STS09039

Examensarbete 20 p
Oktober 2009

Application Pressure for an

Electronic Brake System
Estimation and implementation of a recursive
filter

David Moody
 Abstract
Application pressure for an Electronic Brake System:
Estimation and implementation of a recursive filter
David Moody

Teknisk- naturvetenskaplig fakultet

UTH-enheten Scania CV AB is one of the major heavy truck manufacturers in the world. Many of
their trucks use a brake system that is electronic where the driver demands
Besöksadress: retardation and the brake power is generated pneumatically. The objective of this
Ångströmlaboratoriet
Lägerhyddsvägen 1 thesis is to re-search if it is possible to determine the pressure at which the brakes
Hus 4, Plan 0 start giving a retard-ing force, the application pressure, for the Electronic Brake
System (EBS) and if a pre-dictive estimation of the application pressure can be
Postadress: implemented.
Box 536
751 21 Uppsala
Three different ways of determining the application pressure are presented in the
Telefon: thesis and a signal processing method where the mean of pre-braking acceleration is
018 – 471 30 03 used for change detection gives the most reliable results. This method is then
Telefax: implemented in a recursive filter where a predictive estimation of the application
018 – 471 30 00 pressure is developed. Due to the lack of precision in the signals the estimation does
not give a result that can be used and until this is solved by Scania it is difficult to see
Hemsida: how a good way of pre-dicting the application pressure can be developed.
http://www.teknat.uu.se/student

Handledare: Daniel Frylmark

Ämnesgranskare: Alexander Medvedev
Examinator: Elisabet Andrésdóttir
ISSN: 1650-8319, STS09 039
Sponsor: Scania CV AB
Table of contents
Terminology ________________________________________________________________ 3
Abbreviations ___________________________________________________________ 4
Mathematical Notations ___________________________________________________ 5

Part I_______________________________________________________________________ 9

1. Introduction_______________________________________________________________ 9
1.1 Objective ___________________________________________________________ 9
1.2 Disposition __________________________________________________________ 9
1.3 Confidentiality________________________________________________________ 9

2. Method__________________________________________________________________ 11
2.1 Collecting data ______________________________________________________ 12
2.1.1 Plane ground_________________________________________________ 12
2.1.2 Constant velocity______________________________________________ 12
2.1.3 No turning ___________________________________________________ 12
2.1.4 Free rolling __________________________________________________ 13
2.1.5 Assumption that the ground is constant before braking ________________ 13
2.1.6 Retarder and exhaust brake not included___________________________ 13
2.2 Considerations for the Different Methods of Collecting Data___________________ 14
2.2.1 Braking One Wheel____________________________________________ 14
2.2.2 Braking All Wheels ____________________________________________ 14
2.2.3 Braking One Axle _____________________________________________ 15
2.2.4 Altering the Application Pressure of the Wheels _____________________ 15

Part II – Theory _____________________________________________________________ 16

3. What is EBS? ____________________________________________________________ 16

3.1 An Overview________________________________________________________ 16
3.2 The Air Disc Brake in the Electronic Brake System __________________________ 17
3.2.1 Brake Chamber_______________________________________________ 18
3.2.2 Brake Calliper ________________________________________________ 19
3.2.3 Friction Pair__________________________________________________ 20
3.2.4 Summary of the Air Disc Brake System ____________________________ 21

4. Wheel and Vehicle Dynamics _______________________________________________ 22

4.1 Moment of Inertia ____________________________________________________ 22
4.2 Angular Velocity and Angular Acceleration ________________________________ 22
4.3 Torque ____________________________________________________________ 22
4.4 Accelerating Forces __________________________________________________ 23
4.4.1 Engine______________________________________________________ 24
4.4.2 Clutch ______________________________________________________ 24
4.4.3 Transmission_________________________________________________ 24
4.4.4 Propeller shaft________________________________________________ 24
4.4.5 Final drive ___________________________________________________ 25
4.4.6 Drive shafts __________________________________________________ 25
4.4.7 Wheels _____________________________________________________ 25
4.5 Decelerating Forces __________________________________________________ 26
4.5.1 Aerodynamic drag resistance ____________________________________ 26

1
 4.5.2 Rolling resistance _____________________________________________ 26
4.5.3 Slope inclination ______________________________________________ 27
4.5.4 Brakes______________________________________________________ 27
4.5.5 Engine brake_________________________________________________ 27
4.6 Summary of the forces acting on the vehicle _______________________________ 27

5. Statistical theory/change-point analysis ______________________________________ 29

5.1 Statistical Terminology ________________________________________________ 29
5.1.1 Mean _______________________________________________________ 29
5.1.2 Standard Deviation ____________________________________________ 29
5.1.3 Moving Average Filter __________________________________________ 30
5.2 Statistical Process Control _____________________________________________ 30
5.2.1 Shewhart Control Chart ________________________________________ 30
5.2.2 Western Electric Company Rules for Detecting Non-Random Behaviour __ 31

Part III – Empirical studies____________________________________________________ 33

6. Data considerations _______________________________________________________ 33

6.1 Signals ____________________________________________________________ 33
6.2 Estimating Acceleration _______________________________________________ 34
6.2.1 Numerical Estimation of the Acceleration ___________________________ 34
6.3 Signal noise ________________________________________________________ 36
6.4 Filtering data _______________________________________________________ 36
6.5 Limitations in data ___________________________________________________ 37
6.5.1 Phase Shifts Between Signals ___________________________________ 37

7. Methods of determining the application pressure ______________________________ 39

7.1 Using acceleration to determine application pressure ________________________ 39
7.2 Model Based Method _________________________________________________ 39
7.2.1 Choosing a Model to Describe the System _________________________ 39
7.2.2 Model Based Diagnosis ________________________________________ 41
7.2.3 Results from Model Based Method________________________________ 41
7.3 Signal Processing Methods ____________________________________________ 42
7.3.1 Method one: Modelling the Pre-Braking Acceleration _________________ 42
7.3.2 Results from Modelling the Pre-Braking Acceleration _________________ 44
7.3.3 Method two: Detecting when the Acceleration Deviates from the Mean of the
Pre-Braking Acceleration ____________________________________________ 45
7.3.4 Results from Determining the Mean of the Pre-Braking Acceleration _____ 46
7.4 Finding the Deviation from the Pre-Braking Acceleration _____________________ 46
7.4.1 Modifying the Western Electric Company Rules _____________________ 47

8. Implementing an Estimation of the Application Pressure ________________________ 49

8.1 Recursive Filter _____________________________________________________ 49
8.2 Considerations when Implementing Recursive Filter_________________________ 49
8.3 Results from the Observer Implementation ________________________________ 50

9. Conclusion ______________________________________________________________ 51
9.1 Suggestions for Scania _______________________________________________ 52
9.2 Suggestions for further research ________________________________________ 52

2
Terminology
This Section explains some of the terms used in the thesis. Abbreviations and mathe-
matical notations are also described. Further explanations are made in the sections
where the terms are used more frequently.

Acceleration
The rate of change of velocity (Light and Matter, 2008-04-23).
Application Pressure
Pressure at which point a wheel starts to brake.
Brake Calliper
Device in a disc brake system with which the pads are applied to the brake disc (Scania
CV AB:2, 2008-01-18).
Brake Chamber
Cylinder for the piston in a vacuum and compressed air brake system
(Scania CV AB:2, 2008-01-18).
Brake Disc
Disc in the disc brake system, rotating with the wheel (Scania CV AB:2, 2008-01-18).
Brake Pads
Pad made of friction material which is applied to the brake disc when braking. Brake
pads come in pairs - one pad on each side of the brake disc (Scania CV AB:2, 2008-01-
18).
Brake Lever
Arm that transmits the cylinder's motion to the brake camshaft (Scania CV AB:2, 2008-
01-18).
Clutch
Device which disconnects the transmission from the engine enabling the gears to be
changed (Scania CV AB:2, 2008-01-18).
Deceleration
In this thesis it is considered equal to negative acceleration.
Diaphragm
A circular device in rubber that is attached on the inside periphery of the brake chamber.
When the air pressure is large enough it expands and causes movement in the rest of the
Air Disc Brake system (Harju, 2008-01-28).
Drive Shaft
A driveshaft, driving shaft, or Cardan shaft is a mechanical device for transferring
power from the engine or motor to the point where useful work is applied (Sveriges
Trafikskolors Riksförbund, s 87-95).
Driveline
Synonymous with powertrain.
Electronic Brake System (EBS)
Brake system that is controlled using electronic actuators and sensors (Palmertz, 2003).
Electronic Control Unit (ECU)
Unit for the controlling of electrically controlled functions or accessories, such as axle
weight limiter, speed limiter, auxiliary heater, brake pressure etc
(Scania CV AB:2, 2008-01-18).
Electro-Pneumatic Modulator (EPM)
The EPMs are electro-pneumatic actuators. The EPMs receive the demanded brake
pressure via brake-CAN from the EBS-ECU and sends out the brake pressures
(Palmertz, 2003).

3
Engine
Machine which converts a non-mechanical energy into a useful movement energy
(Scania CV AB:2, 2008-01-18).
Exhaust brake
Valve in the exhaust pipe that reinforces engine brake by stopping up the exhaust flow
(Scania CV AB:2, 2008-01-18).
Final drive
The final drive transfers the driving force from the propeller shaft to the drive shafts
(Sveriges Trafikskolors Riksförbund, s 87-95).
Flywheel
Heavy cast iron machine element mounted on the crankshaft which levels power im-
pulses of the pistons (Scania CV AB:2, 2008-01-18).
Foot Brake Module (FBM)
FBM is connected to the brake pedal in the pedal assembly. The FBM sends two elec-
tric analogue signals and one switch signal to the ECU (Palmertz, 2003).
Gear Ratio/Conversion Ratio
Ratio between the rotational speeds of connected shafts (Scania CV AB:2, 2008-01-18).
Powertrain
Totality of components making up the power transmission system of a motor vehicle
(Scania CV AB:2, 2008-01-18).
Propeller Shaft
Shaft which transfers torque from the power source to the central gear
(Scania CV AB:2, 2008-01-18).
Push Rod
Rod that transmits axial movement from one part of a system to another when in com-
pression (Scania CV AB:2, 2008-01-18).
Retardation
Like deceleration, it is considered equal to negative acceleration.
Retarder
Additional braking device which supplements the ordinary service brake system to
avoid the fatigue failure of the latter when in continuous use
(Scania CV AB:2, 2008-01-18).
Transmission
Device which allows to change rotation speed of the engine down or up and thus to
choose between gear ratios (Scania CV AB:2, 2008-01-18) .

Abbreviations
ADB Air Disc Brake
CAN Controller Area Network
EBS Electronic Brake System
ECU Electronic Control Unit
EPM Electro Pneumatic Modulator
FBM Foot Brake Module
WECO Western Electric Company (rules)

4
Mathematical Notations
Symbol Description
Ad Area of the diaphragm in the Air Disc Brake.
Afront Area of the front of the truck.
a Generic term for acceleration.
aveh Acceleration of the vehicle.
α Angle of slope inclination.

c Spring damping constant

cairdrag Aerodynamic drag coefficient

E(X) The expected value of random variable X.

