Master Thesis in Electronic Engineering

Analysis and development of Battery Management

System for electric vehicles

Author: Floridia Stefano Professor: Passerone Claudio

Supervisor: Leandro Malara

Academic year 2021/2022

 Contents

List of Figures 3

Abstract 5

1 Battery Management System 6

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2 Accumulator design . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3 BMS functionalities . . . . . . . . . . . . . . . . . . . . . . . . 19

2 State of Charge 23
2.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.2 SoC Evaluation methods . . . . . . . . . . . . . . . . . . . . . . 25
2.2.1 Coulomb Counting . . . . . . . . . . . . . . . . . . . . . 25
2.2.2 Open Circuit Voltage method . . . . . . . . . . . . . . . 26
2.2.3 Model-Based methods . . . . . . . . . . . . . . . . . . . 28
2.2.4 Kalman filter-based methods . . . . . . . . . . . . . . . 30
2.2.5 Neural networks-based methods . . . . . . . . . . . . . . 31
2.3 Algorithm implemented . . . . . . . . . . . . . . . . . . . . . . 32
st
2.3.1 1 grade model equation . . . . . . . . . . . . . . . . . 40
2.3.2 2nd grade model equation . . . . . . . . . . . . . . . . . 41
rd
2.3.3 3 grade model equation . . . . . . . . . . . . . . . . . 42
2.3.4 4th grade model equation . . . . . . . . . . . . . . . . . 43
th
2.3.5 5 grade model equation . . . . . . . . . . . . . . . . . 44
2.3.6 6th grade model equation . . . . . . . . . . . . . . . . . 45
2.3.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . 46

i
 CONTENTS

3 Firmware 48
3.1 Master-Slave data exchange . . . . . . . . . . . . . . . . . . . . 50
3.1.1 UART message structure . . . . . . . . . . . . . . . . . 51
3.1.2 BMS data management function . . . . . . . . . . . . . 56
3.2 SoC value calculation and validation . . . . . . . . . . . . . . . 59

4 Conclusions 61

Bibliography 63

ii
 Acronyms

ADC Analog to Digital Converter. 19

BMS Battery Management System. 5, 6, 9, 16, 18–21, 24, 46

BP Battery Pack. 5, 8, 9, 11, 12, 19, 21, 22, 30, 46, 47

CAN Controller Area Network. 7

CC Counting Coulomb. 26

CRC Cyclic Redundancy Check. 8

DLC Data Length Code. 8

DMA Direct Memory Access. 54, 55

EChM Electrochemical Model. 28, 29

ECM Electrical Cicruit Model. 28

ECU Electronic Control Unit. 6, 19, 21, 46

EV Electric Vehicle. 5, 6, 9, 11, 13, 23, 24, 30

GLVS Grounded Low Voltage System. 8

HEV Hybrid Electric Vehicle. 5, 6, 9, 24

HV High Voltage. 8, 9, 19, 21

HVIL High-Voltage Inter Lock. 9, 21

1
 Acronyms

I2C Inter Integrated Circuit. 6

IDE Identifier extension. 8

IMD Isolation Monitoring Device. 8

LV Low Voltage. 9, 21, 22

MCU Microcontroller. 48

NTC Negative Temperature Coefficent. 19

OCV Open Circuit Voltage. 26–28

RMSE Root Mean Square Error. 39, 41, 43, 45, 46

SAR Successive Approximation Register. 19

SoC State of Charge. 5, 18, 20, 23–30, 32, 33, 39, 46, 58–60, 62

SoH State of Health. 20, 23, 24, 30, 62

SPI Serial Peripheral Interface. 6

UART Universal Asynchronous Receiver Transmitter. 6, 21, 50–52, 54, 56

VCU Vehicle Contol Unit. 6, 8, 21

2
 List of Figures

1.1 CAN bus topology . . . . . . . . . . . . . . . . . . . . . . . . . 7

1.2 Standard CAN message frame . . . . . . . . . . . . . . . . . . . 8
1.3 Battery Voltage characteristic . . . . . . . . . . . . . . . . . . . 9
1.4 Battery schematic . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.5 Parallel (on the right) and Series (on the left) configurations . . 11
1.6 Tesla Model S Module: 6 groups containing 74 cells in parallel 12
1.7 Li-ion cells features . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.8 Li-ion batteries comparison . . . . . . . . . . . . . . . . . . . . 14
1.9 Cell comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.10 Passive cell balancing . . . . . . . . . . . . . . . . . . . . . . . 16
1.11 Active cell balancing . . . . . . . . . . . . . . . . . . . . . . . . 17
1.12 Software controlled balancing schematic . . . . . . . . . . . . . 18
1.13 Current sensor with Hall effect . . . . . . . . . . . . . . . . . . 20
1.14 Cell operative range . . . . . . . . . . . . . . . . . . . . . . . . 21
1.15 Low voltage battery management system . . . . . . . . . . . . . 22

2.1 Cell hysteresis in Open Circuit Voltage condition . . . . . . . . 25

2.2 OCV SoC curve [12] . . . . . . . . . . . . . . . . . . . . . . . . 27
2.3 Ideal battery model . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.4 Caption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.5 Kalman filter schematic . . . . . . . . . . . . . . . . . . . . . . 30
2.6 Neural network for SoC estimation . . . . . . . . . . . . . . . . 31
2.7 Molicel Lithium-ion Cell characteristics . . . . . . . . . . . . . 34
2.8 Example of non linear dataset . . . . . . . . . . . . . . . . . . . 35

3
 LIST OF FIGURES

2.9 Polynomial Regression model . . . . . . . . . . . . . . . . . . . 36

2.10 Overfitting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.11 Dataset division . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.12 Input dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.13 Grade 1 model equation . . . . . . . . . . . . . . . . . . . . . . 40
2.14 Grade 2 model equation . . . . . . . . . . . . . . . . . . . . . . 41
2.15 Grade 3 model equation . . . . . . . . . . . . . . . . . . . . . . 42
2.16 Grade 4 model equation . . . . . . . . . . . . . . . . . . . . . . 43
2.17 Grade 5 model equation . . . . . . . . . . . . . . . . . . . . . . 44
2.18 Grade 6 model equation . . . . . . . . . . . . . . . . . . . . . . 45
2.19 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

3.1 Firmware structure . . . . . . . . . . . . . . . . . . . . . . . . . 48

3.2 Firmware flow chart . . . . . . . . . . . . . . . . . . . . . . . . 50
3.3 UART frame structure . . . . . . . . . . . . . . . . . . . . . . . 51
3.4 Frame initialization byte . . . . . . . . . . . . . . . . . . . . . . 52
3.5 Frame initialization byte fields table . . . . . . . . . . . . . . . 53
3.6 UART command message acquisition . . . . . . . . . . . . . . . 54
3.7 Prototype Electrical system . . . . . . . . . . . . . . . . . . . . 59
3.8 Battery charging curves . . . . . . . . . . . . . . . . . . . . . . 60
3.9 Data flowchart . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

4
 Abstract

Our times are characterized by numerous electronic applications that are chang-
ing the lifestyle of many. There are devices designed for various situations and
one of the biggest challenges is making sure these applications last as much as
possible. For this purpose, rechargeable batteries are the best solution until
now, guaranteeing a permanent and clean energy storage if we talk about the
automotive industry. In fact, batteries are starting to be the most common
electrical energy storage system for an Electric Vehicle (EV) or Hybrid Elec-
tric Vehicle (HEV). In Electric Vehicles, a Battery Pack can provide several
hundred Volts, since various functions within a car depend on it, including
powering the electric motor itself. The key aspect of all these applications is
that the batteries must be constantly monitored, to have better energy con-
sumption or to take safety measures.
A Battery Management System (BMS) is an integrated system, designed to
closely monitor battery circuits, implementing protective measures in case the
battery is operating outside its normal operating area, but also acquiring data
that can be used for better power management. The State of Charge (SoC) is
one of those data that must be tracked in a battery to optimize its performance
since it is essential to know how much capacity is available.
The purpose of this thesis work is to describe a set of functions of the Battery
Management System, with particular attention to the development of an algo-
rithm for the evaluation of the State of Charge, but still having the possibility
of being able to apply these functions to generic management systems of a bat-
tery in vehicles. The system used for data validation is the one developed for
the Formula SAE ”Squadra Corse PoliTo” prototype.

