VEHICLE DYNAMICS

COMPENDIUM

unit
rear
coupling on
front unit
coupling on
unit

Control algorithm:

Version 2020 for course academic year autumn 2020 and spring 2021
Latest draft available at https://tinyurl.com/VehDynCompDraft
Bengt Jacobson et al
Vehicle Dynamics Group, Division Vehicle and Autonomous Systems,
Department of Mechanics and Maritime Sciences, Chalmers University of Technology, www.chalmers.se
2
Preface 2020
From and including 2020, Chalmers does not print the compendium on paper. Therefore, the two vari-
ants, “Printed on paper” and “Digital only” introduced 2019, are not equally motivated. So, the material
marked with “#Digital only” from 2019 is kept, and still with blue text, but now marked “§ ”. The
reader should understand the extra material as possible to jump over for a basic understanding but
recommended if aiming for more advanced knowledge.
Many contribute to this compendium, and new for version 2020 is that contributors are mentioned in
the beginning of the section where their contribution was primarily done:
Jelena Andric and Majid Astaneh, Chalmers
Pinar Boyraz, Vehicle Safety at Chalmers
Adam Brandt, Vehicle AeroDynamics at Chalmers and CEVT
Fredrik Bruzelius, Vehicle Dynamics at Chalmers and VTI
Niklas Fröjd, Volvo Trucks
Toheed Ghandriz, Vehicle Dynamics at Chalmers
Anders Hedman, Volvo Trucks
Inge Johansson, Volvo Trucks
Ingemar Johansson, CEVT and Vehicle Engineering at Chalmers
Mats Jonasson, Volvo Cars and Vehicle Dynamics at Chalmers
Mikko Karisaari, Oulo University, Finland
Waltteri Koskinen, student at Tampere University, Finland
Leo Laine, Volvo Trucks and Chalmers
Mathias Lidberg, Vehicle Dynamics at Chalmers
Luigi Romano, Vehicle Dynamics at Chalmers
Dragan Sekulić, Vehicle Dynamics at Chalmers
Alexey Vdodin, Vehicle AeroDynamics at Chalmers
My apologies to contributors I have forgotten. Remind me, and I’ll add you to next version.
/Bengt Jacobson, Göteborg, Last Modification: 2020-10-31 10:19

Preface 2019
The compendium 2019 is published in 2 variants: “Printed on paper” and “Digital only”. Both are available as pdf-file. The
“Digital only” variant contains some additional material; search for “#DigitalOnly”. The numbered items (figures, equations,
etc) in the “Digital only” variant does not have numbering, in order to keep same numbers of each item between the variants.
However, note that page numbering varies between the variants.
Thanks to Tobias Brandin, Fredrik Bruzelius, Edo Drenth, Niklas Fröjd, Toheed Ghandriz, Patrick Gruber, Mats Jonasson, Ma-
thias Lidberg, Anders Lindström, Oscar Ljungcrantz, Peter Nilsson, Luigi Romano, Juliette Utbult and errata reporting from
students. Sorry, if I forgot some contributor! /Bengt Jacobson, Göteborg, October 2019

Preface 2018
One large rearrangement is done: Subsystem descriptions has been collected from Chapter 3, 4 and 5 to 2.3-2.7. Also, minor
changes and additions has been done throughout all chapters. Many thanks to, among other, Niklas Fröjd Volvo GTT, Toheed
and Fatemeh Ghandriz, Ingemar Johansson CEVT, Mats Jonasson VCC, Mathias Lidberg, Simone Sebben, Alexey Vdovin.
Thanks also to many students that have found and reported errors in the previous edition.
/Bengt Jacobson, Göteborg, October 2018

Preface 2017
This edition has various smaller changes and additions. Thanks to Fredrik Bruzelius (VTI), Tobias Brandin (VCC), Niklas Fröjd
(Volvo GTT), Assar Jarlsson (Kinnarps), Pär Pettersson (Chalmers), among others. Thanks also to many students that have
found and reported errors in the previous edition. /Bengt Jacobson, Göteborg, October 2017

Preface 2016
This edition has various changes and additions. Some of these are: Chapter 1: Control engineering, Chapter 2: Tyre models,
Driver models, Chapter 3: Propulsion systems, Varying road pitch, Non-reactive truck suspensions, Chapter 4: Track-ability,
Articulated vehicles, and Cambering vehicles.
Thanks to Cornelia Lex (TU Graz), Fredrik Bruzelius (VTI), Niklas Fröjd, Anders Hedman, Kristoffer Tagesson, Peter Nilsson,
Sixten Berglund (Volvo GTT), Tobias Brandin, Edo Drenth, Mats Jonasson (VCC), Mathias Lidberg, Artem Kusachov, Anton
Albinsson, Manjurul Islam, Pär Pettersson, Ola Benderius (Chalmers), Mats Sabelström, and Roland Svensson among others.

3
 /Bengt Jacobson, Göteborg, October 2016

Preface 2015
This edition has various changes and additions. Some of these are: brush model with parabolic pressure distribution, typical
numerical data for heavy vehicle, added “2.2.3 Tyre”, “4.5.3.2 Example of explicit form model”, more about tyre relaxation,
introduction of neutral steering point, introduction of steady state roll-over wheel lift diagram. Thanks to Anton Albinsson,
Edo Drenth (VCC), Gunnar Olsson (LeanNova), Manjurul Islam, Mathias Lidberg, Mats Jonasson (VCC), Niklas Fröjd (Volvo
GTT), Ola Benderius, Pär Pettersson, and Zuzana Nedelkova among other. /Bengt Jacobson, Göteborg, 2015

Preface 2014
This edition has various small changes and additions. The largest changes are: Function definitions added and major update
of sections 2.3, 4.1.1, 0, 6.1.1.
Thanks to Lars Almefelt from Chalmers, Jan Andersson from VCC, Kristoffer Tagesson from Volvo GTT and Gunnar Olsson
from Leannova and Karthik Venkataraman. /Bengt Jacobson, Göteborg, 2014

Preface 2013
This edition has various small changes and additions. The largest additions were in: Functional architecture, Smaller vehicles,
Roll-over, Pendulum effect in lateral load transfer and Step steer.
Thanks to Gunnar Olsson from LeanNova, Mathias Lidberg, Marco Dozza, Andrew Dawkes from Chalmers, Erik Coelingh from
Volvo Cars, Fredrik Bruzelius from VTI, Edo Drenth from Modelon, Mats Sabelström, Martin Petersson and Leo Laine from
Volvo GTT. /Bengt Jacobson, Göteborg, 2013

Preface 2012
A major revision is done. The material is renamed from “Lecture notes” to “Compendium”. Among the changes it is worth
mentioning: 1) the chapters about longitudinal, lateral and vertical are more organised around design for vehicle functions,
2) a common notation list is added, 3) brush tyre model added, 4) more organised and detailed about different load transfer
models, and 5) road spectral density roughness model is added.
Thanks to Adithya Arikere, John Aurell, Andrew Dawkes, Edo Drenth, Mathias Lidberg, Peter Nilsson, Gunnar Olsson, Mats
Sabelström, Ulrich Sander, Simone Sebben, Kristoffer Tagesson, Alexey Vdovin and Derong Yang for review reading.
/Bengt Jacobson, Göteborg, 2012

Preface 2011
Material on heavy vehicles is added with help from John Aurell. Coordinate system is changed from SAE to ISO. Minor addi-
tions and changes are also done. /Bengt Jacobson, Göteborg, 2011

Preface 2007
This document was developed as a result of the reorganization of the Automotive Engineering Master’s Programme at
Chalmers in 2007. The course content has been modified in response to the redistribution of vehicle dynamics and power
train education.
These lecture notes are based on the original documents developed by Dr Bengt Jacobson. The text and examples have been
reformatted and edited but the author is indebted to the contribution of Dr Jacobson. /Rob Thomson, Gothenburg, 2007

Keywords: Vehicle Engineering, Automotive Engineering, Vehicle Dynamics, Vehicle Motion, Modelling, Modelica, Simulation

Cover:
Left column, from top: Steering control testing in Jokkmokk, Testing A-double combination vehicle in real transport operation
in the Autofreight research project (photo: Borås Stad), and Alexander Rasch, Chalmers, tests slalom with e-scooter. Right
column: Figures from compendium additions during 2020.

This compendium is also available as pdf file at https://research.chalmers.se/en/person/bengtja#publications

Paper version printed at Repro-centralen, Chalmers, Göteborg, Sweden, 2020

© Copyright: Chalmers University of Technology, Bengt Jacobson

4
 CONTENTS
1 INTRODUCTION 9
1.1 Definition of Vehicle Dynamics 9
1.2 About this compendium 9
1.3 Automotive engineering 10
1.3.1 Vehicle Dynamics Engineers’ Industry Roles 10
1.4 Requirement Setting 11
1.4.1 Attributes 11
1.4.2 Functions 12
1.4.3 Requirements 14
1.4.4 Models, Methods and Tools 17
1.5 Engineering 18
1.5.1 Model Based Engineering 18
1.5.2 Mechanical/Machine Engineering 38
1.5.3 Control Engineering 47
1.5.4 Tools 54
1.6 Vehicle Engineering 61
1.6.1 Vehicle Motions and Coordinate Systems 61
1.6.2 Complete Vehicle Modelling Concepts 67
1.6.3 Vehicle Dynamics Terms 71
1.6.4 Vehicle Architectures 73
1.6.5 Verification Methods with Real Vehicle 77
1.6.6 Verification Methods with Virtual Vehicle 78
1.7 Heavy Trucks 79
1.7.1 General Differences 80
1.7.2 Vehicle Dynamics Differences 80
1.7.3 Definitions 80
1.8 Smaller Vehicles 81
1.9 Notation List 83
1.10 Typical Numerical Data 86
1.10.1 For Passenger Vehicles 87
1.10.2 For Heavy Vehicles 88

2 VEHICLE INTERACTIONS AND SUBSYSTEMS 89

2.1 Introduction 89
2.1.1 References for this Chapter 89
2.2 Wheels and Tyres 90
2.2.1 Introduction 90
2.2.2 Rolling Resistance of Tyres 98
2.2.3 Longitudinal Force 106
2.2.4 Lateral Force of Tyre 122
2.2.5 Combined Longitudinal and Lateral Slip 133
2.2.6 Summary of Tyre Force vs Slip Models 142
2.2.7 Vertical Properties of Tyres 143
2.2.8 Tyre Wear 143
2.3 Suspension System 144
2.3.1 Components in Suspension 145
2.3.2 Axle and Wheel Rates 146

5
 2.3.3 Suspension -- Heave and Pitch 148
2.3.4 Suspension -- Heave and Roll 148
2.4 Propulsion System 150
2.4.1 Modelling Concepts 151
2.4.2 Prime movers 152
2.4.3 Transmissions 156
2.4.4 Clutches and Brakes in Transmission 158
2.4.5 Hydrodynamic Torque Converters 163
2.4.6 Energy Storages and Energy Buffers 163
2.4.7 Special Topology Propulsion Systems 164
2.5 (Wheel) Braking System 167
2.6 (Wheel) Steering System 170
2.6.1 Chassis Steering Geometry 170
2.6.2 Steering System Forces 171
2.7 Environment Sensing System 174
2.8 Vehicle Aerodynamics 174
2.8.1 Longitudinal Relative Wind Velocity 174
2.8.2 Lateral Relative Wind Velocity 175
2.8.3 § Variation of Relative Wind 175
2.9 Driving and Transport Application 176
2.9.1 Mission, Road and Traffic 176
2.9.2 Driver 176

3 LONGITUDINAL DYNAMICS 183

3.1 Introduction 183
3.1.1 References for This Chapter 183
3.2 Steady State Functions 183
3.2.1 Traction Diagram 183
3.2.2 Power and Energy Losses 186
3.2.3 Functions After Start 188
3.2.4 Starting with Slipping Clutch 189
3.2.5 Steady State Vertical Force Distribution between Axles 190
3.2.6 Friction Limit 190
3.2.7 Start Functions 191
3.3 Functions Over (Long) Cycles 193
3.3.1 Description Formats of Vehicle Operation 193
3.3.2 Rotating Inertia Influence on Acceleration 195
3.3.3 Four Quadrant Traction Diagram 197
3.3.4 Functions Over Cycles 197
3.3.5 Load Transfer with Rigid Suspension 202
3.3.6 Acceleration 204
3.4 Functions in (Short) Events 207
3.4.1 Typical Test Manoeuvres 207
3.4.2 Deceleration Performance 208
3.4.3 Pedal Functions 208
3.4.4 Brake Proportioning 209
3.4.5 Heave and Pitch 210
3.4.6 Steady State Heave and Pitch, Non-Trivial Linkage 215
3.4.7 Pitch Functions at Transient Wheel Torques 215
3.4.8 Acceleration and Deceleration 216
3.4.9 Other Functions 218
3.5 Control Functions 219
3.5.1 Longitudinal Control 219
3.5.2 Longitudinal Control Functions 219

6
 3.5.3 Longitudinal Motion Function Architecture 227

4 LATERAL DYNAMICS 229

4.1 Introduction 229
4.1.1 Lateral Model Categorization 229
4.1.2 References for this Chapter 230
4.2 Low Speed Manoeuvres 230
4.2.1 Low Speed Model, Ackermann, without Forces 230
4.2.2 Low Speed Functions 230
4.2.3 Low Speed Model, Ackermann, with Forces 232
4.2.4 Low Speed Model, Non-Ackermann 234
4.2.5 Articulated Vehicles 237
4.2.6 Reversing 238
4.3 Steady State Cornering at High Speed 239
4.3.1 Steady State Driving Manoeuvres 239
4.3.2 Steady State One-Track Model 241
4.3.3 Under-, Neutral- and Over-steering * 246
4.3.4 Required Steer Angle 249
4.3.5 Critical and Characteristic Speed * 250
4.3.6 Steady State Cornering Gains * 252
4.3.7 How Design Influences Steady State Gains 255
4.3.8 Manoeuvrability and Stability 259
4.3.9 Handling Diagram 259
4.3.10 Lateral Load Transfer in Steady State Cornering 260
4.3.11 High Speed Steady State Vehicle Functions 268
4.3.12 Roll-Over in Steady State Cornering 270
4.4 Stationary Oscillating Steering 275
4.4.1 Stationary Oscillating Steering Tests 275
4.4.2 One-Track Model for Two-Axle Vehicle for Transient Dynamics 275
4.4.3 Using the One-Track Model 283
4.5 Transient Driving 292
4.5.1 Transient Driving Manoeuvres * 292
4.5.2 One-Track Models, without Lateral Load Transfer 293
4.5.3 Two-Track Models, with Lateral Load Transfer 306
4.5.4 Step Steering Response * 316
4.5.5 Phase Portrait 319
4.5.6 Long Heavy Combination Vehicles High Speed Functions 320
4.6 Lateral Control Functions 322
4.6.1 Lateral Control 322
4.6.2 Lateral Control Functions 322

5 VERTICAL DYNAMICS 327

5.1 Introduction 327
5.1.1 References for this Chapter 327
5.2 Stationary Oscillations Theory 327
5.2.1 Time as Independent Variable 328
5.2.2 Space as Independent Variable 331
5.3 Road Models 333
5.3.1 One Frequency Road Model 333
5.3.2 Multiple Frequency Road Models 333
5.4 1D Vehicle Models 336
5.4.1 1D Model without Dynamic dofs 336
5.4.2 1D Model with 1 Dynamic dof 338
5.4.3 1D Model with 2 Dynamic dofs 342

7
 5.4.4 One-Mode Models 346
5.5 Functions for Stationary Oscillations 347
5.5.1 Ride Comfort * 347
5.5.2 Fatigue Life * 352
5.5.3 Road Grip * 353
5.5.4 Other Functions 353
5.6 Variation of Suspension Design 354
5.6.1 Varying Suspension Stiffness 355
5.6.2 Varying Suspension Damping 356
5.6.3 Varying Unsprung Mass 356
5.6.4 Varying Tyre Stiffness 357
5.6.5 § Varying Skyhook Damping 358
5.7 Two Dimensional Oscillations 358
5.7.1 Heave and Roll 358
5.7.2 Heave and Pitch 359
5.8 Three Dimensional Oscillations 362
5.9 Transient Vertical Dynamics 363
5.10 Control Functions 363

BIBLIOGRAPHY 365

MODULES IN THE COURSE MMF062 VEHICLE DYNAMICS 369

Modules on Chapter 1 INTRODUCTION 369
Module 1:1 The Subject of Vehicle Dynamics 369
Module 1:2 Modelling 369
Module 1:3 Vehicle Coordinates, Types, Terminology 370
Module 1:4: Virtual Verification 371
Modules on Chapter 2 VEHICLE INTERACTIONS AND VEHICLE SUBSYSTEMS 371
Module 2:1 Tyre Design and Rolling 371
Module 2:2 Tyre Longitudinal Force and Slip 372
Module 2:3 Tyre Lateral and Combined Slip 372
Module 2:4 Tyre Vertical and other Vehicle Interactions 373
Modules on Chapter 3 LONGITUDINAL DYNAMICS 373
Module 3:1 Longitudinal: Steady State 373
Module 3:2 Longitudinal: Long Cycles 374
Module 3:3 Longitudinal: Short Events 374
Module 3:4 Longitudinal: Control Functions and Heavy Vehicle Brake System and ABS 374
Modules on Chapter 4 LATERAL DYNAMICS 375
Module 4:1 Lateral: Linear One-Track Model 375
Module 4:2 Lateral: Low Speed, Ackermann 375
Module 4:3 Lateral: Low Speed, Non-Ackermann, Articulation 376
Module 4:4 Lateral: Steady State Cornering without Load Transfer 376
Module 4:5 Lateral: Steady State Cornering Functions 377
Module 4:6 Lateral: Steady State Cornering with Load Transfer 378
Module 4:7 Lateral: Stationary Oscillating Steering 378
Module 4:8 Lateral: Transient Manoeuvres, incl. Step Steer 379
Module 4:9 Lateral: Vehicle Simulator as Verification Tool 380
Module 4:10 Lateral: Control Functions 380
Modules on Chapter 5 VERTICAL DYNAMICS 381
Module 5:1 Vertical: Suspension and Stationary Oscillation theory 381
Module 5:2 Vertical: Road Models, 1D Vehicle Models 381
Module 5:3 Vertical: Comfort and Grip 381
Module 5:4 Vertical: Heavy Vehicle Suspension 382

8
 Introduction

1 INTRODUCTION
1.1 Definition of Vehicle Dynamics
Vehicle Dynamics is an engineering subject about motion of vehicles in user-relevant operations. The
subject is applied, and applied on a certain group of products, i.e. vehicles. Vehicle Dynamics always
uses terms, theories and methods from Mechanical/Machine engineering, but often also from Con-
trol/Signal engineering and Human behavioural science.

1.2 About this compendium

This compendium is initially written for the course “MMF062 Vehicle Dynamics” at Chalmers Univer-
sity of Technology. The compendium covers more than included in that course; both in terms of sub-
system designs and in terms of some teasers for more advanced studies of vehicle dynamics. There-
fore, the compendium can also be useful in general vehicle engineering courses, e.g. in the Chalmers
course “TME121 Engineering of Automotive Systems”; and as an introduction to more advance
courses, which at Chalmers is the course “TME102 Vehicle Dynamics Advanced”. The reader of this
compendium is assumed to have knowledge of mathematics and mechanics, to the level of a Bachelor
of Engineering degree. Previous knowledge in dynamic systems, often taught well in Control Engineer-
ing courses, is useful.
The overall objective of the compendium is to educate vehicle dynamists, i.e. engineers that under-
stand and can contribute to development of good motion functionality of vehicles. The compendium
focuses on road vehicles, primarily passenger cars and commercial vehicles. Smaller road vehicles,
such as bicycles and single-person cars, are only very briefly addressed. It should be mentioned that
there exist a lot of ground-vehicle types not covered at all, such as: off-road/construction vehicles,
tracked vehicles, horse wagons, hovercrafts or railway vehicles.
Chapter 1 introduces automotive industry and the overall way of working there (1.3, 1.4). Then, 1.5
defines required pre-knowledge from “product-generic” engineering. From 1.6, the Chapter comes
back to specific for vehicles.
The vehicle is a component or subsystem in a superior transport system consisting of other road users,
roads, and transport missions. A vehicle is also, itself, a system within which many components or sub-
systems interact. Chapter 2 describes what interacts with a vehicle from outside, like aerodynamics
and driver.
Vehicle dynamics is about how to quantitatively predict and analyse the complete vehicle’s behaviour.
To do that, a vehicle dynamist has to quantitatively understand how most of the vehicle’s subsystems
works, but not how to design these subsystems. Therefore, Chapter 2 is there to describes the subsys-
tems relevant for vehicle dynamics:
• Wheels and Tyre in 2.2
• Suspension System in 2.3 (and 3.4.5.2 and 4.3.10)
• Propulsion System in 2.4
• (Wheel) Braking System in 2.5
• (Wheel) Steering System in 2.6
• Environment Sensing System in 2.7
Functions are needed for requirement setting (1.4), design engineering and verification (1.6.5, 1.6.6).
The overall order within the compendium is that models/methods/tools needed to understand each
function are placed before the functions. Chapters 3-5 describes (complete vehicle) “functions”, organ-
ised after vehicle motion directions:
• Chapter 3: Longitudinal dynamics
• Chapter 4: Lateral dynamics
• Chapter 5: Vertical dynamics

9
 Introduction

1.3 Automotive engineering

This section is about the context where Vehicle Dynamics is mainly applied, i.e. the automotive indus-
try. OEM means Original Equipment Manufacturer and is, within the automotive industry, used for a
vehicle manufacturer. OEM is a legal status in some countries. In the automotive industry, the word
Supplier means supplier to an OEM. A Tier1 supplies directly to an OEM. A Tier2 supplies to a Tier1
and so on. Primarily, suppliers supply parts and systems to the OEMs, but suppliers can also supply
competence, i.e. consultant services.
From an engineering view, an OEM does Product Development and Manufacturing. But it is good to
remember that there is also Purchasing, Marketing & Sales, After Sales, etc. However, Product Develop-
ment is the main area where the vehicles are designed. It is typically divided into Powertrain, Chassis,
Body, Electrical and (Complete) Vehicle Engineering. Vehicle Dynamics competence is mainly needed
in Chassis, Powertrain and Vehicle Engineering.
On supplier side, Vehicle Dynamics competence is mainly needed for system suppliers that supplies
propulsion, brake, steering and suspension systems. Additional to OEMs and suppliers, Vehicle Dy-
namics competence is also needed in authorities for legislation and testing as well as research insti-
tutes.
There are engineering associations for automotive engineering. FISITA (Fédération Internationale des
Sociétés d'Ingénieurs des Techniques de l’ Automobile, www.fisita.com) is the umbrella organisation
for the national automotive societies around the world. Examples of national societies are IMechE
(United Kingdom), JSAE (Japan), SAE (USA), SAE-C (Kina), SATL (Finland), SIA (France), SVEA (Swe-
den, www.sveafordon.com) and VDI FVT (Germany). There is a European level association also, EAEC.

1.3.1 Vehicle Dynamics Engineers’ Industry Roles

The activity type that sets the pace in automotive industry are vehicle programme or projects. It de-
fines the technology to be developed, the time and cost aspects. The work is organised around such
programmes, both at vehicle manufacturers and their sourced subsystem supplier. One way to exem-
plify such is Figure 1-1. Engineers with vehicle dynamics profile are typically active at departments
called Chassis, Complete Powertrain, Electrical or Vehicle Engineering. They are responsible for deliv-
erables to vehicle programme in the form of:
• Hardware: Geometry, Strength and function per subsystem, ECUs
• Software: SW per subsystem, Functions such as ABS, ESC, ACC
• Requirement setting and verification: Handling, Driveability, Brake performance, Ground
clearance, Ride comfort, Energy consumption. Verification in real and virtual tests.

year1 year2
Pre-Series1 Pre-Series2 Pre- Start
(mule=old (new chassis) Series3 Of
chassis) (final Production
chassis)
Research
projects
Geometry package, Changing geometry, Changing bushings,
Strength of parts, Function content, Tuning SW,
Signal interface, … …
… within subsystems within subsystems
between subsystems (hi-&lo mu tests) (hi&lo-mu tuning)
Figure 1-1: An example of vehicle programme and Vehicle Dynamics related activities.

1.3.1.1 § Vehicle Programme Size and Life Length

A vehicle programme typically aims on a volume of 10..100 thousands of vehicles per year, manufac-
tured over 3-10 years, where the smaller volumes and the longer manufacturing times are more typi-
cal for heavy vehicles. An alternative of traditional vehicle programmes is to develop smaller series, for

10
 Introduction

few selected customers, pilot projects. Some claim that, for paradigm shifts like automated driving and
electromobility, smaller commercial pilots are more suitable than large vehicle programmes. The soft-
ware in the vehicles, can even be updated after sales, which can be seen as a continuous development,
leaving the concept of only develop towards a certain production start date.

1.4 Requirement Setting

Contribution from Ingemar Johansson, CEVT and Vehicle Engineering at ChalmersIngemar Johansson

Development of a vehicle is driven by Requirements, coming from:

• Manoeuvres/Vehicle operations/Use cases, representative for the need of the customers/users
• Legislation from the authorities and Rating from consumer organisation, and
• Engineering constraints from the manufacturer’s platform/architecture on which the vehicle
should be built.
One way of organising requirements is to define Attributes and Functions. The terms are not strictly
defined and may vary between vehicle manufacturers and over time. With this said, it is assumed that
the reader understand that the following is an approximate/exemplifying and simplified description.
In this compendium, both attributes and functions concern the complete vehicle; not the subsystems
within the vehicles and not the superior level of the transportation system with several vehicles in a
road infrastructure.
Attributes and Functions are used to establish processes and structures for requirement setting and
verification within a vehicle engineering organisation. Such processes and structures are important to
enable a good overall design of such a complex product as a vehicle intended for mass production at
affordable cost. Figure 1-2 gives an overview, with reference to the well-known V-process, of how a
vehicle is developed. Note that these kinds of figures are very idealized, and one should neither trust
the process too much nor neglect them. A vehicle is a very complex product. First, one has several lev-
els of functions and subsystems. Secondly, it is difficult to keep a clean hierarchical order between
functions and subsystems. Thirdly and most conceptually difficult, is the fact that each subsystem gets
requirements from many functions as showed by the dashed lines; this makes the complexity explode.
Traditional V-process V-process for a vehicle programme
for product development Product Customer/User
planning perception
Describe Complete Validate
Vehicle Attributes Complete Vehicle
Design loops on
Design Complete Vehicle level
(pre-series, virtual and real)
Set Requirements on Verify Requirements
Complete Vehicle Functions on Complete Vehicle

Decompose in subsystem. Design loops Verify Requirements

Break down to Requirements on Subsystems per Subsystem on Subsystems
Drawn as a “W-process” to stress
the multiplicity of subsystems

Figure 1-2: V-process for a vehicle program. The more design loops are utilized, the more ”agile”.

1.4.1 Attributes
An attribute is a high-level aspect of how the users perceive the vehicle. Attributes which are espe-
cially relevant for Vehicle Dynamics are listed in Figure 1-3. The table is much generalised and the at-
tributes in it would typically need to be decomposed into more attributes when used in the engineer-
ing organisation of an OEM. Also, not mentioned in the table, are attributes which are less specific for

11
 Introduction

vehicle dynamics, such as Affordability (low cost for user), Quality (functions sustained over vehicle
lifetime), Styling (appearance, mainly visual), etc.
A set of Attributes is a way to categorise or group functions, especially useful for an OEM organisation
and vehicle development programs. A set of Functions is a way to group requirements. Legal require-
ments are often, but not always, possible to trace back to primarily one specific attribute. Require-
ments arising from OEM-internal platform and architecture constraints are often more difficult to
trace in that way. Hence, “platform/architecture” is a “requirement container”, beside the attributes.
Attribute Description
This attribute means to maximize output from and minimize costs for transportation.
Transport output can be measured in 𝑝𝑒𝑟𝑠𝑜𝑛 𝑘 𝑡𝑜𝑛 𝑘 or 3 𝑘 . The costs are
Transport Efficiency

mainly energy costs and time, but also wear of vehicle parts influence. The attribute is most
important for commercial vehicles but becomes increasingly important also for passenger
vehicles. The attribute is mainly addressing long-term vehicle usage pattern, typically 10
min to 10 hours. There are diverse ways to define such usages, e.g. (Urban / Highway /
Mixed) driving cycles. So far, the attribute is mainly required and assessed by the vehicle
customers/users.
The attribute can also be seen to include “Environmental Efficiency”, which means low us-
age of natural resources (mainly energy) and low pollution, per performed transport task.
This is to a substantial extent required and assessed by society/legislation.
Minimizing risk of property damages, personal injuries and fatalities both in vehicle and
Safety

outside, while performing the transportation. This attribute is to a considerable extent re-
quired and assessed by society/legislation. In some markets, mainly developed countries, it
is also important for vehicle customers/users.
How the occupants (often the driver) experiences the vehicle during transport; from relaxed
transport (comfort) to active driving (sensation). This attribute contains sub-attributes as:
• Ride comfort. Ride comfort often refers to vibrations and harshness of the occu-
pants’ motion, primarily vertical but secondly longitudinal and lateral. So, V and H in
User Experience (Driver Experience)

NVH (=Noise, Vibration and Harshness) is included. If expanding to “comfort” it

would include also N (noise) and compartment air conditioning, but these sub-attrib-
utes are less related to vehicle dynamics.
• Performance describes how the vehicle can perform at the limits of its capabilities;
acceleration, deceleration or cornering. Most often, it refers to longitudinal limita-
tions due to propulsion and brake systems limitations.
• Driveability, Handling and Road-holding describes how the vehicle responds to in-
puts from driver and disturbances, and how driver gets feedback from vehicle motion
e.g. through steering feel. It is also the corresponding response aspects for a “virtual
driver”, i.e. a control algorithm for automated driving. Driveability often refers to lon-
gitudinal (acceleration, braking gear shifting). Handling and Road-holding often refer
to lateral manoeuvres.
• Trust in automated driving becomes increasingly important and needs to be bal-
anced; high trust but not over-trust.
This attribute is to a considerable extent required and assessed by the vehicle custom-
ers/users, both through own experience but also indirectly via assessments by experts, e.g.
in motor journals.
Figure 1-3: Attributes relevant for Vehicle Dynamics / Vehicle Motion.

