7.4 Step 2a: Estimator Gain Matrix
3.7.5 Step 2b: State Estimate Measurement Update
3.7.6 Step 2c: Error Covariance Measurement Update

Chapter 4: Active Cell Balancing 33-38

4.1 Need Cell Balancing
4.2 Methods of Cell Balancing
4.3 Circuit Complexity and Cost
4.4 Mechanism of Active Cell Balancing
4.5 Detailed Model 1
4.6 Detailed Model 2

Chapter 5: Observations and Results 39-40

5.1 Results
5.1.1 Cell Equivalent Circuit
5.1.2 State-of-Charge Estimation
5.1.3 Active Cell Balancing
5.2 Justification of Objectives Achieved
5.3 Future Work

Chapter 6: Project Metrics 41-44

6.1 Challenges Faced and Troubleshooting
6.2 Relevant Subjects
6.3 Interdisciplinary Aspect
6.4 Student Outcome (A- K) Mapping

References 45-46
Annexure A: Daily Diary 47-48
Annexure B: Data Used in Models 49-50
B1. Data for Open Circuit Voltage
B2. Variables Used in SoC Estimation Model
B3. Different Parameters of the Battery
Plagiarism Report

3|Page
 DECLARATION

I hereby declare that the project work entitled “Development of Battery Management System model
for electric and hybrid vehicle” is an authentic record of my own. This work has been carried out
at FEV India Pvt Ltd., Pune, as per the requirements of project semester. The work has been done
under the supervision of Mr. Rouble Singh Sandhu and Dr S.K. Jain during Jan-June 2018.

Date: __________ Pranav Mathur

101504089

The above mentioned is correct as per the best of my knowledge and belief.

Dr S.K. Jain Mr. Rouble Singh Sandhu

Faculty Supervisor Sr. Engineer,
TIET FEV India Pvt. Ltd.
Host Mentor

4|Page
 ACKNOWLEDGEMENTS

I would like to extend my heartfelt gratitude to all those who have contributed towards the successful
completion of the project especially the project mentors Mr. Rouble Singh Sandhu and Dr S.K. Jain.
I would also like to take this opportunity to thank the colleagues at FEV India specially Mr. Emran
Ashraf and Mrs. Sarah Ahmed, for their constant guidance and motivation.

Pranav Mathur
101504089

5|Page
 ABSTRACT

The rise of electric vehicles is seen as the next big technology change which is going to
affect how humans move around. With the advancement in internet-of-things (IOT) and
autonomous vehicle technology, EVs have proved to be an ideal platform to bring additional
technologies to our vehicles. The main source of energy for EVs is the batteries. From the
different chemistries available, Li-ion batteries have proved to be more advantageous than
others. Though they have many advantages, they require advanced monitoring circuitry to
optimise their performance. This optimization means more power and energy available to
your vehicle.

This project has been divided into three different parts. The simulations are performed in the
MATLAB/Simulink environment. They have been performed using Modern Indian Drive
Cycle (MIDC) which the legislative norm for emission requirements in India. The model for
cell equivalent circuit (CEC) has been developed which takes power supplied by the battery
and manufacturer data for Open Circuit Voltage (OCV), to provide terminal voltage and
output current. The output of CEC has been used in the model developed for estimation of
State-of-Charge (ESOC). It uses Kalman filter algorithm to predict and correct the values of
SoC. The third model developed is of Active Cell Balancing which balances the SoC of cells
in order to optimise battery usage. The output of all the models have been verified with data
from existing models.

6|Page
 LIST OF TABLES

Figure No. Caption Page No.

1.1 Commonly used cell chemistries 15
1.2 Standards Used 19-20
6.1 Troubleshooting chart 41-42
6.2 Relevant subjects 42
6.3 A-K mapping 43-44
B1 OCV data using SOC at 20o C and 40o C 49
B2 Different parameters used inside model 49-50
B3 Different parameters of battery 50

7|Page
 LIST OF FIGURES

Figure No. Caption Page No.

1.1 Company logo 11
1.2 Categories of hybrid vehicles according to powertrain configuration 12
1.3 Chevrolet Bolt’s electric motor 13
1.4 Tesla Model S motor cut image 13
1.5 Induction motor characteristics 14
1.6 Li-ion battery pack 14
1.7 Charge/discharge process inside Li-ion cell 15
1.8 Different cell structures 16
1.9 Comparison in energy density of different cell chemistries 17
1.10 Battery Management System in embedded form 18
1.11 BMS communication using master-slave bus topology 18
1.12 Modified Indian Drive Cycle (MIDC) 19
2.1 Cell equivalent circuit 21
2.2 Cell equivalent model at the top level 22
2.3 Detailed cell model: 2nd layer 23
2.4 Different components of voltage 24
2.5 Model for diffusion voltage calculations 24
2.6 Model for calculation of hysteresis 25
3.1 Flowchart of electric powertrain 26
3.2 Different steps of extended Kalman filter 28
3.3 Step 1a: State estimate time update 29
3.4 Step 1b: Error covariance time update 29
3.5 Step 1c: Predict system output 30
3.6 Step 2a: Estimator Gain Matrix 31
3.7 Step 2b: State Estimate Measurement Update 31
3.8 Step 2c: Error Covariance Measurement Update 32
4.1 Electrical circuit for passive cell balancing 33
4.2 Electrical circuit for Active cell balancing 34
4.3 Model 1 for active cell balancing 35
4.4 Control functions of the ACB1 model 36
4.5 Signal processing blocks 36
4.6 Simscape part of ACB1 model 37

8|Page
 4.7 Model 2 for active cell balancing 37
5.1 Comparison between OCV and terminal voltage 39
5.2 SOC comparison waveforms 39
5.3 Balancing of different cells over the period of time 40
A1 Daily diary snippet 1 47
A2 Daily diary snippet 2 48
A3 Daily diary snippet 3 48

9|Page
 LIST OF ABBREVIATIONS

Abbreviation Description

EV Electric Vechile
AC Alternating Current
ICE Internal Combustion Engine
BMS Battery Management System
Li-ion Lithium-ion
CC Constant Current
SOC State of current
SOH State of health

10 | P a g e
 CHAPTER-1
INTRODUCTION

1.1 About FEV India

FEV (Forschungsgesellschaft für Energietechnik und Verbrennungsmotoren) India, Pune

is the Indian subsidiary of FEV Europe GmbH, an automotive technology company based
in Aachen, Germany. It was founded by Prof. Franz Pischinger in 1978 who headed the
Institute for Applied Thermodynamics at the Technical University of Aachen. FEV India
provides automotive services like engine testing, noise-vibration-heat testing, design,
simulation and calibration of different aspects of powertrain.

We have grown into an internationally-recognized leader in the design and development

of internal combustion engines, conventional, electric, and alternative vehicle drive
systems, energy technology, and a major supplier of advanced testing and instrumentation
products and services to some of the world's largest OEMs.