F Generic term for force.

Facc Term that groups the accelerating forces of the vehicle.
Fair Force generated by the air pressure and diaphragm in the Air Disc Brake
Fairdrag Force generated by aerodynamic drag.
Fbrake Braking force generated by the braking system.
Fd Spring damping force.
Fdec Term that groups the decelerating forces of the vehicle.
Fengbrake Braking force generated by the engine brake.
Ff Frictional force.
FN Normal force.
Fpiston Force generated by the brake piston in Air Disc Brake.
Fres Resulting force of Fair and Frod.
Frod Force holding the push rod in the Air Disc Brake in place.
Froll Roll resitance force generated by the contact between the tyres and the
ground.
Fs Generic term for spring force.
Fslope Force generated due to road inclination.
Fw Driving force generated by the powertrain.
fc Function that defines the torque coming from the clutch in the powertrain.
fd Function that defines the torque coming from the drive shaft in the power-
train.

5
Symbol Description
ffd Function that defines the torque coming from the final drive in the power-
train.
fp Function that defines the torque coming from the propeller shaft in the
powertrain.
fr Constant used in determining the roll resistance force.
ft Function that defines the torque coming from the transmission in the pow-
ertrain.

g Gravitational constant.
I Generic term for inertia.
Ieng Inertia of the engine.
Iw Inertia of a wheel.
it Transmission ratio.
ifd Final drive ratio.

k Spring constant.
kstd Standard deviation constant used in Chebyshev’s inequality.

L Generic term for angular momentum.

m Generic term for mass.

mveh Mass of the vehicle.

µf Frictional constant.

pair Air pressure

pest(t) Estimated pressure. Used in pressure estimation method.
preal(t) Actual pressure. Used in pressure estimation method.

r Generic term for radius/length.

r1 Length of brake lever.
r2 Length of brake lever from where brake piston is connected.
r(t) Residual of estimation and actual value.
rw Wheel radius.

6
Symbol Description
ρair Density of air.

σ Standard deviation.

T Generic term for torque.

Tbrake Brake torque.
Tc Torque coming from the clutch in the powertrain.
Td Torque coming from the drive shaft in the powertrain.
Teng Torque coming from the engine in the powertrain.
Tfd Torque coming from the final drive in the powertrain.
Tf:eng Frictional torque in the engine.
Tf:fd Frictional torque in the final drive.
Tf:t Frictional torque in the transmission.
Tp Torque coming from the propeller shaft in the powertrain.
Tt Torque coming from the transmission in the powertrain.
Tw Resulting torque on the wheels from the powertrain.

θ Generic term for an angle.

θc The angle of the clutch shaft generated by torsion.
θd The angle of the drive shaft generated by torsion.
θeng The angle of the shaft connected to the engine generated by torsion.
θfd The angle of the final drive shaft generated by torsion.
θp The angle of the propeller shaft generated by torsion.
θt The angle of the shaft connected to the transmission generated by torsion.
θw The angle of the shaft connected to the wheel generated by torsion.

θ& Generic term for angular velocity

θ&c Angular velocity related to the clutch.
θ&d Angular velocity related to the drive shaft.
θ& eng Angular velocity related to the engine.
θ& fd Angular velocity related to the final drive.
θ& p Angular velocity related to the propeller shaft.
θ&t Angular velocity related to the transmission.
θ&w Angular velocity related to the wheel.

7
Symbol Description
θ&& Generic term for angular acceleration.
θ&&eng Angular acceleration related to the engine.

Var(X) The variance of random variable X.

v Generic term for velocity.
vpr Velocity of the push rod’s movement in the Air Disc Brake.

X Indicated random variable.

X The mean of X.

x Generic term for distance travelled by object.

xpr Distance travelled by push rod.
x& pr Velocity of the push rod written as a derivative of distance.

y(t) Measured value in model based diagnosis.

yest(t) Estimated value in model based diagnosis.

8
 Part I

1. Introduction
Scania CV AB is one of the major heavy truck manufacturers in the world and sells ve-
hicles in over a hundred countries (Scania CV AB:1, 2008-04-23). The company builds
most of the parts in-house and assembles the vehicles themselves. One of the available
brake systems on Scania’s trucks and buses is today an electronic system where the
driver demands retardation instead of brake power. The actual brake power is generated
pneumatically.

The control of the system depends on many different factors and one thing that makes it
more difficult is that the exact pressure at which the brakes start producing brake force,
the application pressure, is not known. At the moment this pressure is assumed to be
constant for the brake components used in all vehicles, but tests have indicated that the
application pressure could vary between vehicles and in the brake components of a spe-
cific vehicle. The value of the application pressure is used in several different functions
that are used in the brake system and affect the brake performance of the vehicle. It
would therefore be of interest for Scania to see if it is possible to make a better estima-
tion of the application pressure than the constant that currently is used.

1.1 Objective
The objective of this thesis is to research if it is possible to determine the application
pressure for the Electronic Brake System that Scania develops and if a predictive esti-
mation of the application pressure can be implemented using the measurable signals that
are provided by the onboard computer.

1.2 Disposition
In order to reach the goal that has been set, several different tasks were performed in
different phases and these phases outline the disposition of the thesis.

Firstly the method explains in detail how the objective should be reached and by what
means. Secondly a theoretic base is presented and it explains more about how a brake
system works from different aspects and theory about how the task can be solved. Fi-
nally the empirical results of the thesis, which have been a product of the theory and
method, are presented and from this an analysis of the results are made. This leads to
conclusions and suggestions for developmental decisions that Scania can make in the
future.

1.3 Confidentiality
Scania is a company that conducts research which the company uses to improve its
products. In order to have a competitive edge compared with other manufacturers of
similar products it is essential to protect the ideas and inventions that the company de-
velops. Because this thesis is conducted as a part of the brake systems development at
Scania certain aspects have to remain confidential. This means that for example specific
values that have been determined during the development process will not be revealed.

9
Instead these were altered to constants or values that only show the relations of the
measurements.

10
2. Method
The process for developing an estimation of the application pressure goes through a few
different phases. The initial goal is to build up a model of the system that is going to be
analyzed so that an understanding of the problem can be reached. This was done by in-
terviewing the experts on the different parts of the system at Scania and by dismantling
and studying the brakes in the workshop where repairs and modifications of the brakes
take place. The procedure was fairly informal where the various experts on the different
areas could be reached at the workplace and questions could be answered when they
arose. More formal meeting sessions were also conducted where the brake system was
explained more in detail, but this was done as a presentation of the system where ques-
tions could be asked freely throughout the session. This type of interaction and constant
close proximity to people with expert competency within the area of brake systems lead
to the understanding of the problem and how to approach it.

The next phase was to consider how the objectives of the thesis could be reached. This
was deemed to be most appropriate to divide into two steps: first to develop a method
that can find a point of deviation for each wheel and second to use this to develop a way
to estimate the application pressure for each wheel. In order to find a suitable method
for solving the first step, three different methods were tried and evaluated before the one
that was considered the most suitable was implemented in the second step. The methods
that were examined were considered due to their ability of implementation in the brake
system and through discussions with experts at Scania and literature study.

Parallel with the development of the methods data was collected in order to test the
methods and see how well they work.
Data Collection/
Theoretic background

Evaluate three different ap-

plication pressure detection
methods.

Implement observer using

one detection method.

Conclusion and suggestions

for Scania

Figure 1.The figure shows the method that was used in the thesis. It displays the steps that were carried
out and what order they followed.

11
2.1 Collecting data
The data collection process was conducted on the Scania testing grounds, a test circuit
that simulates different types of roads, and on regular roads in the Södertälje and Stock-
holm area. The testing was performed on the truck Morbror, which can be seen in
Figure 2. It had three wheel axles and the second axle was driven. The tag axle, the third
axle, was not in contact with the ground when testing and therefore the truck only had
four wheels that could brake. The truck carried weights on it to simulate real transport-
ing scenarios. The brake system’s electronic control unit (ECU), which controls the
brake pressure, was connected to a laptop computer that made it possible to feed the
wheels with different signals all depending on the type of test that was being conducted.
Through the ECU the signals from different parts of the truck were collected and stored
in a file on the computer. This meant that the signals were subjected to some filtering
and were not raw, something that is discussed further in Section 6.3. The sampling fre-
quency of the signals was set at 100 Hz. This level of accuracy was set after discussions
with experienced testers at Scania and was considered to be appropriate for the study. A
larger sampling frequency would mean that the data files would be too large to handle
and tests would have to be shortened down.

When conducting the data collection some simplifications were made in order to reduce
the number of variables that could influence the retardation. This made the distinction of
the true application pressure easier to detect, but it was also important to be aware of
that the reliability of the results was reduced. This was taken into consideration and the
simplifications are presented in Section 2.1.1 to 2.1.6. The simplifications were not ap-
plied to all the data collection occasions and were minimized when possible.

2.1.1 Plane ground

The data collection was conducted on a regular asphalt road that did not have any hills
or bumps of any greater magnitude. This was due to the fact that a hill would affect the
retarding forces on the vehicle, explained further in Section 4.5.3, and bumps will also
generate irregularities in the data that will make the detection harder. This isolates the
retarding forces when braking occurs to only be dependent on aerodynamic drag and
rolling resistance.

2.1.2 Constant velocity

As aerodynamic drag increases by the square of velocity when the truck moves, ex-
plained in Section 4.5.1, the velocity was also held at a constant before the brake pres-
sure was applied. This reduced the disturbance from the aerodynamic drag on the retar-
dation of the vehicle and made isolation of the forces that influence the vehicle less
complicated.

2.1.3 No turning
When a vehicle turns the wheels on each side of it do not travel the same distance, be-
cause their radius when turning is different. The vehicle maintains the same speed but
because the distance travelled is not the same, the velocity of the wheels is not the same.
This lead to a disturbance in the retardation signal and in order to reduce this distur-
bance turning was minimized when conducting tests.

12
2.1.4 Free rolling
When a truck is driven forward a torque is generated in the powertrain, which gets
transferred to the driven wheels of the truck. When taking the foot off the gas pedal and
not applying any forward driving torque to the wheels the powertrain exerts a torque
that generates a retarding force on the driven wheels, the engine brake as it is commonly
known. The engine brake is explained further in Section 4.5.5. This affected the retarda-
tion and in certain data collection occurrences the retarding forces wanted to be isolated
to reduce modelling and data errors. In order to reduce those influences the clutch was
pushed down in order to disconnect the powertrain from the driven wheels and thus
make the truck free rolling.

An assumption was made that there would be a certain time between the moment that
the foot left the gas pedal until it pushed down the brake pedal. An empirical study
showed that the fastest time was 0.7 seconds. One can therefore assume that the truck
almost free rolled, if the forces from the powertrain were not included, before the brake
pedal was pushed down. The free rolling simplification was therefore reasonable, even
if it did not take into account the powertrain retardation.

2.1.5 Assumption that the ground is constant before braking

It was assumed that the ground, which the truck rolled on prior to braking had the same
properties as when the truck started braking. That meant that the inclination, surface
texture and material were the same during the 0.7 seconds before braking as they were
after this period when the application pressure point was found.