1.4.2 Functions
In this compendium, a function is more specific than an attribute. A function should define measures of
something the (complete) vehicle does, so that one can set (quantitative) requirements on each meas-
ure, see 1.4.3. The function does not primarily stipulate any specific subsystem. However, the realisa-
tion of a function in a certain vehicle programme, normally only engages a limited subset of all

12
 Introduction

subsystems. So, the function will there pose requirements on those subsystems. Hence, it is easy to mix
up whether a function origin from an attribute or a subsystem. One way to categorise functions is to let
each function belong to the subsystem which it mainly implies requirements on rather than the source
attribute. Categorizing functions by subsystems tends to lead to “carry-over” function realisations
from previous vehicle program, which can be good enough in many cases. Categorizing functions by
source attribute facilitates more novel function realisations, which can be motivated in other cases.
The word “function” has appeared very frequently lately along with development of electrically con-
trolled systems. The function “Accelerator pedal driving” in 3.5.2.1 has always been there, but when
the design of it changed from mechanical cable and cam to electronic communication and algorithms
(during 1990’s) it became much more visible as a function, sometimes referred to as “electronic throt-
tle”. The point is that the main function was there all the time, but the design was changed. The change
of design enabled, or was motivated by, improvement of some sub-functions, e.g. idle speed control
which works better in a wider range of engine and ambient temperature.
At some places, the compendium emphasizes the functions by adding an asterisk “*” in section head-
ing and a “Function definition” in the following typographic form:
Function definition: {The Function} is the {Measure} … for {Fixed Conditions} and certain {Parameter-
ized Conditions}.
The word “conditions” should be understood as a manoeuvre, operation or use case. It is often possible
and efficient to define multiple measures from one “condition”.
The {Measure} should be one unambiguously defined measure (such as time, velocity or force) of
something the vehicle does. The {Measure} is ideally a continuous, objective and scalar physical quan-
tity, subjected for setting a requirement on the vehicle. The {Fixed Conditions} should be unambigu-
ously defined and quantified conditions for the vehicle and its surroundings. The keyword “certain”
identifies the {Parameterized Conditions}, which need to be fixed to certain numerical values or proba-
bilistic distributions, before using the Function definition for requirement setting, see 1.4.3.
Since the term “Function” is defined very broadly in the compendium, these definitions become very
different. One type of Function definition can be seen in “3.2.3.1 Top Speed *”, which includes a well-
defined measure. Another type of Function definition is found in “3.5.2.3 Anti-Lock Braking System,
ABS *” and “4.3.3 Under-, Neutral- and Over-steering *”. Here, the definitions are more on free-text for-
mat, and an exact measure is not so well defined.

1.4.2.1 § Organising Functionality

There is, of course, not any single physically correct or incorrect way to organise the total functionality
in a vehicle into a finite list or structure of functions. However, defining functions is helpful when or-
ganising the development in a vehicle programme or when inheriting functionality between vehicle
programmes. Examples of categorisation of functions are:
• Functions break-down in tree-structure from attributes. This facilitates development from
user needs and is often argued as the best.
• Functions assigned to the (sub-)system which the function mainly belong (purchase-wise or
technology-wise). Such categories can be, e.g., Brake functions and Steering functions.
• Functions sorted in “Customer functions” and “Support functions”. A Support function is typi-
cally used by many customer functions, see “1.6.4.2 Vehicle Function Architecture”.
• Functions sorted in Standard and Option in the product to end customer.
• Dynamic vs Driver-informing functions, differentiated by whether the function actuates vehi-
cle motion or only inform/warn driver.
An example of tree-structure, starting from the attributes in Figure 1-3, is given below. The attributes
are broken down towards functions. Which level to be called function can be as where a measurable
requirement can be formulated. To indicate this, a unit and a brief description of something measura-
ble is given on the function level which is the lowest end of the branch in the tree. In some places, “sub-
jective assessment” can replace the unit, because far from all functions/requirements are measurable.
The structure is given to give a feeling of approximate width needed for requirement setting of a

13
 Introduction

vehicle. However, the structure below does neither reach the full level of details, nor the amount, of
functions and requirements. Also, the structure does not show any requirements broken down to sub-
systems.
• Transport Efficiency
(These measures require a certain transport operation to be defined, see 3.3.1.)
The inverse of Transport efficiency is Transport cost in [€/ ] or [€⁄(𝑡𝑜𝑛 𝑘 )]. This is the sum
of several terms:
o Energy consumption; which depends on energy [€⁄𝑁 ] or fuel price [€⁄ 3 ], see
3.3.4.1
o Transport time costs; which depends on driver salary for goods transports and on occu-
pant travel time cost [€⁄ℎ] for passenger transports, see 3.3.4.3
o Component wear; e.g. tyre wear (2.2.8), battery degradation, etc. The cost can be quan-
tified via a life quantity (e.g. [ ] tread thickness for tyres) and a DegradationQuantity
(e.g. [ /𝑠] for tyres) and a Replacement cost [€/𝑟𝑒𝑝 𝑒 𝑒𝑛𝑡].
• Safety
(These measures requires a certain manoeuvre or traffic scenario. When adding several
measures, one have to consider severity and probability of those.)
o Speed reduction before collision; in [m/𝑠]
o Avoidance speed; in [m/s] maximum entry speed to the traffic scenario, or to a cone
track, lane change or tightening curve
o Longitudinal Deceleration; in [s] for decelerating from one speed to another speed in a
straight-line or curve.
• User Experience
o Ride comfort
▪ Vertical stationary oscillations; in vertical amplitude per road displacement am-
plitude [(m/𝑠 2 )/m] for a certain irregularity of road at certain speed
▪ Vertical and longitudinal transient shock; in [m/𝑠 3 ] vertical and longitudinal
peak when driving over certain cleat on road in certain speed
o Performance
▪ Longitudinal Acceleration ([s] for accelerating from certain speed to another
certain speed)
▪ Lateral SteadyState Acceleration ([m/𝑠 2 ] when driving in certain radius)
o Driveability, Handling and Road-holding
▪ Longitudinal Acceleration margins ([m/𝑠 2 ] or subjective assessment peak or av-
erage during certain transport task)
▪ Transport time ([s] or subjective assessment around a certain handling track)
▪ Pedal response ([(m/𝑠 2 )/m] or [(m/𝑠 2 )/N]) for step pedal apply
▪ Steering wheel response ([(rad/s)/deg]) or subjective assessment for step or
oscillating steering
o Trust (requirements here are far from well established)
▪ Subjective assessment after longer use (maybe ≥a week) of the vehicle. Many
and/or well-trained test users needed. A questionnaire can be used to organise
subjective assessment in different traffic situation, low/high speed, light/dark,
dry/slippery road, longitudinal interaction with other road users, lateral inter-
action with lane edges, etc.

1.4.3 Requirements
A requirement shall be such that it is possible to verify how well a product fulfils it. A requirement on
the complete vehicle is typically formulated as:
“The vehicle shall … do something or have measure … < 𝑜𝑟 > 𝑜𝑟 ≈ … number [unit] …
… under certain conditions.”
Examples: The vehicle shall…
• … accelerate from 50 to 100 km/h in <5 s when full acceleration pedal. On level road.

14
 Introduction

• … decelerate from 100 to 0 km/h in <35 m when brake pedal is fully applied, without locking
any rear wheel. On straight and level road.
• … turn with outermost edge on a diameter <11m when turning with full steering at low speed.
• … have a characteristic speed of 70 km/h (10 km/h). On level ground and high-friction road
conditions and any recommended tyres.
• … give a weighted RMS-value of vertical seat accelerations < .5 ⁄𝑠 2 when driving on road
with class B according to ISO 8608 in 100 km/h.
• … keep its body above a 0.1 m high peaky two-sided bump when passing the bump in 50 km/h.
To limit the amount of text and diagrams in the requirements it is useful to refer to ISO and OEM spe-
cific standards. Also, it is good to document the purpose and/or use cases with the requirement.
The above listed requirements stipulate the function of the vehicle, which is the main approach in this
compendium. Alternatively, a requirement can stipulate the design of the vehicle, such as “The vehicle
shall weigh <1600 kg” or “The vehicle shall have a wheel base of 2.5 m“. The first type (above listed)
can be called Performance based requirement. The latter type can be called Design based requirement or
Prescriptive requirement and such are rather “means” than “functions”, when seen in a function vs
mean hierarchy. It is typically desired that requirements are Performance based, else they would limit
the technology development in society.

1.4.3.1 § Absolute or Relative Requirements

The requirements above are expressed as absolute, in the sense that they can be assessed for one sin-
gle vehicle. This is necessary for legal requirements. However, it is very difficult to develop verification
methods, especially virtual methods, as good as needed for absolute requirement verification. Hence, it
is common to compare, which means relative requirements. The comparison is typically a comparison
of the same defined measure with a competitor vehicle or previous development loop. Rewriting the
first listed requirement from above gives this example of relative requirement:
The vehicle shall…
• … accelerate from 50 to 100 km/h in smaller time than a certain competitor (or our vehicle un-
der development, in previous development loop) when full acceleration pedal. On level road.

1.4.3.2 § Experiment Matrix

This section claims that an Experiment Matrix (or Test Matrix) often is useful and proposes how to rea-
son when generate such a matrix.
In the previous description of requirements it was assumed that one set of numerical parameters can
represent all relevant manoeuvre/operation. In most cases this could lead to overfitting of design pa-
rameters, so that the vehicle does not perform well in real use. Therefore, one often need to vary the
manoeuvre/operation parameters over multiple experiments. We can use an Experiment Matrix to se-
lect variations within the manoeuvre/operation space. Before deciding the experiment matrix it is of-
ten good to start from a Manoeuvre/Operation Parameter Span Matrix. The figure shows an example:

15
 Introduction

Manoeuvre/Operation Parameter Span Matrix

Manoeuvre Road Use
initial speed acceleration (down) radius road-to-tyre road friction Payload
… … …
[km/h] [m/s^2] grade [m] peak friction variation [ton]

80 -8 0 0.8 (hi) constant 0

split,
-2 5 deg 0.2 (lo) 25
left/right=lo/hi
step up
step down

Experiment Experiment Matrix (or Test Matrix)

Name1 22 -8 0 0.8 constant 25
Name2 22 -8 0 0.8 constant 0
Name3 22 -8 0 0.2 constant 25
Name4 22 -8 0 ⁄ 0.2 constant 25
Name5 22 -8 0 ⁄ 0.2 step up 25

Figure 1-4: § An example of “Span Matrix” and “Experiment Matrix” for brake performance. Span
matrix could generate 4 experiments. Experiment matrix shows that only 4 of
them are chosen.
Note that variation of (vehicle) design parameters is very essential to find best product design. So, each
experiment needs to be performed for many vehicle designs. This can be documented either as adding
more columns and rows to the experiment matrix or in a vehicle design alternative matrix, separate
from the experiment matrix. The latter way is proposed with the motivation that vehicle design alter-
natives rather belongs to the design/optimization process than the requirement setting, see Figure
1-8. If separate experiment matrix with 𝑛 rows and vehicle design alternative matrix with rows, the
total number of experiments to carry out will be maximum 𝑛 .
Also note that the required numerical value of the requirement measure (e.g. braking distance) typi-
cally change between the experiments (e.g. differently long braking distance is required for different
road friction). It is even so that, in situations where we have use for an experiment matrix, it is typi-
cally more relevant with comparisons of the measure between different vehicle designs than checking
if the measure is > or < an absolute value.

1.4.3.3 § Requirement Setting in Hierarchical Products

In Figure 1-2, it is indicated that requirements on the vehicle is decomposed into requirements on sub-
systems. There are several examples of this in vehicles; requirements on vehicle might have to be bro-
ken down into requirements on:
• Actuators (prime mover, transmission, brake, steering)
• Environment sensors (video cameras, radars, GPS receivers, etc)
• Suspensions (front axle suspension, rear axle suspension),
• Body (for aerodynamic and strength and compliance)
Of these subsystems, the suspensions are probably the ones which is most complex to handle as sub-
systems in the development process, because the designs of the suspensions influence very many re-
quirements on the vehicle. Therefore, the suspension is used as example below.
To decompose a requirement on the vehicle in requirements on two or more subsystems is a design
activity in itself. No concrete design parameters (“hardpoints” etc) are decided, but functionality (“K&C
data”) is decided and distributed between subsystems, in the example: between front and rear axle.

16
 Introduction

Requirements Requirements Requirements on

on vehicle: on subsystems: sub-subsystems
(typically to supplier)

Reqmnt
Front axle Design
Reqmnt
suspension design Decision Reqmnt
(High level)
Vehicle activity by “front
Reqmnt Design suspension engineers”
design Decision
activity by Reqmnt
Rear axle Design
“attribute Reqmnt
engineers” suspension design Decision
activity by “rear Reqmnt
suspension engineers”
1 2 2
3

Figure 1-5: § Requirement decomposition as a design activity. A design decision on one level
generate the requirement to the next level. Here exemplified with front and rear axle. Red arrows
indicates different design loops.
Note that different experience and different computation/simulation models are useful in the first and
second design step. In each design step, the models needed has to verify how well a certain design of a
system fulfils the requirements on the same system, where the system can be ether complete vehicle,
front suspension or rear suspension. When all low-level design is done, a 3rd type of model is useful, to
verify how well certain designs of all subsystems fulfil the requirements on the complete vehicle. The
last-mentioned verification is traditionally done in prototype vehicles, but with a good virtual verifica-
tion architecture in at the vehicle manufacturer, it becomes increasingly possible to do virtually. The
following figure is a development of Figure 1-2 to show the 3 verification/design loops, which typically
requires different models and computations. Note that number 3 also requires organisation and data-
bases and tools, i.e. a virtual verification architecture.
Product planning Customers / Users
Describe Complete Use of
Vehicle Attributes Vehicle
Complete Vehicle Attributes Designed Vehicle
Quantify Requirements on Validate
Complete Vehicle Complete Vehicle
Requirements on Complete Vehicle Functions 3 Designs of all Subsystem
1 Verify Subsystem Design
Decompose for Requirements on
Requirements Complete Vehicle

Requirements on each Subsystem

Designs of each Subsystem
2
Design Front
2
Suspension
Design Rear
Suspension

Figure 1-6: § V-process where 3 types of verification is shown: 1,2 and 3. Compare with Figure 1-2.
Red arrows indicates the same design loops as in previous figure.

1.4.4 Models, Methods and Tools

The attributes, functions and requirements are top level entities in vehicle development process. But
to design and verify, the engineers need knowledge in form of models, methods and tools.
As mentioned above, some sections in the compendium have an asterisk “*” in the section heading, to
mark that they explain a function, which can be subject for complete vehicle requirement setting. The
remaining section, without an asterisk “*”, are there to give the necessary knowledge (models, meth-
ods and tools) to verify (including understand) the function and the requirements on it. It is the inten-
tion that the necessary knowledge for a certain function appears before the description of that

17
 Introduction

function. One example is that the “2.4 Propulsion System”, “3.2.1 Traction Diagram” and “3.2.2 Power
and Energy Losses” are placed before “3.2.3.1 Top Speed *”. Functions only appear in Chapters 3, 4 and
5.

1.5 Engineering
Engineering (or Design Engineering, in Swedish often “Ingenjörsvetenskap”, in German “Ingenieurwis-
senschaft”), as a science has an important portion of Synthesis. As support for synthesis, it also relies
on Analysis, Inverse analysis or (Nature) Science, see Figure 1-7. Figure 1-8 distinguishes between Anal-
ysis and Synthesis, which shows that Analysis is a step in the whole design loop. The actual Design step
requires Synthesis. The overall purpose is to propose a design, e.g. numerical values of Design parame-
ters of a product. The distinction between Analysis and Inverse Analysis can be made when there is a
natural causality (“cause-effect-direction”).
In 1.5 some useful general methods and tools are described. Parts of these are probably repetition for
some of the reader’s previous education, in mechanical and control engineering. In the end of 1.5,
there is a stronger connection to the vehicle engineering.

1.5.1 Model Based Engineering

An analysis step as in Figure 1-8, can use either real or virtual verification. For real verification one
need to build prototypes. For Virtual Verification, models are needed. A (dynamic) model is a repre-
sentation of something from real-world varying over a time interval, such as a car during longitudinal
acceleration from 0..100 km/h. Models are always based on assumptions, approximations and/or sim-
plifications. However, when using models as a tool for solving a particular problem, the models at least
have to be able to reproduce the engineering problem one is trying to solve. Also, models for engineer-
ing have to reflect design changes in a representative way, so that new designs can be evaluated. Too
detailed models tend to be a less useful, since they require and produce a lot of data.
The models typically used in vehicle dynamics can be called physical dynamic models (many alterna-
tive names: cyber-physical, functional, lumped, discretized, system, DAE-model, ODE-models, etc.). The
models are typically multi-domain type, involving mechanics, hydraulics, pneumatics, electrics, chemi-
cal, control algorithms, electronic, computation, driver’s actions, etc. Examples of modelling methods
which more seldom are used directly in vehicle dynamics are: Finite Elements, Computational Fluid
Dynamics, CAD geometry models, etc.
One can identify modelling in 3 stages in the overall process: 1.5.1.1.2 Physical Modelling, 1.5.1.1.3
Mathematical Modelling, and 1.5.1.1.4 Explicit Form Modelling. See Figure 1-9. The compendium
spends most effort on the first 2 of those 3.
Physical models are assumed. As opposite to this, one can think of formulating a mathematical model
without motivation from established physics. (Without a physical model, the parameters are generally
not interpretable to real design parameters; an indication of this is when the modeller does not know
the units of parameters and variables in the model, and the parameters needs real tests to be found.
Methods for such modelling can be regression, machine learning, etc and such models can only be used
for interpolation, not extrapolation.) However, for vehicle engineering, the vehicle model should be
physical, so that its real-world design parameters can be identified. However, a driver model can be
useful also without strict physical model, since the driver is not to be designed during vehicle engi-
neering. But it is important to consider that the driver do change its behaviour when vehicle design
parameters are changed.

18
 Introduction

Input Output
System

Laws of Nature
Given Find Process
Input, Laws of Nature, System Output Analysis
Output, Laws of Nature, System Input Inverse Analysis
Input, Output, System Laws of Nature (Nature) Science (induction)
Input, Output, Laws of Nature System Engineering Design (Synthesis)
Dixon, J.R., (1966) Design Engineering, Inventiveness, analysis and Decision Making
Figure 1-7: Engineering Design and related activities. Picture from Stefan Edlund, Volvo Trucks.

Vehicle Design
Requirements (Required output)

Requirements Design or Virtual verification (Analysis)

Design
(Required output re-design parameters
for certain input) (Synthesis)
Model
certain (incl. Laws of Require- yes Proposed
achieved
input Nature)
output ments design
fulfilled? parameters
no
Experience/understanding

Design parameters are varied to find best design Operation/Manoeuvre parameters are
(optimization) or good enough design (satisficing). varied in order to make a robust design.

Figure 1-8: How Analysis/Synthesis and Design/Operation parameters appears in Vehicle Design.

1.5.1.1 Stages in (Dynamic) Model Based Engineering

1.5.1.1.1 Formulating the Engineering Design Task
Based on a problem description, one formulates the engineering design task, which describes the re-
quired design decisions (if possible, appoint design parameters) for an existing vehicle or a concept.
Also, requirements on how the output of the system should react in certain vehicle operations (or ma-
noeuvres or situations) have to be identified. Requirements can be either constraints (some-
thing<number) and/or optimization (some scalar to be minimized).
The conceptual idea with requirements in vehicle industry is that they are set independently of the de-
sign solution. However, the general requirements are seldom neither enough for a certain engineering
design task; one often have to reformulate or add requirements. The operation parameters often have
a range/spread/stochasticity to design for, see Figure 1-8. Examples of such parameters describe traf-
fic situation, driver, tyre/road characteristics, weather, see 1.5.1.3.1. The range can be searched for in
feedback or logged data from customers and accident statistics. But it is also very important with expe-
rienced engineering judgment, to forecast how the problem will appear in a future context, with a fu-
ture fleet of vehicles and a future road infrastructure. Design methodology can be used to reason about
and categorize parameters (e.g. Taguchi’s: “signal, response, noise and control factors”).
One design task is typically influenced by several requirements and can therefore need multiple mod-
els of the same system. This stresses the importance of parameterization, which help to secure that the
same design is assessed in the different models.

19
 Introduction

“Modelling for Computation/Simulation

design guidance”
Interpret results,
Explicit form modelling including judge
model validity
Mathematical modelling
Theoretical world
Physical modelling
Real world

Formulate
Evaluate
engineering Design/Re Real-world requirement
OK Propose
task -design testing fulfilment Design
(problem)
NOK

Figure 1-9: (Dynamic) Modelling stages. Real versus Theoretical world. The dashed boxes in the
background indicate that same design has to fulfil multiple requirements, which might require
multiple models.
1.5.1.1.2 Physical Modelling
In this stage, one should generate a physical model, which in this compendium means sketches and
text. Some call this an engineering model and some do not call it a model at all. Free-body diagrams
(1.5.2.1), data flow diagrams (1.5.1.8) and operating conditions (1.6.2) are always important ingredi-
ents in the physical model of a vehicle, but also other domains than mechanical are often represented,
such as hydraulics, electrics, driver and control algorithms. The physical model shall clarify assump-
tions/approximations/simplifications, e.g. rigid/elastic, inertial/massless, continuous/time sampled
algorithms, small angles, etc. Discrete dynamics, see 1.5.1.4, such as ideal dry friction, backlash or end
stop does not exist in real world, so such assumptions are typical examples of clear differences be-
tween real system and physical model. How to model here should be based on what phenomena is
needed to be captured and what variations (e.g. which design parameters) are intended for simula-
tions with the model. The parameters and variables are first defined in the drawig, but might need ad-
ditional text. To distinguish between whether a physical quantity is a parameter or a variable, see
1.5.1.3, is a very important modelling decision. A physical model is often not executable, but it can be.
One example of executable format is a model build in an MBS tool, see 1.5.4.4 and Figure 1-39.
1.5.1.1.3 Mathematical Modelling
In this stage, one should generate a mathematical model, which means equations. Basically, it is
about finding the right equations; equally many as unknowns. For dynamic systems, the unknowns are
unknown variables of time. In the mathematical model, the assumptions are transformed into equa-
tions. For dynamic systems, the equations form a “DAE” (Differential-Algebraic system of Equations). It
is seldom necessary to introduce derivatives with respect to other independent variables, such as posi-
tions, i.e. one does seldom need PDE (Partial Differential Equations). The general form of a DAE:
• 𝒇𝑫𝑨𝑬 (𝒛 𝒛 𝑡) 𝟎
(Here, the function 𝒇𝑫𝑨𝑬 is an operator which include only algebraic operations. An alternative nota-
tion would be 𝒇𝑫𝑨𝑬𝟐 (𝒛(𝑡) 𝑡 𝒑) 𝟎 , but then the function 𝒇𝑫𝑨𝑬𝟐 is an operator which include both al-
gebraic and differential operations.)
The 𝒛 are the (dependent) variables and 𝑡 is time (independent variable). The mathematical model is
complete only if there are equally many (independent) equations in 𝒇𝑫𝑨𝑬 as there are variables in 𝒛,
dim(𝒇𝑫𝑨𝑬 ) dim(𝒛). (Since DAE, we don’t count 𝒛 and 𝒛 as different variables.) Note that one can be
interested if treating also cases where dim(𝒇𝑫𝑨𝑬 ) < dim(𝒛). It is when dim(𝒛) dim(𝒇𝑫𝑨𝑬 ) of the var-
iables in 𝒛 can be considered as known, so called, input variables (𝑡). The notation “(𝑡)” marks that ,
generally, might have to be known DAE variables, i.e. their time derivatives also have to be known.
It is not only a question of finding suitable equations, but also to decide parameterisation, which is how
parameters are defined and related to each other. Parameterisation should reflect a “fair” comparison

20
 Introduction

between different design parameters, which often requires a lot of experience of the product and the
full set of requirements on the vehicle. It is here the engineering science comes in, while the equations
is rather physics and mathematics. To underline the parameters, 𝒑, they can be included in the DAE
form as 𝒇𝑫𝑨𝑬 (𝒛(𝑡) 𝒛(𝑡) 𝑡 𝒑) 𝟎 .
Selection of output variables is important so that output variables are enough to evaluate the require-
ments on the system. Selecting more might drive unnecessary complex models.
The Mathematical model is acausal, i.e. describes relations between the variables, not how and in
which order they are computed. A Mathematical model can be equations on a piece of paper or in a
word processing format and then it is not executable. However, writing the Mathematical model in
Modelica format makes it executable using Modelica tools, see 1.5.4.5.
§ Differentiation Order
Above, the ODEs are assumed to be of first order. There is a strong tradition to in mechanical engineer-
ing to model with 2nd order differential equations, where accelerations of inertial bodies appear as 2 nd
derivative of position. However, numerical methods for solving first order differential equations are
much more mature and Vehicle Dynamics mainly aims at such. Hence, the compendium aims at first
order differential equations.
It is easy to go from few equations to higher order to many first order differential equations; variables
appearing in 2nd order derivative ̈ are replaced with 𝑛 𝑤 and one equation 𝑛 𝑤 is added.
The opposite is not generally as easy, but often possible. An example is a linear model 𝒛 𝑨 𝒛 , which
can be converted from many 1st order to fewer 2nd order becomes as follows:
𝒛 𝑨 𝑨𝟏𝟐 𝒛𝟏 differentiate eliminate
[ 𝟏 ] [ 𝟏𝟏 ] [𝒛 ] ⇒ { }⇒{ }⇒
⏟𝒛𝟐 ⏟𝑨𝟐𝟏 𝑨𝟐𝟐 ⏟ 𝟐 equation for 𝒛𝟏 𝒛𝟐 (and 𝒛𝟐 )
𝒛 𝑨 𝒛
⇒ 𝒛̈ 𝟏 (𝑨𝟏𝟏 + 𝑨𝟏𝟐 𝑨𝟐𝟐 𝑨𝟏𝟐𝟏 ) 𝒛𝟏 + 𝑨𝟏𝟐 (𝑨𝟐𝟐 𝑨𝟏𝟐𝟏 𝑨𝟏𝟏 𝑨𝟐𝟏 ) 𝒛𝟏 𝟎
Another way to go from higher order to lower order differential equations is to go via Laplace trans-
form. This is exemplified on same equations as above, but only for scalar 𝒛𝟏 and 𝒛𝟐 :
𝒛 𝐴 𝐴2 Laplace 𝐴 𝐴2
[ 𝟏] [ ] [ ] ⇒{ }⇒ 𝑠 [ ] [ ] [ ] ⇒
⏟𝒛𝟐 ⏟𝐴2 𝐴22 ⏟ 2 transform 2 𝐴2 𝐴22 2
𝒛 𝑨 𝒛
𝐴 𝐴2
⇒ ([ ] 𝑠 𝑰) [ ] 𝟎 ⇒ {eliminate 2 } ⇒
𝐴2 𝐴22 2
⇒ 𝑠2 (𝐴 + 𝐴22 ) 𝑠 + (𝐴 𝐴22 𝐴 2 𝐴2 ) 0 ⇒
inverse
⇒ { Laplace } ⇒ ̈ (𝐴 + 𝐴22 ) + (𝐴 𝐴22 𝐴 2 𝐴2 ) 0
transform

§ Affine and Linear form (ABCD form)

Contribution from Fredrik Bruzelius, Vehicle Dynamics at Chalmers and VTI

In some cases, the model can be written on an Affine (Explicit) Form:

• 𝒙 𝒌𝒙 + 𝑨 𝒙 + 𝑩 𝒚 𝒌𝒚 + 𝑪 𝒙 + 𝑫 where 𝑨 𝑩 𝑪 𝑫 are matrices.
Note that the notation 𝒙 𝒚 , introduced in 1.5.1.1.4, is used already here in 1.5.1.1.3.2. This is to com-
ply with notation well established in control engineering, which is relevant for 1.5.1.1.3.2.
If the constant terms 𝒌𝒙 and 𝒌𝒚 are zero the special case Linear form (or ABCD form) is at hand:
• 𝒙 𝑨 𝒙+𝑩 𝒚 𝑪 𝒙+𝑫 where 𝑨 𝑩 𝑪 𝑫 are matrices.
To find the numerical solution 𝒚(𝑡) it does not matter much which of the above forms that is used. To
detect instability, one compute eigen-values of 𝑨 and 𝑨 is same in both forms. So far, the affine form is
enough. However, many methods in frequency analysis, control engineering and signal processing re-
quires the linear form. The property linear, which we are after with the form linear, can be described
as that the effect from two superimposed input signals superimposes as: [𝑥 𝑦] 𝑓 (𝑢 + 𝑢2 )
{𝑢𝑠 𝑛 𝑡ℎ𝑒 𝑠𝑝𝑒 𝑝𝑟𝑜𝑝𝑒𝑟𝑡𝑦} 𝑓 (𝑢 ) + 𝑓 (𝑢2 ) . The constant terms would ruin this property,

21
 Introduction

because the effect from constant terms would obviously appear twice in 𝑓 (𝑢 ) + 𝑓 (𝑢2 ) but only
once in 𝑓 (𝑢 + 𝑢2 ). So, it is often motivated to look for a linear form.
If 𝑨 is non-singular we can find a form which is linear in state derivatives by offsetting 𝒙 by replacing 𝒙
with (𝒙𝟎 + 𝒙𝒐𝒇𝑨 ), where 𝒙𝟎 𝑨 𝟏 𝒌𝒙 . The result becomes:
• 𝒙𝒐𝒇𝑨 𝑨 𝒙𝒐𝒇𝑨 + 𝑩 𝒚 (𝒌𝒚 + 𝑪 𝒙𝟎 ) + 𝑪 𝒙𝒐𝒇𝑨 + 𝑫
⏟
𝒌𝒚𝒐𝒇𝑨