In 1985, FEV expanded its business into North America, initially in California, and then
relocating in 1987 to suburban Detroit, Mich., where FEV’s North American Technical
Centre, constructed in 1997, is located.

After significant growth and expansion in Europe and the U.S., FEV entered the Chinese
market with the establishment of an office in Beijing, China in 1998. Entering the new
millennium, with its Asian business growing rapidly, FEV established its Asian Technical
Centre in Dalian, China, providing the company with a powertrain and vehicle
development facility to support its customers in that region.

FEV provides services to leading automobile manufacturers like Daimler AG, TVS, Tata
Motors, Ashok Leyland etc. Engine simulations are performed on GT Suite while
MATLAB/Simulink is used for electronics and calibration purposes.

Fig.1.1 Company logo

11 | P a g e
1.2 Capabilities at FEV India

With a technical centre in Pune and an office in Chennai, FEV offers the full range of
services to our Indian clients. These services include:

▪ Comprehensive combustion and mechanical development, emission and endurance test

programs
▪ Development of hybrid and electric vehicles
▪ Vehicle application and vehicle development
▪ Engine design & CAE solutions
▪ Manufacture / assembly of test benches for customized solutions and requirements
▪ of OEMs
▪ Available skilled manpower for FEV advanced test system commissioning and after
sales services

1.3 Categories of hybrid vehicles

Fig.1.2 Categories of hybrid vehicles according to powertrain configuration

▪ P0: When the motor is connected to provide only start/stop torque in parallel with the
IC engine.
▪ P1: When the motor is connected in series with IC engine to provide additional boost
throughout the vehicle motion.
▪ P2: When the motor is placed after clutch either in series or parallel
▪ P3: When the motor is connected in series/parallel at the output
▪ P4: When the motor is connected directly to rear axle

Depending on the type of hybrid vehicle, the power demand is distributed between the
engine and the motor.

12 | P a g e
1.4 Role of electric motor
Induction motors are used in hybrid and electric vehicles provide additional torque and
power to the wheels. In hybrid vehicles there are 5 different commonly used configurations
which we have seen in the previous section. The rating of motor in hybrids depend on their
configuration. The power and torque output of a motor in P0 category will be much lesser
than P2 which in turn will be less than P3, P4 and P5. The corresponding size of battery
will also keep on increasing as the power and torque output increases.

Fig.1.3 Chevrolet Bolt’s electric motor[1]

In some cases, permanent magnet electric motor is also used such as Tesla Model 3. As in
induction motors where electricity is used to generate magnet field, the permanent magnet
electric motors do not require this step and hence are more efficient. The only drawback is
that the power output of permanent magnet type motors is limited while AC induction
motors can provide higher power outputs.

Fig.1.4 Tesla Model S motor cut image[1]

As a rule of thumb 100kW is almost equivalent to 134 hp in mechanical terms. Companies

like Toyota, Nissan, Chevrolet and Tesla are testing various technologies to improve the
overall efficiency of motor.

The advantage of electric motor is that we get high torque when the vehicle start from a
standstill position. Hence, it can accelerate much faster as compared to using only internal

13 | P a g e
 combustion engine. Similarly, to come to halt, the vehicle can decelerate much faster. This
can be seen in the characteristics below:

Fig.1.5 Induction motor characteristics [2]

1.5 Cells and batteries
Cells: The smallest individual unit in battery pack which outputs a voltage based on its
chemical properties and charge state:

Two basic classes of cells:

▪ Primary cells: These cells are non-rechargeable. Once discharged they have to be
replaced with another cell.
▪ Secondary cells: These cells are rechargeable and have a certain number of life-
cycles. They have to be replaced after they complete all life-cycles. For charging, they
need a special circuit attached.

When different cells are combined in different configurations, we get battery or battery
packs.

Fig.1.6 Li-ion battery pack

1.6 Cell chemistries

There are different cell chemistries developed at different points of time. The battery
pack used in gasoline vehicles are lead-acid type. They serve as an auxiliary purpose like
powering headlights, auto-start, mp3 player etc. While in electric vehicles the main
purpose of battery is to provide energy for locomotion.

14 | P a g e
 Table 1.1 Commonly used cell chemistries
Electrochemistry Negative Positive Electrolyte Nominal
Electrode Electrode Voltage
Lead acid Pb 𝑃𝑏𝑂2 𝐻2 𝑆𝑂4 2.1V
Dry cell Zn 𝑀𝑛𝑂2 𝑍𝑛𝐶𝑙2 1.6V
Àlkaline Zn 𝑀𝑛𝑂2 KOH 1.5V
Nickel Cadmium Cd NiOOH KOH 1.35V
Nickel Zinc Zn NiOOH KOH 1.73V
Zinc air Zn 𝑂2 KOH 1.65V

1.7 Li-ion batteries

Li-ion batteries have replaced other battery chemistries due to several advantages. Their
main advantage is their high charge density which means more energy can be stored inside
them. This makes them very useful when it comes for applications where high amount of
power is required like mobility. They also operate on higher voltages(around 4.1V) hence
more power output as compared to other types like lead-acid(around 3.2V) and nickel
hydride(1.2 V). Their minimum voltage limit is 2.8V.

Fig.1.7 Charge/discharge process inside Li-ion cell[2]

Structure of Li-ion cells:

▪ Cylindrical, they are cylindrical in design made by rolling.
▪ Prismatic, they are rolled around a flat plat.
▪ Pouch, these are made by stacking plates on top of another, hence maximum space
efficient.
Depending on the format of cell being used, the other functions like cooling methods,
pack stacking design is affected.

15 | P a g e
 Fig.1.8 Different cell structures[1]
Advantages of Li-ion cells:
▪ Good energy density.
▪ Simple charging algorithm
▪ High Battery Capacity
▪ Low resistance
▪ Comparatively Short charging time
▪ More number of charging cycles

Disadvantages of Li-ion cells:

▪ Protection circuitry is required for maintenance
▪ It degrades easily when overcharged or undercharged.
▪ Fast charging is not possible at temperatures less than 0 degrees.
▪ Expensive

Different chemistries of Li-ion cells:

▪ Lithium Cobalt Oxide:
These cells contains cobalt oxide at cathode and graphite carbon at anode. These cells a
regenerally used for mobile phones, cameras and laptops. The cathode of these cells hav
e a layeredstructure. During discharging the the Liion move from cathode to anode. The
se cells havecomparatively less life span.

▪ Lithium Manganese Oxide (Li-Mn2O4 cathode, graphite anode):

These cells hve a three dimensional spinal structure which helps in flow of ions. It provi
de faster charging and higher discharging current. The have lower internal resistances.