2.1.6 Retarder and exhaust brake not included

The brake system in a truck does not only entail the wheel brakes. It also includes the
retarder, the engine brake and the exhaust brake. The retarder is an extra braking device
that is used to reduce the wear upon the main brake system. The engine brake uses the
internal energy from the engine to slow down the vehicle. The exhaust brake comple-
ments the engine brake by reducing the exhaust and thereby braking the vehicle. In this
thesis the retarder and exhaust brake will not be included when conducting tests and
collecting data, but the engine brake is taken into account.

13
Figure 2. Morbror – the truck that was used for collecting the data used in the thesis.

2.2 Considerations for the Different Methods of Collecting Data

During the data collection process various different methods of collecting data were
tried in order to see how the data responded to these modifications and if it could be
used in the development process. In this section these different methods of data collec-
tion are presented.

2.2.1 Braking One Wheel

For purposes of detecting when the pressure was enough to start braking a vehicle the
method of braking just one wheel was the easiest. This was because only one of the
wheels demanded retardation and should therefore be the one that reacted first when the
pressure was enough. The negative aspect of braking just one wheel was that it did not
emulate an actual braking procedure when a person puts his or her foot on the brake
pedal. It was however a good starting point for seeing how the acceleration of the vehi-
cle responded to braking.

2.2.2 Braking All Wheels

When feeding all the wheels of the truck with the same pressure it created an environ-
ment that was closer to the true environment when braking. A wheel can start its retar-
dation before the brakes have applied due to the fact that other brakes have applied to
their wheels. This made it harder to detect when the pressure for a certain wheel was
large enough for the brakes to apply and it demanded a greater precision from the meth-
ods of detection.

14
2.2.3 Braking One Axle
Braking one axle was another method that was used in this project and was somewhere
in between the two previously explained ways of collecting data. It was more similar to
braking one wheel as it just added one more wheel to the ones that were braked. But it
has the advantage that it was possible to see how one wheel affected the other one’s
retardation if it braked first and made the tuning of the detection algorithm easier.

2.2.4 Altering the Application Pressure of the Wheels

A major concern when trying to detect when the brakes apply was the effect on the re-
tardation of a wheel when another wheel has begun to brake. To extract data that
showed this type of behaviour the properties of the brake chamber were altered so that
the wheels on the front axle would need a larger pressure to begin to brake. This was
done by shortening the length of the spring in the brake chamber by inserting discs that
compressed it. In Section 3.2 the components of the Air Disc Brake are further ex-
plained. The intention was that when braking, the wheels on the drive axle would be
caused to brake first and it would be possible to study how the wheels on the front axle
behave when this happens.

15
 Part II – Theory
In the second Part of this thesis the theoretical framework is presented. This
Part will explain deeper how the brake system works and the theories that
later are used in order to reach the objective.

3. What is EBS?
This section describes the Electronic Brake System (EBS) which is the type of braking
system that will be able to use the more accurate representation of the application pres-
sure. This will give the reader a better understanding of how the brakes work and there-
fore a better understanding of the rest of the thesis.

3.1 An Overview
The electronic brake system used by Scania consists of a few important parts that work
together in order to get the vehicle to brake. These are the Electronic Control Unit
(ECU), the Foot Brake Module (FBM), the Electro Pneumatic Modulators (EPM), air
tanks and Air Disc Brake (ADB).Figure 3 shows an overview of a truck with the differ-
ent parts marked out and the signals that are sent between them.

ECU EPM Air Disc

Air Disc Brake
Brake

FBM

Air Reservoirs

Air flow
Electrical signals

Figure 3. Schematic overview of the Electronic Brake System (EBS) in a truck. The red dotted lines show
how the electrical signals are transmitted between the components and the blue lines show how the air
flows from the reservoirs and is controlled from the EPMs.

The Scania EBS is a brake system that uses electrical impulses rather than hydraulics or
pneumatics, which is common in other automotive braking systems, to control how

16
much the vehicle should brake. The system is initially controlled from the FBM where
the driver sends a signal demanding a certain retardation of the vehicle by pushing
down the foot brake pedal. That signal is then processed by an ECU, which is a com-
puter that translates the signal from the FBM to a certain pressure that should be applied
to the brakes. This information about the pressures is sent to EPMs which apply the
needed pressures by controlling the correct amount of air to the brake chambers in the
pneumatic part of the system. The air is provided from air tanks located under the vehi-
cle. Air is then sent at a certain pressure to the brakes which apply the force needed to
brake the vehicle in the way the driver requested (Palmertz, 2003).

Unlike a hydraulic system in a car or a regular pneumatic system, which uses brake flu-
ids or air to control the brake mechanics all the way from the pedal to the wheels, the
force that has to be applied to the brake pedal is always the same in an EBS for the same
retardation and not dependent on the weight of the vehicle (Amberkar et al, 2000).

3.2 The Air Disc Brake in the Electronic Brake System

The latter part of the system, the Air Disc Brake (ADB), was not electronic but pneu-
matical and mechanical. Pneumatics is the use of pressurized air to effect mechanical
motion (Wikipedia, 2007-11-08). This part gives a more detailed view of the part of the
EBS where the pressure influences the braking, which was pivotal for this thesis.

The Air Disc Brake could be divided into three parts: Brake Chamber, Brake Calliper
and Friction Pair. Each of the three parts consisted of different components which are
presented in Figure 4.

Fbrake
Brake lever
Fres
pair

Fpiston

Diaphragm Push rod

Brake piston

Brake Chamber Brake Calliper Friction Pair

Figure 4. Schematic picture of the Air Disc Brake. The dashed lines mark the boundaries for the different
parts of the system; Brake Chamber, Brake Calliper and Friction Pair (Based on Harju, 2008-01-28).

3.2.1 Brake Chamber

The brake chamber contained three important parts: diaphragm, return-spring and push
rod. When the air pressure was fed from the air tanks and distributed from the EPM this

17
made the diaphragm expand and a force made the push rod move forward. The push rod
was fairly free moving but was controlled by a spring that was wrapped around it,
which generated a force that had to be overcome by the force from the diaphragm if the
push rod was going to be able to move (Harju & Löfstrand, 2007-10-10).

The dynamics that affected the brake chamber were primarily pressure and the move-
ment of the push rod, which got its movement controlled by the return-spring wrapped
around it. Pressure can be explained as the force applied on a unit area of a surface di-
rected perpendicular to that surface (Pytel & Kyusaalas, 1999a, p 401-405). The surface
was in this case the diaphragm in the brake chamber. The equation is written as

Fair
p air = ⇔ Fair = p air Ad , (1)
Ad

where pair was the pressure from the air tanks, Fair the force on the diaphragm and Ad the
area of the diaphragm. The force that could move the push rod was therefore Fair, but
this had to overcome the force that held the bar in place before braking. The force Fair
was not actually linear as it might appear because the area of the diaphragm changed
when the force created by the pressure was large enough to make the push rod move.
The area of the diaphragm could be written as a function of the length the push rod
moved (Harju & Löfstrand, 2007-10-10). In this schematic representation this was sim-
plified as it was not necessary with a more elaborate way of modelling.

A damped spring is a mechanical object that has different properties that also applied to
the push rod in the brake chamber. The push rod itself was a body that did not act like a
spring. However since it had a spring wrapped around it this caused its movement to act
in a spring-like manner (Harju, 2008-01-28). A spring that is not fully stretched or
compressed beyond its elastic limit obeys Hooke’s law. The law says that the force with
which the spring pushes back is linearly proportional to the distance from its
equilibrium length and is written as

Fs = -kx, (2)

where k is the spring constant which is individual for each spring, x is the displacement
vector with information about the distance and direction about the springs deformation
and Fs is the resulting force vector which states the direction and magnitude of the force
the spring exerts (Pytel & Kyusaalas, 1999a, p 488).

Damping is an effect that reduces the amplitude of oscillations of an oscillatory system.

This also effects the spring and the push rod that its wrapped around. It is
mathematically modelled as a force proportional with the velocity of the object but in
the opposite direction

Fd = -cv, (3)

where c is the viscous damping coefficient, v is the velocity of the object and Fd is the
resulting force vector. (Pytel & Kyusaalas, 1999b, p 545-547)

18
When combining Hooke’s law with the damping abilities of a spring, a representation of
the physical properties of the force holding the push rod in place, Frod could be made

Frod = Fs + Fd = − kx pr − cv pr = − kx pr − cx& pr . (4)

This force had to be overcome by the force generated by air pressure to generate a
resulting force, Fres, acting on the brake lever. The resulting force, Fres, was the force
that was taken to the next part of ADB-system and is written as follows

Fres = Fair – Frod. (5)

3.2.2 Brake Calliper

The brake calliper had two main parts that influenced the braking of the vehicle: the
brake lever and the brake piston. When the push rod moved it pushed with a force, Fres,
on the brake lever in the brake calliper. The brake lever had the same sort of characteris-
tics as a regular lever. The lever is an object that is used with an appropriate pivot point
to multiply the mechanical force that can be applied to another object, as can be seen in
Figure 5 (Pytel & Kyusaalas, 1999a, p 33-86).

F1

F2
r1

r2

Figure 5. Explanation of how a lever works.

The equation is

F1r1 = F2r2
Fresr1 = Fpiston r2, (6)

where F1 is the force applied at one end of the lever at distance r1 from the pivot point
and F2 is the force at the distance r2 from the pivot point. This implies that if r1 > r2 then

19
F2 > F1 and this phenomenon is used by the brake lever in the brake calliper (Pytel &
Kyusaalas, 1999a, p 33-86).

3.2.3 Friction Pair

The third part of the Air Disc Brake consisted of the two brake pads and the brake disc.
In Figure 6 the friction pair can be viewed together with the other components in the
electronic brake system. The brake piston pushed the brake pads, which came in contact
with the brake disc, and frictional force caused the wheel to slow down.

The brake pads moved towards the brake disc with a force equal to Fpiston and friction
between the pads and disc was generated. Friction is the force between two surfaces in
contact, which was what occurred in this case. The formula for frictional force is

Ff = µ fFN = µ fFpiston, (7)

where µ f is the frictional coefficient, which depends on the materials in contact, and FN
is the normal force, which in this case was Fpiston. Heat was also generated, but since this
was an ideal representation that gave a simplified view it was excluded (Pytel &
Kiusaalas, 1999b, p 303-306).

20
Figure 6. The figure shows some of the components in the air disc brake in the electronic brake system.
The brake disc (26) and the brake pads (13), in this figure only one brake pad, make up the friction pair.
The brake chamber(29)and the brake calliper (1) make up the other parts of the air disc brake.

3.2.4 Summary of the Air Disc Brake System

When all the above information was summarized the formula for the brake force looked
as follows

r1
Fbrake = F f = µ f ( p air Ad − kx pr − cx pr ) , r1 > r2 . (8)
r2

The summary was simplified and losses due to friction or heat in the pneumatic system
were excluded because they were not deemed necessary for the overall understanding of
how the system works.