5
 CHAPTER 1

Battery Management System

1.1 Introduction
Inside Electric Vehicles, the battery pack contains one or more ECUs that are
able to perform for monitoring and protection for the vehicle and the user. This
system is what is usually referred to as Battery Management System (BMS)
made of a group of ECUs exchanging information in a Master-Slave fashion.
The Slave system is a group of ECUs connected to each other forming a Daisy
Chain and connected with the Master using standard digital protocols like
UART, SPI, I2C. Its work is to perform data acquisition from the battery such
as Voltage, Current and Temperature, and sometimes it is integrated with a
circuitry for balancing. Cell Balancing is a very useful technique that can be
implemented in order to increase the available capacity of the battery, extend-
ing its life. It is a typical solution for multi-cells systems including laptops,
phones, HEV and EV.
The Master is the main ECU, this why the device is referred most of the
time as the actual BMS, performing the most important algorithms according
to the received data and it is connected to the majority of protection devices
inside this circuitry. The BMS exchanges information with the vehicle, usu-
ally communicating with a Vehicle Contol Unit (VCU) that elaborates what
is important to know about the Battery. The most common communication
protocol implemented for information exchange in automotive, between ECUs,

6
 CHAPTER 1. BATTERY MANAGEMENT SYSTEM

is the Controller Area Network (CAN) [16]. It is a multi-master bus which

has good noise immunity and allows for a large number of node connected,
consisting of only a twisted-pair cable (differential communication) that is ter-
minated on the two sieds by a 120 Ω resistances avoiding signal reflection that
would bring further interference [20]. Figure 1.1 provides an example of CAN
network with 3 nodes.

Figure 1.1: CAN bus topology

CAN protocol is defined in the ISO-11898 [4] as CSMA/CD+AMP, that is:

• Carrier-Sense Multiple-Access: every node can access the bus, wait-

ing for a certain amount of time before trying to send a message;

• Collision Detection and Arbitration on Message Priority: each

node can detect a collision between the actual bus state and the attempted
transmission, by a predefined message priority.

The protocol consists of frames subdivided in fields, as shown in Figure 1.2,

sent at a maximum speed rate of 1 Mbit/s.

• Arbitration field: The Identifier, whose length may be 11 or 29 bits, is

the actual element that defines the priority of a message, the lower it is
the higher is the priority. RTR is a 1-bit length for remote request, it is
always ’0’ in data frames.

7
 CHAPTER 1. BATTERY MANAGEMENT SYSTEM

• Control field: this is the field where the length and the type of frame
are defined. The IDE is 1-bit long, it defines whether the Identifier is:

– Standard: 11 bit long

– Extended: 29-bit long

The Data Length Code is 4-bit long portion which defines the number of
Bytes of the Data field

• Data field: It is the portion containing the actual data bytes to be sent,
its length goes from 0 to 8 bytes

• CRC and ACK fields: CRC is used to detect data corruption during
transmission. ACK is the receiver acknowledgment

Figure 1.2: Standard CAN message frame

For protection purposes the High Voltage Battery circuit is implemented with
devices that may be controlled by the BMS master.

• Isolation Monitoring Device (IMD) which make sure that the Gal-
vanic Isolation between HV and the Grounded Low Voltage System (GLVS)
is ≥ 500Ω/V in AC and ≥ 100Ω/V in DC. It can operate on its own and
perform protection procedure or it can be connected to the Master send-
ing a signal when the Battery Pack is no more safe.

• HV Contactors. These are switches that provide fast switching and

there must be one for each terminal, Positive and Negative. In this way
it is assured that the vehicle is switched off, whenever it is requested
by the VCU, since the car is stopped, or because an hazard occurred.
There’s a third relay that is the one involved in the precharge procedure.
Power supply going out from the Accumulator needs to go through an

8
 CHAPTER 1. BATTERY MANAGEMENT SYSTEM

Inverter in order to reach the Motors, since they work in AC. Inverter
are provided with Capacitive elements that are used for energy storage.
With High Voltage systems, inrush currents may stress a lot the capacitive
components during power up. The solution adopted was to have a parallel
channel in which a resistive component is connected limiting the amount
of current delivered to the load.

Figure 1.3: Battery Voltage characteristic

The outcome, illustrated in Figure 1.3, is that during power-up, the

BMS closes the Negative and Precharge relay and the output Voltage
increases slowly until inrush current has subsided. Then Positive relay is
closed in order to provide the right amount of current during operating
stage.

• Fuses that are used for cable management. They may be connected to
both or one of the Terminals for over-current protection.

• HVIL. High-Voltage Inter Lock loop is a safety feature typical of Hybrid

Electric Vehicles and Electric Vehicles. HVIL is a LV powered cable that
inside a BP is connected between the HV connector and the BMS. If the
connection is interrupted the BMS goeas in safe mode, opening every
contactors in order to cut-off any kind of supply. This is a very impor-
tant feature that come in handy during maintainance on High Voltage
components or in situation like a high-voltage connection becomes loose,
disconnected or damaged during the operation of the vehicle.

Figure 1.4 shows a simplified architecture of the circuitry embedded in a

Battery Pack, including those elements enlisted previously.

9
CHAPTER 1. BATTERY MANAGEMENT SYSTEM

Figure 1.4: Battery schematic

10
 CHAPTER 1. BATTERY MANAGEMENT SYSTEM

1.2 Accumulator design

Before going on talking about the functionalities of a BMS, it is important
to know something about batteries. A Battery is a device that transforms
chemical energy into electrical energy and it is rechargeable. The basic unit
inside Battery for Electric Vehicles is a cell; to obtain proper voltages and
currents, multiple cells are connected following the xSyP configuration.

Figure 1.5: Parallel (on the right) and Series (on the left) configurations

Looking at Figure 1.5, in parallel configuration the Output voltage is still the
same while the Capacity increases. The opposite occurs with series configu-
ration. Usually a set of cells in parallel is called Group connected in series
forming a Module. A set of module in series assemble a Battery Pack. This
is important for the Design since whichever it is the number of parallel cells in
a group or the number of groups in series in a module, the amount of Energy
a Module can deliver can change. Emodule = gVcell × nCcell where g = groups
in series in a module and n = cells in parallel in a group. If you take
into account that a Battery Pack contains m number of modules, the amount
of Energy delivered is EBP = mEmodule .

11
 CHAPTER 1. BATTERY MANAGEMENT SYSTEM

So a Battery Pack having xSyP configuration means that x = g ×m and y = n.

The example shown in Figure 1.6 [22] is a single Module of a Tesla Model S
which has a configuration of 6s74p. It is usually covered by a metallic plate,
typical for accumulator designs, called Bus Bar and it is the element working
as connecting node between two groups in series.

Figure 1.6: Tesla Model S Module: 6 groups containing 74 cells in parallel

Single rechargeable cells must be chosen taking into account different main
features

• Capacity [Ah]: the amount of current deliverable by the cell completely

charged

• Gravimetric Energy density [Wh/Kg]: the amount of electric energy

stored per unit mass

• Power density [w/Kg]: the amount of stored power per unit mass

• Nominal Voltage [V]: Voltage of a fully charged cell

• number of Cycles: number of charge-discharge full cycles before the

cell starts to deteriorate

• recharge time: the lower, the better

• cheap, safe and reliable

12
 CHAPTER 1. BATTERY MANAGEMENT SYSTEM

Lithium ions cells are the most employed for low power solutions like smart-
phones and lately their appearance is common when designing accumulators
for EVs.

Figure 1.7: Li-ion cells features

The Anode of a Li-ion cell is made of Carbon while the cathode is a Li-metal
Oxide, which means that, whatever the metal chosen for the Cathode, the cell
performance, costs and safety features may change as shown in Figure 1.8 [2]

13
 CHAPTER 1. BATTERY MANAGEMENT SYSTEM

Figure 1.8: Li-ion batteries comparison

There exist some alternatives to Li-ion cells that have been exploited, some of
them are still used, like Lead Acid batteries. They are often used as auxiliary
batteries for vehicles, in order to supply low voltage circuitry, but not involved
in the traction system. Their Energy density is low and the Nominal Voltage
is usually around 2V, nevertheless they are very durable and reliable. Nickel
batteries are another example, their Energy density is higher then the previous
one, but not as much as Li-ion batteries. Less durable than Lead-acid batteries
but their charge time is very low. Nickel-Cadmium batteries are now out of
production due to their toxicity. Li-ion batteries, compared with all these
alternatives, are far better in terms of Energy density, since it is possible to
deliver the same amount of Energy with far less unit mass, as shown in Figure
1.9 [22], saving space and improving the vehicle design.

14
 CHAPTER 1. BATTERY MANAGEMENT SYSTEM

Figure 1.9: Cell comparison

Nevertheless Li-ion cells may be very dangerous if not monitored correctly.

For multi-cells systems, even if cells are all the same in terms of the product,
they have differences. Some cells may have different reaction to temperature
changing, some of them may discharge faster or charge slower, all of this leads
to cells having not equal output voltages. Let’s consider Lead-Acid batteries,
as already said they are very durable and strong, so in order to completely
charge the battery, even if some cells reach a full state of charge, it possible to
wait for other cells to reach the same condition, given that the disequilibrium is
not so wide. For Li-ion cells this kind of solution may lead to thermal runaway
damaging the battery. The same is valid in case of discharging of operation, if a
cell goes far beyond the lower limit the reaction is far more dangerous, bringing
the cell to short circuit [2]. The issue concerns a lot for elements connected
in series, for cells in parallel inside a group tend to self balance themselves
having the same voltage drop, exchanging charge.

