POLITECNICO DI TORINO

Department of Control and Computer Engineering

Master’s Degree in Mechatronic Engineering

Master Degree Thesis

Modeling of a Roller Test Bench

for Car Homologation

Supervisors
Prof. Fabrizio Dabbene
Prof. Elisabetta Punta
Ing. Riccardo Cescato
Candidate
Simone Cavallo

April 2021
Sommario
Questa tesi è stata svolta in collaborazione con Control Sistem, azienda leader
nello sviluppo di soluzioni ed apparecchiature nel settore automobilistico. Pres-
so questa azienda è installato e viene utilizzato un banco prova a rulli per auto
per l’omologazione. Al fine di migliorare le prestazioni di questa strumentazione,
Control Sistem ha inteso l’utilità di disporre di un accurato modello matematico
dinamico e di sviluppare un simulatore dinamico per il banco prova a rulli. La
disponibilità di uno strumento di simulazione sia della dinamica del veicolo che
di quella del banco prova è infatti di fondamentale importanza per analizzare le
proprietà dinamiche più importanti del sistema auto/strada, ricavare informazioni
utili dalla simulazione stradale e ottimizzare il lavoro di ricerca e sviluppo. Per la
modellazione del sistema sono stati applicati i concetti fondamentali della meccani-
ca classica sia al veicolo che al banco prova. Il modello dinamico risultante è stato
sviluppato in Matlab/Simulink e validato con dati sperimentali misurati.

i
ii
Summary
This thesis was carried out in collaboration with Control Sistem, a leading com-
pany in the development of solutions and equipment in the automotive sector. Up
to now, a car roller test stand has been installed and used at this company for
homologation. In order to improve the performance of this instrumentation, Con-
trol Sistem intended the utility to have an accurate dynamic mathematical model
and to develop a dynamic simulator for the roller test bench. The availability of
a simulation tool for both vehicle and bench dynamics is in fact of fundamental
importance for analyzing the most important dynamic properties, obtaining useful
information from road simulation and optimizing research and development work.
For the modeling of the system, the fundamental concepts of classical mechanics
were applied to both the vehicle and the bench. The resulting dynamic model was
developed in Matlab/Simulink and validated with measured data.

iii
iv
Contents

List of Tables vii

List of Figures ix

1 Introduction 1
1.1 Composition of a roller test bench . . . . . . . . . . . . . . . . . . . 1
1.1.1 Hardware components . . . . . . . . . . . . . . . . . . . . . 1
1.1.2 Software components . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Fundamental quantities . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 The importance of simulating in a controlled environment . . . . . . 6

2 State of Art 7
2.1 Single-track model . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2 Double-track model . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.3 Longitudinal model . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.3.1 Aerodynamic friction force . . . . . . . . . . . . . . . . . . . 11
2.3.2 Longitudinal force at the wheels . . . . . . . . . . . . . . . . 12
2.3.3 Rolling resistance . . . . . . . . . . . . . . . . . . . . . . . . 13

3 Modeling of Vehicle and Test Bench 15

3.1 Model car on the road . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.1.1 Model car on the road: Newtonian approach . . . . . . . . . 17
3.1.2 Model car on the road: Lagrangian approach . . . . . . . . . 24
3.2 Car on bench . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2.1 FM and the road equation . . . . . . . . . . . . . . . . . . . 28
3.2.2 Fp , roller losses and calibration . . . . . . . . . . . . . . . . 29
3.2.3 Fx , the driving force . . . . . . . . . . . . . . . . . . . . . . 30
3.2.4 Data processing . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.2.5 Verification of RLS: The Coastdown . . . . . . . . . . . . . 34
3.2.6 Model car on bench: Newtonian approach . . . . . . . . . . 36
3.2.7 Model car on bench: Lagrangian approach . . . . . . . . . . 42
v
 Contents

4 Matlab-Simulink model 47
4.1 Coast-down simulation . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.1.1 Coast-down simulation: validation method . . . . . . . . . . 53
4.1.2 Coast-down simulation: results . . . . . . . . . . . . . . . . 55

5 Conclusions and Future Developments 61

A Nomenclature 63

Bibliography 65

vi
List of Tables
4.1 Theoretical coast-down test table . . . . . . . . . . . . . . . . . . . 55
4.2 Comparison table simulated coast-down test with car on road model 57
4.3 Comparison table of simulated coast-down test with car on bench
model partially real . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.4 Comparison table of simulated coast-down test with principal model
of car on bench . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

vii
viii
List of Figures
1.1 Bench equipped with master and slave rollers . . . . . . . . . . . . 2
1.2 Front roller axis diagram . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Translating module . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Communication network between layers . . . . . . . . . . . . . . . . 4
1.5 Load cell . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.6 Encoder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1 Single track model of a car on the road . . . . . . . . . . . . . . . . 8
2.2 Double track model of a car on the road . . . . . . . . . . . . . . . 9
2.3 Longitudinal model of a car on the road . . . . . . . . . . . . . . . 10
2.4 Longitudinal model of a car on a sloping road . . . . . . . . . . . . 10
2.5 Graph µL -sL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.6 Contact patch zoom and normal force distribution to the wheel . . 14
3.1 2D car in longitudinal movement . . . . . . . . . . . . . . . . . . . 16
3.2 Diagram of the forces of a car on the road in Newton’s method . . . 17
3.3 Force diagram for driving wheels . . . . . . . . . . . . . . . . . . . 18
3.4 Force diagram for driven wheels . . . . . . . . . . . . . . . . . . . . 19
3.5 Diagram of the forces of a car on the road in Lagrange’s method . . 24
3.6 Car on bench diagram on front axle bench . . . . . . . . . . . . . . 28
3.7 Force diagram of the traction axis . . . . . . . . . . . . . . . . . . . 31
3.8 Double load cell reading scheme . . . . . . . . . . . . . . . . . . . . 32
3.9 Single load cell reading scheme . . . . . . . . . . . . . . . . . . . . . 33
3.10 Speed-time graph of a coastdown test . . . . . . . . . . . . . . . . . 35
3.11 Diagram of the forces of a car on a bench in Newton’s method . . . 36
3.12 Force diagram for the front axle of the bench . . . . . . . . . . . . . 38
3.13 Force diagram by dynamic equation . . . . . . . . . . . . . . . . . . 40
3.14 Diagram of the forces of a car on a bench in Lagrange’s method . . 42
4.1 Matlab-Simulink model origin summary . . . . . . . . . . . . . . . . 47
4.2 RLS subsystem of the main model . . . . . . . . . . . . . . . . . . . 48
4.3 Load cell subsystem . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.4 Carontheroad dynamic equation subsystem . . . . . . . . . . . . . . 49
4.5 Angular velocity SET calculation . . . . . . . . . . . . . . . . . . . 50
4.6 Main model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
ix
 List of Figures

4.7 Partially real model . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

4.8 RLS subsystem of the partially real model . . . . . . . . . . . . . . 52
4.9 Model car on the road . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.10 Theoretical braking curve of the bench . . . . . . . . . . . . . . . . 54
4.11 Comparison of theoretical and simulated braking curve of car on road
model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.12 Comparison of theoretical and simulated braking curve of car on
bench model partially real . . . . . . . . . . . . . . . . . . . . . . . 58
4.13 Comparison of theoretical and simulated braking curve of pricipal
model of car on bench . . . . . . . . . . . . . . . . . . . . . . . . . 59

x
Chapter 1

Introduction
The key to this thesis work is to create a mathematical model of the roller test
bench for homologation of vheicle, which would allow to anticipate the results of
the system, in order to modify its design according to what is desired, and save
time in setting the fundamental parameters of the Bench. For this purpose, it will
also be necessary to use Matlab/Simulink to create the model.

A roller test room was born from the necessity of the vehicle manufacturers to
test a car already complete and assembled. Initially the vehicles were driven on
the road and tested directly on the field. Bringing the testing phase inside a hall
allows to perform tests in repeatable working conditions not affected by external
environmental factors (otherwise impossible to achieve in the field), and still per-
fectly simulate the conditions that the vehicle would encounter if the vehicle were
to drive on an infinitely flat road, without wind and with constant temperature and
constant humidity.

1.1 Composition of a roller test bench

This chapter looks at the bench’s components, the software and hardware part, and
explains the usefulness of these test benches.

1.1.1 Hardware components

A 4WD test bench for testing four-wheel driving cars must obviously have a roller
system for each axle of the vehicle. Such a system is usually composed of a central
engine twin-shaft, connected to two external drums, on top of which the wheels of
the car are placed.
In some systems, there is not a single drum per wheel, but two rollers are used.
The main one is called master roller and it is the one directly driven by the engine,
1
 Introduction

the second is referred to as the slave roller and is moved accordingly. accordingly.
In this way, the wheel of the car will rest on both rollers. Having provided this
information on the existence of benches equipped with a single roller or double
roller it is necessary to underline the fact that in the following thesis project we
have worked on a 2WD test bench with a single roller. A 2WD test room for the
testing of two-wheel drive cars has a roller system only for the driving axle of the
vehicle.

Figure 1.1: Bench equipped with master and slave rollers

The whole system rests on a ball bearing support to be totally released from the
ground. In doing so, however, the engine would also be free to rotate together
with the drums. To overcome this problem, the engine is anchored to the ground
by means of a mechanical arm connected to a load cell. This brings a double
advantage because it allows the reading of the torque that the motor generates.
This type of motor is called a tilting casing motor.

Figure 1.2: Front roller axis diagram

All of the above applies to both, front and rear axes. It is also good to know that the
latter is shifted to adapt the distance from the first according to the wheelbase of
2
 1.1 – Composition of a roller test bench

the vehicle being tested. Therefore there is a mechanism that allows the movement
of only the rear axle, in order to move it closer or further away from the front one.