If 𝑨 is singular but 𝑩 is non-singular, we can find a form which is linear in state derivatives by offsetting
by replacing with ( 𝒐𝒇𝑩 + 𝟎 ), where 𝟎 𝑩 𝟏 𝒌𝒙 . The resulting form becomes:
• 𝒙 𝑨 𝒙+𝑩 𝒐𝒇𝑩 𝒚 (𝒌𝒚
⏟ 𝑫 𝑩 𝟏
𝒌𝒙 ) + 𝑪 𝒙 + 𝑫 𝒐𝒇𝑩
𝒌𝒚𝒐𝒇𝑩

It is often enough to find a form which is linear in state derivatives, as shown above. However, if both 𝑨
and 𝑫 are non-singular, we can find a Linear Form, i.e. linear in both state derivatives and outputs, i.e.
without both 𝒌𝒙 and 𝒌𝒚 . This is done by offsetting by replacing 𝒙 with (𝒙𝒐𝒇𝑨𝑫 + 𝒙𝟎 ) and with
𝒙𝟎 𝟏 𝒌
𝑨 𝑩
( 𝒐𝒇𝑨𝑫 + 𝟎 ), where [ 𝟎] [
𝑪 𝑫
] [ 𝒙]
𝒌
. The resulting form becomes:
• 𝒙𝒐𝒇𝑨𝑫 𝑨 𝒙𝒐𝒇𝑨𝑫 + 𝑩 𝒐𝒇𝑨𝑫 𝒚 𝑪 𝒙𝒐𝒇𝑨𝑫 + 𝑫 𝒐𝒇𝑨𝑫
(The used terms, affine and linear, can be confusing because the common meaning of the word linear
can include a constant term, while absence of constant term would instead be called proportional.)
If the model is affine (or linear), 𝑨 𝑩 𝑪 𝑫 can be found by identification of the coefficients for the com-
ponents of 𝒙 and . Manual manipuations or symbolic tools can be used. If the model is non-linear, one
can sometimes motivate to neglect small terms so that the model becomes affine. A more general and
automated way to find an affine model is to make a Taylor expansion (symbolic or numerical) around
(𝒙𝟎 𝟎 𝑡 ). If the model is affine and there is no explicit dependency of 𝑡, the Taylor expansion will be
exact:
𝒙 𝒇𝑶𝑫𝑬 (𝒙 ) {affine}
𝜕𝒇𝑫𝑨𝑬 𝜕𝒇𝑫𝑨𝑬
𝒇𝑶𝑫𝑬 (𝒙𝟎 𝟎) +( )|𝒙=𝒙𝟎 (𝒙 𝒙𝟎 ) + ( )|𝒙=𝒙𝟎 ( 𝟎) {affine}
𝜕𝒙 𝜕
= 𝟎 = 𝟎
𝒇𝑶𝑫𝑬 (𝒙𝟎 𝟎 ) + 𝑨 (𝒙 𝒙𝟎 ) + 𝑩 ( 𝟎) where 𝑨 and 𝑩 independent of 𝒙𝟎 and 𝟎 ⇒
⇒ 𝒙𝒐𝒇 𝒇𝑶𝑫𝑬 (𝒙𝟎 𝟎 ) + 𝑨 ⏟ (𝒙 𝒙𝟎 ) + 𝑩 ⏟
( 𝟎)
𝒙𝒐𝒇 𝒐𝒇

If (𝒙𝟎 𝟎 ) is a steady state condition it means same as that 𝒇𝑶𝑫𝑬 (𝒙𝟎 𝟎 ) 0, so the constant term line-
arization gets the form linear in state derivatives 𝒙𝒐𝒇 . Similar Taylor expansion can be done with 𝒈 for
outputs , to find a linear (or at least affine) from also for outputs.
Models with explicit dependency of 𝑡 are treated in (Bruzelius, 2004).
Note that the model was assumed to be affine. This meant that 𝑨 and 𝑩 becomes independent of 𝒙𝟎
and 𝟎, as opposed to a non-linear model, see 1.5.1.1.9.1.
1.5.1.1.4 Explicit Form Modelling
In this stage, one should generate an Explicit Form model, which means equations rearranged to as-
signment statements, i.e. to an explicit form (algorithm) which outputs the state derivatives. The ex-
plicit form model is generally causal. You probably recognize this formulation as “ODE” (Ordinary Dif-
ferential Equation) or “IVP” (Initial Value Problem). So, the general form is:
• 𝒙 ← 𝒇𝑶𝑫𝑬 (𝒙 𝑡) 𝒚 ← 𝒈(𝒙 𝑡)
(Here, the function 𝒇𝑶𝑫𝑬 is an operator which include only algebraic operations.)
We have essentially the same set of equations in the Mathematical model as in the Explicit Form
model. However, seen as Explicit Form model, we count 𝑥 and 𝑥 as different variables. So far, this
means that we have too few equations. To keep equality between number of variables and equations,
we can count in also the equations 𝑥 ← ∫ 𝑥 𝑡 .

22
 Introduction

The 𝒙 is the state variables, is the input variables (𝑡), and 𝒚 the output variables. In the mathemati-
cal model, there was no distinction between different dependent variables in 𝒛. However, to reach the
explicit model, each variable in z has to be identified as belonging to either of 𝒙 𝒚 or . Simply speak-
ing, states 𝒙 are the variables which appears both as and , inputs are variables that cannot be
solved for within the system of equations and outputs 𝒚 are the remaining variables. Since the state
variables are identified, the Explicit Form is sometimes called State Space Form.
A dataflow diagram (see Figure 1-16 and Figure 1-36) is a graphical representation of the explicit
form. It is drawn using blocks with input and output ports and arrows representing signals between;
integration is represented by integration blocks with 𝒙 as input and 𝒙 as output.
§ Semi-Explicit form
If the explicit form cannot be found, there are time integration methods also for solving the semi-ex-
plicit form: 𝒙 ← 𝒇 𝒆 𝒊 (𝒙 𝒚 (𝑡) 𝑡) 𝟎 𝒈 𝒆 𝒊 (𝒙 𝒚 (𝑡) 𝑡) .
A generalisation of the semi-explicit form is when we cannot even find explicit expression for all state
derivatives 𝒙. We can still formulate an Explicit form model, but then using “implicit form expres-
sions”. As extreme example, when no variables can be solved for:
• [𝒙 𝒚] ← 𝑠𝑜 𝑒(𝒇𝑶𝑫𝑬 𝒊 𝒑𝒍 (𝒙 𝒙 𝒚 (𝑡) 𝑡) 𝟎)
The operator “𝑠𝑜 𝑒” is here meant to be implemented as a (numerical) iteration during the Computa-
tion (compare 1.5.1.4.3). The only difference to the original 𝒇𝑫𝑨𝑬 is that the states 𝒙 are identified
among the variables 𝒛. But, it should also be noted that this is no general cure to avoid doing Mathe-
matical manipulations, because far from all iterations succeed; so, explicit form expressions should be
strived for as far as possible! Often, some parts of the Explicit form can be explicit form and only a few
variables appear in a 𝑠𝑜 𝑒 expression. When implicit expressions appear in the Explicit form model,
we have a situation where causality is not defined for the involved variables, although the model is on
an overall Explicit form.
§ Implicit form
If neither explicit or semi-explicit form cannot be found, there are time integration methods also for
solving the implicit form: 𝒓𝒆 𝒊𝒅 𝒂𝒍 ← 𝒇𝒊 𝒑𝒍𝒊𝒄𝒊𝒕 (𝒙 𝒙 𝒚𝟏 𝒚𝟐 𝒚𝟐 (𝑡) 𝑡) . The difference with the model
in 1.5.1.1.3 is that the states 𝒙 are identified. The 𝒓𝒆 𝒊𝒅 𝒂𝒍 𝟎 is solved numerically and iteratively
yielding {𝒙 𝒚𝟏 𝒚𝟏 𝒚𝟐 } at each model evaluation. Then 𝒙 ← ∫ 𝒙 𝑡 . Both 𝒚𝟐 and 𝒚𝟐 are found without
time integration, which means that dim(𝒚𝟏 ) extra equations are needed compared to 𝒇𝑫𝑨𝑬 in 1.5.1.1.3.
These are found through differentiation of some equations, see more 1.5.1.6.
An alternative solution method time integrates also 𝒚𝟐 ← ∫ 𝒚2 𝑡 without fulfilling the extra equa-
tions. Instead the extra equations are used to correct 𝒙 and 𝒚𝟐 to eachother between each time step.
This is used in many MBS tools, such as Adams, see 1.5.4.4.
Time Sampling and Time Events
In vehicle dynamics, the dynamics occurring in the digital control is essential. A control algorithm can
of course be modelled with continuous equations, but the effect of time sampling and signal communi-
cation delays are often essential to include. A mathematical method to model this phenomenon is
“time events”.
Time events are similar to, but not same as, discrete dynamics in 1.5.1.4. The similarity is that both
time events and state events (used for discrete dynamics) uses transition conditions, 𝒉 > 𝟎. But for
time events, the transition conditions are only a function of time 𝒉(𝑡). Time events appear since algo-
rithms are executed on digital processors or are delayed sin signal communication. Time events are
generally more established and easier to get physically consistent than the more general state events,
𝒉 𝒉(𝒙𝒄 𝒙𝒅 𝑡).
In Modelica, a continuous model with its time sampled controller can be modelled as in Figure 1-10.
The example is speed control of a vehicle in uphill. The longer sample time gives an unstable solution.

23
 Introduction

Sampl_100ms.vx Sampl_1100ms.vx Sampl_100ms.axReq Sampl_1100ms.axReq

25
model SampledAlgorithm
parameter Real m=1500, RollGradeResist=(0.01+0.05)*m*9.80665;
20
Real vx, axReq, Fx;
parameter Real SampleTime, PGain=1/0.5; //0.5 is seconds
initial equation
vx = 10;
15
solid: SampleTime=100 ms
equation 10
dashed: SampleTime=1100 ms
//vehicle (continuous DAE)
m*der(vx) = min(Fx, 0.8*m*9.80665) - RollGradeResist;
5

//control and actuation (time sampled discrete DAE)

when sample(0, SampleTime) then 0
axReq = Pgain*(20 - pre(vx));
Fx = m*axReq; //Model, without resitances -5
end when;
end SampledAlgorithm;
-10
0 1 2 3 4 5

Figure 1-10: Dynamic model with time sampled part.

Discretization of Continuous Models
When a simulation tool solves a dynamic model it typically does time discretization, but as engineers
we still see the model as continuous. However, continuous dynamic models often have to be intention-
ally (time) discretized; as they shall become parts of the product, the vehicle. The integral part of a PID
controller is one example. Another example is when an algorithm is derived as continuous, which is
often the case with filters and estimators, 1.5.3.5. Yet another example is when the algorithms shall
perform simulations of dynamic models which are derived as continuous. An example of converting a
continuous dynamic model to a time discrete is shown in Figure 1-11.
1.5.1.1.5 Computation
Computation, is to find a numerical solution for all variables, also called (time) trajectories. Several
ways can be of interest, e.g.:
• (Time domain) Simulation (e.g. Initial value problem, IVP or End value problem, EVP). There are
many advanced pre-programmed integration methods which one can use without knowing the
details. It is often enough to understand the concept of the simplest “Euler forward with con-
stant time step”, in which the state variables are updated in each time step, Δ𝑡, as follows:
𝒙𝒏𝒐𝒘 ← 𝒇𝑶𝑫𝑬 (𝒙(𝑡𝑛 𝑤 ) (𝑡𝑛 𝑤 ) 𝑡𝑛 𝑤 ) (Explicit form model)
𝒙(𝑡𝑛 𝑤 + Δ𝑡) ← 𝒙(𝑡𝑛 𝑤 ) + Δ𝑡 ∙ 𝒙𝒏𝒐𝒘 (Derivative approximation)
Typically, the time step Δ𝑡 is varied by the integration methods itself to minimize computation
time, but maintaining a certain accuracy. The Δ𝑡 should neither be mixed up with the output in-
terval (how dense in time the simulation results are logged) nor the sample time in time discrete
systems, see 1.5.1.4. A general advice to check that you have got a correct enough solution is to
tune the tolerance to as large as possible without changing the character of the solution. Exam-
ples of, so called forward and backward methods are given in Figure 1-11.

24
 Introduction

model DiscreteIntegrator u yc ydF ydB

parameter Real SampleTime=0.5; 0.25
Real u, yc,
initial equation
udF, ydF, prederydF, udB, ydB, derydB; 𝑦 𝑜𝑛𝑡 𝑛𝑜𝑢𝑠
0.20
yc=0.2; pre(ydF)=yc; pre(ydB)=yc;
equation 𝑦 𝑓𝑜𝑟 𝑟 𝑢 𝑒𝑟
//input:
u=0.1*sin(2*time);
0.15
𝑦 𝑘 𝑟 𝑢 𝑒𝑟
//model (continuous):
der(yc)=-yc+u; 0.10

//discrete, EulerFwd: pre(deryd)=(yd-pre(yd))/SampleTime

when sample(0, SampleTime) then 0.05
udF=u;
prederydF = -pre(ydF)+pre(udF); //model
ydF=pre(ydF)+prederydF*SampleTime; 0.00

end when;
𝑢
-0.05
//discrete, EulerBwd: deryd = (yd-pre(yd))/SampleTime
when sample(0, SampleTime) then
udB=u; -0.10
derydB = -ydB+udB; //model
ydB=pre(ydB)+derydB*SampleTime;
end when; -0.15
end DiscreteIntegrator; 0.0 2.5 5.0 7.5

Figure 1-11: A continuous dynamic model, 𝑦 𝑦 + 0. 𝑠 𝑛( 𝑡) and two

discretizations: Euler Forward (𝑦 ) and Euler Backward (𝑦 ).

• For linear or linearized models and simple excitation one can find the solution without time inte-
gration, by using the “matrix exponential”, 𝑒 𝑴𝒂𝒕𝒓𝒊𝒙 : The solution to a linear system without input,
𝐴 , starting from a given initial state 𝒙(0) 𝒙𝒊𝒗 is found as 𝒙(𝑡) 𝑒 𝑨 𝒙𝒊𝒗 . An alterna-
tive way, using a solution ansatz, is shown in 4.5.4.1.
• Frequency domain analyses. Linear models or models linearized around an operating condition
by differentiating and 𝐴 𝑓 ⁄ 𝑥 and 𝐵 𝑘 𝑓 ⁄ 𝑢𝑘 . The matrices are very useful for eigen-
modes, eigen-frequencies, step response, stability, or use as model base in model-based control
design methods.
• Stability analysis is to study when small disturbances lead to large response and how. (In linear
models, when limited disturbances lead to unlimited solutions, typically oscillating or exponen-
tially increasing.) Stability analysis is normally done for one operating point [𝒙𝟎 𝟎 ], around
which one linearizes. Instability is detected as when any eigenvalue to 𝑨 has positive real parts.
If there are inputs (dim( ) > 0), the stability is “for the open loop system”, e.g. for the vehicle
without driver. If the equations for is included in the model, there will be no 𝑩 matrix, and the
stability is then “for the closed loop system”, e.g. for the vehicle with driver. Adding a driver can
make the vehicle with driver more or less stable.
• Optimization. Either optimizing a finite number of defined design parameters or time trajecto-
ries, e.g. 𝑢 (𝑡). There are many optimization methods, ranging from trial-and-error to mathemati-
cally/numerically advanced gradient based or evolutionary inspired methods.
§ Closed Form Solution of Model on Affine Form
It will here be shown that a closed form solution can be found for the affine form. This is an alternative
to simulation. The following example shows how to find the solution for 𝒙 𝒌 + 𝑨 𝒙 with 𝒙(0) 𝒙𝟎 .
𝑇
Make an ansatz: 𝒙 𝒙𝒄𝒐𝒏 𝒕 + 𝑽 [𝑒 𝜆1 𝑒 𝜆2 ⋯ 𝑒 𝜆dim(𝒙) ] 𝒙𝒄𝒐𝒏 𝒕 + 𝑽 𝒆𝝀 ⇒
⇒ 𝒙 𝑽 𝐃𝐢𝐚𝐠([𝜆 𝑒 𝜆1 𝜆2 𝑒 𝜆2 ⋯ 𝜆dim(𝒙) 𝑒 𝜆dim(𝒙) ]) 𝑽 𝐃𝐢𝐚𝐠(𝜆 𝜆2 ⋯ 𝜆dim(𝒙) )
𝒆𝝀 𝑽 𝑫 𝒆𝝀
Insert: 𝑽 𝑫 𝒆𝝀 𝒌 + 𝑨 (𝒙𝒄𝒐𝒏 𝒕 + 𝑽 𝒆𝝀 )
Identify constant terms: 𝟎 𝒌 + 𝑨 𝒙𝒄𝒐𝒏 𝒕 ⇒ 𝒙𝒄𝒐𝒏 𝒕 𝑨 𝒌
Identify exponential terms: 𝜆 𝑰 𝑨 ∀ ⇒ [𝑽𝒏𝒐𝒓 𝑫] eig(𝑨) 𝑛 𝑽 𝑽𝒏𝒐𝒓 𝐝𝐢𝐚𝐠( )
𝑽𝒏𝒐𝒓 is a matrix with normalized eigenvectors as columns. is a column vector with magnitudes of
eigenvectors.
The solution before fulfilling initial conditions becomes: 𝒙 𝑨 𝒌 + 𝑽𝒏𝒐𝒓 𝐃𝐢𝐚𝐠( ) 𝒆𝝀

25
 Introduction

Fulfilling initial conditions 𝒙𝟎: 𝒙𝟎 𝑨 𝒌 + 𝑽𝒏𝒐𝒓 𝐃𝐢𝐚𝐠( ) 𝒆𝝀 𝑨 𝒌 + 𝑽𝒏𝒐𝒓 ⇒

𝑇
𝑽𝒏𝒐𝒓 (𝒙𝟎 + 𝑨 𝒌) 𝑽𝒏𝒐𝒓 (𝒙𝟎 + 𝑨 𝒌)
So, the solution becomes: 𝒙 𝑨 𝒌 + 𝑽𝒏𝒐𝒓 𝐃𝐢𝐚𝐠(𝑽𝑇𝒏𝒐𝒓 (𝒙𝟎 + 𝑨 𝒌)) 𝒆𝝀
§ Inline Integration or Difference Equation Formulation
The methods in mind above, e.g. “Euler forward with constant time step”, assumes a loop over time
where each a time-sequential step includes an update of time and state variables. In this context there
is one important assumption, the derivative approximation. If one, already in the Mathematical model,
assumes a time step Δ𝑡 (or any numerical grid 𝒕𝒈𝒓𝒊𝒅 [𝑡 𝑡 ⋯ 𝑡𝑁 ]) and replaces each occurrence of
the state derivatives with its derivative approximation one get a system of difference equation instead
of a system of differential equations. So, the solution over a whole/longer time interval (a macro step)
consisting of multiple micro time steps, is iteratively found at once, i.e. without the stepwise looping
over time. It is useful for co-simulation. The method is also used in predictive control, 1.5.3.2, i.e. opti-
mization over a time interval.
Using the derivative approximation for “Euler forward” gives 𝒙(𝑡𝑛 𝑤 + Δ𝑡) 𝒙(𝑡𝑛 𝑤 ) + Δ𝑡 ∙ 𝒙𝒏𝒐𝒘 (or
shorter 𝒙𝒊+𝟏 𝒙𝒊 + (𝑡 + 𝑡 ) ∙ 𝒙𝒊 ). So, one can reformulate:
• 𝒇𝑫𝑨𝑬 (𝒛 𝒛 𝑡) 𝟎 where 𝒛 [𝒙 𝒚]
to:
• 𝒇𝑫𝑨𝑬 ([(𝒙𝒊+𝟏 𝒙𝒊 )⁄(𝑡 + 𝑡 ) [𝒙𝒊 𝒚𝒊 ]] 𝒛𝒊 𝑡 ) 𝟎 for 0 ⋯ 𝑁
This is a system of ordinary algebraic equations. The 𝒚𝒊 exists only if we have “High order” (see
1.5.1.6). We can eliminate such as described in 1.5.1.6. Then, with 𝒇𝑫𝑨𝑬 [𝒇𝒙𝑫𝑨𝑬 𝒇𝒚𝑫𝑨𝑬 ]:
• 𝒇𝒙𝑫𝑨𝑬 ((𝒙𝒊+𝟏 𝒙𝒊 )⁄(𝒕𝒊+𝟏 𝒕𝒊 ) 𝒙𝒊 𝒚𝒊 𝑡 ) 𝟎 for 0 ⋯ 𝑁
• 𝒇𝒚𝑫𝑨𝑬 (𝒙𝒊 𝒚𝒊 𝑡 ) 𝟎 for 𝑁
We assume initial values 𝑥 and time grid [𝑡 𝑡 ⋯ 𝑡𝑁 ] are known. Then there are N dim(𝒙) +
(𝑁 + ) dim(𝒚) unknowns and equally many equations. An example (same as in Figure 1-36) is here
given. The and are parameters. The Mathematical model is the following DAE with 3 unknowns
( 𝑥 ) and 3 (differential) equations:

𝑥
𝑥
Rewritten to system of system of difference equations with 𝑁 + (𝑁 + ) 𝑁 + unknowns
( 2 ⋯ 𝑁 𝑥 𝑥2 ⋯ 𝑥𝑁 ⋯ 𝑁 ) and 𝑁 + (𝑁 + ) 𝑁 + equations:
𝑣𝑖+1 𝑣𝑖
+
𝒇𝒙𝑫𝑨𝑬 [ 𝑖+1
𝑖+1
𝑖
𝑖
] 𝟎 for 0 ⋯ 𝑁
𝑖+1 𝑖

𝒇𝒚𝑫𝑨𝑬 [ 𝑥] 𝟎 for 0 ⋯ 𝑁
There are many ways to solve such (algebraic) system of equation. One do not need a dynamic tool,
since the independent variable time is removed. In case of a linear dynamic model 𝒙(𝑡) 𝑨 𝒙(𝑡) , one
can assembly a large algebraic system of equations with “another 𝑨 matrix”, here called 𝑨𝒍𝒂𝒓𝒈𝒆 :
𝑨𝒍𝒂𝒓𝒈𝒆 [𝒙𝒂𝒓𝒓𝒂𝒚 𝒚𝒂𝒓𝒓𝒂𝒚 ] 𝒃
Note that 𝒙(𝑡) [𝑥 (𝑡) ⋯ 𝑥dim( ) (𝑡)] means an array of variables, while 𝒙𝒂𝒓𝒓𝒂𝒚 means a longer array
(or column vector) 𝒙𝒂𝒓𝒓𝒂𝒚 [[𝑥 (𝑡 ) ⋯ 𝑥 (𝑡𝑁 )] ⋯ [𝑥dim( ) (𝑡 ) ⋯ 𝑥dim( ) (𝑡𝑁 )]]. Our example, with
𝑁 , becomes:

26
 Introduction

0 0 0 𝑡 +𝑡 0 0
[ ] [ ] [ ] [ ] [ ]
0 0 0 𝑡 + 𝑡2 0 2 0
0 0 0 0 0 0 𝑥 𝑥 + (𝑡 𝑡 )
[ ] [ ] [ ] [𝑥 ] [ ]
𝑡
𝑡2 0 0 0 0 2 0
0 0 0 0 0 0 𝑥
[0 0] [ 0] [0 0] [ ] [ 0 ]
[
⏟ ] ⏟ 2 ]
[ [
⏟ 0 ]
0 0 0 0 0
𝑨𝒍𝒂𝒓𝒈𝒆 𝒙𝒂𝒓𝒓𝒂𝒚 𝒃

However, a Modelica tool is a dynamic tool and it can be useful, because a Mathematical model in Mod-
elica can, with minor code editing, be changed to a difference equation. The “traditional” Modelica
model in Figure 1-37 can be edited to a variant with difference equations as shown in the following
figure. (Some Modelica tool have this rewriting as a build-in and selectable automatic feature, typically
used for real time simulation.) Another derivative approximation is used in the figure, to show that
also “implicit” approximations as “midpoint” can be used.
v F x model ExampleModel_InlineIntegration
4 parameter Real m=2, c=3;
Real v, x, F;

with N=10 parameter Real F0=4, v0=0, x0=-1/3;

2
Real v_array[N+1], F_array[N+1], x_array[N+1];
Real derv_array[N-1 +1], derx_array[N-1 +1];
parameter Integer N=20; //=2; //
parameter Real t0=0, tend=5, t_array[:]=linspace(t0,tend,N+1);
//Note: Modelica indexes vectors from 1, while the compendium indexes
0 initial value with 0. Therefore "+1"
equation
//Equations with "constant variables":
𝑥 v_array[0 +1]=v0; x_array[0 +1]=x0;
-2 //Derivative approximations:
for i in (0 +1):(N-1 +1) loop
if i==1 then
//Forward 1st order derivative aproximation:
-4 derv_array[i]=(v_array[i+1]-v_array[i])/(t_array[i+1]-t_array[i]);
0.0 v F x 2.5 5.0 derx_array[i]=(x_array[i+1]-x_array[i])/(t_array[i+1]-t_array[i]);
else
4 //Midpoint 1st order derivative aproximation:
derv_array[i]=(v_array[i+1]-v_array[i-1])/(t_array[i+1]-t_array[i-1]);
with N=20 derx_array[i]=(x_array[i+1]-x_array[i-1])/(t_array[i+1]-t_array[i-1]);
end if;
2 end for;
for i in (0 +1):(N-1 +1) loop
m*derv_array[i] = -F_array[i]; //instead of m*der(v)=-F;
derx_array[i] = +v_array[i]; //instead of der(x)=v;
0 end for;
for i in (0 +1):(N +1) loop
F_array[i] = F0 + c*x_array[i]; //instead of F = F0 + c*x;

-2
𝑥 end for;

//Equations with "variable variables":

v = Modelica.Math.Vectors.interpolate(t_array,v_array, time);
x = Modelica.Math.Vectors.interpolate(t_array,x_array, time);
F = Modelica.Math.Vectors.interpolate(t_array,F_array, time);
-4 end ExampleModel_InlineIntegration;
0.0 2.5 5.0

Figure 1-12: § Example from Figure 1-37 implemented as difference equation model in Modelica.
Finally, it should be reminded about that the form described above is “contaminated” by containing
the derivative approximation, so it is a mixture of “1.5.1.1.3 Mathematical Modelling” and “1.5.1.1.5
Computation”.
§ Parameter vs Trajectory Optimization, Cost vs Constraints
So far, we have assumed that the model is used to predict how the system will behave (simulation). An
alternative use of the model is optimization. As engineering task, one can differ between parameter op-
timization or trajectory optimization:
• 𝑛 𝒑 𝑡ℎ 𝑡 𝑛 𝑒𝑠 𝑓 (𝒑) 𝑠𝑢 ℎ 𝑡ℎ 𝑡 𝒈𝒍𝒊 𝒊𝒕𝑪𝒐𝒏 𝒕𝒓 (𝒑) < 𝟎
• 𝑛 𝒛 𝒛(𝑡) 𝑡ℎ 𝑡 𝑛 𝑒𝑠 𝑓 (𝒛) 𝑠𝑢 ℎ 𝑡ℎ 𝑡 𝒈𝒍𝒊 𝒊𝒕𝑪𝒐𝒏 𝒕𝒓 (𝒛) < 𝟎
Parameter optimization has (a finite number of) design parameters 𝒑, while trajectory optimization
has (a finite number of) design trajectories 𝒛(𝑡). A trajectory has an infinite number of values, which
makes a big difference. The first can be to find the best gear ratios (𝒑 [𝑟 𝑟2 … 𝑟6 ]) and the cost
function 𝑓 includes a simulation of the dynamic system, e.g. modelled by a driver model and a vehicle
model. The latter can be to find the best pedal and gear selection trajectories (𝒛
[𝑝𝑒 (𝑡) 𝑒 𝑟(𝑡)]). The cost function also includes a simulation, e.g. a simulation of a the same vehi-
cle model but without the driver model. The simulation represents a look into the future to find out

27
 Introduction

what to do from present time instant and some time on, until a new decision is computed. An optimiza-
tion problem can be both types, i.e. include both design parameters 𝒑 and design trajectories 𝒛(𝑡).
The cost 𝑓 can sometimes be evaluated without simulation, 𝑓 𝑓 (𝑡 ) 𝑓 (𝒛𝟎 ). Then the optimi-
zation problem collapses to what to do at present time instant 𝑡 . We can then find a 3rd categorhy of
optimization, momentaneous variable optimization:
• 𝑛 𝒛𝟎 𝒛(𝑡 ) 𝑡ℎ 𝑡 𝑛 𝑒𝑠 𝑓 (𝒛𝟎 ) 𝑠𝑢 ℎ 𝑡ℎ 𝑡 𝒈𝒍𝒊 𝒊𝒕𝑪𝒐𝒏 𝒕𝒓 (𝒛𝟎 ) < 𝟎 ∈ ℝ
This might look very similar to the parameter optimization because both 𝒑 and 𝒛𝟎 are finite numbers of
scalars. However, 𝒑 is to be decided during product development while 𝒛𝟎 (typically requests to actua-
tors) has to be computed during product use, which is also true for 𝒛(𝑡).
The engineering task has to be reformulated to a computation task. Then, some adjustments are often
made to be able to use a particular existing mathematical/numerical method for optimization:
• Objectives can be moved from constraints to cost. Instead of demanding (𝒑) < 0, one add
terms (𝒑) to the cost function 𝑓 (𝒑). The weigth is then chosen so large so that the
original constraint is approximatively fulfilled.
• For trajectory optimization:
o The trajectories often have to be discretized, either as 𝒛 values over a grid in time or as
parameters of curve fits of 𝒛(𝑡). So, when coming to the computation task, the differ-
ence between parameter and trajectory optimization might vanish so same computa-
tional methods can be used.
o The dynamic model can be used also as constraint, which changes computation task to:
𝒈 (𝒛) < 𝟎 𝑛
𝑛 ⋯ 𝑡ℎ 𝑡 𝑛 𝑒𝑠 𝑓 (⋯ ) 𝑠𝑢 ℎ 𝑡ℎ 𝑡 { 𝒍𝒊 𝒊𝒕𝑪𝒐𝒏 𝒕𝒓
𝒈 𝒐𝒅𝒆𝒍𝑪𝒐𝒏 𝒕𝒓 (𝒛) 𝟎
This formulation is necessary if the method is of collocation type, see 1.5.1.1.5.5.
In the engineering task the distinctions cost vs constraints and cost vs dynamic model are often im-
portant. They are often mixed up in the computation task. Misunderstandings between vehicle engi-
neers and computation engineers might therefore occur.
§ Single- or Multi-Objective Optimization
One can say that optimization can only optimize a scalar cost. One can, of course, weight several costs
together to one cost: 𝑓 𝑓 + 2 𝑓2 . This can be called be called multi-objective optimization. It is
then important to understand the units. Consider 𝑓 as having the unit [€], which leads to that you
have to be able to set a number on the weigths [ 2 ⋯ ] with units [€⁄𝑢𝑛 𝑡𝑓 €⁄𝑢𝑛 𝑡𝑓2 ⋯ ]. This is
often difficult to motivate, which makes the interpretation of the found optimum difficult to use for di-
rect design decisions.
However, for parameter optimization, there is a useful variant of multi-objective optimization. It is to
not take the optimization all the way to one optimum, but stay at a relation between design parame-
ters. One can then draw a Pareto front, i.e. a relation between design parameters which forms an enve-
lope in a diagram with the scalar costs on the axes. The same relations can be drawn in a diagram with
design parameters on the axes, which reduces the dimension of the space in which one should look for
the best design parameters.
§ Shooting and Collocation Methods for Optimization
A divider among (trajectory) optimization methods is between those who iteratively find the solution
𝒙 through simulation (time integration) and iteratively update between those simulations (shooting
methods) and those who solves iteratively for both 𝒙 and at the same time (collocation methods).
Using collocation methods, one uses the mathematical model to generate equality constraint
𝒈 𝒐𝒅𝒆𝒍𝑪𝒐𝒏 𝒕𝒓 , see 1.5.1.1.5.3. One also have to discretize the model in time, which converts the differen-
tial equations into difference equations, using a certain derivative approximation, such as 𝒙𝒌+𝟏 ← 𝒙𝒌 +
Δ𝑡 ∙ 𝒙𝒌 𝒙𝒌 + Δ𝑡 ∙ 𝒇(𝒙𝒌 (𝑡𝑘 ) 𝑡𝑘 ) 𝒙𝒌 + Δ𝑡 ∙ 𝒇(𝒙𝒌 𝒌 𝑘 Δ𝑡) . From computational point of view, the
problem is then converted from trajectory optimization to an optimization of a finite number of trajec-
tory values (𝑡 ). Typically, very simple derivative approximations are used, compared to today’s inte-
gration methods for simulation.