15
 CHAPTER 1. BATTERY MANAGEMENT SYSTEM

Cell Balancing is a solution implemented by the BMS, monitoring cells during

charge and discharge operations, equalizing the charge reaching as much as
possible the maximum battery capacity, avoiding cell damage and extending
the battery life. Cell balancing methods have been classified in two classes:

• Passive balancing: it is the easiest method to implement that com-

monly works on cell voltage indicator to its status of charge. The classic
method is based on discharging cells dissipating charge upon a load placed
in parallel. There are systems stopping the charging operation, discharg-
ing the cells with higher voltage until it is equalized with weaker cells.
There are also systems that are able to do that without stopping the
charge, until all cells are fully charged, bypassing the cell with higher
voltage or discharging them on a load. These common methods are very
wasteful, since it implies energy dissipation, and reaches balance after a
number of cycles, due to the fact that voltage is not a good indicator
for defining how charged is a cell since they may have same capacity but
different voltage drop, because of different internal resistance. Figure
1.10 shows a typical example of passive balancing. When the controller
measures a cell voltage higher then the others, it opens the MOSFET
connecting the cell to the resistance in parallel, called bleeding resistors
[21].

Figure 1.10: Passive cell balancing

16
 CHAPTER 1. BATTERY MANAGEMENT SYSTEM

• Active balancing: it implements capacitive, inductive circuits or dc/dc

converters for charge exchange between cells. Far more efficient then the
previous one, shifting energy where is needed with infinitesimal dissipa-
tion. Capacitors and inductors work as charge storage and of course it is
very time consuming since charge is distributed between cells sequentially.
Since it needs more components than a simple load it’s more expensive
and harder to implement. There are a variety of methods based on the
concept just introduced. Figure 1.11 shows a simple schematic of one
of the most common technique called Flying capacitors.

Figure 1.11: Active cell balancing

The capacitive elements are the one that are physically shifting energy from
one cell to the next one, this implies a sequential action between adjacent cells,
very time consuming considering also the time needed for the capacitor to be
charged. This circuit have been improved in terms of spent time and efficiency
having different configuration involving DC-DC converters like buck-boost or
fly-back converters, but increasing a lot the circuitry complexity. Nowadays
software controlled solution tried to solve the problem aiming to reduce Hard-
ware usage, gaining in space and complexity [14].

17
 CHAPTER 1. BATTERY MANAGEMENT SYSTEM

The idea is based on cell bypassing having as decision criterion the SoC of the
single cell calculated by the BMS. While charging, cells that reaches full charge
are bypassed, letting other cells to join. When discharging cells with low SoC
are disconnected, having minimal loss since those cells have very low capacity.

Figure 1.12: Software controlled balancing schematic

18
 CHAPTER 1. BATTERY MANAGEMENT SYSTEM

1.3 BMS functionalities

The main functions of a BMS are:

• Battery monitoring: the ECU is constantly controlling those features

of a battery that are very important in order to know in which state the
accumulator is. Voltage, current and temperature is what the Master
needs in order to perform its calculations. For this thesis work, the slave
system has an ECU using the bq76pl455 micro-controller, which is able
to measure up to 16 cell connected in series and it is possible to connect
up to 16 devices in Daisy chain. The acquisition is driven by a 14-bit
Successive Approximation Register (SAR) ADC, then the value is stored
into a 16 bit register ready to be sent to the master [16]. The same ac-
quisition system is used for temperature measurement, with the slight
difference that this time the micro can handle only 8 analog inputs, that
are designed as auxiliary channels. NTC thermistor are placed in strate-
gic positions inside the module or the Battery Pack in order to observe
better temperature mutations. The NTC works as a resistance which de-
creases its value exponentially as temperature increases. Mathematically
speaking this behaviour is described as

1 1
RN T C = R0 eβ( T − T0 ) with R0 = R(@T0 = 25C)

The analog value measured by the ADC is the voltage drop on the NTC
itself. The simpler way to connect the NTC is to use a pull-up or pull-
down resistors, but there are Hardware solutions that may help forcing
a linear relation between temperature and voltage acquired by the slave.
Current sensing is usually done exploiting Faraday’s law of electromag-
netic induction, as illustrated in Figure 1.13 [15]. These kind of sensors
are designed to have a circular element, in which the magnetic field is in-
duced, looping around the HV cable and having two terminals forcing
a voltage upon a plate, called Hall plate, connected to an operational
amplifier whose output voltage is the one read directly by the Master for
data managing.

19
 CHAPTER 1. BATTERY MANAGEMENT SYSTEM

Figure 1.13: Current sensor with Hall effect

• Battery protection and optimization: Smart battery usage is a key

feature a Battery Management System in order to keep the battery ve-
hicle alive as much time as possible. As already said battery cells, es-
pecially Li-ion cells, are very dangerous if not closely monitored risking
reaction that can damage the battery, the vehicle and therefore the user.
Battery must perform safety action when over-current, over-voltage or
over-temperature occurs, but it should be far better if it is possible to
prevent those events keeping an eye on the status of the battery. The
same data monitored is useful for battery status data evaluation like SoC
or SoH, which are good parameters for knowing how much charge can be
delivered to single cells, how much capacity the battery is still able to
guarantee or how old is the battery in terms of number of operative cy-
cles, since cells performance degrades with it. Usually a BMS should keep
the battery to work inside an operative range where damage possibility
is lower, moreover it is common practice that battery manufacturers pro-
vide data sheets with operative range smaller then the actual capacity of
the cell in order to reduce risks further more [1] [9].

20
 CHAPTER 1. BATTERY MANAGEMENT SYSTEM

Figure 1.14: Cell operative range

With a good evaluation of cell status, the balancing operation is definitely

improved letting the master to equalize the working range to each group
of cells connected in series.

• Communication: continuous exchange of information is vital for a

BMS, communicating to the vehicle useful data for the VCU or the user.
For example alerting when the battery charge is complete or it needs
charge. It is also helpful for diagnosis purposes, since the master is able
to distinguish different kind of hazard, like over-current or HV connector
not placed correctly with HVIL open, so whenever a problem occurs it is
possible to discover what caused that faulty behaviour. CAN bus is the
most used for Automotive purposes due to its robustness to noise and
ability to cover long distances with very low probable data communica-
tion errors, mostly used in order to connect ECUs belonging to different
subsystems inside the vehicle. Inside the battery circuitry, simple com-
munication signals, belonging to the Low Voltage system, are very often
hardwired like supply signals, the HVIL presence or contactors status
(open/closed). More complex information exchange, like measured cell
voltages and BP temperatures or cell balancing commands, are managed
by digital communication protocols like UART.

21
 CHAPTER 1. BATTERY MANAGEMENT SYSTEM

For this thesis work the system implemented, shown in Figure 1.15, is far
simpler then the one shown before (see §1.4), also because the Battery Pack
monitored this time is a Low Voltage accumulator (output voltage < 60VDC ,
according to ISO-6469 [5]), so some of those system protection components and
strategies presented previously are not needed.

Figure 1.15: Low voltage battery management system

22
 CHAPTER 2

State of Charge

2.1 Definition
State of Charge (SoC) represents the available battery capacity with respect
to its rated capacity. The preferred SoC reference should be the rated capacity
of a new cell rather than the current capacity of the cell. This is because the cell
capacity gradually reduces as the cell ages. For example, towards the end of the
cell’s life its actual capacity will be approaching only 80% of its rated capacity
and in this case, even if the cell were fully charged, its SoC would only be 80%
of its rated capacity [10]. State of Charge (SoC) is one of the most important
states that need to be monitored to optimize the performance and extend the
lifetime of batteries, its evaluation is not direct as it involves to take under
consideration voltage, current and temperature of a cell. In EVs, the number
of battery has series-parallel combinations to match the load requirement [7].
Due to manufacturing procedures, not all cells simultaneously attain full volt-
age during charging resulting in voltage imbalance among different cells and,
consequently, lower capacity from the entire battery. Knowing the amount of
energy that a battery can deliver is vital for EVs applications requiring to con-
stantly monitor the actual capacity of the cell in order to evaluate the actual
and practical capability of the battery in order to fit the different road con-
ditions and driving patterns of the vehicle. Another parameter important to
evaluate is the State of Health (SoH) whose purpose is to provide an indica-

23
 CHAPTER 2. STATE OF CHARGE

tion of the performance which can be expected from the battery in its current
condition or to provide an indication of the how much of the useful lifetime of
the battery has been consumed and how much remains before it must be re-
placed. In critical applications such as standby and emergency power plant the
SoC gives an indication of whether a battery will be able to support the load
when called upon to do so. Knowledge of the SoH will also help the engineers
to anticipate problems to make fault diagnosis or to plan replacement. This is
essentially a monitoring function tracking the long term changes in the battery
[11]. A BMS must be not only a protection circuit but also a thorough and
accurate device that can predict the SoC, SoH, capacity, and available power to
increase the efficiency and the safety of the battery. By continuously measuring
current, voltage, and temperature in batteries, the aforementioned parameters
can be estimated. The estimation of the SoC is key in a BMS, but its online and
accurate estimation remains a challenge due to strong nonlinear and complex
electrochemical reactions in the battery and because of battery characteristics
change with aging [3] [13]. There is a dependency between the actual capacity
of a cell and the rate at which the battery is charged and discharged, due to the
fact that a cell’s internal electrochemical reactions take some time, this means
that the battery cannot easily follow instantaneous charge-discharge pulses, as
it happens in EV and HEV. The consequence of this is that when charging is
complete and a load is applied to the battery to discharge it, there is a delay
before the full current can be delivered through the load. This phenomenon is
called Hysteresis and leads to SoC evaluation errors, even for direct methods
like Counting Coulomb, that involves the monitoring of instantaneous capac-
ity. There’s still the possibility to minimize this error taking in account the
behaviour, implicating more complex calculation [10].