21
4. Wheel and Vehicle Dynamics
When causing a vehicle to move several different forces and other physical phenomena
influence the direction and speed of the vehicle. The forces that were of interest in this
thesis were the ones that were longitudinal and that drove the vehicle forward, the ac-
celerating forces, and the forces that braked the vehicle, the decelerating forces. The
forces that affect the vehicle laterally were for the sake of simplification not included.

In order to drive the vehicle forward the accelerating forces, Facc, had to be larger than
the decelerating forces, Fdec. This gave a resulting force in the direction that the vehicle
travelled that could be estimated with Newton’s second law of motion:

ma = Facc – Fdec (9)

(Eriksson & Nielsen, 1998, p 109-111).

In Figure 8 these forces can be viewed. To fully understand how these forces are gener-
ated, some other physical phenomena have to be explained.

4.1 Moment of Inertia

Moment of inertia is analogous to mass but in rotational form. It is the inertia of a rotat-
ing body with respect to its rotation. The moment of inertia is defined as

I = mr2, (10)
where m is the mass of the object and r is the distance from the rotational axis to the
point of mass (Pytel & Kiusaalas, 1999b, p 347-351).

4.2 Angular Velocity and Angular Acceleration

When a body moves in a circular motion it has an angular velocity. If the radius of the
constant path it takes is r and the angle of the motion is θ then the velocity v of the body
is
v = rθ& (11)

and the acceleration a that is tangent with the circle is

a = rθ&& , (12)

where θ& is the angular velocity and θ&& is the angular acceleration
(Pytel & Kiusaalas, 1999b, p 85-89).

4.3 Torque
Torque describes how a force can rotate a body around an axis. It is defined by

22
 T = rF, (13)
where T is the torque, r is the length from where the force, F, is perpendicular to the
axis.

Torque is also the time derivative of angular momentum, L, which depends on the mo-
ment of inertia, I, and angular velocity and this gives the following relationship

dL Iθ&
T= = = Iθ&& (14)
dt dt

(Pytel & Kiusaalas, 1999a, p 247-249, 344-346).

4.4 Accelerating Forces

The accelerating force, Facc, is generated from the engine and through the driveline a
forward driving motion of the vehicle can be obtained. In order to understand how Facc
is transferred from the engine to the wheels a schematic picture of the powertrain is pre-
sented in Figure 7. The powertrain consists of an engine, clutch, transmission, propeller
shaft, final drive, drive shafts and wheels. The equations that explain how the driveline
works are derived by using the generalized Newton’s second law of motion:

F = ma. (15)

Some of the equations that relate to the forces on the wheel are acquired by using the
complete dynamics of the vehicle. This will mean that for instance the vehicle mass will
be included in the equations describing the wheels (Eriksson & Nielsen, 1998, p 109).

Figure 7. A rear-driven vehicular driveline (Based on Eriksson & Nielsen, 1998, p 110).

23
4.4.1 Engine
The outgoing torque from the engine is distinguished by the driving torque, Teng, result-
ing from combustion, the engine’s internal friction, Tf:eng, and the external load from the
clutch, Tc. By using Newton’s second law of motion the following equation is given

I engθ&&eng = Teng − T f :eng − Tc , (16)

where Ieng is the engine’s mass moment of inertia and θeng is the angle of the flywheel,
which is connected to the pistons in the engine (Eriksson & Nielsen, 1998, p 111).

4.4.2 Clutch
In vehicles which are equipped with manual transmission a clutch connects the flywheel
of the engine with the transmission’s input shaft. When the clutch is engaged, and the
internal friction is assumed to be zero, Tc = Tt is obtained where Tt is the torque of the
transmission. The transmitted torque from the engine to the clutch and transmission is

Tc = Tt = f c (θ eng − θ c ,θ&eng − θ&c ) , (17)

which means that it is a function, fc, of the angular difference θ eng − θ c and the angular
velocity difference θ& − θ& over the clutch (Eriksson & Nielsen, 1998, p 111).
eng c

4.4.3 Transmission
A transmission consists of a set of gears and these each have a conversion ratio or gear
ratio, it. The incoming torque is multiplied by this ratio, which is larger for the lower
gears. This means that the engine has to make more revolutions for every revolution that
the transmission makes and therefore more force is transferred for lower gears than for
higher gears, which instead can get more revolutions from the transmission than from
the engine. The relation between the input torque and the output torque of the transmis-
sion, Tp, is

T p = f t (Tt , T f :t ,θ c − θ t it ,θ&c − θ&t it , it ) , (18)

where ft defines what the torque is a function of and Tf:t is the internal friction of the
transmission. The angular difference, θ c − θ t it , is considered due to the possibility of
having torsional effects in the transmission (Eriksson & Nielsen, 1998, p 111).

4.4.4 Propeller shaft

The propeller shaft functions as the connector between the transmission and the final
drive. When no friction is assumed, Tp = Tfd, the function, fp, for the propeller shaft’s
torque is

T p = T fd = f p (θ t − θ p ,θ&t − θ& p ) , (19)

24
where θ t − θ p is the angular difference between the transmission and propeller shaft and
θ& − θ& is the difference in angular velocity (Eriksson & Nielsen, 1998, p 111).
t p

4.4.5 Final drive

The final drive connects the propeller shaft with the drive shafts and has a conversion
ratio, ifd, in the same sense as in the transmission, but it only has one “gear” and the
conversion is therefore constant. The output is as follows

Td = f fd (T fd , T f : fd ,θ p − θ fd i fd ,θ& p − θ& fd i fd , i fd ) , (20)

where ffd defines what the final drive torque is a function of, Td is the outgoing torque to
the drive shaft, Tfd is the torque from the final drive, Tf:fd the internal friction,
θ p − θ fd i fd the angular difference and θ& p − θ& fd i fd the difference in angular velocity
(Eriksson & Nielsen, 1998, p 111).

4.4.6 Drive shafts

With the drive shafts the wheels are connected to the final drive. In this case it is as-
sumed that the wheel speed is the same for the two wheels and the drive shafts can be
modelled as one shaft. If the vehicle turns and the speed is different for the two wheels,
both drive shafts have to be modelled. With no friction assumed, Tw = Td, the function
of the torque, fd, is

Tw = Td = f d (θ d − θ w ,θ&d − θ&w ) , (21)

where Tw is the resulting torque on the wheels and θ d − θ w is the angular difference and
θ& − θ& is the difference in angular velocity (Eriksson & Nielsen, 1998, p 111).
d w

4.4.7 Wheels
In Figure 8 the forces that act on the vehicle with mass m and speed v can be viewed.
Using Newton’s second law in the longitudinal direction gives

mveh a veh = Fw − Fairdrag − Froll − Fslope − Fengbrake − Fbrake , (22)

where Fw is the driving force from the powertrain, mveh the mass of the vehicle, aveh the
acceleration of the vehicle, Fairdrag the aerodynamic drag resistance, Froll the rolling resis-
tance, Fslope the gravitational force due to the inclination of the road. Fengbrake is the en-
gine brake which brakes the vehicle when rolling and Fbrake is the force from the brakes
if they are used.

And with

Tw
Fw = , (23)
rw

25
where rw is the wheel radius an expression for the forces on the wheels can be gathered
from equations (16) – (21) and by the following force expressions (Eriksson & Nielsen,
1998, p 111).

Fw

-Fairdrag -Froll - Fslope

-Fbrake - Fengbrake mg
a
Figure 8. The figure illustrates the forces that affect the truck (Based on Eriksson & Nielsen, 1998, p
112).

4.5 Decelerating Forces

4.5.1 Aerodynamic drag resistance

Aerodynamic forces have an important impact on the motion of the truck. They can be
divided into three components: drag, lift and side force. As they do not act on the centre
of gravity they cause pitching, rolling and yawing moments on the vehicle (Eriksson &
Nielsen s 197). As was explained in the beginning of the thesis a simplification was
made which means that only the drag force will be of interest in the further studies as it
works in the longitudinal direction. This force can be calculated as

ρ air
Fairdrag = c airdrag A front v2 , (24)
2

where Afront is the front area of the vehicle, v is the velocity of the vehicle, ρair is the
density of the air that the vehicle travels through and cairdrag is the drag coefficient
(Eriksson & Nielsen, 1998, p 197 - 198).

4.5.2 Rolling resistance

The rolling resistance of the vehicle is caused by the deformation of the tyres. It de-
pends on different factors such as the load on that tyre, tyre design, tyre pressure and
velocity (Eriksson & Nielsen, 1998, p 198). The factors that have been considered in
this case are the load on the tyre and velocity, the other ones have been considered con-
stant in the tests made.

Modelling of the rolling resistance can be done with the equation

Froll = f r FN = f r mveh g , (25)

26
where FN is the normal force of the vehicle on the tyres and fr is dependent on the above
mentioned factors and can usually be found in the range 0.01-0.03 for a passenger car
and less for a truck (Eriksson & Nielsen, 1998, p 198).

4.5.3 Slope inclination

The slope of the road has an effect on the motion of the vehicle. If the vehicle is driving
uphill this will cause a decelerating force and the opposite will occur if the vehicle is
going downhill due to gravitation. The weight of the vehicle and the slope α effect this
force, where α = 0 is a flat road and α = 90° is equivalent of a wall and the equation is

Fslope = mveh g sin(α ) (26)

(Eriksson & Nielsen, 1998, p 112).

4.5.4 Brakes
The force that the brakes have on the wheel is the generated in the way that was pre-
sented in Section 3.2 about the pneumatic part of the brake system and the equation is
the same as in equation (8).

4.5.5 Engine brake

The engine brake has a decelerating effect on the vehicle and occurs when the driver is
not pushing down the accelerator pedal but the engine is still producing energy in the
opposite direction compared with when the driver pushes the accelerator pedal. This
means that the engine generates torque on the wheels that works in the same way as was
presented in equations (16) – (21), but in the opposite direction and the torque of the
engine brake is equal to equation (21) and the force is the same as in equation (23). If
the accelerator pedal is pushed down and the engine is exerting “positive” torque on the
rest of the powertrain the force from the engine brake will be zero (Lu & Hedrick,
2005).

4.6 Summary of the forces acting on the vehicle

If the clutch, propeller shaft and drive shaft are assumed to be stiff, this will mean that
from the engine to the wheel only the conversion ratios in the transmission and the final
drive will be included in the calculation of the torque on the wheel. Stiffness means that
the angular differences that occur between the components in the powertrain are not
included in the expression. This will give a formula for the torque on the wheel due to
the engine which is

Tw = it i fd Teng (27)

and a force that is

Tw it i fd Teng
Fw = = (28)
rw rw

When summarizing all the accelerating and decelerating forces this gives

27
 it i fd Teng ρ air
mveh a veh = − c airdrag A front v 2 − f r mveh g − Fengbrake − Fbrake (29)
rw 2

ρ air
it i fd Teng c airdrag A front v2 Fengbrake
2 Fbrake
a veh = − − fr g − − (30)
rw mveh mveh mveh mveh

(Eriksson & Nielsen, 1998, p 112 - 114).