28
 Introduction

1.5.1.1.6 Evaluate Requirement Fulfilment

This is often called Requirement Verification and means to find out how well the vehicle fulfils the re-
quirements. Also, Model Validity should be evaluated. (One can also do “Requirement Validation” which
is to judge whether a suitable selection of requirements was selected in 1.5.1.1. This requires much
experience.)
Often, there is an old design which was the cause for the whole engineering task and then there might
be test data to validate model with. Such validity check will be for one certain design. If directly com-
paring one simulation with one test it is an absolute model validity. If comparing the change between
two tests where something has been changed, it is a relative model validity, e.g. telling if the model
reflects design parameter changes well. The relative model validity is often more important than the
absolute. Since one often don’t know exactly how the model is going to be used, it can be a good habit
to include a boolean auxiliary variable for validity. One clear reason to set the validity “false” is when
the physical model assumptions are not met, such as if a wheel lifts from ground (assuming equations
to handle wheel lift is not included). In other cases, it can be more of an engineering judgement on
what threshold to set the validity false. Then, if validity is false during only a tiny fraction of the whole
studied time, it is again a judgement whether to disqualify the simulation result or not.
Evaluation of requirement fulfilment (or Analysis, as opposed to Synthesis) involves understanding
and interpretation of the computation results to real world. New requirements can appear, typically
when new constraints become active with the new design.
1.5.1.1.7 Redesign
Redesign (or Synthesis, as opposed to Analysis) is a creative part where experience and intuition is
important. In some rare cases, one can inversely calculate the new design or use numerical methods
optimize to a new better design.
1.5.1.1.8 Final Design
Final design, meaning a concrete proposal of design, such as a numerical design parameter value or a
drawing or control algorithm.
1.5.1.1.9 Modelling for Design Guidance
The activity ”Modelling for Design Guidance” in Figure 1-9 marks another “branch” than the straight-
forward computation/simulation for virtual verification described in 1.5.1.1 to 1.5.1.1.8. It refers to
activities that adjust the mathematical model with purpose to generate knowledge that can support
the design. Examples of such adjustments are:
• Linearization of the models, helps control or estimator designs and frequency and modal anal-
yses,
• Optimization, both parameter and trajectory optimization,
• Simplified, reduced order/modal restricted models which can be simulated in the vehicle for
on-line/off-line predictions/predictive optimization, and
• General understanding of how parameters (both physical such as suspension hardpoints,
spring stiffnesses, actuator capabilities and SW parameters such as control gains) influence the
vehicle in the studied manoeuvres.
To do these model adjustments, one often aims at a Closed form expression or Analytical expression, 𝒙
𝒇𝑪𝒍𝒐 𝒆𝒅𝑭𝒐𝒓 (𝒙 ). In some cases, one can even find Closed form solution or Analytical solution, 𝒛
𝒇𝑪𝒍𝒐 𝒆𝒅𝑭𝒐𝒓 (𝑡) . Examples where closed form solutions are possible are linear models with inputs (𝑡)
expressed as a simple time functions, such as step, harmonic and/or exponential functions, see 0 and
5.4.
To reach closed form expressions/solutions, one often has to, after the mathematical model, reduce
the number of equations by eliminating variables. This can help to understand the behaviour of the
whole system. The Mathematical model is still important since it documents better which physical
phenomena has been modelled. The differentiation order can often increase when eliminating varia-
bles, so that 𝒛̈ 𝒛
⃛ ⋯ appears in the equations.
§ Linearization of Non-Linear Model

29
 Introduction

A general model is not affine and consequently also not linear. They are often called non-linear (but
non-affine would be more correct, because an affine model is normally not called non-linear). There
are many reasons to why models are non-linear: examples of continuous non-linearities are terms of
the type 𝑥 2 , 𝑥 .5, 𝑥 𝑥 , 𝑥 𝑢 , sin(𝑥 ). There are also discrete non-linearities, such as time sampling or
discrete physical phenomena like dry friction, back-lash, etc. Note the difference between an affine
model resulting from physical model without additional approximations, and a linearization of a
non-linear model. The latter means that there are more approximations than what was done in the
physical model. These additional approximations normally have less physical interpretation.
In order to find the numerical solution, one typically does not gain anything on finding an affine form.
But there are reasons for such form; basically the same reasons as mentioned in 1.5.1.1.3.2 for finding
the linear form of a linear model.
If non-linear model, but only continuous non-linearities, one can linearize the models via Taylor expan-
sion and get an approximate model, a linearization on affine form. A Taylor expansion to find an ap-
proximate model in affine form, corresponding as in 1.5.1.1.3.2, would look like:
If the model is non-linear but there is no explicit dependency of 𝑡, the Taylor expansion will be exact:
𝒙 𝒇𝑶𝑫𝑬 (𝒙 ) ≈
𝜕𝒇𝑫𝑨𝑬 𝜕𝒇 non-
≈ 𝒇𝑶𝑫𝑬 (𝒙𝟎 𝟎 ) + ( 𝜕𝒙 )|𝒙=𝒙𝟎 (𝒙 𝒙𝟎 ) + ( 𝑫𝑨𝑬 )|𝒙=𝒙𝟎 ( 𝟎) { }
=
𝜕
=
𝑛𝑒 𝑟
𝟎 𝟎

𝒇𝑶𝑫𝑬 (𝒙𝟎 𝟎 ) + 𝑨(𝒙𝟎 𝟎 ) (𝒙 𝒙𝟎 ) + 𝑩(𝒙𝟎 𝟎 ) ( 𝟎) ⇒

⇒ 𝒙𝒐𝒇 𝒇𝑶𝑫𝑬 (𝒙𝟎 𝟎 ) + 𝑨(𝒙𝟎 𝟎 ) ⏟ (𝒙 𝒙𝟎 ) + 𝑩(𝒙𝟎 𝟎 ) ⏟
( 𝟎)
𝒙𝒐𝒇 𝒐𝒇

If (𝒙𝟎 𝟎 ) is a steady state condition it means same as that 𝒇𝑶𝑫𝑬 (𝒙𝟎 𝟎 ) 0, so the linearization gets
the form linear in state derivatives 𝒙𝒐𝒇 and inputs 𝒐𝒇 , i.e. without constant term. Similar Taylor expan-
sion can be done with 𝒈 for outputs , to find a linear (or at least affine) from also for outputs.
Note that the model was assumed to be non-linear. This meant that 𝑨 and 𝑩 became dependent of 𝒙𝟎
and 𝟎. For an affine model this was not the case, see 1.5.1.1.3.2. The consequence is that the model
can be stable around one (𝒙𝟎 𝟎 ) and instable around another. Therefore, many methods for fre-
quency analysis, control engineering and signal processing is not so applicable if the system is strongly
non-linear.
§ Stability Analysis
Stability is when all components in 𝒙 stays limited (finite) with limited (finite) and over long but fi-
nite time. Stability for a linear model or linearization of a non-linear model is guaranteed if the real
part of all eigenvalues of 𝑨 are negative. For a linear model, the 𝑨 matrix does not depend on 𝒙, so the
stability is relatively easy checked. For a non-linear model one would typically linearize around the
initial value and check sign of real part of all eigenvalues. However, this is not enough, since the solu-
tion can “slip away” to a regime where the signs of real part of the eigenvalues changes. The following
picture gives some examples.

30
 Introduction

𝑥 𝑥 x[1, 1] x[1, 2] x[1, 3] x[1, 4] x[1, 5] x[1, 6] x[1, 7]

42
𝑥 𝑥 0 ∀ e 𝜆 <0
41 all 𝑥 stable
40
39
38
0 1 2 3
x[2, 1] x[2, 2] x[2, 3] x[2, 4] x[2, 5] x[2, 6] x[2, 7]
𝑥 + 𝑥 0 34
32 ∀ e 𝜆 >0
all 𝑥 unstable
30
28
26
0 1 2 3
𝑥 0. sign 𝑥 0 𝑥 0 3/2 + 𝑥 0 x[3, 1] x[3, 2] x[3, 3] x[3, 4] x[3, 5] x[3, 6] x[3, 7]
22
e 𝜆 >0

21
20 all 𝑥 stable
19
18
0 1 2 3
𝑥 0. + sign 𝑥 40 𝑥 40 3/2 𝑥 40 x[4, 1] x[4, 2] x[4, 3] x[4, 4] x[4, 5] x[4, 6] x[4, 7]
12
e 𝜆 <0

11
10
9 all 𝑥 unstable
8
𝑥 0 1 2 3 𝑡

Figure 1-13: § 1st and 2nd rows: Linear models. Stability or instability same for all 𝑥 and follwos
what 𝑅𝑒(𝜆) indicates. 3rd: Non-linear model. Stability regimes are larger than 𝑅𝑒(𝜆) indicates. 4th
row: Non-linear model. Stability regimes are smaller than 𝑅𝑒(𝜆) indicates.
The influence of varying will now be treated. Stability is easiest understood for a model without .
However, if we have a , one often assumes is limited, slowly varying and independent of 𝒙, e.g. is
constant. Then it still holds to look only on eigenvalues of 𝑨 in 𝑨 𝒙 + 𝑩 . The only additional note
here is that 𝑨 can depend on even if is constant. An example is the influence from propulsion force
on lateral dynamics, 𝑓 𝑤 in 4.4.2.
However, if is depending on 𝒙, e.g. 𝑪𝑳 𝒙 we have a closed loop system. Then we have to look
on eigenvalues of 𝑨𝑪𝑳 in 𝑨 𝒙 + 𝑩 𝑨 𝒙 + 𝑩 𝑪𝑳 𝒙 (𝑨 + 𝑩 𝑪𝑳) 𝒙 𝑨𝑪𝑳 𝒙. This can change
the stability totally. A vehicle dynamics related case of this is that we can either assess lateral stability
for vehicle alone, with input steering angle limited and independent or, opposite, assess lateral stabil-
ity for vehicle with a closed loop driver, who steers as a direct function of the vehicle motion, 0. An-
other situation when the 𝑨 matrix changes is when a drvier model includes an own state variable, e.g. a
time delay. Then, both 𝒙 and 𝑨 grows with one dimension.

1.5.1.2 Approximations
As mentioned above, assumptions (or simplifications or approximations) are made during Physical
modelling. These can be directly motivated by vehicle design, such as assuming that some part is rigid
or massless. They can also be directly motivated by the manoeuvre to be studied with the model. Ex-
amples of approximations during Physical modelling are massless bodies, steady state conditions and
small angles. These have implications on the Mathematical modelling stage: some fictive forces do not
appear in the equations, and 𝑛 𝑒 replaces sin( 𝑛 𝑒) in the equations.
We can also assume that some variables or parameters are small. If a term in the final Explicit form
model gets a high order, one can approximate mathematically by removing it. For example, a term
𝑠 𝑁
where 𝑁 ≥ can be removed if the other terms have 𝑁 < . Note that such approximations
can not be done in an equation (during Physical or Mathematical modelling), but in the final expression

31
 Introduction

(in Explicit form model). For example, a term with order 𝑠 𝑁

in Mathematical model, it can end up
in denominator in final Explicit form model, which correspond to order 𝑠 𝑁
. Removing terms can
also be done by keeping only the first terms in a Taylor expansion of 𝒇𝑶𝑫𝑬 in the Explicit form: 𝒇 ≈
𝒇𝟎 + (𝒅𝒇⁄𝒅𝒙)|𝒙𝟎 𝟎 (𝒙 𝒙𝟎 ) + (𝒅𝒇⁄𝒅 )|𝒙𝟎 𝟎 ( 𝟎) .

1.5.1.3 Constants, Parameters, Variables and Signals

Variables and Signals vary with time during an experiment with the model. Constants and parame-
ters do not. In the following, it is understood that all have physical interpretation, e.g. a corresponding
(physical) quantity with a known unit. Units are central for engineering. A {𝑞𝑢 𝑛𝑡 𝑡𝑦} 𝑒 𝑠𝑢𝑟𝑒
[𝑢𝑛 𝑡] such as {𝐿𝑒𝑛 𝑡ℎ} 𝐿 [ ] [ ] . If using a consistent set of units, such as SI-units, each
equation can be seen as either a relation between quantities or relation between measures. The com-
pendium assumes SI-units, (ISO 80000-3, 2013),where else is not stated. Angle units are not stipulated
by SI, but the recommendation is to use radians. It is recommended to not use numerical constants di-
rectly in the equations unless they are dimensionless=unitless, such as 𝜋 or 2.
1.5.1.3.1 Constants and Parameters
A parameter can be changed between experiments. A constant does not even change between experi-
ments. Typical constant is 𝜋 or gravity constant . Typical parameters are vehicle wheelbase and road
friction coefficient.
From vehicle engineering point of view there is a significant difference also between Design parame-
ters and Operation/Manoeuvre parameters. The Design parameters are varied to fulfil the requirements
as good as possible. (Note that not all of the design parameters are subject to variation in all engineer-
ing tasks; e.g. if designing one part of the vehicle, the other parts are considered as fixed.) The Opera-
tion/Manoeuvre parameters should also be varied, but as disturbances for which the vehicle need to be
robust.
It is often natural to have different parameter sets to describe the design and operation/manoeuvre
and another for use in the equations. For example, the kerb weight (design) and load (operation) are
good for description while the total weight (sum of kerb and load) is the parameter that appears in the
equations. Therefore, the parameters that appears in the equations cannot always be categorized as
either design or operation/manoeuvre parameters. It is easy to forget that the selection of parameters
and the relations between parameters are often an equally important part of the engineering problem,
as the relations between variables (the equations) is for the dynamic system. A good question to ask
oneself is often: Is the comparison “fair” if I vary parameters as I do?
1.5.1.3.2 Variables
Variables vary during the studied time interval. If not else is mentioned, we assume continuous varia-
tion, but variables can also vary in discrete levels, 1.5.1.4. Time derivatives of (continuous) variables
cannot be assumed to zero, which is important when going from DAE to ODE formulation since some
DAE equations might need to be differentiated. In Mathematical model (DAE), a variable and its time
derivatives are counted as one variable. In Explicit model (ODE), one can count each time derivative of
a variable as an additional variable, but then one also should count each integration ( ∫ 𝑡 ) as
one equation.
From vehicle engineering point of view, the Operation/Manoeuvre can stipulate a constant value of
some variables. Mathematically, it is then a parameter (parameter 7 ), but as vehicle engineers it
can be better to think of it as a variable with an initial value, coming from the Operation/Manoeuvre,
such as 7 and an extra equation 0 . The initial value is then an Operation/Manoeuvre param-
eter and not a Design parameter. A typical example could be longitudinal speed, .
The compendium mainly deals with simulation, but it can be noted that, during parameter optimiza-
tion, 𝑝 𝑝 min𝑝 (𝑓𝑢𝑛 𝑡 𝑜𝑛 (𝑀𝑜 𝑒 (⋯ 𝑝))) , one do several simulations with same model. Some
model parameters are varied between simulations and these parameters could therefore potentially
be called variables. To avoid such semantic confusion, one can use a prefix and call these parameters
optimization variables when talking about them in the optimization context.

32
 Introduction

A very important special case of variables are state variables or states. State variables are given ini-
tial values and then updated through integration along the time interval studied. (Continuous) States
are updated through time integration. Which variables to use as states is not uniquely defined by the
physical (or mathematical) model. See 1.5.2.1.2.
1.5.1.3.3 Signals
Since vehicle dynamics so often requires models of the control algorithms, one often use the words
variable and signal interchangeable. It is suggested to, at least, reflect over the difference:
• A signal can represent an ODE variable with prescribed causality. So, a variable cannot be
represented by a signal before the modelling stage “1.5.1.1.4 Explicit Form Modelling”.
• Signals can appear already in mathematical model (DAE), typically as interface on models of
mechatronic subsystems or subsystems consisting purely of algorithms (software). For such
signals, the causality is normally prescribed already in the mathematical model and one have
to differ between differentiation orders, so that and are counted as two variables. A
strict way to implement this is given in the Modelica modelling format, see 1.5.4.5.

1.5.1.4 Discrete Dynamics

Contribution from: Fredrik Bruzelius, Vehicle Dynamics at Chalmers and VTI

Only continuous dynamics, as opposed to discrete dynamics, were considered in 1.5.1.1.3 to 1.5.1.1.5.
Discrete dynamics modelling can be used for computational efficient models of, e.g., ideal dry friction
or ideal backlash but also to model time sampled (digital) systems. See Figure 1-14. The way the states
evolve over time differs between continuous and discrete dynamics: During the modelling, the contin-
uous states, 𝒙𝒄, should be thought of as changing continuously. In the computation there has to be a
discretization in time steps since computers are digital (not analogue). Hence the 𝒙𝒄 are updated only
between each time step but thought of as continuously changing in between. The update is based on
the state derivatives, 𝒙𝒄, and a derivative approximation, e.g. 𝒙𝒄 (𝒙𝒄 (𝑡 + Δ𝑡) 𝒙𝒄 )⁄Δ𝑡 . Discrete
states, 𝒙𝒅 , should be thought of as constant except that they stepwise change value at the time instants
when one of the transition conditions becomes true (not between these time instants!). These time in-
stants are called (state) events, and can be implemented as when event indicators, 𝒉(𝒙𝒄 𝒙𝒅 (𝑡) 𝑡) >
0, becomes true.

Solid show
𝑥 mathematically
correct solution.

Dashed show
computed solution,
Shows that 𝑥 can be here linear
reset during the event. interpolation over
time steps.
𝑥
Initial Event
states
ℎ 𝑡

All these time steps are …but this time step is adjusted to find the exact
constant or adopted to error time when the event’s transition condition
estimate in 𝑥 … becomes true, i.e. when ℎ becomes > 0.

Figure 1-14: A visualization of a solution with both continuous and discrete states.
At a first glance, the 𝑥 should not change in the events, but are two reasons why one sometimes model
them to change (stepwise) at an event: one is to reinitialize to physically obvious values (e.g. a bounc-
ing ball need to start from the surface it bounces at when the bounce happens) and the other is if one
change physical model at an event, e.g. introduce an inertia or elasticity. The latter can be powerful to
solve an engineering task, but the engineer has to be especially observant on how credible the numeri-
cal solution is, because the simulation tools can typically not estimate errors in such model. The latter

33
 Introduction

way is here described as one model, but in another context it can be seen as “splicing” of several mod-
els.
A system which includes both types (hybrid dynamics) evolves as follows:
𝒙𝒄 (𝑡 + Δ𝑡) ← 𝒙𝒄 + Δ𝑡 ∙ 𝒙𝒄 𝒙𝒄 + Δ𝑡 ∙ 𝒇𝒄 (𝒙𝒄 𝒙𝒅 (𝑡) 𝑡) and
ℎ𝑒𝑛 𝒉(𝒙𝒄 𝒙𝒅 (𝑡) 𝑡) > 𝟎 𝑡ℎ𝑒𝑛 𝒙𝒅 (𝑡 + ) ← 𝒇𝒅 (𝒙𝒄 𝒙𝒅 (𝑡) 𝑡) .
Note that a “knee” (discontinuous derivative but continuous value) in an equation does generate an
event, but no discrete state, since no “memory” is needed. But a step (mathematical discontinuity in
value) or a hysteresis (mathematical function with multiple values depending on which branch is ac-
tive) generate state event with discrete state. See examples in 2.4.4.
One can see discrete dynamics as if the physics get stuck on the steep part of a step or in a branch in a
hysteresis, until when an exit condition becomes true. Examples where this occur is dry friction, one-
way clutches, back-lashes, diodes, etc. Another example appears if one changes physical models, e.g.
remove a mass or compliance, during some parts of the simulation, governed by certain events.
Discrete dynamics is not as well established in most basic engineering education as continuous dy-
namics.
1.5.1.4.1 § Event Iteration and Chattering
Several discrete state transitions can occur during same time instant, which calls for event iteration
one time instant, i.e. during zero time. An example is when a friction element switches state from posi-
tive slipping to sticking and to negative slipping during one time instant. A typical problem with these
models is that transition conditions are such that event iteration does not converge, so called chatter-
ing.
It can be needed to reinitialize continuous state variables (𝒙𝒄 (𝑡 + ) ⋯) in events, e.g. motivated by im-
pact dynamics. There is a risk that the numerical solution gets stuck in undesired high frequency
switching of discrete states which also can lead to chattering.
Chattering can be caused by coding errors of the event handling or inconsistent physical modelling.
1.5.1.4.2 § Finite State Machines
Examples of state events which typically appear in control algorithms or driver models are (finite)
state machines. These are easier in the way that typically do not have event iteration, but only one
state transition should happen in each time sample. This reduces the risk of chattering, or at least
makes the chattering to appear with the sample time, not during a time instant.
1.5.1.4.3 § Discrete Dynamics in Modelica Tools
Here is an example of how discrete states can be modelled (or implemented) in Modelica. The example
is purely mathematical and without physical interpretation.
Model example_DiscreteState
Real y, x_c (start=0);
Integer x_d (start=0);
equation
//Eqs defining continuous variables y and x_c (but can use the discrete variable):
y = x_c^2;
der(x_c) = if (x_d == 0) then -1 else +1;
//Eq defining discrete variable x_d (how it is updated at state events):
when (pre(x_d) == 1 and x_c > +1) then x_d = 0; reinit(x_c,+2);
elsewhen (pre(x_d) == 0 and x_c < -1) then x_d = 1; reinit(x_c,-2);
end when;
end example_DiscreteState;

1.5.1.4.4 § Discrete Dynamics in Dataflow Diagram Based Tools

Here is the same example as in 1.5.1.4.3, but modelled (or implemented) in a Dataflow Diagram based
tool, Simulink. Many tools does not have direct support for discrete dynamics. A way to use an “Inte-
grator block for continuous states” can be used to handle “updates of a discrete state 𝑥 ” is shown in
the figure below.

34
 Introduction

Continuous Eqs inside

Discrete Eqs inside

𝑥

Figure 1-15: § Same model as shown in Modelica format, but here implemented in the Dataflow
Diagram Based Tool “Simulink”. The dashed arrow shows how to handle updates of a discrete
state 𝑥 .

1.5.1.5 Algebraic Loops or Simultaneous Equations

One problem that can appear when generating the Explicit form model is Algebraic loops. The name is
most intuitively understood when Data flow diagram is used for the Explicit form model, see 1.5.4.2.
Then, an Algebraic loop is a signal loop which does not include and integrator blocks; meaning that
each branch of the loop needs input from the other to calculate its output. Algebraic loops can also be
called Simultaneous equations. The smallest possible can actually be one single equation, e.g. a trans-
cendent equation, such as “𝑥 sin(𝑥) ”, which has no closed form solution.
A simple mechanical example where two equations need to be solved before each other is given in Fig-
ure 1-16. The example is a vehicle decelerating with locked wheels; Vertical forces are needed to calcu-
late friction forces, which are needed to calculate acceleration, which is needed to calculate vertical
forces.
Loops appear often between bigger chunks of equations than in Figure 1-16, namely between modules
in the model. The decomposition in certain modules is often desired for model modularity. A typical
other example where a loop occurs is when doing quasi-steady state assumptions about roll moment
distribution between axles (4.3.10.3).
Loops can always be solved by setting up an iteration in each evaluation of time derivatives. This can
sometimes be accepted, but it gives poor computational speed. Other ways are to handle the loops by
sacrificing simplicity in model is to model an additional elasticity or inertia between the modules, or
simply a (non-physical) filter. “Memory blocks” can work but are not recommended, since the simula-
tions will be dependent on solver. Best is if the loop can be symbolically solved, as in Figure 1-16.
Simultaneous equations (Algebraic loop)
If we guess we can
calculate all unknowns,
𝑓 + 𝑓 including itself.
ℎ

+ 𝑓 + 0
If the guess is adjusted
𝑓 ℎ+ 𝐿⁄ 𝐿⁄ 0 until it gives same
𝑓
calculated we have
𝐿/ 𝐿/ 𝑓
∫ 𝑡 an iterative solution.

3 equations, 3 unknowns ( 𝑓 ): Solved by symbolic manipulation of equations:

𝑓 +
+ The loop could, in this case, be
𝑓 + 0 𝑓 ⁄ ℎ/𝐿
symbolically solved, so that we
ℎ + 𝑓 𝐿⁄ 𝐿⁄ 0 𝑓 get this explicit solution.
(with as state, ∫ 𝑡 ) ( ∫ 𝑡 )
Figure 1-16: An example of Algebraic loop, also called Simultaneous equations.

35
 Introduction

1.5.1.6 High Index

A possible problem for a mathematical model is High index. A simple mechanical example of High in-
dex system of equations is when two inertias are rigidly connected, see Figure 1-17, where a working
causality cannot be found with retained modularity. Similarly to algebraic loops, it can be solved with
equation manipulations, but for high index problems, the manipulations include differentiation of
some equations. It is tempting to try a “differentiating” block ( 2 ( ⁄ 𝑡) 2 instead of 2 ∫ 2 𝑡
in Figure 1-17) but that will not work. Other examples where high index occurs is when modelling
purely rolling wheels (3.3.2).
High index No working causality:
2 in-ports for
∫ 2 𝑡 and 2 out/ports
for v2 .
2
2 2

2 𝑡
2 2 ∫ 2 2

3 equations, 3 unknowns ( 2 ): Solved by symbolic manipulations:

1. Differentiate ” 2 ” to ” 2 ”
2 2. Algebra gives:
2 ⁄ + 2
(with as states, ∫ 𝑡 𝑓𝑜𝑟 )
∫ 𝑡
2 and 2 ”
Figure 1-17: An example of a High index model; an accelerating rigidly connected car and trailer.

1.5.1.7 Causality
Systems can be modelled with Natural causality. For mechanical systems, this is when forces on the
masses (or motion of the compliance’s ends) are prescribed as functions of time. Then the velocities of
the masses (or forces of the compliances) become state variables and have to be solved through time
integration. The opposite is called Inverse dynamics and means that velocities of masses (or forces of
compliances) are prescribed. For instance, the velocity of a mass can be prescribed and then the re-
quired forces on the mass can be calculated through time differentiation of the prescribed velocity. Cf.
Analysis and Inverse Analysis in Figure 1-7.

1.5.1.8 Drawing
Drawing is a very important tool for engineers to understand and explain. Very often, the drawing con-
tains free body diagrams, FBD, see 1.5.2.1, but also other diagrams are useful. On top of normal draw-
ing conventions for engineering drawings, it is also important to draw motion and forces. The notation
for this is proposed in Figure 1-18.
It is often necessary to include more than just speeds and forces in the drawings. In vehicle dynamics,
these could be: power flow and signal or data flows. These can preferably be drawn as arrows, but of
another kind than the motion and force arrows.
When connecting components with signal flow, the resulting diagram is a data flow diagram or block
diagrams. Physical components and physical connections can be included in such diagram, and if ar-
rows between them it would represent data flow or causality. Physical components can also be con-
nected by “Physical connections”, which does not have a direction, see 1.5.1.8.1. It should be noted that
a (computation) flow charts and (discrete) state diagram represent something quite different from data
flow diagrams, even if they may look similar; in state diagrams, an arrow between two blocks repre-
sents a transition from one (discrete) state or operation mode, to another.

36
 Introduction

Vector (e.g. motion (position, velocity or

acceleration) or force (or moment or torque)): motion In data flow diagrams:
Along axis in plane of paper:
force Variables, Signals
Perpendicular to paper, pointing out of paper: with prescribed causality:
Perpendicular to paper, pointing in to paper:
Physical Connection
Rotation (e.g. rotational motion or moment) (no prescribed causality):
Around axis in plane of paper:
Parametrization
Around axis perpendicular to paper: (sometimes relevant to show):

Quantity flow (e.g. Energy flow = Power):

Figure 1-18: Arrow-like drawing.