▪ Lithium Nickel Managanese Cobalt Oxide:

These cells are made by combining Nickel and Manganese. These cells contains the adv
antages ofboth Nickel and Manganese and provides more specific energy density and le
ss internal resistance. These cells have some stability problem.

16 | P a g e
 ▪ Lithium Iron Phosphate (Li-Fe-PO4):
These cells provide better electrochemical properties and also less resistances. These cell
s are moreestable and do not get easily stressed under abnormal conditions like overcharg
e and over-discharge. They are often used at the place of lead acid batteries.

Out of these LiFePO4 is the most popular cell chemistry with uses in vehicles, space
exploration, grid storage etc.

Fig.1.9 Comparison in energy density of different cell chemistries[2]

1.8 Definitions
Nominal Capacity: Cell nominal capacity specifies the amount of charge that the cell is
rated to hold(in Ah or mAh)
▪ It is printed on the body of cell outer shell
▪ It is an average quantity specified by the manufacturer

C-rate: This is defined as the constant-current charge/discharge rate a cell is able to

provide for one hour.e.g. A 30Ah cell can provide 30A current for 1 hour or it can provide
3A for 10 hours.

Energy: It is cell nominal voltage multiplied by the cell capacity(kWh)

1.9 Battery Management Systems

We use Battery Management Systems (BMS) along with battery packs to perform different
functions. It is an embedded system with computational power varying according to
dimensions of the battery pack and the algorithms being used to calculate several variables.

We need BMS for the following reasons:

▪ To protect the user/operator
▪ To protect battery pack from damage and avoid replacement
▪ To maximise the battery output(power/energy)
▪ To improve the battery life cycles and its usage

17 | P a g e
 Fig.1.10 Battery Management System in embedded form

All applications do not need a BMS as it is an additional investment. If the battery is cheap
enough, it might be easier to change it rather than investing in costly BMS. Usually for
applications such as electric vehicles we need an advanced BMS while for applications
such as mobile batteries, a BMS with basic functionality will suffice.

A BMS is required to perform following tasks:

▪ Monitor voltage, current and temperature of cells through slave BMS.

▪ Estimate state of charge using the data gathered.
▪ While charging/discharging make sure upper and lower charge limits are followed and
cut-off whenever necessary.
▪ Detect any faults like short-circuit which might damage the battery pack.
▪ If temperature rises above a point, BMS needs to warn the user and if necessary, cut-
off the battery.

Fig.1.11 BMS communication using master-slave bus topology[1]

18 | P a g e
1.10 Standards Used

This project has been done following the widely used standards of various components
which are listed as follows:

The Modified Indian Drive Cycle(MIDC) has been used in all the three models developed.
This drive cycle has been developed by Automotive Research Authority of India(ARAI)
for testing of vehicles before they receive commercial licences. It has two components:
▪ Four cycles to represent city driving. The maximum speed during this time is 50kmph.
▪ One cycle to represent highway driving. The maximum speed during this cycle is
90kmph.

Modified Indian Driving Cycle(MIDC)

100
90
80
70
60
50
40
30
20
10
0
239

613

987
1
35
69
103
137
171
205

273
307
341
375
409
443
477
511
545
579

647
681
715
749
783
817
851
885
919
953

1021
1055
1089
1123
1157
Fig.1.12 Modified Indian Drive Cycle (MIDC)

Table 1.2 Standards Used

Standard Title

ISO 12405-3 This part specifies the values to be used in

various tests of over-voltage, short-circuit
currents etc.

ISO 12405-4:2018 Performance testing of Li-ion battery packs,

test specifications.

19 | P a g e
 SAE J2293 Energy Transfer System used in Electric
Vehicles

Embedded system developed for road

vehicles is of appropriate rigour.
ISO 26262

This is the legislative Indian drive cycle for

emission norms. It is used by manufacturers
MIDC Drive cycle for testing and simulation of vehicles in India.

20 | P a g e
 CHAPTER-2
EQUIVALENT CELL MODEL

2.1 Requirement of cell model

A cell model is used in a number of applications:
▪ To estimate various parameters like:
▪ State-of-Charge
▪ State-of-Health
▪ Available power and energy
▪ To simulate different effects and optimise the output at system level.
▪ To co-simulate different designs for evaluation.

2.2 Equivalent circuit of cell

The battery has three major areas of voltage rise/drop:
▪ Internal resistance: This is due to the basic resistance present in electrode material,
electrolyte etc. inside a battery. It is modelled with a resistance whose value is usually
specified by the manufacturer.
▪ Diffusion voltages: Lithium ions tend to get concentrated on one region while
charging/discharging. When the cell rests, these ions then diffuse to different parts of
the electrode and hence we observe a slight gain or drop in the terminal voltage. This
is modelled with two R-C parallel branches.
▪ Hysteresis: This is the effect of user on the battery. Each user has a different style of
driving and its effect on the battery will be different. Many a times, this parameter is
ignored due to its small effect.

Fig.2.1 Cell equivalent circuit[2]

2.3 Input data
We take the following input data in this cell model:
▪ Power required: The ECU of the car divides the total power required between the IC
engine and battery. As we are running Modified Indian Drive Cycle(MIDC) on a P0
hybrid vehicle, we take the power required by the battery only. We ignore the functions
happening in other parts of the vehicle.

21 | P a g e
 ▪ Temperature and SOC: These are the data provided by the manufacturer. Using these
two data, OCV of the battery can be determined.
▪ Current: This is used iteratively from the output and is used in calculating the power
loss due to internal resistance of the battery.

2.4 Output data

There are two important quantities that we obtain after running this model:
▪ Terminal voltage: This is the voltage that we obtain when we take into account
voltage drops due to internal resistance, diffusion voltages and hysteresis.
▪ Current: Using the input data of power required and the terminal voltage calculated,
we obtain output current of the battery. This data is used again in the next iteration to
calculate voltage drop due to internal resistance and diffusion voltages.
This output data is then used in the second model which estimates State-of-Charge.
2.5 Model overview
The model on the top level is shown in the following figure:

Fig.2.2 Cell equivalent model at the top level

The model takes in the input value of power, temperature and SOC from the workspace
and uses the current from last iteration to calculate output voltage of this iteration. This
output voltage is then used to calculate output current of this iteration using the power
required.
2.5.1 Detailed model: 2nd layer
This model shows the working of different components that we are trying to calculate to
find terminal voltage.
▪ Using manufacturer data, we calculate a factor referred in the model as ‘fac’. This factor
is used along with current to calculate final hysteresis value. The hysteresis value is
added to OCV when discharging and subtracted from OCV while charging.
▪ It calculates the value of current in the resistor of RC branch and then uses it to calculate
diffusion voltage drop.
▪ SOC is taken through a lookup table which interpolates and extrapolates OCV values
at temperature 20𝑜 C and 40𝑜 C at different SOC ranging from 0.1 to 1 in steps of 0.1

22 | P a g e
 Section 1.5.2
Section 1.5.4

Section 1.5.3

Fig.2.3 Detailed cell model: 2nd layer

2.5.2 Different components of voltage
We take all the components of voltage and obtain final terminal voltage. Each component
has a different purpose:
▪ The open circuit voltage (OCV) is taken first. This is specified by the manufacturer who
performs several cycles of slow charging/discharging to obtain the specified values in
datasheet.
▪ The hysteresis takes into account the past usage of battery. There are two types of
hysteresis values: static and dynamic, both are accounted in this model.
▪ Diffusion voltages takes into the account the voltage rise or drop after 𝐿𝑖 + ions reach
equilibrium state when the battery is at standstill after being charged or discharged.
▪ Internal resistance drop is due to the extra salts which sometimes get formed due to
unwanted chemical reactions inside a cell.