Figure 1.3: Translating module

1.1.2 Software components

The management architecture of a rollers testing room is composed of a control
software called CSRolls which communicates with an application called DBSrolls.
The latter manages and interfaces directly with a Wago PLC connected to the roller
system and to the various sensors that make it up. The first is a room management
software, which provides the user with the possibility of controlling the room to
perform the various test cycles. Via Ethernet cable than it will drive the DBSrolls
by sending work modes and commands. There is also a second Wago PLC directly
connected to the CSRolls for the management of auxiliary signals for the control of
the room. The blue arrows indicate an ethernet connection, the red arrow indicates
a physical connection with the system components.

3
 Introduction

Figure 1.4: Communication network between layers

1.2 Fundamental quantities

The main measurement sensors of an automotive roller test bench are two:
• load cell;
• encoder.
The load cell is a force sensor connected at one end to the electric motor and at
the other end to the ground. It is necessary to specify that this sensor is distanced
from the electric motor by means of a mechanical arm which avoids that the motor
rotates together with the shaft since it has a tilting casing that would rotate around
the axis coinciding with that of the motor shaft. Therefore, the load cell measures
the force F in Newton (N) necessary to prevent rotation and, since the arm is known
and constant, the torque is univocally determined:

Ce = F̃ b
This instrument can be said to be the most important measurement sensor present
on the roller bench because it is the only one that measures all the force exchanged
between the vehicle and roller, which, suitably processed, gives indications on the
driving force supplied by the vehicle.
4
 1.2 – Fundamental quantities

Figure 1.5: Load cell

The encoder is an angular velocity sensor, it detects the rotation speed of the rollers.
It consists of a disc generally made of plastic material with windows of a known
angle, a photodiode and a photoresistor. The photodiode sends the input signal
through a light signal that will cross the windows of the disk. The photoresistor
receives the light signal and will in turn send a logical output signal (1 if it receives
light, 0 if it does not receive light). The type of encoder present on Control Sis-
tem s.r.l.’s roller benches are relative (or incremental) encoders, which signal only
the detectable increments (variations) with respect to another position taken as
reference. These increments are independent from the direction of rotation which
cannot be detected by this type of transducer.

Figure 1.6: Encoder

Absolute encoders are not used because what is important to know are the impulses
per revolution of the roller, while the absolute position of the roller is not useful
data to detect.
5
 Introduction

1.3 The importance of simulating in a controlled

environment
As previously mentioned, the most important reason for choosing to carry out
tests in a controlled environment is the possibility of performing tests in perfectly
repeatable working conditions. For example, if you want to test the car at various
driving speeds and at a precise wind speed it will be almost impossible to have the
repeatability of working conditions because it is not said that the wind speed will
remain constant.

Possible tests conducted in controlled environment are tests of:

• Aerodynamics: car profile optimization activities;

• Reliability: tests at very low temperatures;
• Durability: synthesis tests on components;
• Performance: speed, traction force, braking force;
• Robustness of accessory equipment: air conditioning tests in roller rooms at
high temperature and artificial sunlight;
• Refinement of calibrations and strategies for vehicle emissions;
• Vehicles homologation for a specific regulation.
The most important test conducted in an automotive roller test room is the RLS,
also called road simulation. This cycle simulates the road course that is imple-
mented on the car during its normal operation. In order to replicate the conditions
that the vehicle would have on the road the rollers must rotate according to a very
precise dynamic, they must be able to apply to the car the same resistant forces
that it would encounter if it were really on the road.

6
Chapter 2

State of Art
The possibility to have available a mathematical model of the physical system
of interest is very useful because it allows to predict and formally describe the
behavior of that system. In an industrial environment this means not only relying
on experience and common sense, but also to use an objective description able
to provide predictions in terms of numerical values. In literature there are many
articles or scientific books where physical systems are studied in order to create a
mathematical model that represents them.
Most of these papers or books follow a certain ladder to arrive at the model. The
various steps followed to create a mathematical model in the paper analyzed [16]
are:

1. definition of a first simplified scheme of the system where by simplified it means

formed only by the main components of major importance;

2. identification of the physical phenomena involved and creation of a free body

diagram of the system;

3. extrapolation of the equation(s) of the system by means of equilibrium equa-

tions of forces and moments (Newton-Euler method) or Lagrange methods;

4. identification of the quantities of Input and Output necessary to the simula-

tion system;

5. model validation.

The two methods mentioned in point 3. are two alterative methods to each other
because they lead to the same dynamic equation and this is a procedure that
7
 State of Art

has been carried out in [1]. The Newtonian method is faster to use with less
complex systems while the Lagrangian method is faster than the Newtonian method
with more complex systems. In the literature some publications choose to use one
method rather than the other for the reason just mentioned, however there are also
cases where both were performed to be sure that they were performed properly.
Searching in the technical-scientific panorama mathematical-physical models that
represent the entire operation of an automotive chassis dynamometer were not
found results but were found ideas from which to start to get the desired model.
Undoubtedly the starting point is the mathematical physical model of a car of
which in the various sources have been found more alternatives, as [14] proposes.
The most recurrent are three:

• double-track model;

• single-track model;

• longitudinal model.

2.1 Single-track model

Single-track model, this model (Riekert and Schunk 1940) allows the approximate,
but physically plausible, description of the lateral dynamics of a vehicle. This
model has the advantage of being simple, which allows for rapid integration. This
simplicity, however, does not allow to consider phenomena which, according to the
situation, can be not negligible, for example the phenomenon of pitching or rolling.

Figure 2.1: Single track model of a car on the road

8
 2.2 – Double-track model

2.2 Double-track model

Double-track model, this model assumes that the vehicle movements are planar
movements and movements such as pitch, roll and vertical movements are neglected.
At this model is arrived by adding longitudinal dynamics to the single-track model,
so it will be able to perform both longitudinal and yaw movements. For this reason
it is often used for estimating longitudinal and lateral states.

Figure 2.2: Double track model of a car on the road

2.3 Longitudinal model

Longitudinal model, the longitudinal model of the vehicle shown in the figure is
applied to describe the longitudinal dynamics of the vehicle during braking and
driving maneuvers. When only the longitudinal movement of the vehicle are con-
sidered at the slope angle of the road, the lateral movements of the vehicle and any
other types are neglected. The left and right wheels of a vehicle can be combined
into a wheel by ignoring the difference in movement between the left and right
wheels.

9
 State of Art

Figure 2.3: Longitudinal model of a car on the road

Since the aim of this thesis is to generate a model that simulates the behaviour of
the chassis dynamometer, it was necessary to look at the industrial and university
engineering scene in order to study the longitudinal dynamics of a car on the road.
Research had to be carried out into the longitudinal dynamics of a vehicle on the
road, since this knowledge would enable the test bench to test the car as if it
were actually on the road. From the various papers analysed, the starting point
is the longitudinal model, and in particular considering the longitudinal model of
the research in [9] on a vehicle moving on a sloping road, the external longitudinal
forces acting on the vehicle are the aerodynamic friction force, the weight force, the
longitudinal forces at the wheels and the rolling resistance forces of the tyres.

Figure 2.4: Longitudinal model of a car on a sloping road

10
 2.3 – Longitudinal model

The balance of forces along the x-axis obtained from this model is:

mvehicle ẍ = T1 + T2 − Faero − R1 − R2 − mvehicle g sin θ

where:

• T1 is the longitudinal force at the front wheel;

• T2 is the longitudinal force at the rear wheel;

• R1 is the rolling resistance force at the front wheel;

• R2 is the rolling resistance force at the rear wheel;

• Faero is the aerodynamic friction force;

• mvehicle is the mass of the vehicle tested;

• g is the gravitational acceleration;

• θ is the slope of the road.

The following subtitles explain how these forces are usually calculated.

2.3.1 Aerodynamic friction force

The aerodynamic friction force is calculated as follows:

1
Faero = ρCd Af (v + vwind )2
2
where ρ is the air density, Cd is the coefficient of aerodynamic friction, Af is the
frontal area of the vehicle subjected to aerodynamic resistance, v = ẋ is the longi-
tudinal velocity of the vehicle and finally vwind is the wind velocity.
As atmospheric conditions affect air density, they also affect drag. As a conse-
quence of this relationship, the conditions to which all aerodynamic tests refer are
a temperature of 15Ľ and an atmospheric pressure of 101.32 kPa. With this set,
kg
the resulting air density ρ is 1.225 m 3 . The frontal area Af is usually calculated

according to the studies carried out in [5], who calculates the area affected by drag
as the area between 79 − 84% of the area calculated with the vehicle width and
height. However, this calculation is only accepted for vehicles with masses between
800 and 2000kg.
11
 State of Art

2.3.2 Longitudinal force at the wheels

The longitudinal wheel forces T1 and T2 are friction forces due to the wheel-road
contact. In the literature in [7] from the study of Pacejka was defined a tyre
model, which is now a standard in the field of vehicle dynamics simulation, is used
to calculate the longitudinal forces at the wheel. The proposed Pacejka model
is an empirical-mathematical model with the aim of reproducing the character-
istic behaviour of the real component, based on mathematical formulae created
ad hoc following experimental characterizations, independent of the physical real-
ity that determining the behaviour acquired through measurements. The Pacejka
tyre model, more commonly called, "Pacejka’s Magic Formula", is therefore an
empirical-mathematical model that attempts to summarise the experimental tyre
performance through mathematical formulas. These have a precise structure in
which quantified coefficients appear based on specific experimental tests.
From the experimental results obtained by Pacejka, the parameters on which the
longitudinal force at the wheels depends were identified:

• sL , longitudinal slip;
• FZ , the weight force acting on the wheel;
• µL , the coefficient of longitudinal friction due to wheel-road contact.