28
5. Statistical theory/change-point analysis
This chapter presents some of the statistical methods that have been used in order to
generate a solution for finding the application pressure. The theories presented deal with
change detection in a data sequence, which means finding the point when the data starts
to change, as shown in Figure 9. The reason for this being interesting is because when
studying the data sequence of the acceleration curve there is always a short time span
before braking that the vehicle rolls, the pre-braking phase which is defined in Section
7.3.1. During this time period the retardation of the vehicle will be fairly constant until
the brakes start to take effect and at that point the application pressure will be able to be
found.

Point of deviation

Figure 9. The red circle shows where the curve deviates and that indicates the point of deviation.

5.1 Statistical Terminology

This Section defines some of the terminology that will be used later in the thesis that is
important to fully comprehend in order to understand how the methods used work.

5.1.1 Mean
The arithmetic mean is calculated by dividing the sum of all numbers in list with the
number of items in the list (About.com: Mathematics, 2008-01-30). The formula for the
mean is
X
X = ∑ i , i = 1,..., n , (31)
n
where X is a vector of numbers, Xi is a number on the i:th position in the vector and n is
the number of items in the vector.

5.1.2 Standard Deviation

The standard deviation measures the spread of the values in a data series from the mean
of the values. The formula for the standard deviation is

29
 σ = E (( X − E ( X )) 2 ) = E ( X 2 ) − ( E ( X )) 2 = Var ( X ) ,
(32)
where X is a random variable, E(X) is the expected value of X and Var(X) is the vari-
ance of X (NIST/SEMATECH e-Handbook of Statistical Methods, 2008-01-13).

5.1.3 Moving Average Filter

The moving average filter is used for calculating the mean of the n latest data points.
When using the moving average filter the number of points, or window size, can be al-
tered depending on how the data is going to be used. It also works as a low-pass filter to
smooth noisy data. There are different types of moving average filters, but in this thesis
the simple moving average filter has been used. It looks just like when calculating the
mean as in equation 31, but it moves over the data series with a time window continu-
ously calculating a new mean (NIST/SEMATECH e-Handbook of Statistical Methods,
2008-01-13).

5.2 Statistical Process Control

Statistical process control (SPC) is usually used in industry as a method of controlling
the quality of the products that are manufactured. Quality is something that varies be-
tween products but when producing a specific product it is of the interest of the pro-
ducer to make sure that the quality stays at a certain level. If the quality of an item that
is produced varies from the rest of the production line this is of interest to detect and if
it is faulty it will be separated from the rest of the produced items. And this is where
SPC comes in. With SPC there is theoretic framework for detecting when data deviates
from the expected or wanted value and this is something that can be applied not only in
quality control but in other areas too as it is deviation that is the common ground (Bas-
seville & Nikiforov, p 23-44). This was the reason for using SPC in the thesis.

Other possible methods to use to detect change are for example CUSUM and Likeli-
hood-ratio test (Schechtman et al, 2007). Due to the limitations of the thesis these were
not further investigated.

5.2.1 Shewhart Control Chart

The Shewhart control chart was developed as the first form of statistical method to de-
termine quality in a process by detecting deviations. The control chart uses upper and
lower control limits to detect whether a deviation has occurred. These control limits
were set at three standard deviations below and above the mean of the data extracted
from the process. The reason for choosing three standard deviations was that according
to Chebyshev’s inequality that states that nearly all the values in any data sample are
close to the mean value according to

1
P (| X − X | ≥ k std σ ) ≤
2
(33)
k std
and the Vysochanskii-Petunin inequality that refined Chebyshev’s inequality for a
random variable with a finite variance in a unimodal distribution according to

4 8
P (| X − X | ≥ k std σ ) ≤ , k std > , (34)
9k std 3

30
where X is a random variable, kstd is the number of standard deviations and σ is the
standard deviation (Vysochanskij & Petunin, 1980).

These inequalities state that the probability of a random variable deviating from the
mean is less than 1/9 = 0.1 for Chebyshev’s inequality and less than 4/81 = 0.05 for the
Vysochanskij-Petunin inequality. Empirical studies also showed that less than one
percent of the values were situated more than three standard deviations from the mean,
which is due to the fact that for a normal distributed data sample about 0.003 of all
values lay three standard deviations from the mean. The basis for the control chart can
therefore be considered fairly conservative in comparison, but with the ability of a more
general application (NIST/SEMATECH e-Handbook of Statistical Methods, 2008-01-
13).

5.2.2 Western Electric Company Rules for Detecting Non-Random Behav-

iour
The Western Electric Company Rules (WECO) for detecting non-random behaviour
were developed as a quality control mechanism at the Western Electric Company and
have there basis in the Shewhart control chart. They are a set of rules that have been
used to detect when a point deviates from the mean. If one of the rules is true, this
means that a deviation can be said to occur. The rules are depicted in Figure 10 and are
stated like this:

1) Any point deviating with three standard deviations from the mean
2) Two out of the last three points deviating from the mean by two standard devia-
tions
3) Four out of the five last points deviating from the mean by one standard devia-
tion
4) Nine consecutive points on the same side of the mean

31
Figure 10. The graphs show the four Western Electric Company Rules and when a curve is considered to
deviate, marked with red. The zones represent the distance from the mean where zone C is within one
standard deviation of the mean, zone B is within two standard deviations and zone C is within three stan-
dard deviations. UCL is the Upper Control Limit and LCL is the Lower Control Limit (Argentieri, 2007-
12-14).

The rules are based on probability, in the same sense that they were for the Shewhart
control chart. For a normal distribution there is a probability of finding a point outside
of ± 3σ of 0.3%, which is quite rare and this is an indication of a shift in the process.
The similar case exists for the other rules that all have a probability of about 0.3% of
occurring (NIST/SEMATECH e-Handbook of Statistical Methods, 2008-01-13).

5.2.2.1 False alarms

In the regular Shewhart control charts there exists a small risk of detecting false alarms,
which are points that are not deviating due to non-random behaviour. For the Shewhart
charts this happens on average every 371 points. When implementing the WECO rules
this risk increases to a false alarm for every 91.75 points. This is something that the user
of the WECO rules has to be aware of and something that is commented later in this
thesis in the implementation chapter (NIST/SEMATECH e-Handbook of Statistical
Methods, 2008-01-13).

32
 Part III – Empirical studies
The theoretical groundwork has now been laid and in the third Part the em-
pirical studies will commence. This Part will go through the methods used
to reach the objective and how an estimation of the application pressure can
be made.

6. Data considerations
This chapter addresses the data that was collected during the development process. It is
central to understand what type of signals that have been used in order to grasp the re-
searched methods and how they have been developed. The chapter also discusses the
limitations that the data consists of and what kind of manipulations that were tried dur-
ing the development process.

6.1 Signals
In this section the signals that were used for data collection are presented.
Velocity
The velocity of every wheel was used, as it was important for determining when the
brakes apply to the wheels. The velocity also enabled a way to determine the accelera-
tion, which is discussed further in Section 6.2.

Brake Pressure - Measured

The air pressure that was fed to the brake chamber was measured as it was crucial to
know what the pressure actually was when the brakes had been determined to apply.

Brake Pressure - Requested

The requested brake pressure signal was the pressure signal that stated how much pres-
sure actually was requested, either by pushing down the brake pedal or by computer in
experimental environments. This was useful as a control to see if the measured pressure
reached the value it was supposed to. This data was not used in the modelling, just as a
reference point.

Clutch Pedal Position

The clutch pedal position indicated how far the clutch pedal had been pushed down at a
certain time. This was an important indicator to see if the truck was free rolling or not,
as the clutch released the powertrain from the gearbox which was connected to the
wheels. This was used in the method for determining the application pressure that is
presented in Section 7.3.1.

Brake Pedal Position

The brake pedal position worked in the same way as the clutch pedal position. The
brake pedal position was crucial in determining the pre-braking phase, which is dis-
cussed further in Section 7.3.1.

Accelerator Pedal Position

The accelerator pedal position signal gave the position of the accelerator pedal. The
signal was used to detect the pre-braking phase, defined in Section 7.3.1.

33
Time
Signal that showed the time in seconds for the data set. The signal was used in several
different contexts, for example in estimating the acceleration, which is described in Sec-
tion 6.2.

Loss of Torque
Loss of torque gave data on how much torque that had been lost due to for example fric-
tion and inertia in the driveline. The information given here was essential in determin-
ing the forces that affected the wheels from the driveline.

Use of Maximum Torque

The use of maximum torque indicated how much of the maximum torque was used by
the engine. This was used together with the signal that determines the loss of torque
from the engine when finding out the forces that were affecting the wheels.

6.2 Estimating Acceleration

A significant signal in detecting when the brakes apply to wheels was acceleration. The
signal was used in the different methods that were tried for determining the application
pressure, discussed in Chapter 7. However the acceleration of the individual wheels was
not one of the signals that could be measured and had to be calculated using the signals
that were present, the obvious one being velocity. As acceleration is equal to the deriva-
tive of velocity this was not too difficult but it required a method for making this esti-
mation of the derivative.

When calculating the derivative of a set of data there were a number of different nu-
merical and analytical methods that could be used. They offered different advantages
but in this case a numerical method was chosen to be used. An observer would also
have been possible to use but due to programming and implementation advantages with
a numerical method this was considered to be the most appropriate.

6.2.1 Numerical Estimation of the Acceleration

The easiest way to calculate the derivative of a function f(x) is to use

f ( x i + h) − f ( x i )
f ′( x) = , (35)
h

where h is close to zero in order to make the approximation accurate (Wolfram Math-
world:3, 2008-01-25).

A more symmetrical derivative can be calculated using

f ( x i + h) − f ( x i − h )
f ′( x) = , (36)
2h

where h is close to zero (Wolfram Mathworld:1, 2008-01-25).

There were other approaches that included more points, but in this context the symmet-
ric derivative was deemed the most suitable and of good enough accuracy. The step size

34
h of the derivative was important when determining the approximation of the accelera-
tion. As the data collected was discrete this meant that the step sizes had to have a value
of at least one, which gave the most accurate acceleration estimation. However this also
resulted in data that was noisy and difficult to study the behaviour of. Therefore a larger
step size was used and through empirical tests this value was set at h = 50. This gave a
smoother data sequence that was easier to interpret and use, but it also lost some of the
information that a smaller step size would contain. This was a compromise that had to
be considered when performing this type of operation.

It was also important to remember that the single that was being sampled had been pre-
filtered during the data collection using a frequency of 100 Hz. This was taken into con-
sideration during the following studies and was not considered as being a major prob-
lem.

Acceleration with h = 1
100

80

60

40

20
Acceleration

-20

-40

-60

-80

-100
0 200 400 600 800 1000 1200 1400
Time (s)

Figure 11. Plot of acceleration as a function of time with a step size of 1.

35
 Acceleration with h = 50
100

50

0
Acceleration

-50

-100

-150
0 200 400 600 800 1000 1200 1400
Time (s)

Figure 12. Plot of acceleration as a function of time with a step size of 50.