𝑣 = 𝑂𝐶𝑉(𝑇, 𝑆𝑂𝐶) + 𝑚 ∗ ℎ + 𝑚0 ∗ 𝑠 − 𝑅𝑃𝑎𝑟𝑎𝑚 ∗ 𝑖𝑟𝑐 − 𝑖 ∗ 𝑅0 𝑃𝑎𝑟𝑎𝑚 (2.1)

where,
v is the terminal voltage of battery pack
OCV (T, SOC) is the open circuit voltage which is dependent on temperature and SoC
m is the parameter provided by the manufacturer to calculate dynamic voltage
h is the hysteresis value updated in every iteration
m is the parameter to calculate dynamic hysteresis
m0 is the parameter to calculate static hysteresis
s is the sign of current which tells us whether the battery is getting charged or discharged
RParam is the resistance value specified by the manufacturer
R0Param is the internal resistance value of the battery

23 | P a g e
 Fig.2.4 Different components of voltage
2.5.3 Calculation of current through diffusion voltage
To calculate voltage, drop due to diffusion of Li-ion inside electrodes, we need to find the
small current which is created dur to this diffusion of ion. Once we have this small value
of current, we can simply multiply it with resistance of electrodes to obtain voltage
drop/rise:
𝑖𝑟𝑐(𝑘 + 1) = 𝑅𝐶𝑃𝑎𝑟𝑎𝑚 ∗ 𝑖𝑟𝑐(𝑘) + (1 − 𝑅𝐶𝑃𝑎𝑟𝑎𝑚) ∗ 𝑖1(𝑘) (2.2)
where,
irc is the output of the block, it is the current due to diffusion of Li ions inside the
electrode
i1 is the output current of the battery model from previous iteration
RCParam is the parameter specified by the manufacturer to model diffusion voltages

(a)

(b)

Fig.2.5 Model for diffusion voltage calculations

24 | P a g e
2.5.4 Calculation of hysteresis value
We calculate hysteresis voltage drop/rise in two steps. They have been modelled using
eq.2.3 and eq.2.4. In equation 2.3, we calculate an intermediate variable referred here as
‘fac’. In the exponential power we calculate the ratio of charge gained/lost by the battery
with its total capacity. The exponential then gives us the rate of change of this
charge/discharge. We use this in eq.2.4 to update the previous hysteresis value. The
function ‘sign’ is being used to find whether the output current is supplied to or from the
battery depending on whether the battery was getting charged or discharged previously.

𝑔∗𝑖𝑘
−| |
𝑓𝑎𝑐 = 𝑒 3600∗𝑞 (2.3)
where,
fac is the intermediate variable which gives us the rate of change during
charging/discharging
g is a parameter given by manufacturer for hysteresis calculations
q is the total capacity of the battery in Ah, which is constant

ℎ = 𝑓𝑎𝑐 ∗ ℎ + (1 − 𝑓𝑎𝑐) ∗ 𝑠𝑖𝑔𝑛(𝑖𝑘) (2.4)

where,
h is the value of hysteresis which updated in every iteration
ik is the value of output current of battery
sign is the function being used to detect whether the current is supplied to or from
battery

(a)

(b)

(c)

Fig.2.6 Model for calculation of hysteresis

25 | P a g e
 CHAPTER-3
STATE-OF-CHARGE ESTIMATION

3.1 Introduction
State of charge (SOC) is defined as the ratio of available capacity of the battery to its
rated capacity. To put check on different conditions, a parameter called State-of-Charge
(SOC) is used. This parameter cannot be directly obtained like voltage or current. It has
to be estimated indirectly using different mathematical algorithms. There are several
algorithms to estimate SOC.

▪ Coulomb Counting method

▪ Extended Kalman Filter method
▪ Unscented Kalman Filter method
▪ Using neural network method

Fig. 3.1 Flowchart of electric powertrain [2]

3.2 Coulomb counting method

Coulomb counting method is the simplest techniques but its accuracy decreases as the
dataset increases. This method has been used at FEV India also in their full-scale vehicle
model, it simply takes integrates the Ampere-hours with respect to time to arrive at total
battery charge (while charging or discharging). The inherent problem with this approach is
that it integrates the error also while calculating total charge through the battery in a given
time. As the dataset increases the error keeps on increasing and hence the result is not
accurate.
In this method, SoC at time t is given by:
1 𝑡
𝑧(𝑡) = 𝑧(0) − ∫ Ƞ(𝑡). 𝑖(𝑡). 𝑑𝑡 (3.1)
𝑄 0
where,
Ƞ(t) is the coulombic efficiency at time t
z(0) is the initial SoC
z(t) is the SoC at time t
i(t) is the output current from battery at time t
Q is the rated capacity of cell specified by the manufacturer

Q has been taken from manufacturer data as 21.0348Ah.

26 | P a g e
3.3 Neural networks
Neural network becomes accurate with each data input but it requires a minimum dataset
to get trained. The problem with their application is that they require lots of data to get
trained and also the computational capacity of processors in BMS need to be high.
The computational capacity is usually measured in terms of ‘computational class’ with
coulomb counting method requiring the least followed by Kalman filter and then neural
networks requiring the highest class among the three.

3.4 Model based estimators

The alternative approach to estimating SOC is by using model-based estimators. Kalman
filter approach provides a fine balance of accuracy and data required, hence it can be
effectively implemented to estimate SOC. Kalman filters perform prediction and correction
in two back-to-back steps to estimate unmeasured states of a process. Extended Kalman
filter approach was developed for non-linear discrete-time processes. It is an iterative
technique which keeps on correcting itself to minimize error.
In this project extended Kalman filter is being used to estimate SOC which gives result
with high precision. The model takes in data at the sampling rate and using the previous
state, estimate SOC for the present iteration.