Longitudinal slip is defined as the difference between the longitudinal speed in the
direction of the wheel axis v and the rotational speed wr, the whole divided by v.
v − w1 r
sL =
v
Once the longitudinal slip has been calculated and the coefficients related to the
asphalt conditions (c1 , c2 , c3 , c4 and c5 ) appropriately chosen, it is possible to apply
the "Pacejka’s Magic Formula" to calculate the longitudinal friction coefficient due
to the wheel-road contact.
The ’Magic Formula’ is given below:

µ(sL ) = (c1 (1 − ec2 sL ) − c3 sL )e−c4 sL v (1 − c5 FZ2 )

The coefficient values for the asphalt conditions are values calculated by Pacejka
derived from experimental characterisations following empirical tests.
Below are the ranges within which the empirical coefficients are chosen depending
on the asphalt conditions:

• 0.05 ≤ c1 ≤ 1.37 from ice to cobblestone dry;

• 6.46 ≤ c2 ≤ 306 from cobblestone dry to ice;
• 0 ≤ c3 ≤ 0.67 from ice to cobblestone dry;
12
 2.3 – Longitudinal model

• 0.002 ms ≤ c4 ≤ 0.004 ms

• c5 ∼
= 0.00015kN −2

The figure below shows the graph µL -sL where it is possible to appreciate the
behaviour of the longitudinal friction coefficient as the longitudinal slip varies.

Figure 2.5: Graph µL -sL

From the graph it is possible to see that the maximum values of the longitudinal
coefficient of friction are in the range of 0.1-0.3 of longitudinal slip, which means
that a minimum amount of longitudinal slip is necessary for good grip.
Finally, after applying the "Magic Formula" to calculate µL , from the definition of
friction force and knowing the load normal to the wheel, it is possible to calculate
the longitudinal force on the wheel:

T1,2 = µL FZ

2.3.3 Rolling resistance

In the end the last force applied to complete the model is the rolling resistance
force. The rolling resistance acts while the tire rotates and it is a dissipative effect.
Both the tire and the road are subject to deformation in the contact patch. The
tire is elastic and new material from the tire continuously enter the contact patch
as the tire rotates. Due to the normal load, this material is deflected vertically as
it goes through the contact patch and then springs back to its original shape after
it leaves the contact patch. Due to the internal damping of the tire material, the
energy spent in deforming the tire material is not completely recovered when the
13
 State of Art

material returns to its original shape. These losses of energy can be represented
by a force on the tires called rolling resistance that acts to oppose the motion of
the vehicle. The loss of energy in tire deformation also results in a non-symmetric
distribution of the normal tire load over the contact patch. When the tires are still,
then the distribution of the normal load FZ in the contact patch is symmetric with
respect to the center of the contact patch.
Nevertheless, when the tires are rotating, the normal load distribution is non-
symmetric, as shown in the figure below:

Figure 2.6: Contact patch zoom and normal force distribution to the wheel

The normal component is then displaced by a quantity u in the direction of mo-

tion and by balancing the rotation around the wheel centre we obtain the rolling
resistance force described by the formula:
u
3 4
R1,2 = FZ = fa FZ
r
u
where fa = r
is the rolling resistance coefficient.

This depth of research into the longitudinal model has been of great help in pro-
viding the basis for modelling a generic road car, which will be used to calculate
the SET speed to be supplied to the rollers so that they can simulate the road
behaviour of the vehicle under test.

14
Chapter 3

Modeling of Vehicle and

Test Bench
In order to obtain a model that simulates the operation of a rollers test bench, it is
necessary to first create a mathematical-physical model, first of a car on the road
and then of a car on the bench. A mathematical-physical model is a set of equations
and other mathematical relationships that represent physical phenomena, explain-
ing hypotheses based on observation of reality. In general, a model is constructed
from general laws and constitutive relations, of an experimental nature. In order
to build mathematical models of cars on road and cars on rollers test bench and
to obtain the related dynamic equations, two approaches have been followed: the
Newtonian approach and the Lagrangian approach. If performed correctly they
validate each other as they lead to the same dynamic equation. These two methods
will be applied to both systems under analysis (car on road, car on test bench).
In order to realize the mathematical models it has been necessary to represent the
two systems according to a concentrated parameter schematization.

3.1 Model car on the road

The on-road car model was constructed based on the Longitudinal model from three
assumptions:
• Absence of wind;
• Lack of rolling friction;
• Flat road.
In order to better understand the longitudinal dynamic behavior of a vehicle, it is
necessary to identify the main forces acting on the car during its travel on the road,
which are:
15
 Modeling of Vehicle and Test Bench

• Driving force;
• Resisting force;
• Inertial force.

Figure 3.1: 2D car in longitudinal movement

The driving force is generated by a set of type processes:

• Chemical
– combustion of endothermic engine
• Electric
– electric or hybrid motors
• Mechanics
– Transformation of torque at the drive shaft into torque at the wheels by
means of gears, clutches and transmission
– Transformation from torque to wheels to force to tire
– Transformation from brake pad torque to tire force.
The resisting force is given by:
• Vehicle conformation
– aerodynamics
16
 3.1 – Model car on the road

– tire size and type

– road condition (uphill or downhill).

Inertial force is given by Newton’s second principle of dynamics:

• F = mvehicle a

3.1.1 Model car on the road: Newtonian approach

The Newtonian approach exploits the free body diagram of the system that you
want to study by reporting the forces acting on the system itself. Once we have cho-
sen a positive direction of the forces and moments we obtain the various equilibria
around an arbitrary pole.

Figure 3.2: Diagram of the forces of a car on the road in Newton’s method

The forces applied at the center of gravity of the vehicle are:

• Fi = inertial force;
• Faero = aerodynamic friction force;
• mvehicle g = weight force.

The car under consideration is front wheel drive so the wheels of the front axle are
driving wheels and the wheels of the rear axle are driven or dragged wheels. For
this reason the two "wheel-systems" will be studied later in a different way in order
to compose the final model.

17
 Modeling of Vehicle and Test Bench

Driving wheels

Figure 3.3: Force diagram for driving wheels

The forces and moments acting on the driving wheels are:

• N1 = ground reaction force on the wheel;

• T1 = dynamic friction force;
• CM = driving force;
• Ir ẇ1 = moment of inertia;
• Cr = resistant torque due to bearings and transmission;
• fa = dynamic friction coefficient.

We apply the assumptions of adherence for a wheel that must not slip and must
move of pure rolling motion so that the system has only one degree of freedom:

T1 ≤ fa N1
ẍ = rẇ1

18
 3.1 – Model car on the road

Driven or dragged wheels

Figure 3.4: Force diagram for driven wheels

The forces and moments acting on the driving wheels are:

• N2 = ground reaction force on the wheel;

• T2 = friction force;

• Ir ẇ1 = moment of inertia;

• Cr = resistant torque due to bearings and transmission;

• fa = dynamic friction coefficient.

We apply the assumptions of adherence for a wheel that must not slip and must
move of pure rolling motion so that the system has only one degree of freedom:

T2 ≤ fa N2
ẍ = rẇ1
19
 Modeling of Vehicle and Test Bench

Calculation N1 and N2
In order to reach the dynamic equation it is necessary to obtain the equations of the
reaction forces of the ground on the wheel and of the dynamic friction forces because
they are terms not obtainable from any type of sensor present inside a generic car.
The reactions of the ground on the wheel have been calculated in the following way:

Equilibrium at rotation around point 1:

CM − 4Cr − 4Ir ẇ1 + N2 L − Fi h − Faero h − mvehicle gL(1 − α) = 0

from the above equation we get :

Fi h + Faero h + mvehicle gL(1 − α) − CM + 4Cr + 4Ir ẇ1

N2 =
L

Equilibrium rotation around point 2:

CM − 4Cr − 4Ir ẇ1 − N1 L − Fi h − Faero h + mvehicle gLα = 0

from the above equation we get:

CM − 4Cr − 4Ir ẇ1 − Fi h − Faero h + mvehicle gLα
N1 =
L

20
 3.1 – Model car on the road

Calculation T1 and T2
Considering instead the wheel subsystem, two equilibrium rotation are performed
around the front wheel center and the rear wheel center to obtain the two dynamic
friction forces. For the wheel subsystem it has been considered the entire axis and
for this reason it will be found in the following equations a coefficient 2 that will
multiply the resistant torque, the moment of inertia of the wheel and the moment
generated by the friction force.

Equilibrium rotation around the point c1 :

CM − 2Cr − 2Ir ẇ1 − 2T1 r = 0

from the above equation we get:
CM
2
− Cr − Ir ẇ1
T1 =
r

Equilibrium rotation around the point c2 :

2T2 r − 2Cr − 2Ir ẇ1 = 0

from the above equation we get:

Cr + Ir ẇ1
T2 =
r

21
 Modeling of Vehicle and Test Bench

Dynamic equation
Now by making an equilibrium to the horizontal translation knowing T1 and T2 we
can obtain the dynamical equation of the car system.