6.3 Signal noise

When conducting tests and gathering data that was going to be analyzed, signal noise
was something that affected the collected information. Signal noise disturbs the actual
signal and is not controllable or chosen. In many systems this signal can have a signifi-
cant influence on the behaviour of the actual signal that is going to be studied (Glad &
Ljung, 2004, p 221). The Electronic Brake System used both raw and filtered signals
and in this case the filtered signals were collected. The reason was that these filtered
signals were the ones used for the control of the vehicle by the EBS. Some noise was
however still present and this could depend on numerous factors; the unevenness of the
road, frictional noise in the driveline, fluctuations of the wind and instability in the
measuring equipment are just a few examples.

It was important to be aware that noise existed in the signals that were studied and it
could be dealt with in different ways. Filtering is a common method of reducing certain
frequencies of the signal that the noise is presumed to have.

6.4 Filtering data

By filtering a signal, the noise that disturbs it can be reduced and the “true” signal can
become more visible. This is done by only allowing certain frequencies to protrude.
There are different forms of filters that have different functions. A low-pass filter only
lets through signals with a frequency below a certain point, a high-pass filter is the op-
posite and band-pass filter lets the signals in a certain frequency band through (All
About Circuits, 2007-11-23). The signals that were used in this study had been low-pass

36
filtered before they were collected. This meant that further filtration of the signals was
not considered necessary because as much information as possible wanted to be used.
The exception was the acceleration signal that was extracted from the velocity signal,
described in Section 6.2. In that case a low-pass filter was applied by extending the step
size for the derivative calculation.

Further filtering tests were conducted where both a low-pass and band-stop Butterworth
filter were applied to the acceleration signal, but the results showed that the unfiltered
signal gave more reliable results. This is because if a filter was applied to a signal this
will mean that a phase shift will be added to the signal and also that information might
be lost.

6.5 Limitations in data

The filtering of signals meant that some information might be lost, as was explained in
the previous Section. Filtering was however not the only reason for the loss of informa-
tion. Many of the signals, for example the pedal position signals and the pressure sig-
nals, had data that was displayed by truncating with a floor function. To truncate with a
floor function means that the function holds the same value until a threshold is exceeded
and a new value is set (Wolfram Mathworld:3, 2008-01-30). A graphical description can
be viewed in Figure 13. This meant that the signals did not display their actual value,
but rather a value that was equal to or less than the actual value.

This was not crucial for the pedal position signals, but for the pressure signal this meant
that the exact value of the pressure was not possible to obtain and this had an impact
when determining the application pressure and making an observer.

Figure 13. Explanation of a floor function. The floor function of x, , remains at the same level until x
reaches a value that is large enough and can change (Wolfram Mathworld:3, 2008-01-30). Trans-
lated to the pressure signal this will mean that the displayed pressure is a floor function of the actual
pressure that will not show the exact value of the pressure.

6.5.1 Phase Shifts Between Signals

When dealing with the signals that had been filtered when they were collected it was
important to be aware of the phase shifts that could be present. Depending on how much
the signals had been filtered this would cause different phase shifts. Especially a phase
shift between the pressure signal and the acceleration signal would have caused prob-

37
lems because these had to be in the same phase in order for the deviation detection to be
accurate, which is discussed further in Section 7.3. This was studied and no phase shift
was detected between the two signals.

38
7. Methods of determining the application pressure
There are several different ways to go about developing a method for determining the
application pressure. Three different methods were tried, getting varying forms of accu-
racy. In this chapter the three methods will be presented and the results from these tests
are analyzed. One of the methods has been implemented and further developed. Other
ways of solving the task at hand are of course possible, but for this thesis the ones pre-
sented here were deemed most suitable.

The methods used can be put in two different categories; model-based and signal-based
methods. Common for them is that the acceleration signal for the wheels is studied.

7.1 Using acceleration to determine application pressure

In order to study when the brakes had applied, the signals presented in Section 6 had to
be used. There was no sensor that for instance detected contact between the brake pads
and the brake disc. Therefore other signals that acted as a result of the contact had to be
studied. The fact that the wheels’ velocities and accelerations decreased when the
brakes applied made these signals the most suitable to study. From these alternatives the
acceleration was chosen as the most appropriate to use, due to the fact that it was easier
to incorporate with physical expressions, such as Newton’s Second Law, and it gave a
more visible reaction when the brakes applied.

7.2 Model Based Method

The first approach was to build a model of the pneumatic system, breaking it down into
smaller and easier pieces that could be described by physical properties, in the same
way as was done in Section 3.2. The model was built around one wheel with its brakes.

To model the pneumatic system, the Matlab modelling and simulation extension Simu-
link was used. One consideration that had to be made was that the system contained
several non-linearities, which could be hard to model. The system would for example
act differently before the brakes had applied than after application was made. The com-
ponents would move easier in the pneumatic system as the resistance was much less
than when the brakes had applied.

7.2.1 Choosing a Model to Describe the System

The system could be described using two different models that were linear. One model
to describe the system before the brake pads are in contact with the discs and one model
to describe the system after the brake pads are in contact. This would avoid a non-linear
description of the system, which would have been more difficult to model. It would also
make it possible to represent in an accurate state space form and from there get an ob-
server. Figure 14 shows schematically how the brake force was non-linear since it
changed characteristics when the brake pads got in contact with the brake disc. When
the characteristics of the acceleration were studied from graphs where the braking oc-
curred, these linear attributes could be detected and therefore seemed reasonable to use
in the method. (Harju & Löfstrand, 2007-10-10)

39
 F brake

Fair
Figure 14. The graph shows how the force applied on the brakes, Fbrake,, changes when the brakes have
applied, which the red circle indicates. Since the acceleration of the wheel depends on the brake force
this means that the acceleration has the same nonlinear properties as the brake force (Based on Harju &
Löfstrand, 2007-10-10).

From these two options it was chosen to model the system prior to contact between the
brake pads and brake disc. State space representation of the acceleration when the
brakes are not applied can be derived from equation (8) by going via torque and mo-
ment of inertia

r1
Fbrake = µ f ( p air Ad − kx pr − cx& pr ) , r1 > r2 (37)
r2

Tbrake = rw Fbrake (38)

r1
rw µ f ( p air Ad − kx pr − cx& pr )
Tbrake r2
aw = = , r1 > r2 , (39)
Iw Iw

where Fbrake is the braking force, µ f the frictional constant, r1 the length of the brake
lever, r2 the length of the brake lever from where the brake piston was connected, pair
the air pressure, Ad the area of the diaphragm, k the spring constant, xpr the length of the
push rod, x’pr the velocity of the push rod, Tbrake the brake torque, rw the wheel radius,
aw the wheel acceleration, Iw the wheel’s moment of inertia.

40
7.2.2 Model Based Diagnosis
Model based diagnosis compares the model with the measured data in order to detect
when a deviation between the two occurs. This means that it has to be possible to meas-
ure the data that is going to be modelled, in this case the velocity, and of course to be
able to build an accurate model. In Figure 15 this is illustrated. By studying the residual,
r(t), between the model that estimates the output, yest(t), and the data, y(t), according to
equation (40) a detection can be made.
r (t ) = y (t ) − y est (t ) (40)
If the model is accurate and there is no noise the residual will be zero. If a change or
fault occurs the model will not be affected, but the measured output will change thus
leading to a residual that will deviate from zero. The change can be detected and not
isolated, but as change detection was what was needed to find the application pressure
this method was appropriate (Nyberg & Frisk, 2002, p 26).

u(t) y(t)
System

r(t)
Model
yest(t) -

Figure 15. Description of model based diagnosis and the study of the residual between the measured
output, y(t), and the estimated output, yest(t). The common input is u(t) (Nyberg & Frisk, 2002, p 26).

7.2.3 Results from Model Based Method

There were two purposes of building a model of the pneumatic system: The primary
aim was to learn and get a proper understanding of the brake system and how it worked,
both in a schematic form and through physically representing the pneumatic system.
The secondary aim was to get a state space representation of the system and from that to
be able to build an observer that would predict the application pressure. The latter being
the purpose of the entire thesis.

The primary aim was hard to present any definitive results from, but the theoretic part of
this thesis can be said to be the outcome of that purpose. During the process of under-
standing the system the physical properties of the system were derived and that has been
the basis of the theoretic section about the pneumatic system and how it influenced the
wheel when braking.

The secondary aim to use the physical representation to build an observer was less suc-
cessful than intended. The model that was built to represent the system did not have the
accuracy that was needed to be able to use any further and build an observer from and
therefore the modelling of the pneumatic system was considered not being useful for
this purpose. The reason for the model not being accurate enough could be that the

41
physical representation was too schematic and missed properties that existed in the sys-
tem, for example losses due to friction or materials acting in an unexpected way.

p_air

From
Workspace1

Air press ure

Velocity of Push R od

x' = Ax+Bu
F _air y = Cx+Du Mov ement of Pus h R od
Product State-Space1

Diaphragm area

R es ulting F orce
x _ts _prim
A_d

From x _ts
Workspace

F _res

Frictional Forc e
F _f

F _piston
F _f
Subsystem 3
Forc e f rom Brake Piston

F _res F _pis ton

T_brake
T_brake 1/I_w a_w
Acc eleration

Wheel Inertia T o Workspace

Subsystem2

Figure 16. The figure shows the Simulink model that was used in the model based method for extracting
the application pressure.

7.3 Signal Processing Methods

In the previous Section a method that based itself on modelling the physical properties
of the system was presented. In this Section a different approach of estimating the ap-
plication pressure will be made. Signal processing methods will be used. The signals
from the Electronic Control Unit (ECU) in the truck were studied and used in order to
find the application pressure. Two methods of using the signals are presented, but there
are of course other ways to deal with the problem too.

The two methods deal with the problem in two different ways but have in common that
they both use the acceleration signal, which was derived using the velocity signal from
the ECU. By studying this signal and focusing on the data sequence before the brake
pressure was above zero, this gave a level of all the decelerating forces in equation (22);
Fairdrag, Froll and Fengbrake, and this caused a negative acceleration that was at approxi-
mately the same level over the entire rolling sequence. When Fbrake has reached a level
that influences the deceleration of the vehicle this caused the curve to diverge from its
previous steady level, as seen in Figure 9. At this point the pressure data could be stud-
ied and that was the application pressure of the vehicle.

7.3.1 Method one: Modelling the Pre-Braking Acceleration

This method approached the acceleration data from a modelling perspective. The data
was modelled using the dynamics of the vehicle that were presented in Chapter 4. If the
decelerating forces were correctly modelled using signals from the ECU this would
have given a representation of the pre-braking resistance that was accurate to use for

42
detecting when the data started deviating from it. The pre-braking resistance was mod-
elled using the results from Chapter 4 and is defined later in this Section. The slightly
modified equation is

− F prebrake = − Fairdrag − Froll − Fengbrake − Fslope =

ρ air (41)
= −c airdrag A front v 2 − Froll − Fengbrake − Fslope = mveh a veh
2


ρ air
− c airdrag A front
a veh = 2 − Froll − Fengbrake − Fslope (42)
mveh mveh mveh mveh


ρ air
− c airdrag A front
a veh = 2 − f g − it i fd Teng − g sin(α ) , (43)
r
mveh mveh rw

where aveh was the pre-braking acceleration that worked as a reference value for the
acceleration when the wheel started to brake.