𝑥(𝑘 + 1) = 𝐴(𝑘). 𝑥(𝑘) + 𝐵(𝑘). 𝑢(𝑘) + 𝑤(𝑘) (3.2)

𝑦(𝑘) = 𝐶(𝑘). 𝑥(𝑘) + 𝐷(𝑘). 𝑢(𝑘) + 𝑣(𝑘) (3.3)

Kalman filters perform prediction and correction in two back-to-back steps to estimate
unmeasured states of a process.
Equations for extended Kalman filter:
𝑥(𝑘 + 1) = 𝑓(𝑥(𝑘), 𝑢(𝑘)) + 𝑤(𝑘) (3.4)

𝑦(𝑘) = 𝑔(𝑥(𝑘), 𝑢(𝑘)) + 𝑣(𝑘) (3.5)

Using these equations on the data from battery, SOC can be estimated with high precision.

3.5 Input data

This model is used after the calculations of model 1 have taken place. We take the
following input data in this cell model:
▪ Current: This is obtained through cell equivalent circuit model developed in chapter 1,
and is used in step 1a of the Kalman filter to obtain initial estimate of SoC.
▪ Voltage: This is also used from the output of cell equivalent circuit model and is used
in step 1c to obtain output value which further is used in step 2b.

3.6 Output data

We obtain SoC of the battery from this model which is used to provide overvoltage and
undervoltage limits to the battery. SoC is also used to calculate State-of-Health (SoH)
which measures the number of life cycles left.
The output data of SoC is again used in the next iteration to generate next estimate of SoC
of the battery pack.

27 | P a g e
 3.7 Model overview
This model is being developed to provide accurate SoC for various purposes. The extended
Kalman filter method involves two major steps which are further divided into three sub-
steps each:
▪ Prediction Step 1a-1c:
This is the ‘prediction’ step in which we predict different values. These values are not
accurate as they involve only the value used in previous steps.
▪ Correction Step 2a-2c:
This is the ‘correction’ step in which we correct the values predicted in step 1a-1c. This
is done with the help of a gain factor whose details we will see in section 3.7.4
The data at the output is used iteratively at every next step to provide accurate estimation.
Previous data is used in the steps 1a and 1b to provide accurate SoC estimation. Step 1c
estimates the output(voltage) using the values of SoC estimated in steps 1a and 1b. This
value of voltage is compared with the value of input voltage. The difference in these values
of voltage is stored in gain variable. The gain variable is used to correct the value of SoC
and its error in step 2b and 2c.

(a)

Section 3.7.1 Section 3.7.3 Section 3.7.6

Section 3.7.4

(b)

Section 3.7.5

Section 3.7.2

Fig.3.3 Different steps of extended Kalman filter

28 | P a g e
3.7.1 Step 1a: State estimate time update
In this step, we take the previous values of SoC and using the value of input current we
predict the SoC value of current iteration. This predicted value of SoC is not yet correct
and we’ll update it in step 2b.
𝑥ℎ𝑎𝑡(𝑘 + 1) = 𝐴 ∗ 𝑥ℎ𝑎𝑡(𝑘) + 𝐵 ∗ 𝑖(k) (3.6)
where,
xhat is the SoC estimated using previous values
A is the matrix which stores the variation in SoC as it is a Gaussian random variable
B is the matrix which stores the change in value of Gaussian random variable noise
i(k) is the input current to the model

Fig.3.4 Step 1a: State estimate time update

3.7.2 Step 1b: Error covariance time update
Similar to step 1a, in this step we predict the covariance value of error. This value is not
correct and we’ll update it in step 2c using a gain factor.
𝛴𝑥 =A ∗ 𝛴𝑥 ∗ 𝐴′ + 𝛴𝑤 (3.7)
where,
𝛴𝑥 is the error covariance matrix of SoC
A is the matrix which stores the variation in SoC as it is a Gaussian random variable
𝐴′ is the error covariance matrix of SoC
𝛴𝑤 is the error covariance matrix of noise

Fig.3.5 Step 1b: Error covariance time update

29 | P a g e
3.7.3 Step 1c: Predict system output
Using the value of SoC for this iteration we predict a value of voltage. This value of voltage
will be compared in the next step with the input value of voltage (which is erroneous but
gives us the real-time variation) to arrive at a gain factor in order to correct the predicted
value of SoC and SoC error covariance in steps 1a and 1b.
𝑦ℎ𝑎𝑡 = 𝐶 ∗ 𝑥ℎ𝑎𝑡 + 𝐷 ∗ 𝑣 (3.8)
where,
yhat is the system output for this timestep based on xhat(k+1)
C is the matrix which stores the change in value of random variable yhat w.r.t. xhat
xhat is the estimated value of SoC in step 1a
D is the matrix which stores the change in value of random gaussian variable v

Fig.3.6 Step 1c: Predict system output

3.7.4 Step 2a: Estimator gain matrix
In Eq.3.9 we calculate the error covariance values of predicted output voltage. Here we
use the covariance of input voltage also. Hence using both the covariance of SoC error
from step 2a and the value of error covariance in input voltage, we finally obtain net error
covariance in predicted value of voltage.
𝛴𝑦 = 𝐶 ∗ 𝛴𝑥 ∗ 𝐶 ′ + 𝛴𝑣 (3.9)
where,
𝛴𝑥 is the error covariance matrix of SoC
𝛴𝑦 is the error covariance matrix of system output
C is the matrix which stores the change in value of random variable yhat w.r.t. xhat
𝐶 ′ is the transpose of C matrix
In Eq.3.10, we calculate the ratio of error covariances of SoC and predicted voltage from
Eq.3.9. This value is the main reason why Kalman filter algorithm is able to correct the
value of SoC on its own which other methods fail to do so.

𝛴𝑥 ∗𝐶′
𝐿= (3.10)
𝛴𝑦

where,
L is the gain which we need to correct the value of estimated output

30 | P a g e
 Fig.3.7 Step 2a: Estimator gain matrix
3.7.5 Step 2b: State estimate measurement update
In Eq.3.11 we update the value of SoC predicted in step 1a using the gain value from step
2a. We use two values of output, ytrue is the one calculated using only SoC and current
and yhat is the one calculated using both SoC and input value of voltage.
𝑥ℎ𝑎𝑡(𝑘 + 1) = 𝑥ℎ𝑎𝑡(𝑘) + 𝐿 ∗ (𝑦𝑡𝑟𝑢𝑒 − 𝑦ℎ𝑎𝑡(𝑘 + 1)) (3.11)
where,
v is the terminal voltage of battery pack
xhat is the SoC estimated using previous values
L is the gain which we need to correct the value of estimated output
ytrue is the true value of system output, here voltage
yhat is the true value of system output(voltage)

Fig.3.8 Step 2b: State estimate measurement update

3.7.6 Step 2c: Error covariance measurement update
Similar to step 2b, in Eq.3.12 we use the gain value to update the value of error covariance
of SoC. Note that the second term is negative which means as we get more data and the
system undergoes more iterations, the covariance value of error decreases. This is another
indictor that the system is moving towards a stable accurate output value of SoC. Unlike
Coulomb Counting method, where the error increases over time, here the error decreases
over time.