Equilibrium at horizontal translation:

2T1 − 2T2 − Faero − Fi = 0

Substituting the values of T1 and T2 we obtain:

CM − 4Cr − 4Ir ẇ1
− Faero − Fi = 0
r
noting that the driving torque and the resisting torque are divided by their own
arm r, for clarity and convenience they are reported using the forces:
Ir ẇ1
Fx − 4Fr − 4 − Faero − Fi = 0
r
Knowing the relationship between angular acceleration and tangential acceleration

a = ẇ1 r

it is possible to obtain an equation having as only unknown the tangential acceler-

ation by replacing the angular acceleration with the tangential acceleration divided
by the wheel radius:
Ir ẇ1 Ir a
= 2
r r

Substituting in the above equation gives the value of the tangential acceleration:
Fx − 4Fr − Faero
a= 3 4
4Ir
mvehicle + r2

Fr is a force due to friction, which is directly dependent on speed, so it is replaced

with a coefficient that multiplies the speed:

F r = k1 v

Faero is the aerodynamic friction force, which is directly dependent on the square
of the velocity, so it can be substituted in this way:

Faero = kaero v 2
22
 3.1 – Model car on the road

The force of inertia is given by Newton’s second principle of dynamics, the recurring
mass by acceleration:
Fi = mvehicle a

Substituting the resistant force, dynamic force, and inertia force with the cor-
responding terms representing them, the equation for tangential acceleration be-
comes:
Fx − 4k1 v − kaero v 2
a= 3 4
4Ir
mvehicle + r2

While the dynamic equation will take this form:

A B
4Ir
ẍ mvehicle + 2 + ẋ2 kaero + ẋ4k1 = Fx
r

where ẍ corresponds to the tangential acceleration and ẋ to the tangential velocity.

23
 Modeling of Vehicle and Test Bench

3.1.2 Model car on the road: Lagrangian approach

This is an analytical, energy-based approach. Once the generalized coordinate
vector is chosen, one exploits the kinetic energy K, the potential energy of the
system P, the knowledge of the external forces and of the dissipative forces to
derive the dynamic equations of the system through derivatives with respect to the
generalized coordinate vector and its derivative with respect to time, as can be seen
in the equation below. This is the so-called Lagrangian equation:
A B
d dL dL
− = Fi − fif ric
dt dq˙i dqi

where L is the Lagrange function which is defined as the difference between the
total kinetic co-energy and the total potential energy of the system
L(q, q̇) = K(q, q̇) − P (q)

Fi represent the external forces acting on the system, while the f f ric represent the
dissipative forces of the system.

Figure 3.5: Diagram of the forces of a car on the road in Lagrange’s method

The car system consists of five masses, where four masses represent the wheels mr
and the fifth represents the mass of the vehicle deprived of the wheels mchassis . It
is characterized by 2 generalized coordinates, defined as follows:
• θ1 wheel rotation angle
• x absolute displacement of the vehicle
The relationship between the generalized coordinates is bound by their derivatives
and holds
ẋ = θ˙1 r
24
 3.1 – Model car on the road

The center of gravity of the rear wheels is at the point of absolute coordinates

[x r 0]T

The center of gravity of the front wheels is at the point of absolute coordinates

[x + L r 0]T

The center of gravity of the chassis is at the point of absolute coordinates

[(x + L)α h 0]T

2
The squared norm of the rotation speed of the wheels is equal to θ˙1 .
The absolute velocity of the chassis is vchassis = (ẋ 0 0) e la sua norma al
quadrato vale ||vchassis ||2 = ẋ2 .
Now that it is known the norms of the velocities of the components of the system
it is possible to calculate the total kinetic energy. The total kinetic energy is the
sum of various components and precisely

Ktot = Kchassis + 4Kwheel

where
1
Kchassis = mchassis ẋ2
2
1 1 2
Kruota = mwheel ẋ2 + Ir θ˙1
2 2
from which we get
1 2
Ktot = mchassis ẋ2 + 2mwheel ẋ2 + 2Ir θ˙1
2
The total potential energy is the sum of gravitational and elastic components and
precisely
Ptot = 4mwheel (−g)r + mchassis (−g)h = mvehicle (−g)h
Generalized forces are obtained by applying the principle of virtual work: there
exists a point where the external force is applied and therefore the virtual displace-
ment must be considered. This virtual work concerns the rotation of the wheels
caused by the driving torque τM = Fx r and it holds

δWM = [0 0 τM ][0 0 δθ1 ]T = τM δθ1

25
 Modeling of Vehicle and Test Bench

from the previous relations is obtained

δWM = Fx δx

so it is easy to see that the generalized force corresponds to the driving force Fx .
Regarding the dissipative forces, it is assumed that they are worth

fif ric = 4Fr + Faero

Knowing the total kinetic energy and the total potential energy of the system,
through their subtraction, it is calculated the Lagrangian function which is worth
1 2
L(q, q̇) = mchassis ẋ2 + 2mwheel ẋ2 + 2Ir θ˙1
2

At this point we get to perform a series of derivatives, through which we can write
the Lagrangian equation.
δL δL
= =0
δq δx
δL δL ẋ
= = mchassis ẋ + 4mwheel ẋ + 4Ir 2
δ q̇ δ ẋ r
A B A B
d δL d δL ẍ
= = mchassis ẍ + 4mwheel ẍ + 4Ir
dt δ q̇ dt δ ẋ r2

The Lagrange’s equation is:

A B
d dL dL
− = Fi − fif ric
dt dẋ dx
ẍ
mchassis ẍ + 4mwheel ẍ + 4Ir = Fx − 4Fr − Faero
r2
The double derivative with respect to time of the absolute displacement of the
vehicle ẍ corresponds to the longitudinal acceleration of the vehicle, so Lagrange’s
equation becomes
a
mchassis a + 4mwheel a + 4Ir = Fx − 4Fr − Faero
r2
Now by collecting the acceleration as a first member, it is possible to combine the
mass of the 4 wheels added to the mass of the chassis into a single term that will
be called the total mass mtot .
A B
4Ir
a mchassis + 4mwheel + 2 = Fx − 4Fr − Faero
r
26
 3.1 – Model car on the road

4mwheel + mchassis = mtot

A B
4Ir
a mtot + 2 = Fx − 4Fr − Faero
r
At this point we isolate at first member the term that represents the tangential
acceleration
Fx − 4Fr − Faero
a=
mtot + 4I
r2
r

Regarding the dynamic equation, the equation that results is

A B
4Ir
ẍ mtot + 2 = Fx − 4Fr − Faero
r

which with the right substitutions of Fr and Faero with their coefficients multiplied
by their velocity dependence becomes
A B
4Ir
ẍ mtot + 2 + ẋ2 kaero + ẋ4k1 = Fx
r

It can be seen, as expected, that the dynamic equations of the car system on the
road obtained by performing two different approaches (Lagrange and Newton) are
perfectly coincident. This leads to affirm that the two methods have been performed
correctly.

27
 Modeling of Vehicle and Test Bench

3.2 Car on bench

Once the longitudinal dynamics of the vehicle has been studied, in order to arrive
at the result of a roller bench model, the same study must be repeated considering
the vehicle positioned inside a Roller Bench. Therefore, by studying this system, it
is found that the resistant force will be divided into FM due to the conformation of
the vehicle and into Fp also called roller loss. This Fp is due to the frictional forces
generated by the bearings that support the electric motor such that all the forces
involved discharge onto the load cell.

Figure 3.6: Car on bench diagram on front axle bench

• Fx : driving force of vehicle tangential to the roller;

• Fi : inertia force referred to the vehicle tangential to the roller;

• Fp : resistant force given by the presence of the roller;

• FM : resistant force given by the aerodynamic and rolling resistance of the

vehicle.

3.2.1 FM and the road equation

In order to provide a more realistic simulation of the forces a vehicle is subjected
to on the road, over the years a more accurate system than the simple inertia
28
 3.2 – Car on bench

simulation has been sought. Studies have therefore led to the formulation of an
equation that takes the name of RLS, which is identified as FM , which is a resistant
force due to the conformation of the vehicle that is calculated through the use of
the equation road (Road Load Simulation). The most widely used method to date
is the RLS or Road Load Simulation that is a second-order equation that takes into
account the inertia of the vehicle, the rolling friction between wheel and asphalt,
and the aerodynamics of the vehicle.
The typical Road Load Simulation equation is as follows:

FM = F0 + F1 v + F2 v 2

Ns m N s2 m2
[N ] = [N ] + []∗[ ]+[ 2 ]∗[ 2 ]
m s m s
Resistantf orce = Inertia + Rollingf riction + Aerodynamics
F0 , F1 and F2 are the three parameters by which each car is characterized and
affect the characteristic curve of the RLS. These parameters are provided by the
vehicle manufacturer, who obtains these data from vehicle models or through real
tests performed on the road, by performing a particular procedure from which to
obtain these parameters, which is called coastdown (which will be explained later
how it is performed on the bench), which on the road is performed in this way: the
vehicle is brought to a constant speed of 135 km/h on a flat road in the absence of
wind and then put in neutral gear is left to decelerate only because of friction. This
deceleration is divided into intervals. In these intervals the deceleration and the
elapsed time are observed and a and the elapsed time are observed and a resistant
force is derived. These found forces are then averaged by interpolation with a least
squares algorithm to find the parameters F0 , F1 and F2 .

3.2.2 Fp , roller losses and calibration

Fp is the intrinsic resisting force of the roller and is due to the frictional forces
generated by the bearings supporting the electric motor such that all the forces
involved are discharged onto the load cell. Calculating this force correctly is a
delicate operation, and is usually repeated several times, constantly. It depends
on many factors, including temperature and rolling friction. rolling friction. To
identify this force as precisely as possible, a calibration procedure is carried out.
The calibration is used to characterize the mechanical system from the point of
view of the intrinsic resistance to rotation by detecting the residual forces of the
bench, it means the contribution that it must impose on the motor in order to
keep the roller moving at the desired speed. Its result is to find three coefficients
that express the quadratic braking curve of the bench. This curve is very important
since each roller system is characterized by a univocal braking curve. This procedure
therefore as a result allows to know the curve representing Fp as a function of the
29
 Modeling of Vehicle and Test Bench

angular velocity of the rollers and can be performed in two different ways. The two
calibration procedures are following:

• Mode 1 - based on constant resistant force:

With unloaded rollers, it means without vehicle, I bring the system to an
agreed speed, after which I impose a constant resistant force in order to de-
crease the speed of the rollers. I acquire the decelerations and compare them
with the expected theoretical ones. From these data and from the difference
between the measured and theoretical values, through a least squares algo-
rithm, I extract the loss curve. It is usually a quadratic system.
• Mode 2 - based on stationary points:
With unloaded rollers, the roller is driven at various preset speeds and the
force values read by the load cell are saved, necessary to keep it at this speed.
Again using a mathematical function, the points are interpolated to calculate
the loss curve.