7.3.1.1 Defining the Pre-Braking Phase

The pre-braking phase was defined as the phase before the brake pressure for a wheel
was above a threshold pressure and when the vehicle did not accelerate. The threshold
was significantly smaller than the estimated application pressure and the acceleration
could be controlled by studying the position of the accelerator pedal. The pre-braking
phase can be viewed upon as a rolling phase and if the clutch pedal was pushed down
the vehicle was free rolling because the powertrain did not have any decelerating effects
on the wheels. This phase always occurred before braking because of the time it took to
move the foot from the accelerator pedal to the braking pedal. This time gap was
enough to find a reference point to compare the acceleration to when the brakes had
applied.

7.3.1.2 Determining the Parameter Values

In the representation of the pre-braking acceleration there were several parameters that
had to be determined in order to adjust the model to the data and the conditions that the
tests were conducted in.

Fairdrag, the aerodynamic drag resistance, contained a constant, cairdrag, that depended on
the aerodynamic properties of the object that travelled through the air. An object with
cairdrag = 1 brings all the air heading towards it to rest and spreads the air pressure
equally on the surface (Eriksson & Nielsen, 1998, p 197 – 198). When designing a vehi-
cle it is of interest to reduce the drag coefficient so as much air as possible is deflected
to the side of the vehicle and the forward driving force of the vehicle becomes as large
as possible. This reduces the consumption of energy. Through information from the
Scania database it was possible to obtain an approximation of the drag coefficient and
through tests the value was further adjusted. The same procedure accounted for the in-

43
formation about the front area of the vehicle. The weight of the truck, mveh, was deter-
mined by weighing the truck before collecting the data. The density of the air was cho-
sen to be 1.25 as it seemed reasonable judging by the temperature when the tests were
performed (The Engineering Toolbox, 2008-01-30).

For the rolling resistance, Froll, the mass of the vehicle, mveh, was eliminated when de-
termining the acceleration of the truck. This meant that only the rolling resistance coef-
ficient, fr, had to be chosen. For a car this value is usually in the range of 0.01 – 0.03,
but for a truck the value is less as the wheels have different properties that cause less
friction (Eriksson & Nielsen, 2001 p 198). From the Scania database and through testing
a suitable value was extracted and applied.

When determining the decelerating force from the driveline, Fengbrake, the torque on the
wheel had to be known. This could be done using two different methods. Firstly by us-
ing the conversion ratios for the transmission and final drive and multiplying this to the
force from the engine and dividing with the wheel radius, just as it is stated in the for-
mula. The second method was to use the internal signals that had information about the
torque acting on the wheel and use these directly. The first way worked in theory, while
the second one was more practical and faster to use in this context and that was why it
was chosen.

The force that affected the retardation due to the slope, Fslope, could be determined by
using an internal signal that calculated the slope of the road, α.

7.3.2 Results from Modelling the Pre-Braking Acceleration

The purpose of modelling the pre-braking acceleration was to get a reference value and
detect when the acceleration deviated from it. It was therefore crucial to get a correct
representation of the pre-braking acceleration before starting with the deviation detec-
tion. In Figure 17 the pre-braking model and the acceleration data before braking can be
viewed, as well as the air pressure when braking.

The approximation of the acceleration was fairly good at certain sections, for example
before and after the 1000 seconds mark. That kind of modelling would have made it
possible to use the method for deviation detection. These data sequences were taken
from a part of the collected data where the ground was plane and this meant that the
influence from Fslope was zero, as sin(α) = 0 if α = 0. If a different section was viewed,
for example around 925 or 1100 seconds, it could be seen that the model did not fit the
acceleration data in a very good way. The reason for this was that the data sequence was
collected when the truck was driving in a hill and the signal used approximated the hill
in a manner that lacked in precision. It was obvious that the model would be hard to
implement if it could not handle hills, as it was quite common for hills to be present
when driving on a road. In order for the modelling of the pre-braking acceleration to
work, the force that came from the slope of the road had to be determined in a better
way, otherwise it would have been of much less use to implement it.

44
 Left Front W heel
Air Pressure Shows the air pressure when braking
40

35

30

25
Pressure

20

15

10

0
850 900 950 1000 1050 1100 1150 1200
Time (s)

Acceleration Left Front W heel

Model of Acceleration Shows the model of the pre-braking acceleration vs actual data
0

-5

-10

-15
Acceleration

-20

-25

-30

-35

-40

-45
850 900 950 1000 1050 1100 1150 1200
Time (s)

Figure 17. The top graph shows the air pressure when braking and the bottom shows the model of the
rolling, pre-braking acceleration, dashed black line, compared with the acceleration data, solid green
line. The model only emulates the sequence prior to braking.

7.3.3 Method two: Detecting when the Acceleration Deviates from the
Mean of the Pre-Braking Acceleration
This was in many ways similar to the first signal processing method, but instead of us-
ing a model for the pre-braking acceleration, the mean of the acceleration before brak-
ing was calculated. This value was then used as reference value. When the acceleration
deviated from the mean the brakes could be considered to have applied, given that the
brake pressure was above the threshold value. This method was not dependent on any
physical modelling but only used the data from the Electronic Control Unit and mini-
mized the uncertainties that modelling could consist of. The major issues were deter-
mining the deviation from the mean and good way of calculating the mean.

7.3.3.1 Determining the Mean of the Pre-Braking Acceleration

To use this method, a way to calculate the mean in real time had to be chosen. If all the
data needed was already accessible this would not have been necessary, but in this case
it was necessary to be able to calculate the mean continuously. Therefore a moving av-
erage method was used. As described in Section 5.1, this meant that a suitable window
size was chosen for the mean to be counted in and this was done throughout the pre-
braking phase. The window size, if measured in time, should at least be smaller than the
minimum brake response time, which is approximately 0.7 seconds for trucks according
to empirical studies at Scania. To get the most accurate result to compare with, the win-
dow size should not be too large as there was a risk that unnecessary information would
have been included in the determination of the mean. If the window size was too small

45
the effect of using the mean would also have lost in significance as not enough values
would influence the mean that should be used as a reference value. To determine the
window size tests were conducted that indicated the appropriate size of the window.

7.3.4 Results from Determining the Mean of the Pre-Braking Acceleration

The problem with modeling the pre-braking acceleration was that handling hilly terrain
was difficult to get accurate. The second signal processing method removed that prob-
lem because of the mean calculation of the pre-braking acceleration. In Figure 18 the
reference value that was created by using the moving average algorithm of the pre-
braking acceleration can be seen. The sequences are the same as for the first signal
processing method and the accuracy in handling hills became greater when the reference
value was set by the mean. It was therefore chosen to use this method in the further de-
velopment of the application pressure estimation process.

Left Front W heel

Air Pressure Shows the air pressure when braking
30

25

20
Pressure

15

10

0
850 900 950 1000 1050 1100 1150 1200
Time (s)

Left Front W heel

Shows the mean of the pre-braking acceleration plotted during braking and the acceleration data
0

-5

-10

-15
Acceleration

-20

-25

-30

-35

-40 Acceleration
Mean Roll Acceleration
-45
850 900 950 1000 1050 1100 1150 1200
Time (s)

Figure 18. The top graph shows the air pressure when braking and the bottom graph shows the mean
pre-braking acceleration, dashed black line, and the acceleration data, solid green line. The mean pre-
braking acceleration is calculated during the pre-braking, but plotted during braking to make the com-
parison with the acceleration data easier.

7.4 Finding the Deviation from the Pre-Braking Acceleration

The WECO-rules were explained in Section 5.2.2 and they were a good starting point
for trying to detect a deviation from the mean of a data sequence. The rules are usually
used for detecting deviations in stationary processes and since that is not the case here it
was interesting to see if the rules could be applied in a different context.

46
The implementation of the rules was done in Matlab. The criteria for a deviation to oc-
cur were that one of the rules came into play and the pressure was above zero. In Figure
19 the rules were applied and the point of deviation could be observed. An ocular over-
view of the curve indicates that the WECO-rules found the points at which the data de-
viated from the mean in each braking sequence in a satisfactory fashion.

Air Pressure Left Front W heel

Point of Deviation Shows at what pressure the acceleration deviates from the mean
30

25

20
Pressure

15

10

0
850 900 950 1000 1050 1100 1150 1200
Time (s)

Acceleration
Mean Roll Acceleration Left Front W heel
Point of Deviation Shows where the acceleration deviates from the mean
0

-5

-10

-15
Acceleration

-20

-25

-30

-35

-40

-45
850 900 950 1000 1050 1100 1150 1200
Time (s)

Figure 19. The top graph shows at what pressure the acceleration deviates from the mean, marked with
black stars. The bottom graph shows the pre-braking mean acceleration, dashed black line, the accelera-
tion data, solid green line, and the points of deviation from the pre-braking mean, marked with a black
star. The deviation is calculated using the unmodified WECO-rules.

7.4.1 Modifying the Western Electric Company Rules

There was a risk for false alarms present when implementing the Western Electric
Company rules, which was mentioned in Section 5.2.2. Therefore the implementation
using the rules was modified through testing. Some of the rules were excluded and
thereby the risk for false alarms occurring was decreased and the accuracy of the result-
ing deviation points was greater. Even if the accuracy of the results would be at a level
of an error on every 91.75 points, or approximately one percent, this would be consid-
ered acceptable as the filter that is presented in Chapter 8 was constructed in a manner
that made it less sensitive to random errors.

The Western Electric rules were developed to see at what point a curve started to devi-
ate. In this context it was however necessary to investigate whether all the rules were
necessary to use, if some were redundant or should be modified. In Section 5.2.2 the
risk of detecting false alarms was discussed and that it increased if more rules were
used. As false alarms should be avoided, tests were done with the goal of seeing if some

47
of the rules could be modified or excluded in order to get a higher reliability for the
chosen points.

Experimental studies were conducted and these showed that if the second and fourth
rules were excluded, the points of deviation did not change and it was still possible to
obtain an accurate result. The reliability of the results was also greater but it was also
important to be aware of that since a few rules have been removed there might be a de-
lay in detecting the correct point of deviation.

By using the WECO-rules and modifying them to suit the purpose of this thesis a simple
and reliable method of determining the point of deviation and thereby the application
pressure was developed. It was a virtual sensor that detected the point of deviation and
was the starting point for developing an estimator of the pressure.

The previous methods that were discussed were crucial in understanding and testing
what path to take in determining the application pressure, but they will from now on not
be used when implementing an estimator of the application pressure.

Left Front W heel

Shows at what pressure the acceleration deviates from the mean
30
Air Pressure
Point of Deviation
25

20
Pressure

15

10

0
850 900 950 1000 1050 1100 1150 1200
Time (s)

Acceleration
Mean Roll Acceleration Left Front W heel
Point of Deviation Shows where the acceleration deviates from the mean
0

-5

-10

-15
Acceleration

-20

-25

-30

-35

-40

-45
850 900 950 1000 1050 1100 1150 1200
Time (s)

Figure 20. The top graph shows at what pressure the acceleration deviates from the mean, marked with
black stars. The bottom graph shows the pre-braking mean acceleration, dashed black line, the accelera-
tion data, solid green line, and the points of deviation from the pre-braking mean, marked with a black
star. The deviation is calculated using the modified WECO-rules.