24
 CHAPTER 2. STATE OF CHARGE

Figure 2.1: Cell hysteresis in Open Circuit Voltage condition

2.2 SoC Evaluation methods

In the following pages will be described some State of Charge (SoC) evalua-
tion methods, showing main processes, advantages and drawbacks. Evaluation
methods are usually classified into Direct methods and Indirect methods.
As their name suggests, Direct SoC estimation methods use and measure physi-
cal battery features such as voltage, current, and temperature, and then manage
them using mathematical equations. Indirect methods are based on finding a
battery model, using battery physical properties, that can help capturing cell
characteristic and predict its behaviour [13].

2.2.1 Coulomb Counting

The energy contained in an electric charge is measured in Coulombs and is

equal to the integral over time of the current which delivered the charge. The
remaining capacity in a cell can be calculated by measuring the current entering
(charging) or leaving (discharging) the cells and integrating (accumulating) this
over time . In other words the charge transferred in or out of the cell is obtained
by accumulating the current drain over time. The calibration reference point is
a fully charged cell, not an empty cell, and the SOC is obtained by subtracting

25
 CHAPTER 2. STATE OF CHARGE

the net charge flow from the charge in a fully charged cell. This method, known
as Counting Coulomb (CC), provides higher accuracy than most other SoC
measurements since it measures the charge flow directly. CC depends on the
current flowing from the battery into external circuits since in some applications
such as automotive batteries the ”continuous” battery current is not monitored.
Instead the current is sampled and the continuous current is reconstructed
from the samples. In such cases the sampling rate must be fast enough to
capture the current peaks and troughs associated with the acceleration and
regenerative braking corresponding to the user’s driving style [10]. At present,
the Counting Coulomb method is the most used for SoC estimation since it
is the most accurate technique for short-term calculations. The CC method
defines SoC as
Z t0 +τ
100%
SoC(t) = SoC(t0 ) + Ibat δτ
Crated t0

where SoC(t0) is the initial SoC, Crated the nominal capacity, and Ibat is the
charging/discharging current. This method cannot estimate the initial value,
so the initial SoC must be known, and has an accumulative error. As it is an
open loop estimation system, small errors will accumulate with time due to the
integration term, being a source of significant inaccuracy. Several factors affect
its accuracy, such as battery age, discharge rate, and sensor precision [13].

2.2.2 Open Circuit Voltage method

Open Circuit Voltage (OCV) can be defined as the voltage drop on the termi-
nals of a battery when no load is connected. SoC estimation methods commonly
impose a characterization of the OCV curve derived through a polynomial or
a look-up table, as they use either a direct OCV curve inversion method, or a
cell model-based methods. SoC is therefore determined as function of the OCV
which relation, shown by the curve in Figure 2.2, is defined as:

SoC = f −1 (OCV )

In the OCV method, the cell’s voltage is continuously measured, and the corre-
sponding SoC is obtained from a table. The method is not practical due to the
requirement of measuring OCV which means SoC estimation is not available

26
 CHAPTER 2. STATE OF CHARGE

while the battery is charging or discharging thus it cannot be used in real time.
Also, the OCV–SoC relationship differs among cells and, therefore results in
unacceptable error [13]. Problems can occur with some cell chemistry, partic-
ularly Lithium which exhibits only a very small change in voltage over most of
the charge/discharge cycle. The following graph shows the discharge curve for
a Lithium-ion cell.

Figure 2.2: OCV SoC curve [12]

As you can see the cell voltage does not fall rapidly as the cell is discharged,
but for the same reason, the actual cell voltage is not a good measure of the
SoC of the cell, since the smallest error in the OCV obtained from a battery
model can lead to divergence of SoC from the actual value. The rapid fall in
cell voltage at the end of the cycle could be used as an indication of imminent,
complete discharge of the battery, but for many applications an earlier warning
is required [10].

27
 CHAPTER 2. STATE OF CHARGE

2.2.3 Model-Based methods

Model-based methods deploy a battery model with advanced algorithms to es-

timate the states of a battery from its measured parameters such as voltage,
current, and temperature. There are two main model used for SoC estimation
Electrical Cicruit Model (ECM) and Electrochemical Model (EChM).

Electrical Cicruit Model-based methods are widely used for real-time battery
monitoring, using a circuit model built linking together parametric resistances
and capacitors. The idea behind is that the dynamic behaviour of the battery
needs to be mathematically modelled in a way that the SoC, derived from its
relation with OCV (§ 2.2), can be evaluated by just using voltage and current
measured at the terminals of the cell, in order to do this the circuit parame-
ters need to be found considering the operating condition of the cell, making
it the very big challenge of this technique. A very simple ECM is shown in
Figure 2.3 that describes the ideal battery as an OCV source in series with a
resistance representing the cell’s internal impedance.

Figure 2.3: Ideal battery model

Starting from the ideal model, the main improvement is given by adding two
parallel RC network that describe better the transient response of the battery
voltage.

28
 CHAPTER 2. STATE OF CHARGE

Figure 2.4: Caption

The primary RC network describes the charge transfer procedures while the
secondary should be modeled to describe electrochemical polarizations. Of
course the number of circuit element implies the increasing accuracy modelling
together with the computational complexity. Moreover the procedure for find-
ing the correct parameters is costly and high time consuming, not being able
anyway to fully describe the electrochemical reactions occurring inside the bat-
tery and the inaccuracies related to those. Nevertheless, once the parameters
are found, the equations for SoC are simple enough to be implemented by low
cost micro-controllers.

Electrochemical Models use the equations that describe physicochemical phe-

nomena like diffusion, intercalation, and electrochemical kinetics occurring in
a battery. The definition of these equations involves a specialized knowledge of
electrochemistry, hence their exploitation in the field of electrical and electron-
ics engineering is not common [13]. The main limitation is the mathematical
complexity to describe those models being so difficult a BMS implementation-
tion, as such equations would have to be solved in real time. To apply those
models for online battery monitoring is still something to fully explore, but
reduced models have been developed to facilitate employment [8]. The biggest
advantage of EChMs is that they inherently include the dependence of the
battery behavior on SoC and temperature, while electrical models must store
their parameters as look-up tables for various SoC and temperature combina-
tions and they need to be continuously updated for accurate estimation.

29
 CHAPTER 2. STATE OF CHARGE

2.2.4 Kalman filter-based methods

Kalman filters belong to those techniques defined as adaptive, combining direct

and model-based methods, adjusting the systems to changes. SoC is affected
by numerous factors and is continuously changing, also according to the user,
especially in EVs. The Kalman filter is able to minimize the uncertainty by
predicting the cell status and correcting that estimation [18].

Figure 2.5: Kalman filter schematic

The key element in this method is the feedback network performing the correc-
tion. As shown in Figure 2.5, the output of the actual system y is compared
with the prediction of the model ŷ. The result is used for correcting the model,
if the error between the prediction and the actual estimation is too wide, apply-
ing the gain K to the model. After a certain transitory period the estimation
converges to the state of the real system, meaning that the model is able to
perform an accurate prediction of the system behaviour. Kalman filter imple-
mentation resulted as an accurate SoC and SoH estimation method for Li-ion
Battery Pack, having nevertheless some drawbacks. Its complexity and compu-
tational cost are still very high, making its implementation difficult for ordinary
micro-controllers. Of course this kind of solutions is also highly dependant on
the battery model design, any inaccuracy in the system modelling may degrade
the performance of the filter, causing slow adaptation.

30
 CHAPTER 2. STATE OF CHARGE

2.2.5 Neural networks-based methods

A Neural Network is a computer architecture modelled upon the human brain’s

interconnected system of neurons which mimics its information processing,
memory and learning processes. They resemble human brain for their ability
to acquire knowledge through learning and for storing knowledge within neural
interconnections, known as weights. A Neural Network is structured in layers
as shown in Figure 2.6. Input layer receives the input parameters distributing
toward the output layer, through the hidden layers, and each node multiplies
the inputs, that come from previous layers, with the weight coefficient [2].