3.2.3 Fx , the driving force

During the roller test it is necessary to measure the force (or torque) generated by
the vehicle, both to correctly control the speed of the rollers, and for possible testing
requirements. To measure this force, one (or more than one in particular cases)
load cell is used in the configuration shown in the figure. The motor is typically
a twin-shaft with the two shafts connected to the two rollers (also called drums or
flywheels). The shaft is decoupled from the ground since its support is typically
equipped with ball bearings. The motor would therefore be free to turn on itself
if it were not for the arm connecting it to the load cell that ties it to the ground.
This causes all torque generated by the motor to be transmitted as a force to the
load cell and measured.
Taking a moving car as an example, the force scheme is as follows:

30
 3.2 – Car on bench

Figure 3.7: Force diagram of the traction axis

The reading of the load cell does not take into account only the driving force of
the vehicle, but reads all the forces acting on the system and therefore the bench
must be able to isolate the value of the driving force of the vehicle in order to
make a correct road simulation. The formula used to derive the driving force is the
following, where the inertia of the vehicle and the resistant forces due to the bench
system are subtracted from the load cell reading:

Fx = −Fcell + Fr + mvehicle a
The purpose of the bench is that one to simulate the road, therefore imagining
not to be over a bench test rollers but to be on the road the components of the
motive force will be

Fx = Fp + FM + mvehicle a
It is also necessary to add that following the request for tests on vehicles that can
generate forces greater than the standard range (0-5kN), it is possible to insert a
double load cell system. They have different scale and are inserted in series as
shown in the figure.

31
 Modeling of Vehicle and Test Bench

Figure 3.8: Double load cell reading scheme

The design is simplifying but the underlying concepts are easily highlighted. The
spring present between the two load cells is designed to start compressing only for
forces higher than 3500 N. The system is therefore predisposed to decouple the
5kN cell when the forces in play are higher, thanks precisely to the compression
of the spring that will make the structure rest on the support plane. The forces
will therefore pass through the 25kN load cell and will be then unloaded on the
ground as shown by the red arrow in the figure. The actual structure is however
more complex because the load cells work in both compression and elongation, but
in both cases the smaller capacity cell is excluded when higher forces come into play.

32
 3.2 – Car on bench

Figure 3.9: Single load cell reading scheme

The most consistent problem with this solution is the reading transition between
the two cells. Since the system is real, and the two cells have different resolutions,
it will be practically impossible to find an instant when both cells report exactly
the same value. an instant in which both cells report exactly the same value. If
you opt, for example for the direct passage of the feedback signal first of one and
then of the other cell, two consecutive instants are created in which the read pair is
not linear, but has a jump, and this could create stability problems to the system.
The solution adopted is a weighted of the two load cell readings. For low force
readings the reading of the load cell with a low full scale will have more weight (it
means it will have more influence on the total reading), in case of larger readings
more importance will be given to the reading of the 25kN load cell. The variation
of these weights is linear with the variation of the detected force.

3.2.4 Data processing

Now that all the forces at play in the system have been identified, it is important
to understand how they are to be used. Remembering that the main task of the
roller test bench is to simulate the behavior that the vehicle would have on the
road, it must be understood how to control the system to achieve the purpose.
Therefore, since the electric motors are speed-controlled, it is necessary to extract
the speed command to be given to the roller system, instant by instant. To do this,
the acceleration that the vehicle would produce on itself if it were actually on the
road must first be calculated.
From the parameters F0, F1 and F2 given by the manufacturer, the system is aware
33
 Modeling of Vehicle and Test Bench

of the resistant contribution acting on the vehicle (road surface resistance, aero-
dynamics, internal mechanics...). that acts on the vehicle (resistance road surface,
aerodynamics, internal mechanics ...), from the calibration procedure also knows
the intrinsic contribution of resistance of the system and through the coastdown is
able to compensate the mechanical limit given by the rollers in order to correctly
simulate the resistive contribution that would act on the car if it were on the road
with the aim to correctly simulate the resistive contribution that would act on the
car if it were on the road.
At this point, of the equation
Fx = Fp + FM + mvehicle a
is known everything except the acceleration. Fx is obtained instant by instant from
the reading of the load cell, Fp and FM both depend on the speed, which is read
by the encoder and their coefficients are obtained from the two procedures already
described and finally the mass of the vehicle is a known data provided by the
manufacturer. Therefore it is now possible to obtain the instantaneous tangential
acceleration:
Fx − Fp − FM
a=
mvehicle
However, normally the rollers are driven in speed or torque, and not in acceleration.
We therefore still have one more step to perform, which in the case of speed control
is :
V = V0 + at
Discretizing the time:
Vt+1 = Vt + a∆t
Where Vt+1 is the speed set point provided to the electric drive to bring the rollers
to that speed, Vt is the speed read by the rollers in the current instant and ∆t is
the refresh time, it means the time that elapses between one reading and another.

3.2.5 Verification of RLS: The Coastdown

As it has been said many times before, the main task of a rollers test bench is to
correctly simulate what the car would do on the road. It is therefore fundamental,
before being able to use the bench for testing, to understand if it is actually able to
correctly simulate what is required. To achieve this goal, the coastdown procedure
is used. The coastdown procedure is used to define the parameters F0, F1 and F2
of the road equation and is conducted in this way: the vehicle positioned on the
rollers is brought to a constant speed of 135 km/h and then put in neutral gear
and allowed to decelerate only due to friction up to 15 km/h. This deceleration is
divided into intervals. In these intervals the deceleration and the elapsed time are
observed and a resistant force is derived.
34
 3.2 – Car on bench

Figure 3.10: Speed-time graph of a coastdown test

∆v
= amean
∆t

F = mvehicle amean

These found forces are then averaged by interpolation with a least squares algo-
rithm to find the parameters that make up FM .
At the end of the coastdown procedure, the resulting force curve on the vehicle
placed on the rollers shall be congruent with the above mentioned curve, commit-
ting a maximum error defined by standard. At the end of the cycle, by means
of appropriate instruments, the difference between the forces is calculated in the
different speed intervals and those obtained from the three road coefficients given
by the customer. If the error is within tolerance, the coastdown is concluded since
the bench correctly simulates the road behavior. If not, it is possible to recalculate
three new coefficients of F0, F1 and F2 that will be used in place of those given by
the customer. This recalculation can be repeated iteratively, as long as the errors
obtained from the coastdown are in tolerance. It is important to understand that
the results obtained from a recalculation must always be compared with the brak-
ing curve given by the original parameters F0, F1 and F2, because that is what
you want to obtain. This recalculation is tolerated by vehicle manufacturers as the
"Roller+Vehicle" combination slightly varies the previous forces that were identified
with the calibration. It is then possible to recalculate the parameters and run a
coastdown again. If the coastdown results in tolerance, it is possible to use the
bench to test the vehicle, and one of the most used methods is precisely to perform
driving cycles with preconfigured speed traces in order to analyze the pollutants
produced by the vehicle.
35
 Modeling of Vehicle and Test Bench

3.2.6 Model car on bench: Newtonian approach

With the purpose of realizing a mathematical-physical model of a car on a chassis
dynamometer, as already done in the previous sub-chapters for the car system on
the road, the Newton-Euler method and the Lagrange method will be applied.
The chassis dynamometer considered in this study is a 2WD, which means that the
cars tested on this dynamometer have only one drive axle, which in this case is the
frontal axle.

Figure 3.11: Diagram of the forces of a car on a bench in Newton’s method

Also this model was built from the basis provided by the longitudinal car model.
The forces and moments acting on the car+roller test bench system are:
• w1 = angular velocity of the wheels;
• w2 = angular speed of the rollers;
• Cp = Roller loss torque due to the resistance of the bearings that support the
roller-motor axis so that all forces are loaded on the load cell;
• Cr = resistant torque due to the bearings and transmission of the car;
• CM = driving torque of the car;
• Ce = driving torque of the electric motor of the rollers;
• F̃ = force reading [N] of the load cell;
36
 3.2 – Car on bench

• mvehicle g = car weight force;

• Ir ẇ1 = moment of inertia due to the inertial mass of the wheel;
• IR ẇ2 = moment of inertia due to the inertial mass of the wheel;
• b = load cell arm, distance between roller center and load cell.

37
 Modeling of Vehicle and Test Bench

Load Cell Equation

Since we don’t have an instrument to know the value of the car engine torque we
will use the load cell reading to get it. As explained in the introduction, the load
cell reads all the torque generated by the electric motor transmitted as a force.
Another consideration that needs to be made is regarding the speed at the point
of contact between wheel and roller, which is considered equal for both rotating
masses. Therefore, the relation that links the angular speed of the wheel and the
angular speed of the roller is
v = w1 r = w2 R
By means of a rotation equilibrium around the center of the roller we derive the
equation for the load cell reading. Before performing the rotation equilibrium it is
necessary to carry all the torques acting on the wheels on the roller, and to do this
we divide the torque of interest by the wheel radius r and multiply it by the roller
radius R.