31 | P a g e
 𝛴𝑥 (𝑘 + 1) = 𝛴𝑥 (𝑘) − 𝐿 ∗ 𝛴𝑦 (𝑘 + 1) ∗ 𝐿′ (3.12)

where,
v is the terminal voltage of battery pack
𝛴𝑥 (𝑘) is the estimated value of error covariance matrix of SoC
𝛴𝑥 (𝑘 + 1) is the corrected value of error covariance matrix of SoC
L is the gain calculated in step 2a
𝐿′ is the transpose of gain variable

Fig.3.9 Step 2c: Error covariance measurement update

32 | P a g e
 CHAPTER-4
ACTIVE CELL BALANCING

4.1 Need of cell balancing

In a battery pack which consist of 𝑛𝑠 cells in series and 𝑛𝑝 cells in parallel, all cells have
different rate of discharge. This causes voltage to be different among different cells. Two
scenarios can occur:
▪ While charging, the cell with highest SoC reaches upper threshold and the BMS cut-
offs the entire pack hence leaving some cells not fully charged.
▪ While discharging, the cell with the lowest SoC reaches lower threshold and the BMS
cut-offs the power output despite some cells having charge left in them.
These scenarios limit the use of battery and we get issues like less power and energy
available which limit the acceleration/deceleration and range of the vehicle. By performing
cell balancing, we ensure that all the cells have similar levels of charge, hence no early
cut-off of the battery pack. This optimises the battery usage.

4.2 Methods of cell balancing

Depending on the application requirement and the cost involved we perform cell balancing
using one of the following methods:

▪ Passive cell balancing

This is a common approach being used in low cost BMS. Passive cell balancing has a
resistor and a switch attached with every cell. If the cell has a higher voltage than
required, then the MOSFET corresponding to this cell is turned on and that cell starts
to provide power to load. As shown in fig.4.2, the control algorithm gives the command
as per the current SoC values. In every iteration the control algorithms find the cell
having highest SoC and switches it on. This is done on different cells until they all have
equal voltages.

Fig. 4.1 Electrical circuit for passive cell balancing

33 | P a g e
 ▪ Active cell balancing
This method uses the inductor windings to shift charge from higher charged cell to
lower charged cell. From charge wastage point of view, this method is more effective
way to approach balancing as it does not dissipate charge which is the major drawback
of passive cell balancing approach.
Fig.4.2 shows the implementation of this approach using synchronous flyback
converter. The circuit has one switch (MOSFET) and one inductor winding
corresponding to each cell. Additionally, there is one switch and inductor winding
connected across the entire row of cells.

Fig. 4.2 Electrical circuit for Active cell balancing

4.3 Circuit complexity and cost

BMS is an additional circuit which adds cost to already expensive Li-ion battery packs.
We can choose wisely whether to choose a BMS with passive or active cell balancing
depending on the application.
Passive cell balancing is the ideal choice inside a BMS for low cost applications. This is
because its circuit is relatively less complex and involves lower cost. Hence for
applications like e-bikes and e-scooters, we can use a BMS which performs passive cell
balancing. For higher applications like cars and trucks we generally use active cell
balancing. This is due to the higher cost involved of the vehicle itself and also the higher
ratings of battery which necessitates better BMS.

34 | P a g e
4.4 Mechanism for Active cell balancing:
The following steps are repeated iteratively to achieve the desired result:
▪ Step-1: The cells with the highest and lowest SOC/capacity are identified. This is done
by the use of funtions in MATLAB editor.
▪ Step-2: The MOSFET corresponding to highest voltage cell is switched on. This
charges the winding corresponding to that cell.
▪ Step 3: Now the MOSFET corresponding to flyback converter is switched on. This
charges the winding corresponding to flyback converter.
▪ Step 4: The MOSFET corresponding to weakest cell is switched on. This charges the
winding corresponding to it which has a higher potential across it’s terminals as
compared to weak cell. Hence the current flows in reverse direction and charges the
weak cell.
▪ Step 5: Go to step-1. Repeat the process until all cells have equal SOC/voltage.

4.5 Developed Model 1: Active Cell Balancing (ACB1)

The model consists of four cells which have been given different initial state-of-charge.
Our aim is to develop a model which brings them to same level of SoC after certain time.
This time is controlled by us as we can change the sampling rate of simulator to perform
this cell equalisation faster.

Section 4.5.1 Section 4.5.2

Section 4.5.1

Fig. 4.3 Model 1 for active cell balancing

35 | P a g e
4.5.1 The control functions
Fig.4.4 shows two MATLAB functions which have been used to find the cells having
maximum and minimum state-of-charge. The output from these functions goes through
switch block which ensures that initially the MOSFET corresponding to the cell having
maximum SoC gets the signal for 1/3rd the timestep, followed by the MOSFET
corresponding to flyback converter and finally the MOSFET corresponding to cell having
minimum SoC gets the gate signal. This process is repeated iteratively and the cells get
balanced as more cycles get completed.

Fig.4.4 Control functions of the ACB1 model

4.5.2 Switch case, transfer function and signal converters

The signal from the part of model described in fig.4.4 foes through a switch case block.
This block triggers the MOSFET of the cell which previous circuit wants to be triggered.
The block described as ‘S PS’ converts the Simulink signal into physical signal to be used
in the Simscape circuit described in fig.4.6.

Fig.4.5 Signal processing blocks

36 | P a g e
4.5.3 Simscape circuit
The part of model shown in fig.4.6 contains four cells and corresponding to each cell there
is a MOSFET, a voltage converter and one winding. There is one flyback winding which
is connected across all the cells and one MOSFET is connected to it also for switching. All
the MOSFETs are receiving their gate signals from the part of the model described in
section 4.5.2. Voltage converter detects the voltage of corresponding cells and sends them
to part of the model described in section 4.5.1 which checks for highest and lowest voltages
in every iteration. Flyback converter circuit is used to move the charge from highest to
lowest charged cell irrespective of its position in the battery pack, as it is connected across
all cells.

Fig.4.6 Simscape part of ACB1 model

4.6 Developed Model 2: Active Cell Balancing (ACB2)

The model shown in fig.4.7 involves a different approach than ACB1. It uses two common
capacitors along with a RL branch for charge storage and movement in between cells. It
also uses a switch similar to ACB1 model. The specifications of the Li-ion cells can be
found in appendix B3.

Fig.4.7 Model 2 for active cell balancing

37 | P a g e
 The model takes three identical Li-ion cells and sets them at different initial SoC-
▪ Cell 1: This cell is given the highest initial SoC of 80%
▪ Cell2: This cell is given the second highest SoC of 60%
▪ Cell3: This cell is given the lowest initial SoC of 20%
It is found that the cells reach similar levels of SoC after a certain time which can be
adjusted using sampling rate. This model can be scaled up for series/parallel combinations
of cells.