Figure 2.6: Neural network for SoC estimation

31
 CHAPTER 2. STATE OF CHARGE

Neural Network techniques are useful in estimating battery performance which

depends on quantifying the effect of numerous parameters most of which can-
not be defined with mathematical precision. Also in this case the calculation
complexity is very high, especially knowing that a big number of nodes are
needed in order to increase accuracy. Moreover it needs some time for learning
about the battery and then adapt to it considering also that in the meantime
battery is aging. The high computational effort makes the method not quite
fitting for online battery monitoring for automotive applications [13].

2.3 Algorithm implemented

For this thesis work the solution implemented for SoC estimation is based on
look-up tables constructed from cell’s measured data. To describe the cell be-
haviour from empirical values is a very good method since it automatically
takes in account many of the factors affecting the State of Charge. These look-
up tables are step-wise approximations derived from cell curves describing their
performance related to discharge rate and other parameters. For this study the
cells chosen are Li-ion Molicel-INR-21700-P42B [17] which characteristics are
shown in Figure 2.7. What is represented in those figures is a general descrip-
tion of how the cell behaves in different conditions. The first two images above
represent the discharge rate as a function of the cell voltage. The difference
between these two images is that the first shows the characteristic of the cell at
constant temperature, discharged at different amounts of current. The second
image, on the other hand, has the same purpose, which shows the behavior
of the cell which discharges at a constant amount of current, but at different
temperatures. The last image, below, is the one that shows the charging char-
acteristics of the cell, representing Capacity, Voltage and Current over time.
The key aspect is that the reload time changes when run at different C-Rates
(C). C-Rate is a way of measuring how fast a battery is charged or discharged,
which means that if a battery has a capacity of 4200 mA/h, 4.2A corresponds
to 1C. This kind of method let us to employ the model we are going to de-
rive on cell belonging to the same chain of production. This means that the
cell characterization from look-up tables should be carried out once with the

32
 CHAPTER 2. STATE OF CHARGE

assumption that the accumulator is always constructed using the same kind
of cell. If cell is changed during design process, the model must be redefined
according to the empirical data describing that cell behaviour. The biggest
drawback is that if the number of data available is not enough the model de-
rived is not able to characterize the cell behaviour in different conditions. Of
course there’s still the possibility to implement advanced methods, like neural
network based once, that are able to extrapolate a model even if the amount of
data does not cover the whole cell performance, but the consequence is that the
implementation for on-line monitoring using a micro-controller is not feasible.
The best solutions is to use a simpler model assuming the availability of a big-
ger data set, which requires a big workload, since it needs to test the cell many
times, and very time consuming to be completed, in order to cover as much
as possible the whole cell performance. In any case using this method those
kind of errors on SoC evaluation, coming from cell internal chemical reaction
or cell external operating conditions, should be kept at minimum. For example
in this case, having at disposal both charge and discharge characteristics let us
minimize the error derived from battery hysteresis.

33
 CHAPTER 2. STATE OF CHARGE

Figure 2.7: Molicel Lithium-ion Cell characteristics

34
 CHAPTER 2. STATE OF CHARGE

The mathematical model is built using polynomial regression analysis, in which

the non-linear relation between multiple input variables is modeled as an nth
degree polynomial, so that the model can be used to predict the general trend
of the system. In order to better explain the model, let’s start from the basic
concepts.
The linear regression algorithm is the simplest way to predict the value of a
dependent variable Y, starting from an independent variable X. In case X is a
set of variables, the model implemented is called Multiple Linear Regression.

Simple Linear Regression model equation: Y = θ0 + θ1 X

Multiple Linear Regression model equation: Y = θ0 + θ1 X + ... + θn Xn

As it is possible to extrapolate from the the mathematical equations, the re-

lationship is modeled as a straight line that is able to characterize the input
dataset and predict future values assuming that the dataset remains linear. As
soon as the relationship between variables is no more linear, this kind of model
is not able to fit, having a high prediction error. Figure 2.8 shows a non linear
dataset and its Linear Regression model which is not able to estimate the best
fit line.

Figure 2.8: Example of non linear dataset

Polynomial regression is similar to multiple linear regression but the relation-

35
 CHAPTER 2. STATE OF CHARGE

ship between the dependent and the independent variables is not linear. This
type of regression is used when data points are present in a non-linear fashion.
The polynomial regression algorithm transforms the data points into polyno-
mial features of specified degrees and models them using the linear technique
[24].

Polynomial Regression model equation: Y = θ0 + θ1 X + ... + θn X n

As shown in Figure 2.9 the model is able to better follow the data-set trend.

Figure 2.9: Polynomial Regression model

With this kind of regression there is the possibility to better follow the data
trend, but risking to overf it. This happens when the dataset has missing or
imbalanced data or contains noisy data, becoming more evident with higher
degree polynomials. The effect is that the model adapts very well to training
data points, like the blue line in Figure 2.10, and not so good to other points,
reducing the model accuracy. A linear regression in this example gives better
results.

36
 CHAPTER 2. STATE OF CHARGE

Figure 2.10: Overfitting

There are still ways to minimize the error derived from overfitted models, by
eliminating those weights(θ) with higher value, since they may destabilise the
model trend, but still keeping the polynomial degree [6]. What has been shown
so far was the situation in which it is necessary to predict the performance of a
system starting from a single input. In real situations the models are based on
look-up tables that contain multiple columns. The easiest situation to present
is that the dependent variable Z, has two types of inputs, X and Y , making it
possible to have a data-set that can be represented in three dimensions.

Polynomial Regression model equation with 2 inputs:

Z = θ0 + θ1 X + θ2 Y + θ3 XY + ... + θn X n + θn Y n

In order to make sure that the model is able to predict the cell trend, the first
important step is to split the dataset in two or three subsets:

• Train set: it is the actual set of data used by our model to learn about
the relation between Y and X. From here the model is training to develop
and predict the output trend trying to minimize the difference between
prediction and real data. This is done by updating the weights θ.

• Validation set: it is the set of data able to validate the predictive

capabilities of the model minimizing overfitting. After that the model

37
 CHAPTER 2. STATE OF CHARGE

evaluated the weight, so after it made a prediction, it is given a portion

of X that till now the model did not know about and it makes a predic-
tion. Now there’s a Y predicted by the model and a real Y belonging to
the input dataset, that are compared to see how much distance there is
between those two variables. If the prediction error is too high the model
parameters must be changed, by increasing the grade of the equation and
restart from training.

• test set: this set is omitted most of the time. This is used to perform
further observation on the model prediction, after that its performed was
considered suitable with train and validation.

Due to its great importance, the train set is usually the one containing the
majority of the variables belonging to our input look-up table. The key is to
define the percentage of variables to be given at the validation set. A too large
validation set would mean stealing learning data, while a set too small is not
able to fully validate the model trend [19]. For this work the test set has been
omitted and the dataset division has been randomly performed between train
set and validation set as shown in Figure 2.11.

Figure 2.11: Dataset division

38
 CHAPTER 2. STATE OF CHARGE

% Dataset randomly divided: Train 60%, Validation 40%

[index_training,index_validation] = dividerand(length(X),0.6,0.4,0);

X_training = X(index_training);
Y_training = Y(index_training);
Z_training = Z(index_training);

X_validation = X(index_validation);
Y_validation = Y(index_validation);
Z_validation = Z(index_validation);

What is shown above is a portion of the code written on M AT LAB, regard-

ing the dataset variables initial manipulation. The model that is going to be
presented, is the one predicting the State of Charge of a cell, starting from a
look-up table containing the relation between SoC, voltage and current. This
dataset has been completed from cell characteristics shown in Figure 2.7 and
then reported in Figure 2.12. The output independent variable to be pre-
dicted, the SoC, is labeled as Z, while voltage and current are the dependent
variables labeled as X and Y respectively. The validation is evaluated calcu-
lating the Root Mean Square Error (RMSE), when it’s too high the training is
repeated with a new set of parameters, increasing grade model equation.

Figure 2.12: Input dataset

39
 CHAPTER 2. STATE OF CHARGE

2.3.1 1st grade model equation

This equation represents also the multi-linear regression model, since it de-
scribes a linear relationship between dependent variable and a set of indepen-
dent variables. As we already know this model is not fitted due to the non
linearity of the dataset. This can be noticed looking at Figure 2.13.

Z = θ0 + θ1 X + θ2 Y

θ0 = −214.95, θ1 = 73.586, θ2 = 0.6411

RM SE = 11.6085

Figure 2.13: Grade 1 model equation

40
 CHAPTER 2. STATE OF CHARGE

2.3.2 2nd grade model equation

Form now on the model starts to have non linear behaviour, as shown in Figure
2.14. Nevertheless this equation is still not suitable with respect to the input
dataset. Moreover the RMSE value is still too high to consider the model as
good.