1.5.1.8.1 Measures in Drawings
To define physical quantities in words is generally ambiguous; different people understand them dif-
ferently and from time to time. So, it is much safer to define them in drawings. Figure 1-19 shows some
examples. One has to differ between a minus sign in front of the variable name in the drawing, 𝐿2 ,
and whether the numerical value of the variable is positive or negative.

(“axis”)
- 𝑓 (vector)

𝐿3 (< 0) 𝐿2 (< 0) 𝐿 0
𝐿 (axis)
𝐿2
𝐿2

Figure 1-19: Different ways to define physical quantities. The length 𝐿2 is defined in 3 ways. Note
also different types of arrows: “axis”, “vector” and other, where other are ”meassurements”.
Defining forces and moments in a drawing is special in the sense that they always appear as action and
reaction, see 1.5.2.1.1.

1.5.1.9 Mathematics and Notation Conventions

Generally, a variable is denoted 𝑥 or 𝑥 (𝑡), where 𝑡 is the independent variable time. In contexts where
one means that variable’s value at a certain time instant, 𝑡 , it can be denoted 𝑥 (𝑡 ). In contexts where
one wants to mark that the variable’s time history over a time interval (infinitely many values over an
infinite or finite time interval) is referred to, it can be denoted 𝑥 ( ).
Differentiation (of x) with respect to time (t): 𝑥 𝑒𝑟(𝑥).
Matrices (two-dimensional arrays) are denoted with bold letters, often uppercase 𝑨. Column and row
vectors (one-dimensional arrays) are denoted with bold, often lowercase 𝒃. When elements in arrays
are written, brackets are used: [𝑥𝑦 𝑦].
Geometric vector (or physical vector, or spatial vector), denoted by . In order to use geometric vec-
tors in calculations, they are often expressed as one-dimensional arrays with components. It has to be
clear for each component, which direction it is expressed in, e.g. by a subscript, e.g. 𝒗𝒙𝒚𝒛 [ ].
Multiplication symbol (* or ∙ or × or •) shall be used to avoid ambiguity, e.g. ∙ 𝑥, not 𝑥 (assum-
ing 𝑥 is one variable, not two). Stringent use of multiplication symbol enables use of variables with
more than one token, cf. programming. Stringent use of multiplication symbol reduces the risk of am-
biguity when using operators, e.g. it shows the difference between 𝑓 (𝑥) and 𝑓 (𝑥). Compendium

37
 Introduction

denotes multiplication between scalars and matrices equally, using “ ”. Multiplication between geo-
metric vectors are denoted: Cross multiplication, denoted × ⃗ and scalar multiplication, denoted •
⃗.
Parentheses shall be used to avoid ambiguity, e.g. ( ⁄ ) ∙ or ⁄( ∙ ), and not ⁄ ∙ .
An interval has a notation with double dots. Example: Interval between and is denoted . . .
An explanation, between two consecutively following steps in a derivation of equations is written
within {} brackets. Example: 𝑥 + 𝑦 { + + } 𝑦+𝑥.
An inverse function is denoted with superscript 𝑛 , 𝑦 𝑓(𝑥) ⇔ 𝑥 𝑓 𝑛𝑣 (𝑦). The use of superscript
can cause an ambiguity whether 𝑓 means inverse function or inverted function value ( ⁄𝑓).
When a name of the inverse function is available, it can be used, e.g. arctan(⋯ ) instead of taninv(⋯ ).
Fourier and Laplace transform of function 𝑓 (𝑡) are denoted ℱ(𝑓(𝑡)) and ℒ(𝑓(𝑡)), respectively:
+∞
𝜔
ℱ(𝑓(𝑡)) ∫ 𝑓 (𝑡) 𝑒 𝑡 ℎ𝑒𝑟𝑒 ∈ 𝑅𝑒
∞
∞
ℒ(𝑓(𝑡)) ∫ 𝑓(𝑡) 𝑒 𝑡 ℎ𝑒𝑟𝑒 𝑠 𝜎+𝑗 ℎ𝑒𝑟𝑒 𝜎 𝑛 ∈ 𝑅𝑒

There are many practical rules for manipulation transformed differential equations, such as
ℱ (𝑓(𝑡)) 𝑗 ℱ(𝑓(𝑡)) and ℒ (𝑓(𝑡)) 𝑠 ℒ(𝑓(𝑡)) 𝑓(0) . For Laplace transform, more examples
are given in Figure 1-20.
Function Function in time
Function in Laplace domain, ℒ(𝑓(𝑡))
description domain, 𝑓(𝑡)
Unit impulse (𝑡) ℒ( (𝑡))
Unit step 𝜎(𝑡) ℒ(𝜎(𝑡)) ⁄𝑠
Unit ramp 𝑡 𝜎(𝑡) ℒ(𝑡 𝜎(𝑡)) ⁄𝑠 2
Exponential 𝑒 𝛼 ℒ(𝑒 𝛼 ) ⁄(𝑠 + )
Sine sin( 𝑡) ℒ (sin( 𝑡)) ⁄(𝑠 2 + 2 )
Cosine cos( 𝑡) ℒ (cos( 𝑡)) 𝑠⁄(𝑠 2 + 2 )
Figure 1-20: Examples of function in time domain and the corresponding in Laplace domain.
An example: The dynamic model 𝑦 (𝑦(𝑡)) 𝑛 𝑢(𝑡) looks as follows, if written Fourier trans-
formed or Laplace transformed: 𝑦( ) 𝑛 𝑢( )⁄( 𝑗 ) or 𝑦(𝑠) 𝑛 𝑢(𝑠)⁄𝑠 , respectively.
The order within subscripts can be debated, e.g. should a longitudinal (𝑥) force ( ) on rear axle (𝑟)
should be denoted or . The intention in this compendium is to order subscript with the physical
vehicle part (here rear axle) as 1st subscript and the specification (longitudinal, x) as 2 nd, leading to .
If there are further detailed specifications, such as coordinate system, e.g. wheel coordinate ( ), it will
be the 3rd subscript, leading to 𝑤 . Further additional specifications, such as case, e.g. without (0) or
with (1) a certain technical solution, will be the 4 th subscript, leading to 𝑤 and 𝑤 . If subscripts
have >1 token, it can sometimes be good to use comma, e.g. 𝑤 𝑙𝑤 .
Naming of variables (signals), parameters and function components (blocks) is often important in
large products as vehicles. Some naming needs to be readable for many engineers, such as CAN signals.
Hence, companies developing their own naming standards. But also, the standardization organisation
AUTOSAR has released a naming standard, with intention to be accepted by both vehicle manufactur-
ers and system suppliers.

1.5.2 Mechanical/Machine Engineering

Vehicle dynamics originates from Mechanical or Machine engineering (in Swedish “Maskinteknik”, in
German “Machinenbau”). Therefore, it is important to be fluent with the following mechanical basic
relations:
• [Torque or Moment] = Force ∙ Lever: 𝑇 or 𝑀 𝑒𝑛 𝑡ℎ

38
 Introduction

• Power = Force ∙ Translational velocity:

or Power = Torque ∙ Rotational velocity: 𝑇
(or in other domains: Power=Voltage Current; Power=Flow Pressure; etc)
• Energy = time integral of Power: ∫ 𝑡 or: ∫ 𝑥 or: ∫𝑇 .
• (Torque) Ratio = Output torque/Input torque: 𝑟 𝑡 𝑜𝑇 𝑇 ⁄𝑇 𝑛
• (Rotational velocity) Ratio = Input velocity /Output velocity: 𝑟 𝑡 𝑜𝜔 𝑛⁄
• (Power) Efficiency = Output power/Input power: 𝜂 ⁄ 𝑛
or, in a wider meaning, Efficiency = Useful/Used;
Interfaces in mechanical systems are recommended to use [ ] or [𝑇 ], which is a “power-factor-
ized interface”, see 1.5.1.8.1. Note that Mechanical or Machine engineering, includes also other physical
domains than mechanical, such as electric, hydraulic, etc, see 1.5.1.8.1.

1.5.2.1 Free-Body Diagrams

In the physical model, see 1.5.1.1.2, a free-body diagram, FBD is often central. See example in Figure
1-22 and Figure 1-23. Also, division into subsystems by introducing cuts or connections. It is often
practical to idenfity moment-free connection points, etc. The subsystem split typically goes through:
• Connection point between towing unit and towed unit (typical interface quantities: 2 positions
(with their derivative, velocities) and 2 forces).
• Driveshaft close to each wheel (typical interface quantities: 1 shaft torque, 3 forces, and 2 an-
gles (steering and shaft rotation), sometimes 1 wheel camber angle).
• Surface between driver’s hand and steering wheel (typical interface quantities: 1 angle and 1
torque).
• Sensed signals and request signals, which cuts out control algorithms as subsystem.
One can draw free-body diagrams with two purposes: Understand the real-world problem/manoeuvre
or as a help to formulate (equilibrium) equations. A FBD with the 1st purpose, typically has force ar-
rows with the actual force directions and named | | or an explicit numerical value. A FBD with the 2 nd
purpose has force arrows as the corresponding forces are defined positive and named (or , if
drawn as defined negative). The reasoning about naming with | | also applies to other signed or
vector quantities: velocities, displacements and distances; in translation and rotation.
It is recommended to not draw two forces acting at the same point of part and having same direction,
and 2 . With such unsuitable drawing, it is very easy to set up wrong equations, e.g. for friction limi-
tation.
A short-cut, which avoids one equation, is to use same notation on two forces in the FBD, e.g. denote
𝑓 as 𝑓 in Figure 1-22. If using this, one should keep in mind that the naming itself includes a physi-
cal assumption between 2 forces. Compare this to action/reaction force in a cut, which actually is the
same force and should have the same name.
When using FBD as a help to set up equilibrium equations, a short-cut that can be to express a force in
other, in a way that equilibrium is already fulfilled in the FBD. This eliminates one equilibrium equa-
tion. An example of this could be to use 𝑓 instead of 𝑓 in Figure 1-22, which would make the verti-
cal force equilibrium for the front axle ( 𝑓 𝑓 0) unnecessary and not useful.

1.5.2.1.1 Forces in a Cut or Connection

There are two ways of denoting forces and moments in a cut (or connection):
• Using single cut force notation: Same force (and moment) notation in both part’s FBD.
• Using multiple cut forces notation: Different force (and moment) notation in each part’s FBD.
The latter is more systematic, but requires extra equations, telling that the cut is in equilibrium. It is
also more general since it can handle more than two parts in a cut (which makes the name connection
more understandable than cut). The latter is also a way to allow notations in local coordinate systems
for each part. However, the first is probably easiest for intuition since it follows the concept of that a
force in a cut always has one action and one reaction. The compendium uses both ways. Example is
given in 1.6.1.1.

39
 Introduction

1.5.2.1.2 Physical Connections and Power-Factorized Interface

It is recommended to use interface variables which factorize power , e.g. force and velocity , be-
cause . Similar factorization is found in all other domains, on a general form effort flow.
Connection via such “power-factorized interfaces” is sometimes referred to as “power bonds”. Effort
and flow often get opposite variable flow causality or signal flow causality in the explicit form model.
If modelling with only as interface variable, it is easier to miss validity limits of the models, such as
supplying a certain non-zero power to a stand-still vehicle requires infinite forces ( ⁄ → ⁄0).
When connecting physical components in a data flow diagram with “Physical connections”, there can
be two main concepts:
• Connecting in “nodes”, see Figure 2-68. Here, each node has typically potential and flow varia-
bles (velocities and forces in Mechanical systems). And there are only two components con-
nected to each node. The node variables appear in both components’ equations. The connec-
tion itself generates these equations: “Potential variable in node and both connected compo-
nents are equal” and “Flow variable in node and both connected components are equal”.
• Connecting with “connections between connectors”, see Figure 2-83. Here, the same variables
are defined in connectors at the components. A connection can be made with lines between the
connectors at the components, possibly connecting more than two components. The equations
in each component is formulated in the connector’s variables. The connection itself generates
these equations: “Effort variables are equal” and “Sum of flow variables are zero”.
The first is often the easiest (least number of equations and variables) for small models with a fixed
topology. The latter is a more systematic which easier handles larger models and models with varying
topology. The latter is supported in the standard modelling language Modelica, see 1.5.4.5 and the
open book on internet in (Tiller, 2019).

1.5.2.2 Choice of State Variables

From 1.5.1.3.2, we know state variables has to be selected and that selection is not unique. For me-
chanical systems, one often selects positions and velocities of (inertial) bodies as state variables, but it
is quite possible and sometimes preferable, to use forces in compliances (springs) and velocities of
bodies as state variables. With mass-spring system as example, the two alternatives become:
• 𝑥 states: [ ⁄ ∙𝑥 𝑥 ] (or [𝑥̈ ⁄ ∙ 𝑥 ]) and
• states: [ ⁄ ∙ 𝑥 ⁄ ], respectively.
The latter alternative is especially relevant when position is not of interest, typically in propulsion sys-
tems. If states, one equation (𝑥 ⁄ ) can then be omitted which can simplify. It is also often eas-
ier to express steady state initial conditions for a pre-tensioned system, e.g. wheel suspension where
spring forces can be states. A drawback might be that, if overdetermined pre-tensioned systems (e.g.
mass suspended in two springs: [ ( + 2 )⁄ ∙ 2 2∙ 𝑥 ⁄ 𝑥 2 ⁄ 2 ]),
more states than necessary will be used, which means that positions (𝑥) can be calculated in two ways
(𝑥 ⁄ 𝑥 2 ⁄ 2 ). These two ways can drift apart. The drift can be eliminated by completely
disregard the original spring constitutions (𝑥 ⁄ 𝑥 2 ⁄ 2 ) and instead use position as a state
(𝑥 ). The mathematical system is then effectively:
• 𝑥 states: [ ⁄ ∙ 𝑥 ].
One can see the latter alternative as if the states are the variables which represents the two energy
forms kinetic and potential energy, as in Lagrange mechanics.

40
 Introduction

1.5.2.2.1 § Varying the Order of the ODE

Spring-parallell damping Spring-seriell damping …both previous and Damping to Ground

𝑥 𝑥
𝑥
𝑥 𝑝
𝑥 𝑥 𝑥 𝑥

1st order ODE: 1st order ODE: 1st order ODE: 𝑝

+ + 𝑝 +

⇒
Single state ODE: Single state ODE:
̈+ + + ⇒ ⇒ 𝑝+
⇒ 𝑥̈ + 𝑥+ 𝑥 ̈+ 𝑝+ + + +
Single state ODE:
𝑥 + 𝑥 + 𝑜𝑛𝑠𝑡 𝑛𝑡 𝑝
̈+ + 𝑝 + + + +

Figure 1-21: § Examples of damped mass and spring systems, as 1st order ODEs and single state
ODEs.

1.5.2.3 Equation Types

The step from physical model (see 1.5.1.1.2) to mathematical model (1.5.1.1.3) means basically to
identify variables and parameter and find the relationship between them, i.e. the equations. Models for
Vehicle Dynamics always includes mechanics, and for these parts we can identify 3 main types of equa-
tions: Equilibrium, Compatibility and Constitution.
§ Equilibrium and Compatibility are generally to be questioned in Physical model; e.g. shall a part be
regarded as inertial and rigid or massless and compliant? Constitution is more debatable when doing
the Mathematical model; e.g. shall a relation be modelled as linear or quadratic relation?
1.5.2.3.1 Equilibrium
Contribution from Dragan Sekulić

Equilibrium gives relations between forces (including moments). For a static system, we can use the
(static) equilibria: Sum of forces in any direction is zero: ∑ 𝑭 𝟎 and Sum of moments around any
axis is zero: ∑ 𝑴 𝟎
In a dynamic system, there is typically inertia effects. Inertia effects can be seen as “fictive forces” and
“fictive moments” (or “d’Alembert forces or moments”). Typically, the fictive forces are 𝑓
(where translational acceleration) and the fictive moments are 𝑀𝑓 . These are counter-
directed to and , respectively. We can find the equations either as “Dynamic equilibria”:
• Sum of forces, including 𝑓 , in any direction is zero: ∑𝒊𝒏𝒄𝒍.𝒇𝒊𝒄𝒕 𝑭 𝟎
• Sum of moments around any axis, including both 𝑀𝑓 and “ 𝑓 𝑒 𝑒𝑟”, is zero:
∑𝒊𝒏𝒄𝒍.𝒇𝒊𝒄𝒕 𝑴 𝟎
or as “Equations of motion” or “Newton’s 2nd law”:
• Sum without fictive in direction have to be equal to : ∑𝒆𝒙𝒄𝒍.𝒇𝒊𝒄𝒕 𝑭 𝒂
• Sum without fictive around axis through CoG in direction of have to be equal to :
∑𝑪𝒐𝑮 𝒆𝒙𝒄𝒍.𝒇𝒊𝒄𝒕 𝑴 𝑱𝑪𝒐𝑮 𝝎 + 𝝎 × (𝑱𝑪𝒐𝑮 𝝎)
The is the mass moment of inertia matrix in CoG along the same coordinate axes as is expressed
in. In this compendium, the alternative with Dynamic equilibria is mainly used. The fictive forces are
introduced in the free-body diagrams with dashed arrows, see Figure 1-22 and Figure 1-23. Ad-
vantages with Dynamic equilibria, as opposed to Newton’s 2 nd law, are easiness to form moment equi-
libria since it can be taken around any axis (example in Figure 1-23). It is also easier to include motion
effects within a part of the system (such as Jf ωf and in Figure 1-22). The general form for the
fictive forces is:

41
 Introduction

𝑓 (𝑡𝑟 𝑛𝑠 𝑡 𝑜𝑛 𝑜 𝑒𝑛𝑡𝑢 ) ( ) ( 𝑟)
𝑡 𝑡 𝑡 𝑡 𝑡
𝒗 ⏟ 𝝎×𝒗 ⏟ 𝝎 × (𝝎 × 𝒓) ⏟ 𝝎×𝒓
𝑙 𝑓 𝑛 𝑓 𝑙𝑓 𝐸 𝑙 𝑓
The 𝑟 is the position vector. The Coriolis and Euler forces can often be assumed as zero in examples in
this compendium due to no motion within the vehicle and rigidity of rotating bodies. The general form
for the fictive moments is:
⃗⃗ 𝑓
𝑀 (𝑟𝑜𝑡 𝑡 𝑜𝑛 𝑜 𝑒𝑛𝑡𝑢 ) (𝑱 ⃗ ) 𝑱𝑪𝒐𝑮 𝝎 + 𝝎 × (𝑱𝑪𝒐𝑮 𝝎)
𝑡 𝑡 𝑪𝒐𝑮
The term 𝝎 × (𝑱𝑪𝒐𝑮 𝝎) can often be assumed as zero in examples in this compendium due to symme-
tries and rotation in one plane at the time.
Figure 1-22 and Figure 1-23 show examples of free-body diagrams (FBDs) with all forces, including
fictive forces. With such FBDs, the equilibrium equations are implicitly defined.
In Figure 1-22, the mass of the wheels can be included in the FBD of the vehicle body and rear axle, but
not the FBD of the front axle. The moment of inertia is marked as , but in practice we can often
neglect these terms in the FBD of the whole vehicle (but not for the FBD of the wheel, where they are
often essential in the equilibrium 𝑇 𝑊ℎ𝑒𝑒 𝑅 𝑢𝑠 ), see more in 3.3.5.2.
acceleration,

acceleration, Draw free body diagrams, 𝑓 𝑇𝑓

selecting the free bodies

𝑓 𝑓

𝑓 𝑓 𝑇𝑓 𝑓
Selecting:
Selecting: • vehicle body together 𝑓 𝑓
𝑓 • whole vehicle body with rear axle 𝑓
𝑓
with both axles • front axle alone 𝑓

Figure 1-22: Free body diagram. The dashed arrows marks “fictive” force or moment. The star is a
way to mark around which point(s) moment equilibrium is taken in the later stage “Mathematical
model”. The fictive moments 𝑓 𝑓 and are often neglectable in equilibrium for whole
vehicle, but significant in equilibrium for individual wheels.
In Figure 1-23, the prescribed motion actuation (steering angle (𝑡)) and good road grip result in 4
states, e.g. [ 𝑀 ] or [ 𝑀 ]. If unsprung parts have only a neglectable portion of
the total vehicle mass, it can be easier to use 0 and replace with the total vehicle mass. How-
ever, should not be changed, because it is still only the sprung body that rolls.

42
 Introduction

Free body diagram of Equilibrium for “sprung body”. Method with with fictive
“sprung body”: forces, e.g.:
Vehicle seen • Lateral: 0
from rear: • Vertical: + 0
• Roll around any point, e.g. the star ( ):
𝑀 + Δℎ + Δℎ 0
Alternative method, without fictive forces. Equally correct, but it
𝑪𝒐𝑮 ( ) only works with roll equilibrium around 𝑜 :
• Lateral in direction:
( ) 𝑀 • Vertical in direction:
• Roll around 𝑜 : 𝑀 + Δℎ + Δℎ
Δℎ
Equilibrium for “unsprung parts”:
• Lateral: + 0
ℎ

RC
• (Vertical and Roll could have been included. However, here, we
( ) 𝑀 do not aim at solving for and 𝑀 .)
ℎ

Compatibility:
ℎ

• ≈0 Δℎ +
ℎ ℎ +
Free body diagram of 𝑀 Constitution for suspension spring:
“unsprung parts” (axles): • 𝑀
Constitution for tyre-to-ground ( is steering angle):
•
Eqs (or dependencies) between parameters: Actuation (steering):
+ • 𝑓 𝑡 (𝑓 =known function of time 𝑡 )
ℎ ℎ + ℎ ℎ 8 eqs, 8 unknown variables ( 𝑀)

Figure 1-23: Free body diagrams for combined translation and rotation. The figure also shows the
two ways of setting up equilibria, with and without fictive forces. Angle is assumed small. The
is written as a general damping coefficient, while, in 2.2.4, it can be found that ⁄
§ A Modelica implementation of the model in Figure 1-23 is here shown:

model RollDynamics_Sprung_Unpsrung with m

delta 0 𝑡𝑜𝑛 : with m
delta 0 𝑡𝑜𝑛 :
1 1
… //declaration of parameters and variables
0 0
equation
//Equilibrium for sprung body: -1 -1
0 2 4 6 8 10 0 2 4 6 8 10
m_s*der(v_sy)-P_y=0; v_uy v_sy v_y v_uy v_sy v_y
-m_s*g-m_s*der(v_z)+P_z=0; 0.15 0.2
M_s -J_s*der(w_sx) +P_z*Dh_s*phi_sx 0.10
0.1
+m_s*der(v_sy)*Dh_s =0; 0.05
0.00 0.0
//Equilibrium for unsprung parts: -0.05
m_u*der(v_sy) -F_y +P_y =0; -0.1
-0.10
F_z=P_z;
-0.15 -0.2
P_y*h_RC +m_u*der(v_uy)*h_u +M_x -M_s =0; 0 2 4 6 8 10 0 2 4 6 8 10
M_x M_s M_x M_s
6E4 6E4
//Compatibility:
4E4 4E4
der(phi_sx)=w_sx;
v_z=0; 2E4 2E4

w_sx*Dh_s=-v_sy+v_uy; 0E0 0E0

w_sx*(h-h_RC)=-v_y+v_uy; -2E4 -2E4

-4E4 -4E4

//Constitution: -6E4
0 2 4 6 8 10
-6E4
0 2 4 6 8 10
der(M_s)=-c*w_sx; phi_sx
phi_x phi_sx
phi_x
F_y=-d*(v_uy-v_x*delta); 5 5
end RollDynamics_Sprung_Unpsrung; 0 0
-5 -5
0 2 4 6 8 10 0 2 4 6 8 10

Figure 1-24: § A Modelica implementation of the model in Figure 1-23 and two simulation results.
§ Centrifugal Forces in General 3D Motion
Centrifugal force terms appear as where ≠ 𝑗, i.e. for rotation perpendicular to translational
velocity:
0 + 0 +
[ ] [ ]+[ 0 + ] [ ] [ ] + [+ 0 ] [ ]
+ 0 + 0

43
 Introduction

+ – 𝑛 𝑜𝑠𝑡 𝑟𝑒 𝑒 𝑛𝑡 0
[ + ] ≈ { 𝑒ℎ 𝑒 𝑜𝑝𝑒𝑟 𝑡 𝑜𝑛𝑠: } ≈ [ ] + [+ ]
+ ≫ 𝑛 ≫
In most vehicle operations, the most important centripetal acceleration terms are: ∙ (see Eq
[4.45]) and ∙ (see Eq [3.24]).
§ Force Equivalence and Coordinate Transformation
Special cases of equilibrium equations are coordinate transformation between force components, e.g.
Eq [1.3]. Another is force equivalence. Note that it is not the same as force equilibrium. Force equiva-
lence is used e.g. in Figure 2-43 and Figure 2-97. A set of forces are equivalent with some other set if
they can be replacing each other in the free body diagram. It is recommended to not draw the two set
of forces in the same free body diagram, since they are alternative to each other.
𝑓 𝑤

Axle forces in 𝑓 𝑤

Wheel directions,
𝑭𝒇𝒙𝒘 𝑭𝒇𝒚𝒘 𝑭𝒓𝒙 𝑭𝒓𝒚
𝑓

Axle force vectors, 𝑭𝒇𝒙 𝑭𝒓𝒙

𝑓 𝑣

Axle forces in Vehicle

directions, 𝑭𝒇𝒙𝒗 𝑭𝒇𝒚𝒗 𝑭𝒓𝒙 𝑭𝒓𝒚 𝑓 𝑣

𝑤
𝑀𝑤

Summed Wheel forces,

𝑭𝒘𝒙 𝑭𝒘𝒚 𝑴𝒘𝒛 𝑤

Summed Vehicle forces including

Body forces, 𝑭𝒙 𝑭𝒚 𝑴𝒛
𝑛

𝑀 𝑛
Summed Vehicle forces including
Inertial forces, 𝑭𝒙𝑰𝒏 𝑭𝒚𝑰𝒏 𝑴𝒛𝑰𝒏 𝑛

Figure 1-25: § Coordinate Transformation and Force Equivalence of forces on a vehicle

1.5.2.3.2 Compatibility
Compatibility gives relations between motions (positions, velocities, accelerations, …). One example is
𝑆𝑝𝑒𝑒 𝑅 𝑢𝑠 𝑅𝑜𝑡 𝑡 𝑜𝑛 𝑠𝑝𝑒𝑒 for a purely rolling wheel or 𝑓 ≈ ( + 𝑓 )⁄ for body
side slip angle over front axle for a vehicle body moving in road plane.
(A special case of compatibility is a coordinate transformation for motion quantities.)
1.5.2.3.3 Constitution
Constitution are relations between forces (including moments) and motions, e.g.
• For a (linear) spring: 𝑜𝑟 𝑒 + 𝐷𝑒𝑓𝑜𝑟 𝑡 𝑜𝑛 or 𝑜𝑟 𝑒 𝐷𝑒𝑓𝑜𝑟 𝑡 𝑜𝑛𝑆𝑝𝑒𝑒 .
Metallic materials typically follow such linear behaviour, cf. Hooks law, if not deformed too
much or too fast.
• For a (linear) damper: 𝑜𝑟 𝑒 𝐷𝑒𝑓𝑜𝑟 𝑡 𝑜𝑛𝑆𝑝𝑒𝑒 . Shearing of thin liquid films typically
follows such linear behaviour, e.g. in lubricated bearing where shear force is proportional to
sliding speed.