48
8. Implementing an Estimation of the Application
Pressure
It was now possible to determine the application pressure by studying the acceleration
data. The next step in the development process could begin: creating an estimator. This
could be done in many different ways depending on what was going to be estimated and
how the data was represented prior to this step. In this case a signal processing approach
was chosen and the data was not modelled physically. This meant that for example con-
structing an observer would not be possible and other ways had to be considered.

8.1 Recursive Filter

The recursive filter used previous measured information to adjust the estimation of the
application pressure. It started with an initial guess of the pressure and updated the es-
timation continuously by multiplying the difference between the previous measurement
and the previous estimation by a scale factor. If the factor was large the estimation de-
pended a great deal on every measurement. A small scale factor would only make small
adjustments after every measurement and was not as sensitive for errors in the meas-
urement. The formula was

p est (t ) = p est (t − 1) + k d ( p real (t − 1) − p est (t − 1)) , (44)

where pest was the estimation of the pressure at time t, preal was the measured pressure at
time t and kd was the multiplying factor that adjusted the estimation.

In order to design a recursive filter that makes a good prediction the initial guess, pest(0),
and multiplying factor, kd, have to be chosen. Because there already existed a constant
estimation of the application pressure at Scania that value seemed like a reasonable ini-
tial guess. When determining the multiplying factor kd this was solved through experi-
mentation where sensitivity and speed of the estimation were weighted in order to make
the estimation work in an adequate fashion.

8.2 Considerations when Implementing Recursive Filter

A concern when determining the application pressure and implementing a recursive
filter was that of the dependence between the wheels. The velocity and acceleration of
one wheel affected the velocity and acceleration of all the other wheels. Since all the
wheels had separate Air Disc Brakes they could have different application pressures and
therefore the time at which they started braking might not be the same. When determin-
ing the application pressure by studying the acceleration signal, this became important.
If one wheel started to brake before all the other wheels, the other wheels would also
slow down somewhat. This made it difficult to detect when a wheel actually started
braking, because it might be influenced by another wheel that braked before it. The
problem has been tried to be solved but not in a completely satisfactory manner. A solu-
tion that was developed was to study the different brake sequences and sort out which
wheel brakes first each time. If this is done information about that wheel should at least
be accurate since no other wheels could have affected its acceleration.

49
In Section 6.5 the limitations of the pressure signal were discussed. The signal was trun-
cated and it did not contain the exact values of the pressure. This became a problem
when the steps in the floor function were about the same size as the approximate spread
of the application pressure that was used as a constant by Scania. It could also present a
difficulty when using the recursive filter as the values that were collected would behave
quite abruptly and the changes that adjust the pressure estimation will also be affected
by this abrupt motion.

The positive thing with checking the deviation from the mean pre-braking acceleration
was that it was easy to implement and that it gave an approximation of the application
pressure. The definite pressure was hard to retrieve as the signal did not have enough
precision, but it would generate a method that can be further developed in the future.

8.3 Results from the Recursive Filter Implementation

In Figure 21 the results of the observer implementation can be viewed. It is clearly visi-
ble that the estimation of the application pressure diverges from the measured applica-
tion pressure. The reason for this was the truncated measurement of the application
pressure. In order for the estimation to work properly and to be useful in the truck the
signals have to be more accurate otherwise the estimation will lack in accuracy.

Figure 21. The graph shows the estimated application pressure in relation to the measured application
pressure.

50
9. Conclusion
The purpose of this thesis was to see if it was possible to make a better and more precise
approximation of the application pressure for the disc brakes in Scania’s Electronic
Brake System and implementing this solution. The objective was also to present differ-
ent ways to address this problem and show which possible paths that are possible to take
in order to estimate the application pressure.

In the thesis three different ways of determining the application pressure were pre-
sented; one model-based and two based on signal processing. When studying and de-
veloping the model-based approach essential information for understanding the system
was generated and the nonlinear properties of the Air Disc Brake system were under-
stood. That presented difficulties in modelling the acceleration of the wheel and there-
fore further paths were studied before a definite method was chosen.

Instead of modelling the mechanics of the brake system, a study of the acceleration sig-
nal was initiated. This would exclude the non-linearities of the brake system, as the re-
sults of the inner and outer effects on the wheels are studied directly. The first approach
that was made tried to model the forces that affect the wheels acceleration in the se-
quence before braking and from this create a reference value that could be used to see if
the actual signal would deviate from. The model that was created proved to be accurate
in certain situations but lacked in precision when the braking occurred in hilly terrain
when the signals from the vehicle’s Electronic Control Unit were not accurate enough.
The method did however prove useful in providing a way to find the application pres-
sure point by using a reference value that the signal could be compared with. This was
something that could be used in the further development when finding the application
pressure.

The second signal processing approach used a different approach of finding the refer-
ence point for the acceleration to deviate from. By applying a moving average filter and
calculating the mean acceleration continuously during the pre-braking phase, this was
used to check when the acceleration signal would deviate and therefore when the brakes
would apply. This method excluded all the difficulties of modelling outer and inner dy-
namics and just focused on the actual signal. By doing that different modelling errors
could be ignored and a reliable way of determining the deviation was the only thing
needed. For this purpose the WECO-rules were implemented. They defined a set of
rules that determine when a curve deviates from its mean. They are primarily used for
stationary processes, but were chosen here to see if they could be applied in a different
situation as well. The implementation gave a reliable way of determining the application
pressure points for the different wheels and this is the starting point for developing a
method of estimating the application pressure.

The goal of the application pressure estimation was to present an estimated value of the
application pressure before every braking sequence and update this value continuously.
A recursive function fitted this description and was therefore implemented. The result-
ing estimation became difficult to use and lacked in precision due to the resolution of
the pressure signal and the fact that the acceleration of the wheels depend on each other.

51
In conclusion a reliable method of determining the application pressure that was based
on the acceleration signal was developed. The choice of using a numerical method for
estimating the acceleration should in hindsight have been revised. More time should
have been put into developing an observer that would have done the same calculations
but in a more reliable fashion. This is something that should be taken into notice for
further studies.

There were also issues regarding the resolution of the signals that were used in control-
ling the brakes and a reliable estimation was at this point not possible to implement. If
an estimation is going to be possible to use, a change in these signals has to be made
and then further studies would be useful for Scania. It is also necessary to develop a
solution that can separate the effects on the acceleration that the wheels have on each
other because otherwise the value of the application pressure found will not be of great
significance. If the other methods that were discussed in the thesis should be possible to
use an improvement of the signals that determine the slope of the road should be made.

9.1 Suggestions for Scania

Improve the resolution of the pressure signal and other signals.

Reduce or provide information about the phase shift between the signals.

A solution for separating the retardation on the different wheels is needed.

Improve the signals that have information about the slope of the road, for example using
GPS.

9.2 Suggestions for further research

Design an accurate model for the brake pressure that can be used for the estimation.

Include some of the simplifications that were used in this study, for example turning in
the model.

Use an observer when getting the acceleration signal.

Develop a method that gives the slope of the road, for example using GPS.

52
References
Printed
Amberkar, S., D’Ambrosio, J.G, Murray, B.T, Wysocki, J. & Czerny, B.J (2000), “A
System-Safety Process For By-Wire Automotive Systems”, SAE 2000 World Congress,
Detroit, Michigan, March 6-9, 2000, SAE 200-01-1056

Sveriges Trafikskolors Riksförbund (2002).”Att köra lastbil/buss”, Sveriges Trafiksko-

lors Riksförbund, Sollentuna

Basseville, M. & Nikiforov, I.V. (1993), “Detection of Abrupt Changes: Theory and
Application”, Prentice-Hall, Englewood Cliffs

Eriksson, L. & Nielsen, L., (2001). ”Course Material Vehicular Systems”, Linköping
Institute of Technology, Linköping

Glad, T. & Ljung, L., (2003). ”Reglerteori: Flervariabla och olinjära metoder”, Student-
litteratur, Lund

Glad, T. & Ljung, L., (2004). ”Modellbygge och simulering”, Studentlitteratur, Lund

Kiusaalas, J. & Pytel, A., (1999a). “Engineering Mechanics Statics”, 2nd Ed.,
Brooks/Cole, Pacific Grove

Kiusaalas, J. & Pytel, A., (1999b). “Engineering Mechanics Dynamics”, 2nd Ed.,
Brooks/Cole, Pacific Grove

Lu, X-Y & Hedrick, J. K., (2005). “Heavy-duty vehicle modelling and longitudinal con-
trol”, Vehicle System Dynamics, 43:9, p653 – 669

Nyberg, M. & Frisk, E., (2002). “Diagnosis and Supervision of Technical Processes”,
Course Material Linköpings universitet, Linköping

Palmertz, J., (2003). “System Description Brake System EBS 5”, Scania Technical
Product Data Document, Södertälje

Schechtman, E., Bandner, G. & Meginy, S. (2007). “Detecting a change in a scale pa-
rameter – a combination of SPC and change point procedures”, International Journal of
Production Research, 45:23, p5535 – 5545

Vysochanskij, D F & Petunin, Y I (1980). ”Justification of the 3σ rule for unimodal

distributions”, Theory of Probability and Mathematical Statistics vol. 21, p25-36

Electronic
About.com:Mathematics (2008). The Mean,
http://math.about.com/library/weekly/aa020502a.htm (2008-01-30)

All About Circuits (2007), What is a filter?,

http://www.allaboutcircuits.com/vol_2/chpt_8/1.html (2007-11-23)

53
Argentieri, R., (2006), Western Electric Rules,
http://en.wikipedia.org/wiki/Western_Electric_rules (2007-12-14)

Light and Matter (2008). Acceleration and Free Fall,

http://www.lightandmatter.com/html_books/1np/ch03/ch03.html#Section3.2
(2008-04-23)

NIST/SEMATECH e-Handbook of Statistical Methods (2006).

http://www.itl.nist.gov/div898/handbook/ (2008-01-13)

Scania CV AB:1 (2008). Om Scania, http://www.scania.se/About_Scania (2008-04-23)

Scania CV AB:2 (2008). Lexicon, http://scania-lexicon.scania.se/termweb/app/termweb

(2008-01-18)

The Engineering Toolbox (2008). Air Properties,

http://www.engineeringtoolbox.com/air-properties-d_156.html (2008-01-30)

Wikipedia (2007). Pneumatic, http://en.wikipedia.org/wiki/Pneumatic (2007-11-08)

Wolfram Mathworld:1 (2008). Derivative,

http://mathworld.wolfram.com/Derivative.html (2008-01-25)

Wolfram Mathworld:2 (2008). Floor Function,

http://mathworld.wolfram.com/FloorFunction.html (2008-01-30)

Wolfram Mathworld:3 (2008). Numerical Differentiation,

http://mathworld.wolfram.com/NumericalDifferentiation.html (2008-01-25)

Oral
Harju, R. & Löfstrand, M., Scania Brake Performance: Brake Callipers and Chambers,
2007-10-10

Harju, R., Scania Brake Performance: Air-Feeding System, 2008-01-28

54