Z = θ0 + θ1 X + θ2 Y + θ3 XY + θ4 X 2 + θ5 Y 2

θ0 = 165.8021, θ1 = −150.8119, θ2 = −0.7859

θ3 = 0.5280, θ4 = 32.2993, θ5 = −0.0063

RM SE = 9.0881

Figure 2.14: Grade 2 model equation

41
 CHAPTER 2. STATE OF CHARGE

2.3.3 3rd grade model equation

At this point equation complexity is starting to grow, together with accuracy,

having higher grade with 10 parameters and the RMSE value is starting to
decrease. It is still not enough, not fitting correctly at both ends of the dataset,
Figure 2.15.

Z = θ0 + θ1 X + θ2 Y + θ3 XY + θ4 X 2 + θ5 Y 2 + θ6 X 2 Y + θ7 XY 2 + θ8 X 3 + θ9 Y 3

θ0 = 2660.1, θ1 = −2351.7, θ2 = −20.265

θ3 = 12.223, θ4 = 670.07, θ5 = 0.0060

θ6 = −1.6780, θ7 = −0.0129, θ8 = −60.835

θ9 = 0.0004

RM SE = 5.4602

Figure 2.15: Grade 3 model equation

42
 CHAPTER 2. STATE OF CHARGE

2.3.4 4th grade model equation

Having increased the degree, the number of parameters that make up the equa-
tion has increased accordingly. RMSE value still not valid and visually the
model in not enough accurate.

Z = θ0 + θ1 X + θ2 Y + θ3 XY + θ4 X 2 + θ5 Y 2 + θ6 X 2 Y + θ7 XY 2 + θ8 X 3 +

θ9 Y 3 + θ10 X 3 Y + θ11 X 2 Y 2 + θ12 XY 3 + θ13 X 4 + θ14 Y 4

θ0 = −7039.1, θ1 = 8945.4, θ2 = 107.31

θ3 = −103.05, θ4 = −4202.1, θ5 = −0.1854

θ6 = 32.54, θ7 = 0.089, θ8 = 862.22

θ9 = 0.0019, θ10 = 4.5 · 10−5 , θ11 = −0.017

θ12 = −3.335, θ13 = −64.89, θ14 = −1.8 · 10−5

RM SE = 3.5718

Figure 2.16: Grade 4 model equation

43
 CHAPTER 2. STATE OF CHARGE

2.3.5 5th grade model equation

Z = θ0 + θ1 X + θ2 Y + θ3 XY + θ4 X 2 + θ5 Y 2 + θ6 X 2 Y + θ7 XY 2 + θ8 X 3 +

θ9 Y 3 + θ10 X 3 Y + θ11 X 2 Y 2 + θ12 XY 3 + θ13 X 4 + θ14 Y 4 + θ15 X 4 Y +

θ16 X 3 Y 2 + θ17 X 2 Y 3 + θ18 XY 4 + θ19 X 5 + θ20 Y 5

θ0 = −29852, θ1 = 43442, θ2 = 241.26

θ3 = −266.14, θ4 = −24896, θ5 = −0.4283

θ6 = 105.46, θ7 = 0.4451, θ8 = 7020.7

θ9 = −0.0037, θ10 = 0.0035, θ11 = −0.1617

θ12 = −17.541, θ13 = −974.48, θ14 = −2.9 · 10−5

θ15 = 1.0196, θ16 = 0.0172, θ17 = −0.0004

θ18 = −1.1 · 10−5 , θ19 = 53.363, θ20 = 4.2 · 10−7

RM SE = 2.9890

Figure 2.17: Grade 5 model equation

44
 CHAPTER 2. STATE OF CHARGE

2.3.6 6th grade model equation

Most accurate model obtained having RM SE < 2. Equation complexity is

very high.

Z = θ0 + θ1 X + θ2 Y + θ3 XY + θ4 X 2 + θ5 Y 2 + θ6 X 2 Y + θ7 XY 2 + θ8 X 3 + θ9 Y 3

+ θ10 X 3 Y + θ11 X 2 Y 2 + θ12 XY 3 + θ13 X 4 + θ14 Y 4 + θ15 X 4 Y + θ16 X 3 Y 2 +

θ17 X 2 Y 3 + θ18 XY 4 + θ19 X 5 + θ20 Y 5 + θ21 X 5 Y + θ22 X 4 Y 2 + θ23 X 3 Y 3 +

θ24 X 2 Y 4 + θ25 XY 5 + θ26 X 6 + θ27 Y 6

θ0 = −469.54, θ1 = 0, θ2 = −2394.8, θ3 = 3652

θ4 = 246.79, θ5 = 11.596, θ6 = −2198, θ7 = −14.991

θ8 = 0, θ9 = 0.0093, θ10 = 0.0051, θ11 = 7.0575

θ12 = 652.35, θ13 = −74.284, θ14 = −0.0005, θ15 = −95.414

θ16 = −1.4556, θ17 = −0.0023, θ18 = 5.7125 · 10−5 , θ19 = 24.504

θ20 = 7.0458 · 10−6 , θ21 = 5.5003, θ22 = 0.1107, θ23 = 0.00035

θ24 = −1.087 · 10−5 , θ25 = 7.816 · 10−8 , θ26 = −2.3131, θ27 = −4.7735 · 10−8

RMSE = 1.9796

Figure 2.18: Grade 6 model equation

45
 CHAPTER 2. STATE OF CHARGE

2.3.7 Conclusions

The last model presented is the one with the lowest value of RMSE, so the
most accurate. The overfitting is minimal in regions where train data is not
present meaning that is not required any additional work and the equation can
be still employed, despite its complexity, for a real time online calculation by
the Battery Management System Master ECU. The main problem is that the
model presented is the one trying to predict the battery trend according to
Voltage and Current with constant temperature at 23 °C. This is not enough
since temperature is constantly changing around a Battery Pack, making the
SoC to change accordingly. For this purpose, there was the attempt to develop
a model using as independent variables voltage and temperature derived from
the cell characteristic from Figure 2.7. The problem was that the little amount
of data and the difficult behaviour of the voltage at low temperature, make the
model too imprecise with many regions where overfitting occurred, especially
where the train data is less dense. Figure 2.19 shows the model obtained
using polynomial regression with 6th grade equation. This one was the most
precise obtained, but the RMSE value is still too high.

Figure 2.19

46
 CHAPTER 2. STATE OF CHARGE

Between these two models presented, the V-I model is the one easier to imple-
ment. There exist in fact solutions to let the temperature around the Battery
Pack to be constant, letting cells to operate at optimal conditions.

47
 CHAPTER 3

Firmware

Figure 3.1: Firmware structure

The firmware has been structured to have three main layers. Further visual
details are shown in the Figure 3.1

• Microcontroller Abstraction Layer: modules, or drivers, belonging

to this layer have direct access to all the MCU peripheral modules and
external devices, which are mapped to memory. In this way upper layers
are able to work independently from the micro.

• Hardware Abstraction Layer: this layer allows the application to

interact with the hardware device at a general or abstract level rather
than at a detailed.

48
 CHAPTER 3. FIRMWARE

Functions belonging to this module are programmed independently to

the physical interface of the device or the actual Hardware features.
At this stage is where data are managed and sent to upper layer.

• Application Layer: this is the upper level of the structure, where all
interactions with an external user take place. This is the upper level
of the structure, where all the interactions with an external user take
place, in this case the battery pack itself. Here the data coming from
the lower levels are acquired and consequently the management of the
battery system takes place according to the processing of this data.

The execution of the functions belonging to the three layers is carried out by
the scheduler inside the main. The idea is to entrust the execution of certain
portions of code to three different tasks that work at three different frequencies.
In order to assure that each code portion is executed at the same frequencies,
the constraints are:

• The frequency at which each task is executed is such that the period of
execution of each task must be an integer multiple of the fastest task

t1 = a · t3

with a, b ∈ N, 1 < a < b.
t2 = b · t3


• The execution period of each piece of code belonging to a task must be

lower then the time lapse between calls of the fastest task in order to
avoid overlapping, which means Tex < min{t1 , t2 , t3 }

The three tasks are executed with frequency of 1 ms, 2 ms, 5 ms each. Before
the execution of the main, during an initialization phase, a timer is set that
counts the number of clock cycles that correspond to the execution frequency
of the tasks. Figure 3.2 shows the flowchart of the pieces of codes belonging
to the three tasks.

49
 CHAPTER 3. FIRMWARE

Figure 3.2: Firmware flow chart

3.1 Master-Slave data exchange

The acquisition of voltage and temperature values to monitor the status of a
battery is managed directly by the slave and then sends everything in a message
via UART. The exchange of information through the UART protocol is instead
managed by the micro through the specific driver that works together with the
DMA in such a way as to speed up the data processing procedure.

50
 CHAPTER 3. FIRMWARE

3.1.1 UART message structure

The Universal Asynchronous Receiver Transmitter (UART) protocol is among

the most used in the embedded world, it is a type of P2P communication in
which the two communicating nodes have two different clock sources, so it is
defined asynchronous. Nevertheless, the data transmission requires periodic
synchronization between the two nodes so that the message is transmitted and
received correctly. The transmission of a message requires a subdivision into
frames. What is shown in Figure 3.3 is the general structure of a frame
transmitted by the master and by the slave for this thesis work.