44
 Introduction

• For a dry friction contact: 𝑜𝑟 𝑒 𝑜𝑛𝑠𝑡 𝑛𝑡 𝑠 𝑛(𝑆 𝑛 𝑆𝑝𝑒𝑒 ) . This is the most common
friction model in mechanical engineering, explained by adhesion between molecules and hys-
teresis when material is deforming over micro level asperities. So, the proportionality constant
depends on both cooperating bodies material and surface roughness. Note: When a friction
contact sticks, the equation switches to a compatibility equation (𝑆 𝑛 𝑆𝑝𝑒𝑒 0 ).
• For a more general model component: 𝑜𝑟 𝑒 𝑓𝑢𝑛 𝑡 𝑜𝑛( 𝑜𝑠 𝑡 𝑜𝑛 𝑆𝑝𝑒𝑒 ) . Even more gen-
eral would include models of actuators: 𝑜𝑟 𝑒 𝑓𝑢𝑛 𝑡 𝑜𝑛( 𝑜𝑠 𝑡 𝑜𝑛 𝑆𝑝𝑒𝑒 𝑆 𝑛 ) , where
𝑆 𝑛 often is a request signal, e.g. 𝑜𝑟 𝑒𝑅𝑒𝑞𝑢𝑒𝑠𝑡.
1.5.2.3.4 Algorithms and Other Equation Types
The listing of equation types in 1.5.2.3.1..1.5.2.3.2 is a help to model but it is not claimed to be com-
plete. There are many other equation types that can appear, e.g. from electrical and chemical science.
Among these others there are two sub-types (control algorithms and driver models) that are espe-
cially important for vehicle dynamics, so they will be discussed here in 1.5.2.3.4.
Via sensors and actuators, control algorithms can operate with the mechanics, mechatronics. The con-
trol algorithms with their interface to sensors and actuators is here included in the equation type “al-
gorithms”. We also include models of how the human driver controls and experience the mechanical
quantities. This equation type cannot be sorted into the traditional 1.5.2.3.1..1.5.2.3.1.2. Conceptually,
any quantities that can be sensed or actuated in, and outside of, the subject vehicle can occur in these
equations. (Finite) State machines are often useful when modelling (and designing) algorithm-based
functions but also the driver, see discrete state machine in 1.5.1.4.
A model of control algorithm can often be the same artefact as the design of it, especially if using a
modelling tool that allows automatic generation of real-time code, like Simulink. However, note that
the algorithms in real system is implemented in a time discrete digital computation platform and digi-
tal communication, so using a time continuous version as model is an approximation in itself. For best
fidelity, the models need to be formulated as time discrete dynamics. Then one can properly represent
the influence from the design parameter sample time on the vehicle functions.
1.5.2.3.5 § Lagrange Mechanics
An alternative to Newton mechanics is Lagrange mechanics, which is sometimes an easier method to
find the same equations. Lagrange mechanics is basically to express kinetic energy (𝑇) and potential
energy (𝑉) in (position) coordinates, Cartesian or Generlized. With Lagrange mechanics, the equilib-
rium is not formulated in forces (or moments) but in partial derivatives of energy (kinetic energy sur-
plus, called the Lagrangian 𝐿 𝑇 𝑉) with respect to these coordinates and their derivatives. One
can see this as “equilibria of projected energy”. Lagrange mechanics does not use the factorisation of
power into , so it undermines the categorization in 1.5.2.3.1..1.5.2.3.1.2; the compatibility and
constitution will be mixed in from start. So, the modularity corresponding to physical components in
the system often gets lost. Also, energy generating and dissipating components has to be treated with
special care, especially dry friction. Next figure shows an example of how the Lagrange method can be
applied on a simple system. The system has no external forces, i.e. it has no non-constraint forces. It
neither has energy exchange nor energy dissipation inside itself. Hence, the energy in the system is
constant. The example is the same as the one which Newton method is applied on in Figure 1-38.
Δ𝑥
Δ𝑥
𝑥 𝑥2
2
𝑥 𝑥2
Lagrange 1st kind Lagrange 2nd kind

45
 Introduction

(Cartesian) Coordinates: 𝒓 [𝑥 𝑥2 ] We have to, intuitively, select Generalized co-

Energies: 𝑇
𝑚1 12 𝑚
+ 2 2
2
,𝑉 + 𝑉 and 𝐿
2
2
𝑇 𝑉 ordinates: 𝒒 [𝑞 ] [𝑥 ]
2 2 2 (𝑚1 +𝑚2 ) 12
We have to, intuitively, identify Constraints: Energies: 𝑇 2
,
𝑓 𝑥2 𝑥 Δ𝑥 0 ( 1 +Δ ) 2
𝑉 2
+ 𝑉 and 𝐿 𝑇 𝑉
Lagrange equations (≈ “equilibria of projected energy”): Lagrange equations
𝜕𝐿 𝜕𝐿 𝜕𝑓 (≈ “equilibria of projected energy”):
( ) + ∑𝜆 𝟎 ⇒ 𝜕𝐿 𝜕𝐿
𝜕𝒓𝒌 𝑡 𝜕𝒓𝒌 𝜕𝒓𝒌 ( ) ⇒
= 𝑡 𝜕𝑞 𝜕𝑞
𝜕𝐿 𝜕𝐿 𝜕𝑓 𝜕𝐿 𝜕𝐿
𝜕𝑥 𝜕𝑥 𝜕𝑥 ⇒ ( ) ⇒
0 𝑡 𝜕𝑥 𝜕𝑥
⇒ +𝜆 [ ] ⇒
𝜕𝐿 𝑡 𝜕𝑓 𝜕𝑓 0 ⇒ ( + 2 ) 𝑞̈ (𝑞 + Δ𝑥)
[𝜕𝑥2 ] ([𝜕𝑥2 ]) [𝜕𝑥2 ]
0 𝑥̈
⇒ [ ] [ ] + 𝜆 [ ] [0]
𝑥2 2 𝑥̈ 2 + 0
Eliminate to independent coordinates, using the Con- Transform to Cartesian coordinates:
straints. Here, eliminate 𝑥2 𝑥 + Δ𝑥 , i.e. solve for 𝑥 and 𝑥̈
(x1 + Δ )
(2nd order ODE)
(the Lagrange multiplier) 𝜆 : 𝑚1 +𝑚2
x1 + Δ
𝑥̈ 𝑚 +𝑚
(2nd order ODE)
1 2
𝑚1
𝜆 (x + Δ𝑥) (algebraic equation)
𝑚1 +𝑚2

Figure 1-26: § Example of deriving Mathematical model using Lagrange method

We can identify 𝜆 as the constraint force from the Newton mechanics. But, 𝑓 can be formulated in
many ways which leads to other interpretations, e.g. multiplying 𝑓 by 2 gives 𝜆 ⁄ instead. So,
more generally, the Lagrange multipliers are linear combinations of constraint forces. Lagrange me-
chanics formulates the equations in kinetic and potential energy in a way which has similarities with
using velocities of bodies and forces in compliances as state variables, as described in 1.5.2.1.2.
1.5.2.3.6 § Relation between Equilibrium and Compatibility for Mechanisms
Consider a mechanism, meaning something which only interacts with its surrounding via velocities ,
including rotational velocities, and forces , including moments. Assume these are pairwise [ ] for
⋯ 𝑁. Examples are a transmission and a suspension linkage. Assume further that the mecha-
nism does neither includes energy storage nor power dissipation. We would typically need velocity
equations and force equations. For a certain position, or pose, these equations are proportional:
𝒗𝑭 𝑹
⏟𝒗 𝒗𝒗 and 𝑭𝒗 𝑹
⏟𝑭 𝑭𝑭 𝟎
𝑁𝐹𝐷𝑜𝑓 ×𝑁𝑣𝐷𝑜𝑓 𝑁𝑣𝐷𝑜𝑓 ×𝑁𝐹𝐷𝑜𝑓

where 𝒗 [ ⋯ 𝑁 ]𝑇 and 𝑭 [𝑭𝑇𝒗 𝑭𝑇𝑭 ]𝑇 [ ⋯ 𝑁 ]𝑇 .

[𝒗𝑇𝒗 𝒗𝑇𝑭 ]𝑇
The 𝑁𝑣𝐷 𝑓 is the number of motion degrees of freedom and 𝑁 𝐷 𝑓 is the number of force degrees of
freedom. The 𝑁 is the number of interfaces and 𝑁𝑣𝐷 𝑓 + 𝑁 𝐷 𝑓 𝑁.
We can derive 𝑹𝒗 and 𝑹𝑭 from compatibility and equilibrium, respectively. However, we can alterna-
tively find one of 𝑹𝒗 and 𝑹𝑭 from the other, using power equilibrium, see (Jacobson B. , 1993). Here is
the proof in short: The power input in each node is 𝑠 where 𝑠 + or if is
power to and fromp the mechanism, respectively. The power equilibrium becomes:

𝑛 ∑ ( 𝑠 ) 𝒗𝑻 ⏟
diag([𝑠 𝑠2 ⋯ 𝑠𝑁 ]) 𝑭 0 ⇒
= ⋯𝑁 =diag([ 𝒗 𝑭])
⇒ [𝒗𝒗 𝒗𝑻𝑭 ] [( 𝒗 𝑭𝒗 )𝑻
𝑻 ( 𝑭 𝑭𝑭 )𝑻 ]𝑻 0
⇒ [𝒗𝑻𝒗 (𝑹𝒗 𝒗𝒗 )𝑻 ] [( 𝒗 𝑹𝑭 𝑭𝑭 )𝑻 ( 𝑭 𝑭𝑭 )𝑻 ]𝑻 0
⇒ 𝒗𝑻𝒗 ( 𝒗 𝑹𝑭 + 𝑹𝑻𝒗 𝑭 ) 𝑭𝑭 0 ⇒ 𝒗 𝑹𝑭 𝑹𝑻𝒗 𝑭

𝑭 or
𝑻 𝑇
⇒ 𝑹𝑭 𝒗 𝑹𝒗 𝑹𝒗 𝑭 𝑹𝑭 𝒗

46
 Introduction

For a transmission, such as a planetary gearbox with many shafts, it can often be most natural to de-
rive the 𝑹𝒗 from compatibility and 𝑹𝑭 from 𝑹𝑭 𝑻
𝒗 𝑹𝒗 𝑭. For a suspension linkage, it can often
be most natural to derive the 𝑹𝑭 from equilibrium and 𝑹𝒗 from 𝑹𝒗 𝑭 𝑹𝑭
𝑇
𝒗 . Note that, for a
suspension, the proportional relations expressed with 𝑹𝒗 and 𝑹𝑭 , are only valid close around a certain
position or pose.
If the system studied has energy storage or power dissipation included, they can often be defined as
outside the mechanism, only leading to more interfaces. However, if the energy dissipation is of dry
friction type, this is not easily possible. Instead, the geometry can be adjusted via a geometry interpre-
tation of dry friction, using a friction angle which defines a corrected geometry, see (Mägi, Jacobson, &
Resev, 1998).

1.5.3 Control Engineering

Vehicle dynamics is more and more influenced by electronics, where algorithms are the main artefact
to engineer. For this reason, this section introduces some relevant theory and methods and their con-
nection to vehicle models as they are described in vehicle dynamics. From Figure 1-27, we realize that
there are many types of Vehicle Level Algorithms, and the Vehicle Motion Controller is maybe the most
central for vehicle dynamics. It should be underlined that the control structure in today’s vehicles are
not as clean and structured as in Figure 1-27, very much depending on that vehicle level control is dis-
tributed in several ECUs, each belonging to its own subsystem.

Vehicle
Vehicle Level Algorithms
requests on requests on
Environment Driver Vehicle
vehicle, 𝒚𝒓𝒆𝒒 actuators, 𝒓𝒆𝒒 Actuated and
(road and Driver Interp- Motion
traffic) reter
Arbitra- Controller Sensing Vehicle
Virtual tor information,
𝒂𝒄𝒕 𝒂𝒍 𝒄𝒂𝒑𝒂𝒃𝒊𝒍𝒊𝒕𝒚
Driver

Estimator
and Sensor
information Fusion

Vehicle variables (positions, velocities, forces, etc.) and

parameters (mass, wheelbase, etc.)

(Open loop) Vehicle Motion (Closed loop) Vehicle Motion Controller (of gain type)
Controller
𝒚𝒓𝒆𝒒 + request for request,
Algorithm
𝒚𝒓𝒆𝒒 Algorithm 𝒓𝒆𝒒 - increase, 𝒚𝒓𝒆𝒒 Δ𝑡 𝒚𝒓𝒆𝒒 𝒓𝒆𝒒

information information
except 𝒚𝒂𝒄𝒕 𝒂𝒍
𝒚𝒂𝒄𝒕 𝒂𝒍 except 𝒚𝒂𝒄𝒕 𝒂𝒍

Figure 1-27: Context for Vehicle Level Algorithms and Vehicle (Motion) Controller. “Virtual Driver”
takes vehicle driver into account and closes the loop in vehicle speed and lateral position.
Control algorithms can be designed without utilizing knowledge about the controlled system, such as
neural networks, tuned only on observations on how the system responds. However, in this section we
only consider Model based controllers. For those, the input and output signals, as well as parameters
inside, has a clear physical interpretation in vehicle quantities, as well as units, see Figure 1-28. For
instance, the requests on vehicle, 𝒚𝒓𝒆𝒒 , is typically forces or acceleration, [ [𝑁] [𝑁] 𝑀 [𝑁 ]] or
[ [ ⁄𝑠 2 ] [ ⁄𝑠 2 ] [𝑟 ⁄𝑠 2 ]] or velocities [ [ ⁄𝑠] [ ⁄𝑠] [𝑟 ⁄𝑠]]. The model base is
helpful when tuning the controller parameters. Moreover, the model base helps to find a consistency

47
 Introduction

between derivatives order in the controller and the controlled system, such as if P, I or D gains should
be used in a PID-controller.
The 𝒚𝑹𝒆𝒒 contains typically vehicle motion variables (positions, velocities, accelerations, forces) but
other vehicle variables, such as energy or power variables can be added, e.g. 𝒚𝑹𝒆𝒒 [ 𝑀 𝑓𝑓 ],
where 𝑓𝑓 is energy level (or state of charge 𝑆𝑜 ) in an energy buffer.
⋯

Vehicle
Motion
𝑣 𝑣
Control e.g. ←

e.g. ← sin

𝑓𝑖
𝑇 𝑝 +𝑇 𝑘
Actuator Coordinator 2
𝑇𝑙 𝑘
𝑤
𝑙 𝑟𝑓 𝑛 𝑙 𝑇 𝑝
𝑤 𝑅𝑤
𝑇 𝑝 𝑇𝑙 𝑘
𝑇 𝑇 𝑘
𝑘
𝑒
Actuators +
𝑙
Prop Brk

Figure 1-28: Request flow in Vehicle Motion Control. Each block has its own Physical model, which
explains the its interface quantities (variables and parameters).
⋯ 2 2
Motion Model
2 2 2
Vehicle
Motion
2 2
Control

2 Force Model 1
2 2 2 2

2 2

2
Motion quantities inherited from Motion Model
but not drawn. Not drawn forces and moments:
𝑀 𝑀 From air on bodies: 𝑀 𝑀 2
2
From ground on bodies: 2
From tyre on each unit : 𝑀 𝑀 2

Force Model 2
Axel Group Coordinator For both units, but only drawn for unit 2.
Force and
moment
𝑀 equivalence
𝑓
⋯ 2
⋯ 𝑀 𝑀 𝑀
2 2𝑓 2

Actuation Models
Actuator Actuator Actuator
Not drawn, but they can typically
Coordinator Coordinator Coordinator
be drawn per axle group.

Requests on Wheel/Axle Torques, Steering angles

⋯ ⋯ ⋯ ⋯ ⋯

Actuators (modularity depends on purchase strategies)

Propulsion Steering TruckBrk TrailerBrk

Figure 1-29: § Similar to Figure 1-28, but for a more complex vehicle.

48
 Introduction

1.5.3.1 § Decision Making or Execution

Nowadays, one can often hear Decision-making algorithms as an expression. It means some kind of
rules for when to transit between discrete states, such as transit from LaneFollowing to LaneChange in
automated driving or switching from Wheel Torque Control to Wheel Speed Control for ABS/TC-like
functions. The complement to Decision making can be called Actuation execution and it would typically
mean the computation of continuous request signals, such as Steering Wheel Angle Request or Brake
Pressure Request, in the two above examples.

1.5.3.2 § Momentaneous or Predictive Control

If the controller considers the future (predicted) of its input 𝒚𝒓𝒆𝒒 and/or output 𝒓𝒆𝒒 , the controller
uses Predictive control. If the controller was designed to consider both its input 𝒚𝒓𝒆𝒒 and output 𝒓𝒆𝒒
at same time instant, the controller uses Momentaneous control. This categorization is not strict, but
often useful. The future of the input cannot be known, so future of inputs 𝒓𝒆𝒒 should be seen as pre-
diction e.g. expressed in time or desired path ahead. A predictive controller generally needs a model of
the environment and vehicle such that the consequences (cost [€] and possibly constraints [pass or
fail] can be predicted for different candidates of 𝒓𝒆𝒒 . See 3.5.2.2.1.
A reason to do predictive control, as opposed to the simpler momentaneous control, is that the control
objective requires some strategic planning over future time. One example can be fuel consumption,
where the objective is an integral over future time, 𝑢𝑒 ∫ 𝑢𝑒 𝑅 𝑡𝑒 𝑡 , where a high 𝑢𝑒 𝑅 𝑡𝑒
can be good initially (momentaneously) to minimize 𝑢𝑒 . Another example is how to keep high
transport speed while negotiating a curve, where it is generally more strategically optimal to brake
before steering. The objective is an integral 𝑜𝑠 𝑡 𝑜𝑛 ∫ 𝑉𝑒 𝑜 𝑡 𝑒𝑠 𝑡 . See (Ross,2015).
In 1.5.3.3 and 1.5.3.4, momentaneous control will be assumed unless else is stated.
An example of difference can be that the driver interpreter can output momentary 𝒚𝒓𝒆𝒒 such as
[ 𝑀 ] or [ ] and an energy management in line with this would output [ 𝑓𝑓 ]. A predic-
tive version of 𝒚𝒓𝒆𝒒 is typically [ (𝑡) (𝑡) 𝑀 (𝑡)] or [ (𝑡) (𝑡) (𝑡)] and [ 𝑓𝑓 (𝑡)], where ⋯ (𝑡)
marks that the request is then a plan ahead in time. The plan can also be in travelled distance ahead,
⋯ (𝑠).

1.5.3.3 § Open or Closed Loop Control

As indicated already in Figure 1-27, a Closed Loop control uses the actual value (𝒚𝒂𝒄𝒕 𝒂𝒍 ) of the re-
quested quantity 𝒚𝒓𝒆𝒒 . Else it is an Open Loop control, which can also be called Feed Forward control
or Inverse Model Control.
A driver is in the loop can correct errors, so Open Loop often works well enough. Especially if the
driver is given good feedback, e.g. good steering feel. When automation of driving there is a motivation
to go towards more Closed-loop control. However, there can be architectural benefits to close the loop
in the Virtual driver instead of the Vehicle Motion Controller, see Figure 1-27, so that same algorithms
can used in Vehicle Motion Controller for manual and for automatic driving.
A more including definition of closed loop controller is that the controller uses any signal from the
controlled object, not exclusively whether it uses 𝑙 . With that definition, almost all controllers
becomes closed loop controllers. An example is an open-loop controller for a cruise control function:
← + . Here, the requested vehicle quantity is acceleration ( ) but the used
information about the vehicle is not 𝑙 but estimated mass ( ) and estimated driving resistance
force ( ). These estimated quantities are typically updated much slower than the request control
loop, but they are still a kind of closed loop. With the initial definition, it is open loop controller, but
with the more including definition it would be a closed loop controller.
The cruise controller example above is an example of Open Loop control where the Inverse model is
very clear: ← ( )/ . The most common implementation of Closed Loop is when the con-
trol error is calculated as an intermediate quantity. In the cruise control example, it would be
← 𝑙 or with opposite sign it is rather a differential request Δ ←

49
 Introduction

𝑙 ( ) . A Gain controller could then be: ← 𝑛 Δ . This can be called a

P-control on acceleration or D-control on velocity. The model connection is not obvious since one can-
not directly see that 𝑛 corresponds to mass and resistance force is not even present. A similar gain
controller could be formulated in another differentiation order, such as velocity instead of accelera-
tion: ← 𝑛 Δ . This can be called P-control on velocity, I-control on acceleration or D-
control on position. The model ( ) combined with the controller ( ← 𝑛 Δ )
points out that 𝑛 should be seen as 𝑠𝑠/𝑡 𝑒, where 𝑡 𝑒 is the time to when should be
reached. Interpretation of gains to physical quantities with units is encouraged, since it facilitates tun-
ing and reuse of the controller. Such interpretation requires a physical model.
It is typical for automotive engineering, that algorithms are reused, and sometimes incrementally de-
veloped, from one platform or vehicle programme to the next. Hence, given a certain controller, it is
very relevant to understand/derive its model base, for efficient parameter setting. However, it is im-
portant to realize that it is not generally possible to derive the model base from a given controller. So,
the controller really has to be developed from a physical model. So, the documentation of which physi-
cal model is used is very important output from the control design, beside the controller itself. The
documentation should include model equations and declarations of variables and parameters so that
next vehicle programme can inherit this as document.

1.5.3.4 § Degree of Over-Actuation

Already before designing the controller, it is often clear what vehicle requests 𝒚𝒓𝒆𝒒 and actuateable
quantities 𝒓𝒆𝒒 there are. The number (or dimension= ) of those enables one categorization:
• If ( 𝒓𝒆𝒒 ) (𝒚𝒓𝒆𝒒 ), the vehicle is (formally) Neutral actuated.
• If ( 𝒓𝒆𝒒 ) < (𝒚𝒓𝒆𝒒 ), the vehicle is (formally) Under-actuated ( 𝒓𝒆𝒒 is over-determined)
• If ( 𝒓𝒆𝒒 ) > (𝒚𝒓𝒆𝒒 ), the vehicle is (formally) Over-actuated ( 𝒓𝒆𝒒 is under-deter-
mined). In a function architecture, this requires a Coordinator in 1.6.4.2.
Consider that the vehicle model is 𝒇𝑫𝑨𝑬 (𝒙 𝒚 𝑡) 0 , or in explicit form [𝒙 𝒚]
[𝒇𝑶𝑫𝑬 (𝒙 𝑡) 𝒈(𝒙 𝑡)] . With such a model, we can include saturations (actuators saturated or road
friction saturated) and dependencies between actuators (e.g. >1 torque actuator acting on one wheel).
Instead of dimension, we define the property “rank of the model” as the number of dimensions that the
vehicle motion can be influenced in by the available actuators. Using a linearization of the model, we
can find 𝑩 such that:
𝒚 ≈ 𝒈(𝒙𝟎 𝟎 𝑡 ) + 𝑩(𝒙𝟎 𝟎 𝑡 ) ( 𝟎) 𝒈𝟎 + 𝑩(𝒙𝟎 𝟎 𝑡 ) ( 𝟎)
𝜕
ℎ𝑒𝑟𝑒 𝑩 ℎ 𝑠 dim(𝒚) 𝑟𝑜 𝑠 𝑛 dim( ) 𝑜 𝑢 𝑛𝑠 𝑛 𝐵
𝜕𝑢
"𝑅 𝑛𝑘 𝑜𝑓 𝑜 𝑒 " rank(𝑩).
The controller should then aim at finding 𝒓𝒆𝒒 so that 𝒚𝒓𝒆𝒒 (𝒈𝟎 𝑩 𝟎 ) 𝑩 𝒓𝒆𝒒 . We identify
“virtual requests” 𝒗𝒓𝒆𝒒 𝒚𝒓𝒆𝒒 (𝒈𝟎 𝑩 𝟎 ) of same dimension as 𝒚𝒓𝒆𝒒 , so 𝒗𝒓𝒆𝒒 𝑩 𝒓𝒆𝒒 .
The rank of 𝑩 (or any matrix) is the number of non-zero diagonal elements in the middle matrix in
the singular value decomposition of 𝑩, 𝑠 (𝑩) 𝑼 𝑽𝑇 ;. Using the rank, we can define degree of
over-actuation in a way which better helps us differ between different control design types, see also
Figure 1-30:
• If dim( 𝒓𝒆𝒒 ) rank(𝑩), the vehicle is (effectively) Neutral actuated.
Conceptually, 𝒓𝒆𝒒 could be calculated from 𝒗𝒓𝒆𝒒 , using the model, which would be an Inverse
model control design: 𝒓𝒆𝒒 ← 𝑩 𝟏 𝒗𝒓𝒆𝒒 .
• If dim( 𝒓𝒆𝒒 ) < rank(𝑩), the vehicle is (effectively) Under-actuated ( 𝒓𝒆𝒒 is over-determined).
If actuators cannot be added and requests cannot be disregarded, one has to apply optimiza-
tion, e.g. to minimize 𝑓 ‖𝒚𝒓𝒆𝒒 𝒈(𝒙 𝒓𝒆𝒒 𝑡)‖ with some weigths on the components.
There is generally a unique optima. If we can use the linear model and set weigth factors, we
can rewrite to minimizing the 𝑓 ‖𝑾𝒗 (𝒗𝒓𝒆𝒒 𝑩 𝒓𝒆𝒒 )‖ , where 𝑾𝒗 is a square weigth
2

50
 Introduction

matrix. Optimizing in Least Square sense (a.k.a. using quasi-inverse) gives 𝒓𝒆𝒒
((𝑾𝒗 𝑩)𝑇 𝑾𝒗 𝑩) (𝑾𝒗 𝑩)𝑇 𝑾𝒗 𝒗𝒓𝒆𝒒 (𝑾𝒗 𝑩)" " 𝑾𝒗 𝒗𝒓𝒆𝒒 .
• If dim( 𝒓𝒆𝒒 ) > rank(𝑩), the vehicle is (effectively) Over-actuated ( 𝒓𝒆𝒒 is under-determined).
Conceptually, there is no way to convert the problem to a Neutral actuated case (without disre-
garding actuators or adding requests), but one can find and approximate solution through opti-
mization. Trying an optimization as for the under-actuated vehicle will not work well, since
there is generally no unique solutions, i.e. ((𝑾𝒗 𝑩)𝑇 𝑾𝒗 𝑩) is singular). One way to handle
it would be to use some random way to find one 𝒓𝒆𝒒 among all non-unique optima. This can
lead to jumping 𝒓𝒆𝒒 in time, which is undesirable. A second way is to convert the problem to a
Neutral actuated case by simply prescribing suitably many components in 𝒓𝒆𝒒 : 𝑢 𝑢 ,
where subscript 𝑒𝑠 means desired. A third way is to to convert the problem into a Neutral- or
Under-actuated case by adding equations for each and all components in 𝒓𝒆𝒒 , typically: 𝒓𝒆𝒒
𝟎. To use zero is difficult to motivate directly from vehicle motion/actuation perspective but it
can be thought of as a way to guaranteeing unique mathematical solutions, which is good.
Combining second and third way gives an optimization problem with cost function as follows:
𝑓 ‖𝑾 𝒐𝒕𝒊𝒐𝒏 (𝒗𝒓𝒆𝒒 𝑩 𝒓𝒆𝒒 )‖ + ∑‖𝑊 (𝑢 𝑢 )‖ + ‖𝑾 𝒐𝒐𝒕𝒉 𝒓𝒆𝒒 ‖ .
2 2 2
The 𝑾 matrices are diagonal matrices with weight factors. For solving such mathematical
problems there are well established subroutines for Control allocation, CA, which can also
take constraints into account.
A visualisation of these cases and solution is given in Figure 1-30.

Neutral actuated vehicle: 𝑩

dim=2 dim=2 dim=2
𝒗𝒓𝒆𝒒 𝒚𝒓𝒆𝒒 Actuated vehicle:
𝒚𝒓𝒆𝒒 𝒗𝒓𝒆𝒒 𝒓𝒆𝒒 𝑩 𝒗𝒓𝒆𝒒 𝒓𝒆𝒒 𝒇𝑫𝑨𝑬 𝒙 𝒚𝑡 𝟎
𝒈 𝑩 𝟎 𝒓𝒆𝒒

Under-actuated vehicle: 𝑩
𝑒 𝑓
dim=3 dim=3 𝒓𝒆𝒒 𝑩" " 𝒗𝒓𝒆𝒒 or dim=2
𝒗𝒓𝒆𝒒 𝒚𝒓𝒆𝒒 Actuated vehicle:
𝒚𝒓𝒆𝒒 𝒗𝒓𝒆𝒒 𝒓𝒆𝒒
𝒈 𝑩 𝟎 𝑢𝒓𝒆𝒒 arg min 𝑜𝑠𝑡 𝒓𝒆𝒒
𝒇𝑫𝑨𝑬 𝒙 𝒓𝒆𝒒 𝒚𝑡 𝟎
𝒓𝒆𝒒

Over-actuated vehicle: 𝑩
𝑒 𝑓
" " 𝒗𝒓𝒆𝒒
𝑩
dim=2 dim=2 dim=3
𝒗𝒓𝒆𝒒 𝒚𝒓𝒆𝒒 𝒓𝒆𝒒 𝑢 0 or Actuated vehicle:
𝒚𝒓𝒆𝒒 𝒗𝒓𝒆𝒒 𝑰 𝟎 𝒓𝒆𝒒
𝒈 𝑩 𝟎 𝒇𝑫𝑨𝑬 𝒙 𝒓𝒆𝒒 𝒚𝑡 𝟎
𝒓𝒆𝒒 arg min 𝑜𝑠𝑡 𝒓𝒆𝒒
𝒓𝒆𝒒

Figure 1-30: § Visualisation of possible conceptual controllers for vehicles with different degree of
over-actuation. Superscript ⋯" " denotes “quasi-inverse”. Weigthing is not shown.
1.5.3.4.1 Momentaneous Optimization
Note that the weigthing is not shown in Figure 1-30. The weighting is probably the most difficult, in
some sense, one have to compare costs for different motion components, so if simple weigth matrix, its
components should typically have the units [€⁄𝑁] in longitudinal and [€⁄𝑁] in lateral which is very
difficult to set and might vary with such as distance to lead vehicle and distance to road edge. Also, if
𝑦 includes energy related requests, the units can be of type [€⁄ ] depending on the fuel or electric
energy price. So, the approach described above is very mathematical and a bit naive when we see the
original engineering problem, see 1.5.1.1.5.3.
Instead of a naive weigthing “outside” the model, we can include the cost as a variable already in our
model. The cost will then appear more intuitive and less artificial. Typically, one see two types of cost
variables: 𝑜𝑠𝑡 in [€] and 𝑜𝑠𝑡𝑅 𝑡𝑒 in [€/𝑠]. For transport costs (fuel, driver salary, etc) these are often

51
 Introduction

very natural. For motion they are still very difficult to grasp, which suggest to look for ways to see mo-
tion requests as constraints (to fulfil exactly) and transport costs as costs (to be minimized).
Looking only at transport costs, one realizes that the overall goal is to minimize the 𝑜𝑠𝑡 [€], but that
has to be done predictive, i.e. by integrating 𝑜𝑠𝑡𝑅 𝑡𝑒 [€/𝑠] over a longer time, see 1.5.3.2 and
1.5.3.4.2. However, if the motion constraints should be fulfilled as constraints over the time, it is prob-
ably a good idea to minimize 𝑜𝑠𝑡𝑅 𝑡𝑒 [€/𝑠] momentanously (at each time instant) such that the mo-
mentaneous motion constraints are fulfilled.
An example of such optimization is given here: A vehicle with two propulsion motors, subsrcitps 1 and
2, is modelled as:

𝑟 𝑡 𝑜 (𝑇 + 𝑇2 )
2 𝑟 𝑡𝑜
𝑓𝑢𝑒 𝑅 𝑡𝑒 𝑓 (𝑇 ) + 𝑓2 (𝑇2 2 )
𝑜𝑠𝑡𝑅 𝑡𝑒 𝑓𝑢𝑒 𝑅 𝑡𝑒 𝑓𝑢𝑒 𝑟 𝑒
The 𝒗𝑹𝒆𝒒 [ ] and 𝑹𝒆𝒒 [𝑇 𝑇2 ].
Minimizing 𝑜𝑠𝑡𝑅 𝑡𝑒 under constraint of fulfilling motion request :
{𝑇 𝑇2 } arg (min( 𝑜𝑠𝑡𝑅 𝑡𝑒(𝑇 𝑇2 )) 𝑠𝑢 𝑗𝑒 𝑡 𝑡𝑜 𝑟 𝑡 𝑜 (𝑇 + 𝑇2 ) )
𝑇1 𝑇2
where 𝑜𝑠𝑡𝑅 𝑡𝑒(𝑇 𝑇2 ) (𝑓 (𝑇 𝑟 𝑡 𝑜 ) + 𝑓2 (𝑇2 𝑟 𝑡 𝑜 )) 𝑓𝑢𝑒 𝑟 𝑒
There are methods to find the optimum for certain forms of the 𝑜𝑠𝑡𝑅 𝑡𝑒 and the constraint. One ex-
ample is quadratic programming (e.g. Matlab command quadprog), which assumes that 𝑜𝑠𝑡𝑅 𝑡𝑒 can
be formulated on a quadratic form:
𝑻
𝑜𝑠𝑡𝑅 𝑡𝑒 𝑹𝒆𝒒 𝑸 𝑹𝒆𝒒 + 𝒇𝑻 𝑹𝒆𝒒 𝑠𝑢 𝑗𝑒 𝑡 𝑡𝑜 𝑩 𝑹𝒆𝒒 𝒗𝑹𝒆𝒒
This form includes both linear and quadratic terms, so coefficient can often be found which makes a
rather good approximation.
1.5.3.4.2 Momentaneous and Predictive Optimization
The above in the 1.5.3.4 could be applied also for predictive control if we think of 𝒓𝒆𝒒 and 𝒚𝒓𝒆𝒒 as hav-
ing 𝑁 times more elements, where 𝑁 is the number of time instants considered in the prediction. The
model can be applied 𝑁 times, and if it contains state variable, a time integration method is needed to
make the sequences of each variable consistent. If model is linear and optimization is to be applied,
there are cooking-book methods, such as Model Predictive Control (MPC), see e.g. Ref (Ross,2015),
describes MPC. Simply adding more time instants adds equally many equations as unknowns ( 𝒓𝒆𝒒 ), so
it does not change the degree of over-actuation. However, if the variation of 𝒓𝒆𝒒 (𝑡) is parameterised
(e.g. 𝒓𝒆𝒒 (𝑡) 𝒑 or 𝒓𝒆𝒒 (𝑡) 𝒑𝟏 + 𝒑𝟐 𝑡 ), the problems becomes as under-actuated, which is more
well-conditioned and can be solved in a least square sense.
Over actuated passenger car can appear if each wheel has controllable actuation, see (Jonasson, 2009).