38 | P a g e
 CHAPTER-5
OBSERVATION AND RESULTS

5.1 Results
5.1.1 Cell equivalent circuit
From this model, we obtain the terminal voltage of the battery under load. Hence, we
expect the output to be a bit lower than the Open Circuit Voltage (OCV) of the battery.
In fig.5.1 we observe that the dashed line which represents OCV is at a higher voltage as
compared to the dotted line which represents terminal voltage of the battery pack.

Fig 5.1 Comparison between OCV and terminal voltage

5.1.2 State-of-Charge Estimation
We compare SoC estimation using Coulomb Counting method and Extended Kalman
filter. The signal in marked in yellow is from coulomb counting method while the signal

Fig 5.2 SOC comparison using Coulomb counting method and Kalman filter method
39 | P a g e
 marked with blue is from Kalman filter model developed in chapter 2.
5.1.3 Active Cell Balancing
Fig.5.3 shows the balancing of three cells over time which were initially at different SoC
levels(0.8,0.6,0.1). It shows the two cells with 0.8 and 0.6 discharging over time while
the cell with 0.1 getting charged over time. The cells finally have equal SoC values of
0.5

Fig 5.3 Balancing of different cells over the period of time

5.2 Justification of objectives achieved

The project involved the development of Battery Management System model. We divided
that into three components which were achieved successfully:
▪ In chapter 2, we developed the model of equivalent cell circuit (fig. 2.3). The result of
the model can be seen in fig.5.1.
▪ In chapter 3, we developed the model of equivalent cell circuit (fig. 3.3). The result of
the model can be seen in fig.5.2.
▪ In chapter 4, we developed the model of equivalent cell circuit (fig. 4.3). The result of
the model can be seen in fig.5.3.
All these three components form an integral part of BMS.
5.3 Future work
The model can be improved in several ways:
▪ Taking thermal effects on battery performance into account. We usually get higher
voltage values at higher temperature and lower voltage values at colder temperatures.
▪ In the model we have taken 13 cells in series but no cells in parallel. We can simulate
the behaviour of series-parallel combination of cells.
▪ In EVs, there is one master BMS and several slave BMS. The flow of different
information between them can be simulated. Different topologies can be simulated and
compared to arrive at the topology which provides optimum output.

40 | P a g e
 CHAPTER-6
PROJECT METRICS

6.1 Challenges faced and troubleshooting

During modelling several challenges were faced:

• Modelling of code and equations for estimating SoC using blocks in Simulink took a lot of
time.
• Adjusting sampling rate of different blocks to arrive at the desired output was a crucial step,
not thought of beforehand.
• As both Simulink and Simscape were used, their compatibility and correct conversion
between Simulink signal and Simscape signal had to performed.
• Adjusting MOSFET switching with the input pulse from max and min functions to arrive at
correct voltage output.

Table 6.1 Troubleshooting chart

S.No. Problem Cause Remedy

1. Input data for simulation Real time data needs to Use MIDC drive cycle
used for Indian roads.

2. Conversion of mechanical to Parameters like velocity, Use electrical

electrical parameters acceleration needs to be equivalent of
converted to voltage, mechanical power
current etc.
3. MOSFET control Firing pulse is required Use controlled
using certain conditions voltage source in
Simscape

4. Configuration settings Simulation needs Specify

appropriate settings to discrete/continuous
run and sample frequency

5. Simultaneous switching of Irregular switching of Use max/min function

MOSFET MOSFET for control pulse and
switching

6. Switch case action sub-system The output of switch case Every switch case
action does not go requires an action-
directly in controlled port sub-system at its
voltage source output terminals

41 | P a g e
 7. Battery voltages Battery type and voltage Simulink library has
custom battery also
where type of battery
can be chosen.

6.2 Innovation
Several steps were taken to ensure the work builds on the existing progress in this field:
▪ Most real-time model use coulomb counting method to estimate SoC, while it’s already
established by literature that extended Kalman filter gives better results. Hence, through
this internship, I have developed a better model which will be incorporated with FEV
model and used in all projects on hybrid and electric vehicles.
▪ MIDC drive cycle for Indian urban roads has been used which ARAI has developed a
standard to be used by all vehicle manufacturers of India. Hence, by using the same
drive cycle in developed models, a close correspondence has been established between
simulation results and actual Indian conditions.
▪ The developed cell model has a hysteresis component as well which makes it more
accurate than generally used cell models which ignore hysteresis due to its small effect.

6.3 Relevant subjects

Table 6.2 Relevant subjects

Subject Code Subject name

UEE501 Electrical Engineering

UMA001 Mathematics -1

UEC001 Electronic Engineering

UEE505 Analog and Digital Systems

UMA007 Numerical Analysis

UEE405 Network Theory and Design

UEE504 Power Electronics

UEI609 Fundamentals of Microprocessors and

microcontrollers

6.3 Interdisciplinary aspect

The project requires in-depth knowledge of power electronics, batteries and estimation
algorithms. It requires good command over modelling in MATLAB/Simulink. The cell

42 | P a g e
 model being developed is commonly used by simulation engineers in companies which
manufacture batteries like Samsung, LG, Panasonic etc. The Extended Kalman filter
algorithm being used is a very common algorithm used in different applications like object
position estimation and temperature detection inside a rocket etc. It is also being applied
in autonomous vehicle to predict the position of vehicle, based on available data and take
corrective actions if necessary. The coming together of power electronics, algorithms and
embedded system design has made this project a unique interdisciplinary product.

6.4 Student outcome (A-K) mapping

Table 6.3 A-K mapping

A1 Applied mathematics to obtain analytical, Concepts of calculus and statistics
numerical and statistical solutions in the project. have been used in various steps of
Kalman filter

A2 Demonstrated and used the use of scientific and Diverse subject knowledge has
engineering basic principles to solve engineering been used during modelling of
problem batteries

B1 Used appropriate hardware equipment to gather The data being used was provided
data by the battery manufacturer.

B2 Analyse and validate experimental results using Comparative study of the

suitable techniques simulation output has been done
for all the three models.

C1 Share information and data collected with other Data like that of MIDC cycle was
members of the team used by all the members of the
team. The model of SoC
estimation will also be used in
vehicle model simulation.

D1 Classify data to find engineering issues The SoC and terminal voltage
date has been analysed to take into
account different issues.

E1 Use analytics, computation and experimental Kalman filter is being used here
methods to find solutions which perform computation of
SoC at very fast rates. It is based
on statistical concepts like
covariance and gaussian error.

F1 Prepare and present a range of documents like Reports related to comparative

laboratory reports or inspection reports using study of different parameters have
discipline specific standards been presented.