Figure 3.3: UART frame structure

• START bit: it is the bit used for node synchronization and to indicate
the start of the frame. To favor a correct coding of the protocol, the
UART line is kept at a high logic level when it is not used, therefore in
idle. The START bit is the first bit to force a transition from high to low
level, for this reason it is used for synchronization.

• STOP bit: the stop bit indicates the end of the frame. In some cases
there may also be two STOP bits and is always kept at a logical high
level, before going into idle state. The two states are differentiated by
the fact that the position, and number of bits that make up the stop of
the frame is known, once the stop bit is identified.

• 8-bit data: this portion of the frame is nothing more than the actual
information to be sent. In this case its length is 8 bits, but it can vary.
The most common cases concern 6 to 9 bit data transmission

51
 CHAPTER 3. FIRMWARE

The UART protocol also provides for the use of an optional bit for parity
checking.
All these are parameters set in the UART driver that manages its interface.
The transmission rate of the message is also set at this level. This parameter is
defined as Baud Rate, which would be the amount of symbols transmitted in
one second, in this case the transmission speed is set to 1 MBaud, meaning that
the transmission of one bit requires 100ns. If the master transmits the message,
this means that the transmission of a bit requires to set a timer that counts
8 clock strokes, the latter having a frequency of 80MHz. The structure of the
UART messages that the master must send to the slave ECU is predefined
by the microcontroller manufacturer, as specified in the datasheet [23]. The
communication via UART by the master, at the beginning consists in having
to program and set the slave microcontroller, following the procedures already
defined by the manufacturer, in order to carry out the actual acquisition of
the battery monitoring parameters. This occurs by interacting directly with
the microcontroller registers, setting how to manage certain peripherals. Some
examples are the setting of the voltage reading thresholds, the sampling rate
and the BAUD rate. For this reason the procedure for communicating with
the slave involves the transmission of five different types of frames:

• Frame initialization: This is the first frame to be sent in a message

and is used to define the type of message, whether it is a response or
a command, to whom it is intended, since the slave system is made up
of several ECUs, which register to set, based on functionality, and the
amount of data to be sent. More details are shown in Figure 3.5.

Figure 3.4: Frame initialization byte

• Device address: It identifies the device or group of devices targeted

by the command. If the message is defined by the previous frame as a
Broadcast message, this frame can also be omitted.

52
 CHAPTER 3. FIRMWARE

• Register address: This frame defines the address of the register to be

overwritten or the one from which we want to read data.

• Data: The number of data Bytes is defined in the first frame. In case the
message is a command this frame contains the amount of data to write in
the destination register. Otherwise it contains the information contained
in the target register.0

• CRC: This portion is 2 bytes long, contained in two frames, useful for
checking the correctness of the transmitted message

Figure 3.5: Frame initialization byte fields table

53
 CHAPTER 3. FIRMWARE

An example of command frame is the one used for setting the Baud rate:

• 0xF2: Setting a Broadcast command without response. Data quantity 2

Bytes. No device address frame is required.

• 0x10: Communication configuration register address.

• 0x30E0: Data to be written inside the register. In this case in order to

set the baud rate, bit 13 and 12 needed to be written as 1.

• 0x26F5: 16-bit CRC

Figure 3.6: UART command message acquisition

As already said previously, the management of the UART device is entrusted to

its driver which sets the dedicated registers to have a transmission of messages
at 1 MBaud rate and 10-bit frames. The UART driver works concurrently with
DMA for both incoming and outgoing messages. The Direct Memory Access
(DMA) is the mechanism that allows other subsystems, such as peripherals, to
directly access the internal memory to exchange data, read and write, avoiding
overloading the CPU with a series of interrupts that could slow it down.

54
 CHAPTER 3. FIRMWARE

When a message is to be sent, the data is stored in a buffer in Direct Memory

Access (DMA), allowing the UART driver to read the data from the memory,
construct the frames and send the message. The reverse process occurs in re-
ception. The UART driver saves the data in Direct Memory Access (DMA),
ready to be read by the functions dedicated to data processing at the HAL level.
This whole procedure is synchronized by an interrupt system that signals the
drivers when to act. The main reason why this solution has been implemented
is because, once the DMA and UART drives have been set, the message trans-
mission and reception procedure takes place automatically, including saving
the data received in the memory in a static variable, allowing the portions of
code belonging to the HAL to continue processing data in parallel. Once the
transmission or reception is finished, an interrupt is sent in order to avoid that
two transmission operations can overlap or that a received data is overwritten
before being read and, furthermore, in this way the exchange of information
takes place at such a high speed.

void DMA1_IRQHandler(void)
{
DMA -> CINT = 1; /*Clear channel 1 interrupt request*/
/*UART Tx management*/
BMSTxDMACallback();
}

void DMA0_IRQHandler(void)
{
/*Interrupt request for channel 0 - UART RX*/
DMA -> CINT = 0; /*Clear channel 0 interrupt request*/
/*Disable Timer*/
LPIT0Ch1Stop();
/*Callback function for managing received values in UART*/
BMSRxDMACallback();
}

The two functions shown above handle the interrupt request when the trans-
mission and reception procedures are finished. The first calls a function that
simply allows the system to be able to transmit again. The second, on the other
hand, starts the reading of a new data received and its subsequent processing.

55
 CHAPTER 3. FIRMWARE

3.1.2 BMS data management function

Before reaching the phase in which the voltage and temperature values are
cyclically acquired from the slave system, the Master ECU must first perform
a series of steps, concerning the transmission of messages via UART, predefined
according to the datasheet, in order to correctly set the functionality of the slave
system. There are two first stages, the first for initialization and a second one
for setting up the analog front-end. In both these states, the messages are only
transmitted by the master, with the possibility, however, of requesting to reply
to some messages to check the correctness of the previous actions. To speed up
the procedure, already knowing the structure of the messages, they have been
built in advance and saved in a matrix, in the order in which they must be
sent. So once the system is in one of the two states, it constructs the message
and saves it in DMA ready to be transmitted.

/*{datasize, respsize, cmdtype, devaddr, regdest, TxData}*/

static uint08t *MsgConfigInit[] = {
/*Set Communication Rate(1MBaud) and no comm fault*/
(uint08t[]){2u, 0u, BroadcastNoResp, 0u, COMCONFIGreg, 0x30u, 0xE0u},
/*Enable Auto-Addressing Mode*/
(uint08t[]){1u, 0u, BroadcastNoResp, 0u, DEVCONFIGreg, 0x10u},
/*Start Auto-Addressing*/
(uint08t[]){1u, 0u, BroadcastNoResp, 0u, DEVCTRLreg, 0x08u},
/*Set Address for each slave*/
(uint08t[]){1u, 0u, BroadcastNoResp, 0u, DEVADDRreg, 0x00u},
(uint08t[]){1u, 0u, BroadcastNoResp, 0u, DEVADDRreg, 0x01u},
/*check that the addresses are correct*/
(uint08t[]){1u, 1u, SingleResp, 0u, DEVADDRreg, 0x00u},
/*disable COMM_LOW high-side*/
(uint08t[]){2u, 0u, SingleNoResp, 0u, COMCONFIGreg, 0x30u, 0x80u},
/*Clear All faults*/
(uint08t[]){2u, 0u, SingleNoResp, 0u, FAULTSUMMARYreg, 0xFFu, 0xC0u}
};

56
 CHAPTER 3. FIRMWARE

/*{datasize, respsize, cmdtype, devaddr, regdest, TxData*/

static uint08t *MsgConfigAFE[] = {
/*Set sample delay 0*/
(uint08t[]){1u, 0u, SingleNoResp, 0u, SAMPLEDELAYreg, 0x00u},
/*Clear all faults*/
(uint08t[]){1u, 0u, SingleNoResp, 0u, STATUSreg, 0x38u},
/*Set sample period for cells to 60us*/
(uint08t[]){1u, 0u, SingleNoResp, 0u, CELLSPERreg, 0xBCu},
/*Set sample period for temp to 60us*/
(uint08t[]){4u, 0u, SingleNoResp, 0u, AUXreg, 0xBBu, 0xBBu, 0xBBu, 0xBBu},
/*Set no oversampling period*/
(uint08t[]){1u, 0u, SingleNoResp, 0u, OVERSAMPLEreg, 0x00u},
/*Set cells used by the device*/
(uint08t[]){1u, 0u, SingleNoResp, 0u, NCHANreg, 0x00u},
/*Set cells/aux included in the measurement*/
(uint08t[]){4u, 0u, SingleNoResp, 0u, CHANreg, 0x00u, 0x00u, 0x00u, 0x00u}
};

The portion of code shown above is the matrix containing the structure of the
messages to be sent to the slave system for the initialization phase. Unlike the
next phase, that is the setting of the acquisition parameters, this requires to
be performed only once, while the second must be performed for each ECU
belonging to the slave chain. Once the slave system has been correctly set up,
the data acquisition phase begins by decoding the UART messages received
from the slave chain. The structure of the received message is quite complex.
In fact it is a concatenation of messages deriving from the single ECU, starting
from the one with the highest address.