Figure 3.12: Force diagram for the front axle of the bench

38
 3.2 – Car on bench

Balance to the rotation around the center of the roller:

R R2 R
Ce + CM − Cp − 2IR ẇ2 − 2Ir ẇ2 2 − 2Cr = 0
r r r
from the above equation we get:
R r
CM = 2Ir ẇ2 + 2Cr + (2IR ẇ2 + Cp − Ce )
r R
then dividing the CM by the wheel radius:
CM
Fx =
r
R 1
Fx = 2Ir ẇ2 2
+ 2Fr + (2IR ẇ2 + Cp − Ce )
r R
where Ce , motor torque of the electric motor corresponds to the load cell reading
multiplied by the arm measured as the distance between the load cell and the center
of the roller.

39
 Modeling of Vehicle and Test Bench

Dynamic equation
After having calculated the equation that uses the reading of the load cell to obtain
the driving force of the vehicle under test, it is necessary to repeat the same calcu-
lations in order to extrapolate the dynamic equation of the system, but this time
not taking into account the load cell. Therefore another equilibrium is performed
on the rotation always around the center of the roller.

Figure 3.13: Force diagram by dynamic equation

Balance rotation around the center of the roller:

R R2 R
Ce + CM − Cp − 2IR ẇ2 − 2Ir ẇ2 2 − 2Cr = 0
r r r
by collecting it is got the dynamic equation of the system:

R2 R
2ẇ2 (−IR − Ir 2 ) + Ce + (CM − 2Cr ) − Cp = 0
r r

From the dynamic equation it is possible to derive the angular acceleration:

Ce + (CM − 2Cr ) Rr − Cp
ẇ2 = 2
2(IR + Ir Rr2 )

are replaced the torques with the relative forces multiplied by their own arm:
40
 3.2 – Car on bench

Fe + Fx − 2Fr − Fp
ẇ2 =
2R( RIR2 + rIr2 )

Multiplying the angular acceleration by the radius of the roller gives the tangential
acceleration of the roller:
Fe + Fx − 2Fr − Fp
a=
2( RIR2 + rIr2 )
Fr and Fp are forces due to friction, which are directly dependent on velocity,
so they are substituted like this:

F r = k1 v
F p = k2 v

Substituting in the equations for tangential acceleration and angular acceleration

we get:
Fe + Fx − 2k1 v − k2 v
a=
2( RIR2 + rIr2 )
Fe + Fx − 2k1 v − k2 v
ẇ2 =
2R( RIR2 + rIr2 )

Exploiting the relationship that binds the angular speed of the wheel to the angular
speed of the roller it is possible to obtain also the equation that represents the
angular acceleration of the wheel obtaining
Fe + Fx − 2k1 v − k2 v
ẇ1 =
2r( RIR2 + rIr2 )

Regarding the dynamic equation of the system car+bench test rolls the equation
that results from above equation is
A B
IR Ir
2 2
+ 2 ẍ + (2k1 + k2 )ẋ = Fe + Fx
R r

41
 Modeling of Vehicle and Test Bench

3.2.7 Model car on bench: Lagrangian approach

The car+roller test bench system is composed of the masses of the four wheels mr ,
the mass of the vehicle deprived of the wheels, and the masses of the two front
rollers mR .
It is characterized by 2 generalized coordinates, defined as follows:
• θ1 wheel rotation angle
• θ2 roller rotation angle

Figure 3.14: Diagram of the forces of a car on a bench in Lagrange’s method

The relationship between the generalized coordinates is related by their tangential

velocity, which is the same for both rollers and wheels, and is worth
v = θ˙1 r = θ˙2 R
R
θ˙1 = θ˙2
r

The center of gravity of the rear wheels turns out to be at the absolute coordinate
point:
[0 r 0]
The center of gravity of the front wheels turns out to be at the absolute coordinate
point:
[L r 0]
42
 3.2 – Car on bench

The center of gravity of the chassis is found to be at the absolute coordinate point:

[Lα h 0]

The center of gravity of the rollers is found to be at the absolute coordinate point:

[Lα − R 0]

The distances along the Z axis between right and left roller and right and left wheel
have not been considered because they are negligible for this method.
2
The squared norm of the rotation speed of the wheels is equal to θ˙1 .
2
The squared norm of the rotation speed of the rollers is equal to θ˙2 .
Now that we know the square rotational velocities of the components of the system
we can calculate the total kinetic energy. The kinetic energy developed in this sys-
tem is exclusively due to rotational movements because the car cannot make any
kind of translation movement when it is on a roller test bench. The total kinetic
energy is the sum of various components and precisely

Ktot = 2Kroller + 2Kwheel

where
1 2
Kroller = Iroller θ˙1
2
1 2
Kwheel = Iwheel θ˙2
2
from which it is got
2 2
Ktot = Iroller θ˙1 + Iwheel θ˙2
and exploiting the relationship between the two generalized coordinates the equa-
tion representing the total kinetic energy becomes
A B
2 r2
Ktot = θ˙1 Iroller 2 + Iwheel
R

The total potential energy is the sum of the gravitational and elastic components
and precisely

Ptot = 4mwheel (−g)r+mchassis (−g)h+2mroller (−g)r = mvehicle (−g)h+2mroller (−g)r

The generalized forces, in this case generalized torques, are obtained by applying
the principle of virtual work and in this case the virtual work concerns the rotation
43
 Modeling of Vehicle and Test Bench

of the wheels and the rotation of the rollers caused by the driving torque of the
endothermic engine of the vehicle and by the driving torque generated by the electric
motor of the bench, and it is worth

δW = CM δθ1 + Ce δθ2

from the previous relations that link the two generalized coordinates it is obtained
r
δW = Fx rδθ1 + Fe R δθ1
R
so it is easy to see that the generalized torque is

τi = Fx r + Fe r

For what concern dissipative torques are concerned, it is assumed that they are
worth
τif ric = 2Cr + Cp
τif ric = 2Fr r + Fp r

Knowing the total kinetic energy and the total potential energy of the system,
through their subtraction, we calculate the Lagrangian function that is worth
A B
2 r2
L(q, q̇) = θ˙1 Iroller 2 + Iwheel − mvehicle (−g)h − 2mroller (−g)r
R

At this point we perform a series of derivations, through which it is derived the

Lagrangian equation.
δL δL
= =0
δq δθ1
A B
δL δL r2
= ˙ = 2θ˙1 Iroller 2 + Iwheel
δq δ θ1 R
A B A B A B
d δL d δL ¨ r2
= = 2θ1 Iroller 2 + Iwheel
dt δ q̇ dt δ θ˙1 R
The Lagrange equation is:
A B
d dL dL
− = τi − τif ric
dt dẋ dx
A B
r2
2θ¨1 Iroller 2 + Iwheel = Fx r + Fe r − 2Fr r − Fp r
R
44
 3.2 – Car on bench

The double derivative with respect to time of the generalized coordinate θ1 corre-
sponds to the angular acceleration of the wheels

θ¨1 = ẇ1

so Lagrange’s equation becomes

A B
r2
2ẇ1 Iroller 2 + Iwheel = (Fx + Fe − 2Fr − Fp )r
R

At this point it is isolated at first member the term that represents the angular
acceleration of the wheels
(Fx + Fe − 2Fr − Fp )r
ẇ1 = 2
2(Iroller Rr 2 + Iwheel )

Fx + Fe − 2Fr − Fp
ẇ1 =
2r( Iroller
R2
+ Iwheel
r2
)
Exploiting the relation that binds the two generalized coordinates it is possible to
obtain also the equation that represents the angular acceleration of the rollers
Fx + Fe − 2Fr − Fp
ẇ2 =
2R( Iroller
R2
+ Iwheel
r2
)

Finally, the resulting dynamic equation obtained from the previous equation, ap-
plying the relative substitutions seen above for Fr and Fp , is
A B
IR Ir
2 2
+ 2 ẍ + (2k1 + k2 )ẋ = Fe + Fx
R r

In the final analysis, as observed for the road car model also for the car model on
roller test bench it can be affirmed that the dynamic equations obtained with the
two methods are correct having led to the same dynamic equation.

45
46
Chapter 4

Matlab-Simulink model
For the simulation of the bench it is employed the use of a model created in Matlab-
Simulink environment so that comparisons can be made between the results ob-
tained with the simulated tests and the results obtained with the empirical tests
performed on real roller benches. The Matlab-Simulink model created was built by
means of the union of the physical-mathematical model of car on the road with the
mathematical-physical model of car on a roller test bench.

Figure 4.1: Matlab-Simulink model origin summary

From the mathematical-physical model of car on a roller test bench the equation
of the load cell has been extrapolated, where in input is given the driving torque
generated by the electric motor of the bench and in output is obtained the driving
47
 Matlab-Simulink model

force generated by the car.

R 1
Fx = 2Ir ẇ2 2
+ 2Fr + (2IR ẇ2 + Cp − Ce )
r R
From the mathematical-physical model of a car on a roller test bench the road
equation was used, where in input is given the driving force generated by the car
on the bench and in output, solving the differential equation, is generated the
tangential speed that the car would have had if it were really on the road.

Fx − 4k1 v − kaero v 2
a=
(mvehicle + 4I
r2
)

The main model is as follows:

Figure 4.2: RLS subsystem of the main model

The load cell subsystem is the block containing the load cell equation, where the
speed information, load cell readings and roller losses are provided as input. As
input data for the various simulations we used real data from a test performed on
a real Roller test Bench. In the following image you can see the content of the load
cell subsystem representing the equation of the load cell in Simulink environment.

48
 Matlab-Simulink model

Figure 4.3: Load cell subsystem

The carontheroad dynamic equation subsystem contains the road equation and
receives as input the driving force produced by the car from the load cell subsystem,
while as output it generates the speed and the tangential acceleration at the wheels
that the car would have had if it was really on the road. The image below shows
the content of the subsystem carontheroad dynamic equation representing the road
equation in Simulink language.