43 | P a g e
 G1 Awareness of society and world changes which Li-ion technology is being
occur with engineering innovation developed to power EVs which
are seen as a cleaner alternative to
gasoline vehicles.

H1 Able to use resources and also to adopt new The project required extensive use
technologies which are not part of curriculum of MATLAB/Simulink which is
used for simulation purposes in
different industries.

I1 Recognize impact of major engineering projects Evolution of battery technology

on present resources and the environment. will help to limit the dependence
on fossil fuels and also limit
pollution levels of cities.

J1 Able to observe engineering issues using software The models have been developed
programs keeping in mind their practical
implementation. The errors
debugged during simulation stage
will be reflected in hardware also.

44 | P a g e
 REFERENCES

[1] G. Plett, Battery Management Systems, Volume 1: Battery Modeling Battery Modeling, Artech
House, 2015.

[2] G. Plett, Battery Management Systems: Equivalent-Circuit Methods Volume II, Artech House,
2015.

[3] G. K. Teja and S. Prabharan, “Smart Battery Management System with Active Cell,” Indian
Journal of Science and Technology, vol. Vol 8(19), August 2015.

[4] S. Arendarik, Active Cell Balancing in Battery, Freescale Semiconductor, Inc, 2012.

[5] G. Consulting, “An Introduction to the Kalman Filter,” [Online]. Available:

http://www.goddardconsulting.ca/kalman-filter.html.

[6] O. Ershag, A balancing and monitoring system for battery cell stacks in electric vehicles, Lulea
University of Technology, 2008.

[7] D. McDonald, “Electric Vehicle Drive Simulation with MATLAB/Simulink,” in North-Central

Section Conference, American Society for Engineering Education, 2012.

[8] S. T. Radu Tarulescu, “Battery Management System of E-Smart Vehicle,” in Springer

International Publishing, 2017.

[9] R. Jackey, M. Saginaw, P. Sanghvi, J. Gazzarri, T. Huria and M. Ceraolo, “Battery Model
Parameter Estimation Using a Layered Technique: An Example Using a Lithium Iron Phosphate
Cell,” Mathworks, 2013.

[10] T. Huria, M. Ceraolo, J. Gazzarri and R. Jackey, “Simplified Extended Kalman Filter Observer for
SOC Estimation of Commercial Power Oriented LFP Lithium Battery Cells,” Mathworks, 2013.

[11] T. Huria, M. Ceraolo, J. Gazzarri and R. Jackey, “High Fidelity Electrical Model with Thermal
Dependence for Characterization and Simulation of High Power Lithium Battery Cells,” IEEE,
2012.

[12] I. Cho, J. Bae, J. Park and J. Lee, “Experimental Evaluation and Prediction Algorithm Suggestion
for Determining SOC of Lithium Polymer Battery in a Parallel Hybrid Electric Vehicle,” Appl. Sci.
2018, 8, 1641; doi:10.3390/app8091641.

[13] M. Murnane and A. Ghazel, “A Closer Look at State of Charge (SOC) and State of Health (SOH)
Estimation Techniques for Batteries,” Analog Devices Inc., 2017.

[14] “State of Charge (SOC) Determination,” Electropedia, [Online]. Available:

https://www.mpoweruk.com/soc.htm. [Accessed 31 1 2019].

45 | P a g e
[15] S. Yang, C. Deng, Y. Zhang and Y. He, “State of Charge Estimation for Lithium-Ion Battery with a
Temperature-Compensated Model,” Energies, vol. 10, no. 1560, 2017.

[16] Z. Yu, R. Huai and L. Xiao, “State-of-Charge Estimation for Lithium-Ion Batteries Using a Kalman
Filter Based on Local Linearization,” Energies, vol. 8, pp. 7854-7873, 2015.

[17] S. M. Rezvanizaniani, Z. Liu, Y. Chen and J. Lee, “Review and recent advances in battery health
monitoring and prognostics technologies for electric vehicle (EV) safety and mobility,” Journal
of Power Sources, no. 256, pp. 110-124, 2014.

[18] V. Pop, H. J. Bergveld, D. Danilov, P. P. Regtien and P. H. Notten, Battery Management Systems-
Accurate State-of-Charge Indication for Battery Powered Applications, Springer, 2008.

[19] B. Müller and G. Meyer, Electric Vehicle Systems Architecture and Standardization Needs,
Springer, 2015.

[20] K. W. E. Cheng, B. P. Divakar, H. Wu, K. Ding and H. F. Ho, “Battery-Management System (BMS)
and SOC Development for Electrical Vehicles,” IEEE TRANSACTIONS ON VEHICULAR
TECHNOLOGY, vol. 60, no. 1, JANUARY 2011.

46 | P a g e
 ANNEXURE-A
DAILY DIARY

Fig A1 Daily diary snippet 1

47 | P a g e
 Fig. A1 Daily diary snippet 2

Fig. A3 Daily diary snippet 3

48 | P a g e
 ANNEXURE-B
DATA USED IN MODELS
B1. Data for Open Circuit Voltage
The following data is the manufacturer specified Open Circuit Voltage(OCV) data at
different SoC levels at 20𝑜 𝐶 and 40𝑜 𝐶. We interpolate and extrapolate this data in our
model to obtain OCV values at any random temperature and SoC.

Table B1. OCV data using SOC at 20𝑜 𝐶 and 40𝑜 𝐶

State-of-Charge OCV in Volts(𝟐𝟎𝒐 𝑪) OCV in Volts(𝟒𝟎𝒐 𝑪)

0.1 3.4472 3.4476

0.2 3.5184 3.5189

0.3 3.5841 3.5848

0.4 3.6164 3.6172

0.5 3.6569 3.6574

0.6 3.7391 3.7394

0.7 3.8395 3.8398

0.8 3.9364 3.9369

0.9 4.0416 4.0419

1 4.1547 4.1551

B2. Variables used in SoC estimation model

These parameters are used in models developed in chapter 2 and 3. These parameters are
used to tune the model.
Table B2. Different parameters used inside model

Model variables Value

RParam 0.000025151

R 0 Param 0.011163

GParam 61.750

QParam 21.0348

49 | P a g e
 SoC_initial 0.8

A, B, C 1

D 0

MParam, M0 Param 1

RCParam 1

SigmaV 1

B3. Different parameters of the battery

These parameters are used in active cell balancing model described in chapter 4.

Table B3. Different parameters of battery

Battery parameter Value

Nominal voltage (V) 100

Rated capacity (Ah) 5.4

Nominal discharge current (A) 2.3478

Internal resistance (Ohms) 0.013333

Capacity (Ah) at nominal voltage 4.8835

Fully charged voltage (V) 110

Cut-off Voltage (V) 150

Discharge current [i1, i2, i3] (A) [6.5 13 32.5]

50 | P a g e