57
 CHAPTER 3. FIRMWARE

The response structure of a single device in the slave chain consists of:

• Header byte: this value represents the number of bytes given in the
response, subtracted by one, considering that for each reading the value
is saved in 2 bytes. This means that if an ECU has read 6 voltage values,
the response contains 12 data bytes representing those valued and the
value of the header byte would be 11.

• Data bytes: as already specified, each voltage and temperature acquisi-

tion channel saves the read values in 16-bit registers. These are then sent
in a specific order, always starting from the channels with the highest
cardinality, sending first the voltage values and then the temperature.

The complexity of the message therefore requires a phase of elaboration. To

speed up this, once the decoding has been carried out, the single values are
saved in static variables and while the data is being processed a new request
is made. In the end, the voltage values of the cell and temperature groups are
saved in data structure containing information to be sent to the application
level. Those shown below are in fact structures containing information on the
generic state of the system and on the conditions of the cells, in terms of voltage
and temperature, the latter useful for calculating the State of Charge.

typedef struct {
uint16t TotalPackV;
e_BMSStatus_t status;
}s_SysInfo_t;

s_SysInfo_t BMSSysInfo;

typedef struct {
sint16t High;
sint16t Low;
sint16t Avg;
uint08t HighId;
uint08t LowId;
}s_cell_t;

s_cell_t BMS_TCELL;
s_cell_t BMS_VCELL;

58
 CHAPTER 3. FIRMWARE

3.2 SoC value calculation and validation

Once all the parameters have been acquired and processed, the voltage, current
and temperature values of the accumulator are given as input to the model for
the calculation of the SoC, which we talked about in the previous chapters.
The model was validated in a low voltage prototype electrical system devel-
oped by the team ”Squadra Corse” at Politecnico di Torino. Each year the
team develops a fully electric racing car to compete in the student formula in
which the best engineering project is awarded. The vehicle’s electrical system
is powered by two accumulators, one at high voltage, for supplying the engines,
and one at low voltage, to power the set of ECUs to monitor using multiple
sensors and to manage some vehicle functions. Figure 3.7 shows the whole
electrical system.

Figure 3.7: Prototype Electrical system

As for the validation of the SoC model, the parameters were acquired by writing
a log file while charging the battery pack. A device for CAN Analyzing has been
used, connected to CAN Network, reading the target values and constructing
the log file. The acquired parameters are the same ones that are supplied in
input to the model and sent to the rest of the vehicle via CAN. In the case of
this thesis work, the temperature value is not considered because the model to
be validated is based on the calculation of the State of Charge having voltage
and current as independent parameters. The log file containing voltage, current

59
 CHAPTER 3. FIRMWARE

and SoC was then reworked on MATLAB in order to graphically represent these
data.Figure 3.8 represents the graphs obtained while Figure 3.9 shows the
path for processing the values up to the final graph.

Figure 3.8: Battery charging curves

Figure 3.9: Data flowchart

60
 CHAPTER 4

Conclusions

This thesis work stems from the idea of developing a management system for
high voltage accumulators in vehicles, hybrid or fully electric, in such a way
as to optimize the use of the cells, prolonging their life times and reducing the
risks deriving from a bad handling, especially if the cells are made up of lithium
batteries. One of the key parameters is what defines the state of charge of a cell.
The main objective is to design and implement a mathematical model that can
calculate the SoC parameter starting from physical characteristics that describe
the battery, such as voltage, current and temperature. This model must be
implementable with a microcontroller for an evaluation of the SoC parameter
in real time, without blocking the execution of other functions that the MCU
manages. The model was built starting from experimental data provided by
the manufacturer of the lithium cells, with which it was possible to develop a
function that describes the performance of the battery according to these tables
and predict its behavior in different situations.The final result is a fairly precise
model, able to adapt to situations other than those used for the realization
of the input data set. The main limitation of the model is that the accuracy
reached with the model is valid when the temperature is around 23 ° C. As noted
in previous chapters, temperature is a vital parameter for monitoring sudden
changes and predicting cell instability. One possible solution is to be able to
have input datasets that describe as many situations as possible. However, this
is a procedure that requires a considerable expense in terms of time, since it

61
 CHAPTER 4. CONCLUSIONS

would involve filling tables with experimentally measured values. However, this
can lead to a mathematical model capable of calculating the soc more precisely
in non-standard conditions. This would favor the implementation of techniques
for good battery management such as cell balancing with control criterion based
on the state of charge of the battery. The feasibility of solutions of this type
increases if the hardware design of the ECU is carried out according to the needs
of the designer. Currently the system has been implemented using demo boards
that mount the micro controllers that were needed. These modules offer the
ability to create a system without the need for hardware development, reducing
costs and saving time, provided you have a system not tailored to implement
any specific application. A further possible improvement is the possibility of
evaluating the State of Health of a cell, which together with the SoC, are very
useful parameters to understand how much the battery is able to deliver in
terms of capacity and how many number of cycles charge and discharge is able
to withstand.

62
 Bibliography

[1] Battery management systems (bms). URL https://www.mpoweruk.com/

bms.htm.

[2] Giuseppe Luca Amoroso. Realizzazione di algoritmo per la stima dello

stato di carica di batterie al litio con identificazione online dei parametri.
Master’s thesis, Università di Pisa, 2015.

[3] Ines Baccouche, Sabeur Jemmali, Asma Mlayah, Bilal Manai, Najoua Es-
soukri, and Ben Amara. Implementation of an improved coulomb-counting
algorithm based on a piecewise soc-ocv relationship for soc estimation of
li-ion battery. International journal of renewable energy research, June
2017.

[4] BSI ISO-11898-2 Road vehicles - Controller Area Network. British Stan-
dards Institution, December 2016.

[5] BSI ISO-6469-3 Electrically propelled road vehicles - Safety specifications.

British Standards Institution, November 2021.

[6] Cristiano Casadei. Apprendimento supervisionato-la regressione polino-

miale e la regolarizzazione. Maggioli developers, Maggio 2019.

[7] K.W.E. Cheng, B.P. Divakar, Hongjie Wu, Kai Ding, and Ho Fai Ho.
Battery-management system (bms) and soc development for electrical ve-
hicles. IEEE transactions on vehicular technology, Vol. 60, January 2011.

[8] Matteo Corno, Nimitt Bhatt, Sergio M.Savaresi, and Michel Verhaegen.
Electrochemical model-based state of charge estimation for li-ion cells.
December 2014.

63
 BIBLIOGRAPHY

[9] Eletropedia. Lithium battery failures, . URL https://www.mpoweruk.

com/lithium_failures.htm#soc.

[10] Eletropedia. State of charge (soc) determination, . URL https://www.

mpoweruk.com/soc.htm.

[11] Eletropedia. State of health (soh) determination, . URL https://www.

mpoweruk.com/soh.htm#top.

[12] Fadlaoui Elmahdi. Fitting the ocv-soc relationship of a battery lithium-ion

using genetic algorithm method. EDP Sciences, 2020.

[13] Henry Omar Sarmiento-Maldonado Juan Pablo Rivera-Barrera, Nicolás

Muñoz-Galeano. Soc estimation for lithium-ion batteries: Review and
future challenges. Electronics MDPI, November 2017.

[14] Barrie Lawson. A software configurable battery. URL https://www.

mpoweruk.com/Software_Configurable_Battery.htm#balancing.

[15] Automotive current transducer open loop technology. LEM, 2018 March.

[16] Leandro Malara. Analysis and modeling of an energy management system

in automotive environment. Master’s thesis, Politecnico di Torino, Marzo
2020.

[17] Lithium-ion rechargeable battery. MOLICEL.

[18] Martin Murnane and Adel Ghazel. A closer look at state of charge (soc)
and state of health (soh) estimation techniques for batteries. Analog de-
vices.

[19] Davide Nardini. Train, validation, test: cosa sono e come si usano nel
machine learning. Pulp Learning, Novembre 2020.

[20] Java T Point. Can (controller area network) protocol. URL https://www.
javatpoint.com/can-protocol.

[21] Aswinth Raj. Cell balancing techniques and how to

use them. URL https://circuitdigest.com/article/
cell-balancing-techniques-and-how-to-use-them.

64
 BIBLIOGRAPHY

[22] Ajit Sharma. Enabling the electric future of mobility: Robotic automation
for electric vehicle battery assembly. IEEE Access, page 32, September
2019.

[23] bq76PL455A-Q1 16-Cell EV/HEV Integrated Battery Monitor and Pro-

tector. Texas Instruments, 2015 April.

[24] Analytics Vidhya. An introduction to simple linear regres-

sion. URL https://www.analyticsvidhya.com/blog/2021/07/
an-introduction-to-linear-regression/.

65