1.5.3.5 § Filtering, Estimation and Differentiation

Filtering is often used to reduce noise in measurements of variables. The result is a smoother signal of
the same quantity. We can also use filters to do variable estimation; then the filter is typically based on
a physical model so that input is one quantity and output is another quantity, e.g.
⁄ 𝑡 ( 𝑛 ) , where ⁄ 𝑡 require a filter to implement.
The result from the filter can either be needed on-line as input to a controller in real-time (during use
of the vehicle) or off-line (post-processing) after the measurement. On-line filters can only use the his-
tory of the signal and delays in the filter are therefore unavoidable. Off-line filters can use the logged
signal both before and after, by which delays can be avoided.
Estimation means to find an approximation of a physical quantity which is not directly measured, so it
requires a model. Having a model, one can differ between estimation of variables and parameters.

52
 Introduction

(Whether a physical quantity is a variable or a parameter is cannot be uniquely decided in the real
world.) Generally speaking, variable estimation is more typically on-line filtering while parameter esti-
mation can have more of off-line character. Off-line can log several data samples and then find (the
constant) parameter value, best fitted for the whole sample period, e.g. in least square sense. Parame-
ter estimation can also be recursive, which means that each new sample is weighted with previous pa-
rameter estimate; so the estimate is a memory, i.e. a state.
Figure shows an example of an on-line filter is a first order filter. Figure also shows a simple differenti-
ator. Figure shows both continuous and discrete implementation in a block diagram format. Imple-
mentation of discrete integration using when statements and the pre() operator in Modelica was
shown in 1.5.1.4.
Continuous:
≈ ∫ 𝑡
+ 𝑡 𝑓𝑓 𝑡 𝑓𝑓
∫
𝑡 𝑓𝑓
Smaller time delay 𝑡 𝑓𝑓 gives more
correct differentiation and less
(Time) discrete: delayed . But, it also requires more
of the continuous time integration
+ 𝑡 𝑚𝑝𝑙 (smaller steps or higher order), or
𝑡 𝑓𝑓 smaller sample time 𝑡 𝑚𝑝𝑙 in the
discrete time integration.
Figure 1-31: § Example of differentiator (output ) and on-line filter (output y) as block diagram.
Upper: Continuous. Lower: (Time) Discrete.
In dynamic models for simulation, one should avoid differentiation because it introduces delays which
can lead to numerically unstable or wrong solutions. Off-line differentiation is much less troublesome.
However, if the derivative is not fed back, the problems become isolated with less risk for troubles.
Physical-model-based variable estimators (or State Observers) are often basically designed as simula-
tion of a model driven by the sensed inputs 𝒆𝒏 .
Physical model: Mathematical model: 𝑇
𝑇 ←
0 𝑇 𝑅 𝑅
𝑇
Explicit form model (𝒇 𝒈) :
𝑇 𝑛 + ∫
← 𝑡 𝑓𝑓
𝑅
(𝑛𝑜 𝑠𝑡 𝑡𝑒 𝑝𝑢𝑟𝑒 𝑦 𝑒 𝑟 )
Differentiator
To align both physical quantities (𝑇 and ) to same real time
instant, it is recommended to filter 𝑇 with same delay 𝑡 𝑓𝑓 .

Figure 1-32: § Concept and example of Physical-model-based variable estimators involving a

differentiator.

53
 Introduction
Treqveh vehicle_OneWheel_Act.w fxEst_Diff_Cont.w _est fxEst
model FxEst_Diff_Cont 70
parameter Real DifferentiationTime=0.1; 0E0 step in brake torque request
60
input Real w; -1E4
Real w_est, T_est, Fx_est; 0 1 2
50
input Real T, R, J; vehicle_OneWheel_Act.mustick vehicle_OneWheel_Act.muslip
equation 2 40
der(w_est) = (w - w_est)/DifferentiationTime; step up in friction
der(T_est) = (T - T_est)/DifferentiationTime; 30 wheel speed
0 • real
0 = T_est - Fx_est*R - J*der(w_est); 0 1 2
end FxEst_Diff_Cont; vehicle_OneWheel_Act.vx Rw
20 • continuously estimated
• discretely estimated
20 10
model FxEst_Diff_Discr vehicle and wheel speed
parameter Real SampleTime=0.05; 0
18
parameter Real DifferentiationTime=0.1; 0 1 2
input Real w;
16
Real w_est, T_est, Fx_est; vehicle_OneWheel_Act.T fxEst_Diff_Cont.T_est fxEst_D
input Real T, R, J; 0
14
Real derw_est, derT_est; wheel speed
equation • real
12 -500
when sample(-1000, SampleTime) then 0 1 2 • continuously estimated
derw_est = (w-pre(w_est))/ DifferentiationTime; vehicle_OneWheel_Act.sx • discretely estimated
w_est=pre(w_est)+ -1000
((pre(derw_est)+derw_est)/2)*SampleTime; 0.00
wheel slip
derT_est = (T-pre(T_est))/DifferentiationTime; -0.05 -1500
T_est=pre(T_est)+
((pre(derT_est)+derT_est)/2)*SampleTime; -0.10 -2000
end when;
0 = T_est - Fx_est*R - J*derw_est;
end FxEst_Diff_Discr; -0.15 -2500
0 1 2 0 1 2

Figure 1-33: § Continuous and discrete implementation of estimator in Figure 1-32. Tested with a
one-wheel vehicle model with ideal, but sampled, sensing of wheel speed and wheel brake torque.
Implicit midpoint or Trapezoidal rule (or Tustin approximation) used as derivative approximation
for the discrete integration.

1.5.4 Tools
This section presents some tools and methods for modelling and computation.

1.5.4.1 General Mathematics Tools

Contributions from Mats Jonasson, Volvo Cars and Vehicle Dynamics at Chalmers

Examples of tool: Matlab, Matrixx, Python

We will take Matlab as example. Matlab is a commercial computer program for general mathematics. It
is developed by Mathworks Inc. Compendium will use some simple Matlab code to describe models in
this compendium. The following are useful for dynamic models:
General help function, here shown for function “inv” >> help inv
Solve linear systems of equations, 𝑨 𝒙 𝒃: >> x=inv(A)*b;
Solve non-linear systems of equations, 𝒇(𝒙) 𝟎: >> x=fsolve('f',...);
Solve ODE as initial value problems, 𝒙 𝒇(𝑡 𝒙): >> x=ode23('f',x0,...);
Find Eigen vectors (V) and Eigen values (D) to systems: 𝑫 𝑽 𝑨 𝑽: >> [V,D]=eig(A);
Find 𝒙 which minimizes 𝑓(𝒙) under constraints 𝑨 𝒙 ≤ 𝒃: >> x=fmincon(f,x0,A,b);
Find the 𝒙 which minimizes 0.5 𝒙𝑇 𝑯 𝒙 + 𝒇𝑇 𝒙: >> x=quadprog(H,f);
Matlab is mainly numerical, but also has a symbolic toolbox:
>> syms x a; Eq=a/x+x==0; solve(Eq,x) %symbolically solve equation
ans = (-a)^(1/2)
-(-a)^(1/2)
>> diff(a/x+x,x) %symbolically differentiate expression
ans = 1 - a/x^2
>> int(x^3+log(x),x) %symbolically integrate expression
ans = (x*(4*log(x) + x^3 - 4))/4
>> F = laplace(f) %F = Laplace transform of the expression f
>> f = ilaplace(F) %f = inverse Laplace transform of the expression F
>> clear all; syms f t F s a w; f=a*sin(w*t); F=laplace(f), f=ilaplace(F)

54
 Introduction

F = (a*w)/(s^2 + w^2)
f = a*sin(t*w)
>> partfrac(f,x) %is the partial fraction decomposition of the expression f
>> clear all; syms f x; f=(7*x+3)/(5*x^2-3*x+0.5); partfrac(f,x)
ans = (14*x + 6)/(10*x^2 - 6*x + 1)
>> clear all; syms f x; f=(7*x+3)/(5*x^2-3*x+0.5); partfrac(f,x,'Factor-
mode','complex')
ans = (0.7 - 5.1i)/(x - 0.3 - 0.1i) + (0.7 + 5.1i)/(x - 0.3 + 0.1i)
>> ABCD=ss(A,B,C,D); %creates an object ABCD representing the linear state
space model “A,B,C,D”, see 1.5.1.1.3.2
>> step(ABCD) %computes step response of the model ABCD, see Figure 1-34.

Figure 1-34: Plot resulting from Matlab cammands ss and step:

>> clear all; ABCD=ss([-0.5,-1;1,0],[1;0],[0,1],[0]); step(ABCD), grid on

Of special interest for dynamic systems is that Matlab has a built-in function for ”matrix exponen-
tial”, mentioned in 1.5.1.1.5. E.g., if 𝒙 𝑨 𝒙 with 𝒙(0) 𝒙𝒊𝒗 the solution is 𝒙(𝑡) 𝑒𝑨 𝒙𝒊𝒗 which can
simply be computed as:
>> x=expm(A*t)*x_iv; %with A as (square) matrix

1.5.4.1.1 § Solving Linear Dynamic System using expm

clear all;

m=1500;
f_heave=1; c=round(m*(2*pi*f_heave)^2,-3)
damping=0.3; d=round(damping*(2*sqrt(c*m)),-2)

figure(1), clf
%derx=A*x; x=[v_z;F_s;z];
%x=expm(A*t)*x_iv;
A=inv(diag([m,1,1]))*[-d 1 0; -c 0 0; 1 0 0];
x_iv=[0;1000;0];
for N=[7,100]
t_vec=[linspace(0,2,N)];
x_mat=[];
for i=1:length(t_vec)
t=t_vec(i);
x_mat(:,i)=expm(A*t)*x_iv;
end
for j=1:length(x_iv)
subplot(length(x_iv),1,j), plot(t_vec,x_mat(j,:)),
hold on, grid on
end
end
ylabel('t /[s]')
subplot(length(x_iv),1,1), ylabel('v_z /[m/s]')
subplot(length(x_iv),1,2), ylabel('F_s-m*g /[N]')
subplot(length(x_iv),1,3), ylabel('z /[m]')

Figure 1-35: § An example in Matlab code, where expm is used to find the solution.
1.5.4.1.2 § Simulation by Programming Integration
In Figure 1-36, the Matlab function f shows an implementation of an explicit form model:
function [derv,derx, F] = f(t, v,x)
m=2; c=3;
F=c*x;

55
 Introduction

derv=-F/m;
derx=v;
end
The following Matlab script simulates it (using simplest possible derivative approximation, Newton
forward) and plots the result.
>> clear all; dt=0.001; t_vec=[0:dt:5]; v_vec=0; x_vec=1; F_vec=[];
for i=1:length(t_vec)
[derv,derx,F_vec(i)]=f(t_vec(i),v_vec(i),x_vec(i));
v_vec(i+1)=v_vec(i)+derv*dt;
x_vec(i+1)=x_vec(i)+derx*dt;
end
figure(1), clf
plot(t_vec',[v_vec(1:end-1);F_vec;x_vec(1:end-1)]')
grid on, legend('v','F','x')
The states x,v has to be given initial values. The output variable F is not a state and has to be treated
slightly different.
1.5.4.1.3 § Simulation of a model on explicit form using built-in Integration
Methods
There are many other built-in integration methods in Matlab, e.g. euler, ode23, ode45 and
ode23s. The same example as above can also be simulated with such, here exemplified with ode23:
>> clear all; [t_vec,States]=ode23('f2',[0 5],[0 1]);
figure(1), clf, plot(t_vec,States), grid on, legend('v','x')
A wrapper function f2 has to be defined to meet the interface format for ode23:
function [derStates] = f2(t, States)
[derv,derx, F] = f(t, States(1),States(2));
derStates=[derv;derx];
end

1.5.4.2 Dataflow Diagram Based Tools

Examples of tools: Simulink, Systembuild, Altair Activate, Xcos, Modelica with Blocks library

In these tools, the Explicit form model (or an ODE) is built with a graphical representation, around “in-
tegrator blocks”, often marked “1/s” or “∫ ”. An example using Simulink is shown in Figure 1-36. Sim-
ulink is designed for designing/modelling signal processing and control design. It can also be used for
modelling the physics of the controlled systems. There are no dedicated vehicle dynamics tools/librar-
ies from Mathworks (but there are in-house developed specific libraries in automotive companies).
From this type of tools, it is often possible to automatically generate real time code, which is more and
more used instead of typing algorithms. It can be used for rapid prototyping of control functions, or
even for generation of executable code for production ECUs.

56
 Introduction

Physical model: Explicit form model:

x
𝑥
v

Mathematical model:

(3 DAE equations, 3
unknown variables)

Figure 1-36: Graphical modelling using Simulink for Explicit form model.

1.5.4.3 Vehicle Dynamics Specialized Simulation Tools

Examples of tools: CarMaker, TruckMaker, veDYNA, CarSim, TruckSim.

These tools are frequently used in professional vehicle development. They contain purpose-built and
relatively advanced models of vehicles, drivers and environments (ground, road, traffic, …). They are
well prepared for parameter changes. However, they are generally less prepared for modelling con-
ceptually new vehicle designs, which can make these tools less useful for vehicle manufacturers and
advanced research. For this reason, many of these tools offer also an interface to Simulink or FMU, so
that the user can add in their own vehicle models or export parts to other tools.

1.5.4.4 MBS Tools

Examples of tools: Adams, Simpack, LMS Virtual Lab, Simscape Multibody, Modelica with Mechanics li-
brary.
These are general 3D mechanics modelling and simulation tools, so called MBS (Multi-Body Simula-
tion) tools. As one example, Adams contains libraries of general bodies, joints and force elements. But
there are toolboxes in Adams for vehicle dynamics, where template models and special components
(such as tyre models and driver models) are available for vehicles dynamics. The models are very ad-
vanced and accurate for 3D mechanics, and there are import/export interfaces to Simulink.

1.5.4.5 Modelica Based Modelling Tools

Examples of tools: Dymola, Maplesim, System-Modeler, AMESim, Optimica Studio, Jmodelica, OpenModelica
(https://openmodelica.org/), OMwebbook (http://omwebbook.openmodelica.org/static/DynamicWebbook/)

Modelica is not a tool but a globally standardized format for lumped dynamic models on DAE form (or
Mathematical form, see 1.5.1.1.3). There are several tools which supports the format. Specification of
Modelica is found at https://www.modelica.org/. When learning Modelica, http://www.xogeny.com/
is helpful. The model format is acausal and all variables and parameters are declared. An example of
model is given in Figure 1-37. Declaration of which are parameters and variables is a necessary part
for a DAE model, see “parameter Real” and “Real”, respectively.
The model format is also object oriented, which means that libraries of model components are facili-
tated. These are often handled with graphical representation, on top of the model code. There are
some open-source libraries for various physical domains, such as hydraulic, mechanics, thermodynam-
ics and control. There are also commercial libraries, where we find vehicle dynamics relevant compo-
nents: Vehicle Dynamics Library and Powertrain Library. Some simple Modelica code will be used to
describe models in this compendium.

57
 Introduction

Physical model:
Simulation in Dymola:

𝑥 (Mathematical) Model
in Modelica:

model ExampleModel
parameter Real m=2;
parameter Real m=2; parameter Real c=3;
parameter Real c=3; parameter Real F0=4;
Real v; Real v;
Real F(start=3); Real F;
equation Real x(start=-1/3); The tool manipulates the 3 equations to
m*der(v) = -F; equation
der(F) = c*v;
this explicit form:
m*der(v) = -F;
v = der(x); 𝑒𝑟( ) /
Corresponding explicit form: 𝑆𝑡 𝑡𝑒𝐷𝑒𝑟 𝑡 𝑒𝑠
F = F0+c*x; 𝑒𝑟(𝑥)
𝑒𝑟( ) / end ExampleModel; 𝑢𝑡𝑝𝑢𝑡𝑠 𝑥
𝑒𝑟( )
Figure 1-37: Example of model in Modelica format (using the tool Dymola). Two alternative
models are given, leading to either [x,v] or [v,F] as states.
Mathematical modelling is more efficient than Explicit form modelling, since the engineer does not
need to spend time on symbolic/algebraic manipulation of the equations. This is especially true when
a model is reused in another context which changes the causality or for so called “higher index prob-
lems”, see Figure 1-38. In this compendium, many models are only driven to Mathematical model,
since it is enough if assuming there are modern tools as Modelica tools available. One the other hand,
an explicit form model has the value of capturing the causality, i.e. the cause-to-effect chain. The cau-
sality can sometimes facilitate the understanding and in that way help the engineer, which is why at
least one and rather complete model is shown as explicit form model, see 4.5.3.2.
One can also declare variables with prescribed causality, i.e. signals, in Modelica. Declaration of input
signal: “input Real z;”. Modelica can also handle sampled signals and discrete states.

Physical model: Simulation in Dymola:

2 2
2 2
2

2
(Mathematical) Model
in Modelica:
model ExampleModel_HigherIndex
parameter Real m1=0.5;
parameter Real m2=1.5;
parameter Real c=3;
Real v1;
Real v2; The tool first realizes need for differentiating, to get
Real F1;
another algebraic equation:
Real F2(start=3);
equation 2 ⇒ 𝑒𝑟( ) 𝑒𝑟( 2)
m1*der(v1)
F1 = -F1;
+ m1*der(v1) = 0; The tool then selects state variables and manipulates the
m2*der(v2)
F2-F1 = F1-F2;= 0;
+ m2*der(v2) equations to this explicit form:
v1=v2; 𝑒𝑟( ) 2 /( + 2)
der(F2) = c*v2; 𝑆𝑡 𝑡𝑒𝐷𝑒𝑟 𝑡 𝑒𝑠
𝑒𝑟( 2)
end ExampleModel_HigherIndex; 2
𝑢𝑡𝑝𝑢𝑡𝑠 /( +
2 2)

Figure 1-38: Example of “higher index problem” model in Modelica format.

58
 Introduction

Physical Model in Modelica, in graphical editor: Simulation in Dymola:

mass1.v [m/s] mass2.v [m/s] mass2.flange_a.f [N] spring.flange_a.f [N]
4

-1

-2

-3

-4

automatic, 0 2 4 6 8 10

by the tool
same, in text editor:
model ExampleModel_ModelicaLibraries
Modelica.Mechanics.Translational.Components.Fixed fixed;
Modelica.Mechanics.Translational.Components.Spring spring(s_rel0=1, c=3);
Modelica.Mechanics.Translational.Components.Mass mass1(m=0.5);
Modelica.Mechanics.Translational.Components.Mass mass2(m=1.5);
equation
connect(mass1.flange_b, mass2.flange_a); model Mass "Sliding mass with inertia"
connect(mass2.flange_b, spring.flange_a); parameter SI.Mass m(min=0, start=1);
connect(spring.flange_b, fixed.flange); parameter StateSelect stateSelect=StateSelect.default;
end ExampleModel_ModelicaLibraries; extends Translational.Interfaces.PartialRigid;
SI.Velocity v(start=0, stateSelect=stateSelect);
SI.Acceleration a(start=0);
equation
v = der(s);
a = der(v);
m*a = flange_a.f + flange_b.f;
end Mass;

Figure 1-39: Example of model in Modelica format, using Modelica libraries of component.
1.5.4.5.1 § OMwebbook
The cloud computation service OMwebbook for the example in Figure 1-37 is shown in figure below.

Figure 1-40: Modelica format, using OMwebbook, for same example model as in Figure 1-37
Contribution from Pinar Boyraz, Vehicle Safety at Chalmers

1.5.4.6 FMI Supporting Simulation Tools

Examples of tools: Most of the tool examples in 1.5.4.5, but also Simulink, CarMaker

FMI (Functional Mock-up Interface) is not a tool but a globally standardized format for dynamic mod-
els on explicit form (see https://fmi-standard.org/). There are several tools which supports this for-
mat. FMI enables model export/import between tools. It also allows to hide Intellectual Property (IP)
by using “black-box format”, i.e. models compiled (non-human readable) for certain processors, which

59
 Introduction

is important in relation between OEMs and suppliers. An FMU (Functional Mock-up Unit) is a model on
FMI format.
1.5.4.6.1 § Model Exchange and Co-Simulation FMUs
FMUs can be either of Model Exchange type (requires external solver) or Co-Simulation type (includes
an own solver for the own states).

Figure 1-41: § FMI standard models. Left: FMU for Model Exchange. Right: FMU for Co-Simulation.
There are also add-on standards for combining several FMUs in an hierarchy (System Structure and
Parameterization SSP) and for co-simulation with FMUs (Distributed Co-Simulation Protocol DCP).

1.5.4.7 Modularity in Models

Another aspect of modelling is the modularity. For complex products as vehicles, it is often important
to keep a modularity in the models, which reflects the subsystems in the product. This is for reuse of
subsystem models and co-operation between engineering teams at vehicle manufacturer and suppli-
ers. Reuse cuts modelling time, but also improves the models over time since bugs are found when
models are used. Modularity is also important for replacing one subsystem model with a real ECU with
software or splitting model between several processors for increasing the computational capability.
Methods for above motivated “co-simulation” are typically based on that each module are on explicit
form but also include its own time integrator. This can create big challenges in numerical stability and
accuracy of the whole system.
1.5.4.7.1 § Co-Simulation and Real-time Simulation
Co-Simulation is when merging models containing their own time integration code. It can be caused by
different modelling environments or that the models have to be kept confidential. Co-simulation
means that several systems, e.g. several FMUs, integrates on their own in small (micro) time steps, but
exchange interface variables only at certain larger (macro) time steps. There is a standard also for this
communication, see https://dcp-standard.org/.
The opposite to co-simulation is Mono-Simulation. In Mono-simulation, there is only one numerical
time integration method (solver) and it can access the whole model at once; all derivatives are calcu-
lated together: 𝒙 𝒇𝑶𝑫𝑬 (𝒙 𝑡). Then, there is a wide range of well-developed codes and the error is
kept under control by automatic time step variation during the simulation.
Real-time simulation (or On-line simulation) is when the model need to be synchronized with some-
thing from the real world, such as a human driver, a test rig with a real subsystem or just an electronic
hardware (1.6.6.2.1). Opposite can be called Off-line-Simulation. Real-time simulation is similar to co-
simulation in the way that there is not one over-arching integration solver for the whole system.

60
 Introduction

Co-Simulation and Real-time simulation very often they appear at the same time. Many challenges ap-
pear, such as keeping integration error under control, computation fast enough for real time, synchro-
nising variables, etc.
1.5.4.7.2 § Co-Modelling
An alternative way where models of sub-systems are connected on Mathematical form, as opposed to
Explicit form. It could be called “co-modelling” (or “acausal co-simulation”) and make it easier to or-
ganize libraries of models since one would not need to keep different versions of part models for dif-
ferent causality on interface variables. However, co-modelling would require a globally accepted and
used standard of Mathematical modelling format, which does not exist yet; Modelica is the closest to
reach such status. Modelica has a proposal for encryption of models, which would be important for
model exchange with kept confidentiality.

1.6 Vehicle Engineering

1.6.1 Vehicle Motions and Coordinate Systems
A vehicle’s (motion) degrees of freedom are named as in marine and aerospace engineering, such as
heave, roll, pitch and yaw, see Figure 1-42. Figure 1-42 also defines the 3 main geometrical planes,
such as transversal plane and symmetry plane. For ground vehicles, the motion in-road-plane (irp) is
the primary motion, and has the dofs longitudinal, lateral and yaw. The remaining degrees of freedom
(dof) describes the out-of-road-plane (oorp) motion. Also, the forces and moments on the vehicle body
can be categorized in irp forces (longitudinal forces, lateral forces and yaw moments) and oorp (verti-
cal forces, roll moments and pitch moments). Along each dof one defines velocity, such as longitudinal
velocity in [ ⁄𝑠] and yaw velocity (or yaw rate) in [𝑟 ⁄𝑠].
The consistent use of parameters that describe the relevant positions, velocities, accelerations, forces,
and moments (torques) for the vehicle are critical. Unfortunately, there are sometimes disparities be-
tween the nomenclature used in different textbooks, scientific articles, and technical reports. It is im-
portant to apply coordinate system from (ISO 8855). It is the predominant coordinate system used
nowadays.
Historically, a coordinate system with other positive directions, Reference (SAEJ670), has been ap-
plied. (For geometric design, e.g. suspension hardpoints and body shapes, the (ISO 4130) is more used.
It has x rearwards, y to the right and z upwards.)
The distinction of vehicle fixed and inertial (= earth fixed = world fixed) coordinate systems is im-
portant. Figure 1-43 depicts the four most relevant reference frames in vehicle dynamics: the inertial,
the vehicle, the wheel corner and the wheel reference frames. All these different coordinate systems
allow for the development of equations of motion in a convenient manner.
or heave
vertical

ground, road
or horizontal ISO 8855
plane z
yaw y
transversal
plane

x
symmetry
plane

Figure 1-42: Left: Vehicle (motion) degrees of freedom and important planes. Right: ISO
coordinate system

61
 Introduction

Velocity of wheel
=v hub i,

lateral=y x
yaw rate= =wz y
𝑤

n =course angle y x 𝑤 d=Road wheel)

CoG Steering angle

z =(body) side slip

y =pz= angle over wheel
Inertial coor- =heading angle
dinate system: NOTE: The ISO8855 tyre lateral side slip is positive for tyre sliding with
x positive lateral speed, giving negative tyre lateral force. (Opposite tyre
z lateral slip definition is sometimes used, but not recommended.)

Figure 1-43: Coordinate systems and motion quantities in ground plane.

The orientation of the axes of an inertial coordinate system is typically either along the vehicle direc-
tion at the beginning of a manoeuvre or directed along the road or lane. Road or lane can also be
curved, which calls for curved longitudinal coordinate.
Origin for a vehicle coordinate system is often centre of gravity of the vehicle, but other points can be
used, such as mid front axle (ground contact or wheel centre height), mid of front bumper, outer edge
of body with respect to certain obstacle, etc. Figure 1-43 defines velocities . Note that they are rela-
tive to inertial system, “velocities over ground”, not relative to the other coordinate systems. However,
the velocity components are depent of which coordinate system is decomposed in. Positions
are often not included in the models. When positions are needed, e.g. for lane markings, road edges,
other moving vehicles and varying friction, they are typically defined in inertial coordinate systems
but algorithms on-board the vehicle use positions expressed in vehicle coordinate system.
In ISO and Figure 1-43, wheel side slip is defined so that it is positive for positive lateral speed. This
means that lateral tyre forces on the wheel will be negative for positive side slip. Some would rather
want to have positive force for positive side slip. Therefore, one can sometimes see the opposite defini-
tion of wheel side slip, as e.g. in (Pacejka, 2005). It is called the “modified ISO” sign convention.
Often there is a need to number each unit/axle/wheel. The numbering in Figure 1-44 is proposed. It
should be noted that non-numeric notations are sometimes used, especially for two axle vehicles with-
out secondary units. Then front=𝑓, rear=𝑟. Also, to differentiate between sides, left= and right=𝑟.
Using these motion dofs and coordinates for modelling is further described in 1.6.2.

𝑜𝑟
𝑜𝑟
Δ 5 or 1,3,1 3 or 1,2,1 1 or 1,1,1
Δ
6 or 1,3,2 4 or 1,2,2 2 or 1,1,2
2 𝑜𝑟 2
If number of units, 𝑛 > , there Δ Δ Dpz=
are >1 couplings. Then use, e.g., =articulation angle of coupling 2 𝑜𝑟 2
𝑗 for ⋯ 𝑛𝑢 and 𝑗 𝑓 𝑟 or
axle 3 axle 2 axle 1
… for ⋯ 𝑛𝑢 .
or 1,3 or 1,2 or 1,1

unit 2 unit 1
Figure 1-44: Proposed numbering of units, axles, wheels and articulation angle. Example shows a
rigid truck with trailer.
If multiple units: 𝑓 , so the rule is “articulation (yaw) angle in front cou-
pling on a unit is the yaw angle of the unit ahead, relative to the unit ”. This can be extended to

62
 Introduction

articulation articulation roll angles: 𝑓 to articulation pitch angles: 𝑓

. See also 4.5.2.2.