Figure 4.4: Carontheroad dynamic equation subsystem

It is necessary to emphasize the fact that the terms related to the rolling friction
force, the resistant force due to the rotating parts present in the vehicle and the
49
 Matlab-Simulink model

aerodynamic friction force acting on the vehicle present in both subsystems are
replaced with the equation representing the braking curve FM :

FM = F0 + F1 v + F2 v 2
The coefficients F0 , F1 and F2 are provided directly by the manufacturer of the
vehicle under test.
Finally, in the next figure you can see the last part of the model, which generates
the final output:

Figure 4.5: Angular velocity SET calculation

These blocks receive as input the tangential acceleration and tangential velocity at
the wheels, and as output produce the SET angular velocity that must be supplied
to the rollers so that they can simulate the road. What these blocks do math-
ematically is exploit the equation of uniformly accelerated motion discretized in
time.

Vt+1 = Vt + a∆t
Receiving from the carontheroad dynamic equation block the information of the
acceleration and tangential velocity that the car would have if it were on the road,
given the driving force Fx , at time instant t the tangential velocity SET at time
instant t+1 is computed that must be supplied to the rollers in order to simulate
the road.

50
 4.1 – Coast-down simulation

4.1 Coast-down simulation

Using the Matlab-Simulink model, coast-down simulations were conducted. The
purpose of the coast-down test is to verify if the roller test bench can perfectly
reproduce the behavior of the car under test just as if it were really on the road,
so this type of test has been chosen to validate the roller test bench model. As
already explained in the previous chapters the coast down test is a cycle in which
the vehicle is brought to a constant speed and then put in neutral and allowed to
decelerate only because of friction. This deceleration is divided into intervals. In
these intervals the deceleration and the elapsed time are observed and a resisting
force is derived. The braking curve during the coast-down cycle is calculated by
evaluating the deceleration between 125 km/h and 15 km/h.

In order to make comparisons between experimental results and simulated results,

two variants of the main model have been created, which will be called the partially
real model and the on-road car model. In the image below, the main model can be
seen:

Figure 4.6: Main model

As it can be noticed all the inputs of the main model derive from the data taken
from the empirical test carried out with a real rollers test bench.
In the following image instead we can see the partially real model:

51
 Matlab-Simulink model

Figure 4.7: Partially real model

In this variant of the main model not all inputs come from data storage, only
the information about load cell reading and roller losses. While the roller speed
informations read from the encoder are calculated within the model.

Figure 4.8: RLS subsystem of the partially real model

Browsing within the RLS subsystem, it is possible to see how the roller speed infor-
mations read by the encoder were calculated. The encoder reading was simulated
by assuming that the speed read by the encoder at instant i is equal to the SET
speed at instant i-1.
angularvelocityF BK(i) = angularvelocitySET (i − 1)
This was enabled by the use of the Simulink delay block, which provides as output
the input received in the previous integration step.
As a last step we go to analyze the vehicle model on the road:
52
 4.1 – Coast-down simulation

Figure 4.9: Model car on the road

In this case, as can be seen, the load cell subsystem containing the load cell equation
is absent, this is because the model does not receive input information taken from
the DBSrolls data store. The input model receives a constant driving force value
of 0 N because what is simulated is a coast-down test and the car is in neutral gear
during this test. Instead in output it produces the speed that the car would have if
it was really on the road and was allowed to decelerate only because of friction from
an initial speed of 125 km/h. The initial condition of 125 km/h for the tangential
speed of the vehicle under test was entered into the integrator block present within
the carontheroad dynamic equation subsystem.

4.1.1 Coast-down simulation: validation method

In order to validate the main model, data concerning speed information from the
encoder, load cell readings and the bench’s leakage curve were taken from a real
coastdown test performed with a real roller test bench and all these data were used
as input for the model. The data taken from the empirical test are data from a
roller test bench that can perfectly simulate the behavior of the vehicle on the road.

The braking curve parameters provided by the manufacturer of the vehicle under
test are:
• F0 = 99.2115

• F1 = 0.7048

• F2 = 0.1907

53
 Matlab-Simulink model

Therefore based on the FM equation of the braking curve

FM = F0 + F1 v + F2 v 2

FM = 99.2112 + 0.7048v + 0.1907v 2

the braking curve that the model should try to follow is:

Figure 4.10: Theoretical braking curve of the bench

For what concern the braking curve simulated by the model it will be calculated
using the output data of the SET acceleration of the rollers, which will be multiplied
by the mass of the vehicle under test obtaining a resistant force.
The mass of the tested vehicle is 3300 kg.

F = aSET mvehicle
From these force values, the braking curve simulated by the model will be extrap-
olated using a second-order least squares interpolation.

In order to validate the model the NEDC (New European Driving Cycle) standard,
requires to verify the percentage time error in the various speed intervals of 10
km/h, and compare it with a maximum value, which is 5% of the theoretical time
value of that interval. Therefore, what you need to do is to calculate the time
54
 4.1 – Coast-down simulation

in the theoretical braking curve, obtained from the parameters provided by the
vehicle manufacturer for each speed delta ranging from 125 km/h to 15 km/h at
10 km/h intervals. So 125 to 115, 115 to 105, and so on. Then calculate the time
in the same intervals as the model curve. Then compare that the error made, for
each interval, is within a range +/- 5%, except for speed ranges speed ranges below
55 km/h where the regulation is less stringent and the tolerance range increases
by +/-10tolerance range increases to +/-10%. If this occurs, it means that the
coastdown curve is accepted as correct.

4.1.2 Coast-down simulation: results

In the table below you can see the data from the coastdown test performed with
a real roller test stand that will be used as a method of comparison with the
simulation.

velocity range [km/h] ∆t theoretical [s] 5% ∆t

124,4-115,4 2,97 0,1485
115,4-105,2 3,96 0,198
105,2-95,1 4,62 0,231
95,1-85,1 5,61 0,2805
85,1-75,2 7,26 0,363
75,2-65,2 8,91 0,4455
65,2-55,1 12,21 0,6105
55,1-45,1 16,5 0,825
45,1-35,1 23,43 1,1715
35,1-25,1 35,64 1,782
25,1-15,6 52,47 2,6235

Table 4.1: Theoretical coast-down test table

Three comparisons were made in order to validate the main model:

• comparison of theoretical braking curve and simulated braking curve using

road car model;
• comparison of theoretical braking curve and simulated braking curve using the
partially real model of car on bench;
• comparison of theoretical braking curve and simulated braking curve using the
principal model of car on bench.

In the image below you can see the development of the simulated braking curve
obtained using the car on road model compared with the theoretical braking curve.
55
 Matlab-Simulink model

Figure 4.11: Comparison of theoretical and simulated braking curve of car on road
model

As can be seen, the two curves are almost completely overlapping. This shows that
the model is correctly simulating the road with respect to the three parameters F0 ,
F1 , and F2 provided by the vehicle manufacturer. In order to better understand
the comparison between the two curves and to attempt a validation using the
competence standard, the NEDC (New European Driving Cycle), as explained
above, a table was created in which the following data was entered:

• the velocity ranges of interest;

• ∆t theoretical time in seconds, which indicates the time it should take the
vehicle to move from one speed to another;
• the 5% of the theoretical ∆t representing the tolerable error in simulation.;
• the ∆t time in seconds of the simulation indicating the time elapsed to pass
from one speed to another;
• the error in seconds committed by the simulation (difference between theoret-
ical and simulated ∆t);
• the percentage values of the absolute error of the simulation.

The table above can be seen below:

56
 4.1 – Coast-down simulation

vel range [km/h] ∆t theoretical [s] 5% ∆t ∆t sim [s] ∆e ∆e %

124,4-115,4 2,97 0,1485 2,706 0,264 8,888889
115,4-105,2 3,96 0,198 3,597 0,363 9,166667
105,2-95,1 4,62 0,231 4,323 0,297 6,428571
95,1-85,1 5,61 0,2805 5,181 0,429 7,647059
85,1-75,2 7,26 0,363 6,501 0,759 10,45455
75,2-65,2 8,91 0,4455 8,283 0,627 7,037037
65,2-55,1 12,21 0,6105 11,827 0,923 7,559378
55,1-45,1 16,5 0,825 15,312 1,188 7,2
45,1-35,1 23,43 1,1715 23,001 0,429 1,830986
35,1-25,1 35,64 1,782 36,069 -0,429 1,203704
25,1-15,6 52,47 2,6235 61,83 -9,36 17,83877

Table 4.2: Comparison table simulated coast-down test with car on road model

The table shows that all the speed intervals are within tolerance, except for the
last interval between 25.1 km/h and 15.6 km/h. As far as the compliance with the
regulations is concerned, simulating this model (car on road model), no problems
have been found, also thanks to the simplicity of the constitutive blocks which do
not receive input from external sources.

The next comparison to be made is that between the theoretical braking curve
and the braking curve simulated using the car on bench model partially real. This
comparison can be seen in the following image:

57
 Matlab-Simulink model

Figure 4.12: Comparison of theoretical and simulated braking curve of car on bench
model partially real

The graph shows a significant deviation of the simulated braking curve from the
theoretical one at low speeds. By analysing the table comparing the two curves,
this error can be quantified.

vel range [km/h] ∆t theoretical [s] 5% ∆t ∆t sim [s] ∆e ∆e %

124,4-115,4 2,97 0,1485 3,209 -0,239 8,047138
115,4-105,2 3,96 0,198 4,123 -0,163 4,116162
105,2-95,1 4,62 0,231 4,941 -0,321 6,948052
95,1-85,1 5,61 0,2805 5,181 0,429 7,647059
85,1-75,2 7,26 0,363 8,052 -0,792 10,90909
75,2-65,2 8,91 0,4455 9,783 -0,873 9,79798
65,2-55,1 12,21 0,6105 13,887 -1,677 13,73464
55,1-45,1 16,5 0,825 18,712 -2,212 13,40606
45,1-35,1 23,43 1,1715 27,001 -3,571 15,24114
35,1-25,1 35,64 1,782 43,069 -7,429 20,84456

Table 4.3: Comparison table of simulated coast-down test with car on bench model
partially real

From the table it can be seen that from the speed range 55.1-45.1 km/h there is
58
 4.1 – Coast-down simulation

an increase in the percentage error which is outside the tolerance allowed by the
regulations. This behaviour of the model is due to the fact that there is not a
perfect synchronisation between the data collected with the empirical test carried
out with a real test bench and the SET speed values calculated by the model. In
the industrial environment, however, it is considered a good coast-down test taking
this issue into account. Synchronisation was searched following a trial and error
procedure by adding dummy inputs of roller loss and load cell readings equal to
each other that correspond to the data taken during the empirical test at 124.4
km/h until the car had approximately reached the corresponding roller loss and
load cell readings of the relative km/h.

Finally, the last comparison that was made was between the theoretical braking
curve and the simulated braking curve using the pricipal model of car on bench.
This is the most important comparison because the car model on the bench is the
one that, together with the validation, would bring advantages to the company in
terms of research and development and to save time in the testing phases. In fact,
for the validation of this model, it was required to fully meet the requirements of
the regulations. The comparison graph is shown below:

Figure 4.13: Comparison of theoretical and simulated braking curve of pricipal

model of car on bench

The graph shows small deviations from the theoretical curve at the beginning and
end of the simulation, but as can be seen in the table below, this does not affect
59
 Matlab-Simulink model

the fulfilment of the NEDC standard.

The results of the comparison with the simulation are shown below:

vel range [km/h] ∆t theoretical [s] 5% ∆t ∆t sim [s] ∆e ∆e %

124,4-115,4 2,97 0,1485 2,86 0,11 3,703704
115,4-105,2 3,96 0,198 3,801 0,159 4,015152
105,2-95,1 4,62 0,231 4,386 0,234 5,064935
95,1-85,1 5,61 0,2805 5,441 0,169 3,012478
85,1-75,2 7,26 0,363 6,879 0,381 5,247934
75,2-65,2 8,91 0,4455 8,564 0,346 3,883277
65,2-55,1 12,21 0,6105 11,735 0,475 3,890254
55,1-45,1 16,5 0,825 15,537 0,963 5,836364
45,1-35,1 23,43 1,1715 22,555 0,875 3,734528
35,1-25,1 35,64 1,782 32,956 2,684 7,530864
25,1-15,6 52,47 2,6235 47,141 5,329 10,15628

Table 4.4: Comparison table of simulated coast-down test with principal model of
car on bench

The table shows that the model behaves very well in simulation, as all ranges
are within tolerance. These small errors of ∆t that are recorded not only in this
simulation but also in the previous one are due to the fact that the data supplied
as input to the models, which were taken from a real coast-down test, are acquired
with a delay mainly due to the processing that these data must undergo between
the various communication layers, therefore the speed at which they are reported
will be slightly different. In conclusion, it can be said that the principal model has
been fully validated in accordance with the NEDC (New European Driving Cycle)
standard.

60
Chapter 5

Conclusions and Future

Developments
The following work is based on the study of a generic roller test bench for car ho-
mologation and in particular on the realization of a Matlab/Simulink model able
to simulate the dynamic behaviour of the bench.

The first step was to study the bench in order to gain knowledge of its operat-
ing conditions. This knowledge was then applied using two methods, Newtonian
and Lagrangian approach, to obtain dynamic equations describing the behaviour
of cars on the road and cars on the bench. After that, the car on the bench model
was implemented in Matlab/Simulink, which was called the pricipal model. This
model, after passing validation, was made available to the company so that it could
help them both to verify that the roller test bench behaves as expected and also to
save time in finding the right trade off in setting the parameters F0 , F1 and F2 .

Despite the results achieved, there are other possible implementations that could
refine the model, for example, one possible analysis is to see if using an observer
can improve some of the measurements calculated within the model, such as the
driving force of the car and the angular velocity SET to be applied to the bench.
Observers are useful in that they make it possible to know quantities which are not
directly measurable, or whose measurement would require too expensive sensors.
An observer makes it possible, by taking the inputs and outputs of the system, to
know the values of other quantities.

In the case of estimating the driving force of the car one could resort to the use of
observer sliding modes, which provide a non-linear input with the estimation error
of the measurements, which is forced to cancel in a finite time, while the estimated
state tends asymptotically to the true value of the state.
61
 Conclusions and Future Developments

Regarding the estimation of the SET velocity to be applied to the bench, one
could try to analyse the model with the addition of an EKF (extended Kalman
filter), which is an observer used in the estimation of the parameters of a system
which allows the Kalman filter theory to be applied and which could minimise the
uncertainties arising from the system inputs.

62
Appendix A

Nomenclature

Symbol Description
T1 longitudinal force at front tires
T2 longitudinal force at rear tires
R1 rolling resistance of the front wheels
R2 rolling resistance of the front wheels
Faero aerodynamic friction force
mvehicle total vehicle mass
mwheel wheel mass
mchassis chassis vehicle mass
v longitudinal velocity of the vheicle
a longitudinal acceleration of the vheicle
g gravitational acceleration
θ angle of road slope
Cd coefficient of aerodynamic friction
Af frontal area of the vehicle subjected to aerodynamic friction
vwind wind velocity
ρ air density
sL longitudinal slip
FZ weight force acting on the wheel
µL longitudinal friction coefficient due to wheel-road contact
fa rolling resistance coefficient
Fi inertia force
N1 normal ground force on the front wheel
N2 normal ground force on the rear wheel
CM driving torque of the vehicle
Cr resistant torque due to bearings and transmission
L wheelbase of the vehicle
Continued on next page
63
 Nomenclature

Continued from previous page

Symbol Description
h height of the vehicle’s centre of gravity
c1 center of front wheel
c2 center of rear wheel
θ1 angle of wheel rotation
θ2 angle of rotation of the roller
r wheel radius
R roller radius
FM driving force of the vehicle
Ce driving torque of the roller
Fe driving force of the roller
b arm of the load cell
F̃ reading the load cell
ẇ1 angular acceleration of the wheel
ẇ2 angular acceleration of the roller
Ir ẇ1 moment of inertia of the wheel
IR ẇ2 moment of inertia of the roller
Cp roller loss torque due to bearings
Concludes from previous page

64
Bibliography
[1] Alessandro Riccio. Ottimizzazione delle prestazioni dinamiche di un banco prova
trasmissioni HIL. Tesi di laurea magistrale, Politecnico di Torino, 2019.
[2] Gabriele Vandi. Simulazione dinamica di un veicolo dotato di powertrain ibrido
endotermico-elettrico. Tesi di Dottorato, Università di Bologna, 2015.
[3] Isacco Battistini. Installazione al banco e prova di un motore motociclistico. Tesi
di laurea magistrale, Università di Bologna, 2014.
[4] C. Ferraresi, T. Raparelli. Meccanica applicata. CLUT, 2007.
[5] Wong, Jo Yung. Terramechanics and off-road vehicle engineering: terrain
behaviour, off-road vehicle performance and design. Butterworth-Heinemann,
2009.
[6] Eula Gabriella. Dispense del corso Meccanica Applicata alle Macchine. Politec-
nico di Torino, 2017.
[7] H. B. Pacejka. Tire and Vehicle Dynamics. Butterworth-Heinemann, 2012.
[8] C. Ferraresi, T. Raparelli. Meccanica applicata. CLUT, 2007.
[9] Rajesh Rajamani. Vehicle Dynamics and Control. Springer, pp87-101, 2012.
[10] Malan Stefano Alberto. Dispense del corso Automotive Control System. Po-
litecnico di Torino, 2019.
[11] Dabbene Fabrizio. Dispense del corso Modeling and Simulation of mechatronic
systems. Politecnico di Torino, 2018.
[12] Violante Massimo. Dispense del corso Model-based software design. Politecnico
di Torino, 2018.
[13] Pierdomenico Ruggieri. Studio del modello CDTire per analisi Handling e Ride
Comfort di un veicolo in relazione ai parametri geometrici e operativi degli pneu-
matici. Tesi di laurea magistrale, Politecnico di Torino, 2018.
[14] Gentiane Venture, Pierre-Jean Ripert, Wisama Khalil, Maxime Gautier,
Philippe Bodson. Modeling and identification of passenger car dynamics using
robotics formalism. IEEE Transactions on Intelligent Transportation Systems,
IEEE, 2006, 7 (3), pp.349-359. ffhal-00401730f
[15] Jin, X., Yin, G., Chen, N. Advanced Estimation Techniques for Vehicle System
Dynamic State: A Survey. Sensors. 2019; 19(19):4289.
[16] Schramm D., Hiller M., Bardini R. Vehicle Dynamics Modeling and Simulation.
Springer, 2014.
65
 Bibliography

[17] Roberto Bardini, Dieter Schramm, Manfred Hiller. Vehicle Dynamics.

Springer, 2014.
[18] Leonid Fridman Yuri Shtessel, Christopher Edwards and Arie Levant. Sliding
mode control and observation. Birkhauser, 2010.
[19] Impinna D., Cescato R.. DYNAMIC BRAKING SYSTEM: Roller. Control
Sistem s.r.l., 2018.
[20] Impinna D., Cescato R.. Banco a Rulli - RAPTOR CS ROLLS: MANUALE
SOFTWARE. Control Sistem s.r.l., 2017.
[21] Impinna D., Cescato R.. DBS ro: CONFIGURAZIONE DEL BANCO . Con-
trol Sistem s.r.l., 2017.

66