1.6.1.1 Couplings
Coupling forces can be denoted using “single cut force notation” or “multiple cut forces notation”, see
1.5.2.1.1. Coupling, in 1.6.1.1, means moment-free 2D connection between bodies. Similar reasoning is
made for transmission coupling in 2.4.1.
1.6.1.1.1 Single cut force notation
front end of unit + : rear end of unit :

Figure 1-45: Coupling force notation, using “single cut force notation”, for the coupling named
“ 𝑅” (for unit , coupling Rear).
For single cut force notation, it can be an alternative to replace subscript “ 𝑅” with a sequential num-
bered identifier of the coupling within the vehicle, e.g. ⋯ (𝑛 ), where 𝑛 are the number of
units.
1.6.1.1.2 Multiple cut forces notation
free body diagram of coupling:

front end of unit + : rear end of unit :

+
+

compatibility of coupling:
+
+ +

Figure 1-46: Coupling force notation, using “multiple cut forces notation”, for the coupling
between unit and + . Subscripts: “ 𝑅” as in Figure 1-45 and “ + ” after unit + ,
coupling Front.
For multiple cut forces notation, introduces more notations than needed to describe the physical real-
world situation. Hence one need extra transformation equations in Eq [1.1] and Eq [1.2]. Also, extra
equations are needed to set the front coupling forces of the first unit and the rear coupling forces of
the last unit to zero.
Compatibility:

63
 Introduction

Compatibility, from unit to unit + :

+ cos( ) sin( )
[ ] [ ]∙[ ]
+ sin( ) cos( )
[1.1]
Same equation, but expressed as from unit + to unit :
cos( ) sin( ) +
[ ] [ ]∙[ ]
sin( ) cos( ) +

Force equilibrium of coupling (or Force equivalence between forces on the two units):
From unit to unit + :
+ cos( ) sin( )
[ ] [ ]∙[ ]
+ sin( ) cos( )
[1.2]
Same equation, but expressed as from unit + to unit :
cos( ) sin( ) +
[ ] [ ]∙[ ]
sin( ) cos( ) +

1.6.1.2 Wheel Orientation

For steered wheels, there are often reason to translate forces and velocities between vehicle coordi-
nate system and wheel coordinate system, see Figure 1-47 and Eq [1.3].
Velocity of body, Velocity components in Velocity components
above the wheel vehicle coordinates in wheel coordinates
𝑤
𝑣
𝑤
𝑣

Force in road plane, Force components in Force components in

acting on the wheel vehicle coordinates wheel coordinates

𝑣 𝑤
𝑤
𝑣

Figure 1-47: Transformation between forces and velocities in vehicle coordinate

system and wheel coordinate system.

Transformation from wheel coordinates to vehicle coordinates:

𝑣 cos( ) sin( ) 𝑤 𝑣 cos( ) sin( )
[ ] [ ]∙[ ] 𝑛 [ ] [ ] ∙ [ 𝑤]
𝑣 sin( ) cos( ) 𝑤 𝑣 sin ( ) cos ( ) 𝑤
[1.3]
Or, same relation, but expressed as from vehicle coordinates to wheel coordinates:
𝑤 cos( ) sin( ) 𝑣 𝑤 cos( ) sin( ) 𝑣
[ ] [ ]∙[ ] 𝑛 [ ] [ ]∙[ ]
𝑤 sin( ) cos ( ) 𝑣 𝑤 sin( ) cos ( ) 𝑣

1.6.1.3 Axle Position Relative to Vehicle Body

It can also be useful to define axle positions (translatory positions and angular orientations). For a
rigid axle, it has a direct physical meaning. For an axle with individual wheel suspension, it can be a
virtual mean value of left and right wheels direction.

64
 Introduction

A sign convention, aligned with articulation angles (between units) and steering angles, would be: Rel-
ative roll angle of an axle= , so the rule is “axle orientation of axle 𝑗 is relative to the ve-
hicle body”. This can be extended to translatory position and orientation in yaw and pitch.

1.6.1.4 Slip of Tyres and Wheels

Slip (longitudinal 𝑠 and lateral 𝑠 ) for a tyre or wheel will be defined in Chapter 2, see Eq [2.1], Eq
[2.28] and 2.2.4.1.1.1. Slip is central for relating tyre forces in ground plane to vehicle and wheel
motion and wheel vertical force . Note that rotational speed of each wheel comes in as additional
variables to Figure 1-43. Slip is a dimensionless speed, “sliding speed divided by a reference speed”. It
appears as an intermediate variable to compute forces: [ ] 𝒇 ( 𝑠 𝑠 )
𝒇_ ( 𝑠 ( )𝑠 ( )) 𝒇𝒗 ( ) . One can sometimes define slip also for an axle or even
for an axle group, but it becomes less strict, since can generally differ between the wheels
within the axle or axle group.

1.6.1.5 Compatibility between Tyre and Vehicle

When implementing lateral tyre models in a vehicle model for lateral dynamics, there are two major
ways of formulating the compatibility that relates vehicle motion (body slip angle ) to wheel side slip
(𝑠 ≈ tan( )) via wheel steering angle ( ). One way is to express longitudinal and lateral velocity
components with sine and cosine, as done in Eqs [4.3], [4.6] and [4.46]. The other way is to use the an-
gle sum + .

2 2 2 2
√ 𝑣 + 𝑣 √ 𝑤 + 𝑤
𝑣 cos( ) sin( ) 𝑤
[ ] [ ]∙[ ] or [1.4]
𝑣 sin( ) cos( ) 𝑤 arctan(
⏟ 𝑣⁄ 𝑣) + arctan(
⏟ 𝑤⁄ 𝑤)
{ 𝛽 𝛼

If approximating with small angles, it is easier to approximate to a linear vehicle model, see 4.3.2.4 and
4.4.2.2. For reversing, the angles are close to ±𝜋 instead of close to 0.
For ≈0 ≈ ≈0⇒ 𝑣 >0 𝑤 >0 For ≈0 ≈ ≈ ±𝜋 ⇒ 𝑣 <0 𝑤 <0
𝑤 𝑤
𝑠 𝑤 𝑠 𝑤
𝑤 𝑤
𝑥

𝑦
vehicle body
+𝑥 𝑣 +𝑥
arctan ≈
𝑣 𝑣 arctan 𝑣 𝑣 ≈𝜋 𝜋+
𝑣 𝑣
𝑦 𝑣 𝑦
𝑣 𝑣
arctan 𝑠 ≈𝑠 𝜋 arctan 𝑠 𝑤 ≈𝜋 𝑠 𝑤
𝑤 𝑤
𝑜𝑡ℎ +𝑥
+ ⇒ ⇒ ≈ +𝑠 𝑤
𝑝𝑝𝑟𝑜𝑥. 𝑦
Figure 1-48: Approximate compatibility relation for small side slip 𝑤 . Left: For forward driving.
Right: For reversing. Both lead to same final equation.

1.6.1.6 Path, Orientation, Pose and Trajectory

A path can be 𝑥(𝑦) or 𝑦(𝑥) for centre of gravity where 𝑥 and 𝑦 are coordinates in the road plane. To
cope with all paths, it is often necessary to use a curved path coordinate instead, 𝑠, i.e. 𝑥(𝑠) and 𝑦(𝑠). A
path do have an orientation of itself , defined through arctan( ) 𝑦 ⁄ 𝑥 ( 𝑦⁄ 𝑠 )
( ⁄( 𝑥⁄ 𝑠)) 𝑦 ⁄𝑥 . However, a vehicle following the path does not need to point along the path.
′ ′

65
 Introduction

The vehicle also has a varying orientation, (𝑥) or (𝑠), which often is often relevant, but the term
“path” does necessarily include this. In those cases, it might be good to use an expression “path with
orientation” instead.
If the vehicle is articulated, it has not only the orientation as additional variable over the path. It also
has the articulation angle in each articulation point. The set of articulation angles 𝜽 describes the
“pose”, so the total description would be [𝑥 𝑦 𝜽] and called “path with orientation and pose”.
A (time) trajectory is a more general term than a path and it brings in the dependence of time, 𝑡. One
typical understanding is that trajectories can be [𝑥(𝑡) 𝑦(𝑡) (𝑡) 𝜽(𝑡)]. But also, other quantities,
such as steer angle or vehicle propulsion force can be called trajectory: (𝑡) and (𝑡), respectively.
The word “trace” is sometimes used interchangeably with trajectory.
1.6.1.6.1 Path with Orientation
The path and path with orientation was y
introduced in 1.5. The path, in global co-
ordinate system, is related to vehicle
speeds, in vehicle fix coordinates, as given
in Figure 1-49 and Equation [1.5].
Knowing ( (𝑡) (𝑡) (𝑡)), we can de-
termine “path with orientation”
(𝑥(𝑡) 𝑦(𝑡) (𝑡)), by time integration of x
the right-hand side of the equation.
Hence, the positions are typically “state Figure 1-49: Model connecting “path with orientation”
variables” in lateral dynamics models. to velocities in vehicle coordinate system.

𝑥 cos( ) sin( )
[ ] [ ]∙[ ]
𝑦 sin( ) cos( ) [1.5]

It should be noted that in some problems, typically manoeuvring at low speed, the real time scale is of
less interest. Then, the problem can be treated as time independent, e.g. by introducing a coordinate, s,
along the path, as in Equation [1.6].

𝑥′ ∙ cos( ) ∙ sin( )
𝑠 𝑠
𝑦′ ∙ cos( )+ ∙ sin( )
𝑠 𝑠 [1.6]
′
𝑠
where prime notes differentation with respect to s
Here, 𝑠 can be thought of like an arbitrary time scale, with which all speeds are scaled. One can typi-
cally choose 𝑠 [ ⁄𝑠]. However, in this compendium we will keep notation t and the dot notation
for derivative.

1.6.1.7 Length and Width Dimensions

For passenger cars, the length and width dimensions are rather constant and simple to define in a way
that they always are positive: This can simplify the understanding and intuition for equations. For
heavy vehicle, the situation is more complex and varying which motivates a more systematic way of
denoting dimensions. The numerical values of the dimensions then often becomes negative. Two such
conventions are shown in Figure 1-50.

66
 Introduction

A passenger car Subscripts:

A unit 𝑢 in a heavy combination vehicle
Subscripts: • 𝑢 unit (1,2,3,…) Longitudinal dimensions are defined as
• coupling • coupling longitudinal positions ahead of front axle.
• 𝑟 𝑓 rear, front • 𝑅 Rear, Front So, most of them becomes < 0 and 0.
• (axle) group

2
𝑊

𝑊
𝑊𝑓
𝑊

𝑓
3 2 0
𝐿 𝐿
Figure 1-50: Two examples of dimension definitions and notation.

1.6.2 Complete Vehicle Modelling Concepts

The (Dynamic) Equilibria (or Equation of Motion) can be seen as the main equations in a mathematical
model of a complete vehicle. A summarizing view of this is given in Eq [1.7].
In the left approximations, only the most important centrifugal effects, , are included see also
Eqs [3.29] and [4.45]. The sums refer to forces and moments acting on the vehicle body; from
tyre/ground caused by actuation and rolling resistance, from gravity and from aerodynamics. Equa-
tions for forces and moments in the summations are not given here, but are needed for the model to be
complete, i.e. “simulate-able”.

≈ ( ) ∑ 𝑓𝑜𝑟 𝑒𝑠
𝐈n-road-plane
( + )≈ ( + ) ∑ 𝑓𝑜𝑟 𝑒𝑠
(irp) equilibria:

{ ∑ 𝑜 𝑒𝑛𝑡𝑠
[1.7]
( )≈ ( + ) ∑ 𝑓𝑜𝑟 𝑒𝑠
Out-of-road-plane
∑ 𝑜 𝑒𝑛𝑡𝑠
(oorp) equilibria:

{ ∑ 𝑜 𝑒𝑛𝑡𝑠
Chapter 3 and 4 requires complete vehicle models that can measure complete vehicle functions ex-
pressed in irp motion.

1.6.2.1 Model Characterizations

A major categorization, see Eq [1.7] and 1.6.2.3 and 1.6.2.4, of vehicle models is whether the model
considers only irp motion or also oorp motion. Another important categorization is to differ between
one-track and two-track models, see 1.6.2.3.2.
1.6.2.1.1 Dynamic Operating Condition
A third important categorization is which “dynamic operating conditions” that characterizes the oper-
ation/manoeuvre. Categories for such are below defined based on the irp motion of the vehicle body.
Static
Static condition means that vehicle does not moves irp. All velocities and accelerations are zero. So, all
terms on left side in Eq [1.7] are neglected. Force play in longitudinal direction can still be non-zero,
due to e.g. gravitational force and pre-tension between axles and wheels.
Low speed

67
 Introduction

Low speed condition means that vehicle moves with low speed irp: forward, reverse or stand-still. Also,
accelerations are low. No fictive forces ( 𝑠𝑠 𝑒 𝑒𝑟 𝑡 𝑜𝑛) is modelled irp. All terms on left side in
Eq [1.7] are neglected.
Steady State
Steady State condition means that time history is irrelevant for the quantities studied. Seen as a ma-
noeuvre over time, the studied quantities are constant. Explained for a certain model, it means that the
influence of the derivatives of the corresponding variables, that else would generate state variables, is
neglected. In physical model of a mechanical system this often means that “mass acceleration ” or
“force derivative⁄stiffness ” is neglected (in mathematical model: equation “ 0 ” is added, in explicit
form model: “ “ is replaced with “0” ). If only some quantities are treated in this way, one might call
the conditions steady state with respect to these quantities. Only the terms (centrifugal
forces) on left side in in Eq [1.7] are kept.
Transient
Transient (or Transient State, as opposed to Steady State) condition means that time history is rele-
vant; i.e. there are delays, represented by “state variables” when simulated. All terms on the left side in
in Eq [1.7]are kept. Note that also cases with low velocities but large accelerations belongs here, and
not in 1.6.2.1.1.2.
Stationary Oscillating
Stationary Oscillating condition is a special case of transient, where cyclic variations continue over
long time with a repeated pattern. Long time means to that all none-cyclic components of the variation
is damped out. The pattern is often modelled as harmonic (sinus and cosine variations in time) with
constant amplitudes and phases. Example is sinusoidal steering with small enough steering amplitude,
see 4.4, but also driving with over an undulated road surface, see Chapter 5. All terms on left side in in
Eq [1.7] are kept, as for transient operation.
Quasi-Steady State
Quasi-Steady State condition have a more diffuse meaning. It can refer to steady state with respect to
some quantities, i.e. some terms on left side in in Eq [1.7] are neglected but not all. Alternatively, it can
refer to that the quantities are prescribed to an explicit function of time, e.g. 𝑓 (𝑡) meaning that
also is known. The latter is sometimes also known as “inverse dynamics”.
1.6.2.1.2 Other Characterizations
Let us also briefly list some other possible categorizations for Chapter 3 and 4:
• Categorization referring to small angles (sin( 𝑛 𝑒) cos( 𝑛 𝑒) 0 ) or not, applied to
steering/articulation angles and/or tyre/body side-slip angles.
• Categorization referring to tyre models are further explained in 2.2.6.
• One can also categorize referring to subsystem models.
o The suspension can add states per wheel or axle , at least the vertical spring force .
o The propulsion and brake system can be actively controlled. They give wheel torques 𝑻(𝑡)
[ 2 ⋯ ]. Such models often add one state per wheel or axle , the rotational speed .
o Chapter 3 uses different models of propulsion, brake and suspension. E.g., the propulsion sys-
tem can add more states: engine delay, gear shifting, torsional shaft, control algorithms.
o Chapter 4 adds the steering subsystem, with its control algorithms. Also, more about model
categorisation is found in 4.1.1.
Chapter 5 is very different since it treats vehicle functions expressed in out-of-road-plane motion. So,
most models in Chapter 5 uses only the oorp equations, i.e. not the irp equations, above.
To make the overall model simulate-able one needs some form of driver model and environment
model. A rather complete example model is described in 4.5.3.2.

1.6.2.2 1D Models
In some cases, it can be enough to model motion in 1 dimension. Some examples are:
• Longitudinal 1D in 3.2 and 3.3, it can be enough to model : ∑ 𝑓𝑜𝑟 𝑒𝑠 .

68
 Introduction

• Roll 1D in 4.3.12, it can be enough to model : ∑ 𝑜 𝑒𝑛𝑡𝑠

• Heave 1D in 5.4, it can be enough to model : ∑ 𝑓𝑜𝑟 𝑒𝑠
1.6.2.2.1 One-Wheel Longitudinal 1D Models
For Longitudinal 1D model it is common that all wheels (or axles) are lumped to one wheel, see Figure
1-51. The lumping means that torques from all wheel torque actuators (propulsion, brake, …) are
summed to one torque which acts on one fictive wheel. The fictive wheel carries the whole vertical
force from all wheels. It should be noted that we also, implicitly, assume that all wheels rotate equally
fast. The rolling resistance coefficient and slip characteristics of the fictive wheel are, for instance,
taken as a weighted mean between all wheels, weighted with vertical load of each wheel. This kind of
vehicle model cannot well represent driving situation with significant wheel slip, different ground
conditions (grade, peak friction, …) under wheels, non-Ackermann steering geometry or articulation
angles between vehicle units.
1.6.2.2.2 Per-Axle and Per-Axle-Group Longitudinal 1D Models
To repair the mentioned short-comings with the One-Wheel Longitudinal 1D Model, more axles and
axle groups can be modelled separately, see Figure 1-51.
1-wheel model per-axle model

𝑅 𝑅

𝑓
𝑓

2 per-axle-group model 2 per-axle model

𝑅2 2
𝑅2
2
2
2

22 𝑅
𝑅
2 2
22 3
2 2
2
𝑓 3
𝑓 2
Figure 1-51: Different longitudinal 1D models.

1.6.2.3 In-road-plane Models, 2D

When modelling longitudinal dynamics, one has to involve at least . When modelling lateral dynam-
ics, one has to involve at least and . So, models in 𝑥𝑦-plane (irp) are often useful. (Other 2D mod-
els, in 𝑥 - or 𝑦-planes, can also be relevant, e.g. for load transfer.)
1.6.2.3.1 Particle and Body Models
A particle model is similar to a 1D-model but can move in 2D and have forces in 2D. In cases such as
threat estimation in 3.5.2.7 and 4.6.2.4, a particle model can be of use. The alternative is a body model,
see Figure 1-52. A particle model cannot resolve per wheel or axle, which means that tyre slip, nei-
ther 𝑠 nor 𝑠 , can be properly captured which is a problem for both longitudinal and lateral dynamics.
The force can be decomposed in global directions [ 𝑋 𝑌 ] or [ ]. Other decompositions are
road [ ], path [ 𝑝 𝑝 ] or vehicle [ 𝑣 𝑣 ]. Road can be tracked using a map. Path can be
tracked by remembering position from previous time history. Vehicle heading is most difficult, since a
particle model do not capture yaw dynamics well.

69
 Introduction

Models of two-axle single-unit vehicles

particle body
1 point contact 1-track 2-track
𝑌 In between here, we can find 𝑓
hybrid models, e.g. “particle 𝑓
𝑌 with and 𝑀 ” which require
𝑋 𝑋 assumptions about vehicle
orientation while following the
curved path.

1-track models of multi-unit multi-axle vehicles

1-track, Axles lumped within axle groups 1-track, Each axle modelled
2 2 2
2
2 2
2 2
2 2
2 2
2 2 2 2

Figure 1-52: Different in-road-plane vehicle model concepts. The forces are from tyre contacts,
acting on vehicle body. Fictive forces ( 𝑒 𝑒𝑟 𝑡 𝑜𝑛) and body forces (gravity and
aerodynamic) are not drawn.
1.6.2.3.2 One-Track and Two-Track Models
The models can assume that each wheel on the axles have their own ground contact (two-track mod-
els) or if there is only one “virtual tyre” modelled per axle (one-track model or single-track model or
“bicycle model”). See Figure 1-52 and Figure 1-53. For multi-axle vehicles, one can even simplify one
step further and model only one “virtual tyre” per axle group, see Figure 1-52. It should be noted that
the simplifications have limitations, they are not suitable when the forces of the different tyres are sig-
nificantly different, e.g. differences in actuated wheel torques or different wheel side slips.

𝑓
𝑓

𝑓
Common
intersection of all
wheels’ axes of
rotation 𝑅
Figure 1-53: Collapsing a two-track vehicle to a one-track model.
Lateral dynamics phenomena which one-track models do not capture well are, e.g.:
• Deviations from Ackerman geometry within an axle, see 0.
• Roll motion
• Lateral load shift (4.3.7.2) and combined tyre slip (4.3.7.6).
• Added yaw moment due to left/right-asymmetric wheel torque, such as ESC interventions.
The in-road-plane (irp) models does not capture the out-of-road-plane motion, . However,
they can still capture the transfer of loads (vertical forces on wheels).

70
 Introduction

For longitudinal dynamics, one-track models are often enough. For longitudinal functions in “3.2
Steady State Functions” and “3.3 Functions Over (Long) Cycles”, even particle models are often
enough. But, it is important to understand the validity limit of the model!

1.6.2.4 Adding Out-of-road-plane Motion, >2D

The expression “>2D” as opposed to “3D” reflects that sometimes it can be enough and preferable to
combine equations from several 2D views (𝑥𝑦-, 𝑥 - and 𝑦 -plane) without adding all cross-coupling
effects. This is as opposed to a full 3D model, following rigid body dynamics theory in all details.
An example of a two-track model including both irp and oorp motion is given in 4.5.3.

1.6.3 Vehicle Dynamics Terms

1.6.3.1 Load Levels
The weight of the vehicle varies through usage. For many vehicle dynamic functions, it is important to
specify this load level.
Kerb weight is the total weight of a vehicle with standard equipment, all necessary operating consum-
ables (e.g., motor oil and coolant), a full tank of fuel, while not loaded with either passengers or cargo.
Kerb weight definition differs between different governmental regulatory agencies and similar organi-
zations. For example, many European Union manufacturers include a 75 kg driver to follow European
Directive 95/48/EC.
Payload is the weight of carrying capacity of vehicle. Depending on the nature of the mission, the pay-
load of a vehicle may include cargo, passengers or other equipment. In a commercial context, payload
may refer only to revenue-generating cargo or paying passengers.
Gross Vehicle Weight/Mass (GVW/GVM) is the maximum operating weight/mass of a vehicle as spec-
ified by the manufacturer including the vehicle's chassis, body, engine, engine fluids, fuel, accessories,
driver, passengers and cargo but excluding that of any trailers.
Other load definitions exist, such as:
• “Design Weight” (for passenger vehicles, this is typically Kerb weight plus 1 driver and 1 pas-
senger, 75 kg each, in front seats)
• “Instrumented Vehicle Weight” (includes equipment for testing, e.g. out-riggers)
• “Road-allowed GVW/GVM” is the maximum GVW/GVM for a certain road, limited by the
strength of the road and bridges. It is applicable for heavy trucks.
For vehicle dynamics it is often also relevant to specify where in the vehicle the load is placed because
it influences how vertical forces under the wheels/axles distribute as well as moments of inertia.

1.6.3.2 Open-Loop and Closed-Loop Test Manoeuvres

Two expressions used in vehicle dynamics are "Open-loop" and "Closed-loop" test manoeuvres.
An open-loop manoeuvre refers to the case where the driver controls (steering wheel, brake pedal
and accelerator pedal) are operated in a specific sequence, i.e. as functions of time. A typical case is a
sine wave excitation of the steering wheel. The time history of the steering wheel angle is defined as a
function that is independent of the road environment or driver input. This type of manoeuvre can be
relevant to design for, and it tells some but not all about the real-world driving cases. Theoretical sim-
ulation and real testing with a steering robot are examples of how such studies can be made.
A closed-loop manoeuvre refers to the case when (human) driver feedback via driver controls is in-
cluded. This represents real-world driving better. In real vehicle or driving simulator testing, a real
driver is used. This enables collection of the drivers’ subjective experience. In cases of simulation, a
"driver-model" is needed. A driver-model can have varying levels of complexity but in all cases simu-
lates the response of a human driver to different effects, such as lateral acceleration, steering wheel
torque, various objects appearing outside the vehicle, etc.

71
 Introduction

A test with real vehicle, carried out with a steering-robot (and/or pedal robot) can also be called
closed-loop if the robot is controlled with a control algorithm which acts differently depending on the
vehicle states, i.e. if the algorithm is a driver model.
With increasing level of automation, there is sometimes a need for distinguishing between
closed/open loop with respect to human driver or automated driver.
1.6.3.2.1 § Closing Loop via Human and/or Virtual Driver
The concepts of open- and closed-loop control are demonstrated in figure below. The important differ-
ence to note is that in open-loop control there is no feedback to the driver. The figure also shows a
“Virtual driver” which represents today’s automated driving functions based on environment sensing,
such as adaptive cruise control (section 3.5.2.2) and lane keeping aid (section 4.6.2.1). Closed loop con-
trol incorporates the different types of feedback to the driver. Drivers automatically adapt to the dif-
ferent feedback. Understanding driver response to different feedback, and expressing it in mathemati-
cal descriptions, is an active area of research, particularly for driver support functions.

Road irregularities, Road friction, …

Road banking and inclination, Aerodynamic forces, Collision forces, …
Relative Position of road edges, lanes, traffic signs, other road users, …

Open-loop system
Environment sensing
(radar, camera, GPS, V2X, …)

“Virtual driver”
Requests (signals)
Visual
Steering
Suspension,
Surroundings / system,
Environment Human Propulsion Vehicle
Linkages,
(air, road, other Driver Steering system, Road wheel Forces, body
road users, …) wheel, Brake angles, Wheel Moments
torques, … Wheels
Pedals, … system
Absolute
Position, Warnings, Info
HMI (lamps,
Speed sounds, …)
Sensed quantities
Noise,
vibrations
Steering wheel torque Inertial
forces

Figure 1-54: § Open and closed loop representations of vehicle dynamic systems

1.6.3.3 Objective and Subjective Measures

Two main categories of finding measures are:
• An Objective measure is a physical measure calculated in a defined and unique way from data
which can be logged in a simulation or from sensors in a real test.
• A Subjective assessment is a rating measure set by a test driver (e.g. on a scale 1-10) in a real-
vehicle test or driving simulator test.
One generally strives for objective measures. However, many relevant functions are so difficult to cap-
ture objectively, such as Steering feel and Comfort in transient jerks, so subjective assessments are
needed and important.

1.6.3.4 Stable and Unstable

Often, in the automotive industry and vehicle dynamics, the words “stable” and “unstable” have a
broad meaning, describing whether high lateral slip on any axle is present or not. Sometimes it is used

72
 Introduction

for roll-over instability. Also, the articulation angle can be unstable when reversing in low speed with a
trailer. Stability analysis in a stricter physics/mathematical meaning is touch upon in 0.
It is useful to know about this confusion of words. An alternative expression for the wider meaning is
“loss of control” or “loss of tracking” or “directional unstable”, which can include that vehicle goes
straight ahead, but road bends.

1.6.3.5 Subject and Object Vehicle

The subject vehicle is the vehicle that is studied, and often under development. It is practical to have a
name for it, since one often studies one specific vehicle, but it may interact with other in a traffic situa-
tion. Alternative names are host vehicle, ego vehicle or simply studied vehicle.
Other vehicles are called object vehicles or opponent vehicles. A special case of object vehicle is lead
vehicle which is ahead of, and travels in same direction as, subject vehicle. Another special case is on-
coming vehicle which is ahead of, and travels in opposite direction as, subject vehicle.

1.6.3.6 Active Safety, ADAS and AD

The expression Active Safety is used in Automotive Engineering with at least two different meanings:
• Active Safety can refer to the vehicle’s ability to avoid accidents, including both when the
driver is in control (such as ABS and ESC, but also steering response) and functions with auto-
matic interventions based on sensing of the vehicle surroundings (such as AEB and LKA). See
http://en.wikipedia.org/wiki/Active_safety. Active Safety can even include static design as-
pects, such as designing the wind shield and head light for good vision/visibility.
• Alternatively, Active Safety can refer to only the functions with automatic interventions based
on sensing of the vehicle surroundings. In those cases it is probably more specific to use Ad-
vanced Driver Assistance Systems (ADAS) instead, see http://en.wikipedia.org/wiki/Ad-
vanced_Driver_Assistance_Systems. ADAS does not only contain safety functions, but also com-
fort functions like CC and ACC.
Functions that off-load the driver normal tasks during driving can be sorted under Automated Driving,
AD. Fully automated driving, e.g. transportation without human driver on-board, is probably a far fu-
ture vision. On the other hand, it is already a reality that some portion of the driving tasks are auto-
mated in the latest vehicle on the market, such as Adaptive cruise control, see 3.5.2.2, and Lane Keep-
ing Aid, see 4.6.2.1. If both those are active at the same time, we have already automated driving. Defi-
nitions of automation levels 0-5 is found in Reference (SAE_J3016, 2016). Today’s version of these sys-
tems normally has a way to hand-back responsibility to driver rather immediately in hazardous situa-
tions. Future automated driving functions will need to always have a safe-stop function. The way to
compete between vehicle manufacturers will probably be to avoid hand-backs (maximize up-time)
and to allow as long “hand-back times” as possible. So, vehicle dynamics will be important in the devel-
opment, especially for safety reasons for automated driving in higher speeds; hazardous situations
where human driver selects to take back the driving.

1.6.4 Vehicle Architectures

Vehicles are often designed in platforms, i.e. parts of the design solutions are reused in several vari-
ants. Typical variants may be different model years or different propulsion system. To be able to reuse
solutions, the vehicles have to be built using the same architecture.
A mechanical (or geometric) architecture may include design decisions about certain type of wheel
suspension on front and rear axle. An electrical and electronic architecture may include design deci-
sions about electric energy supply system (battery voltage etc.) and electronic hardware for comput-
ers (Electronic Control Units, ECUs) and how they are connected in networks, such as Controller Area
Network, CAN.
The mechanical architecture influences vehicle dynamics functions. However, it is noteworthy that
also the electronic architecture also is very important for the vehicle dynamics, through all electronic
sensors, actuators and control algorithms. One example of this importance is the ABS control of the
friction brake actuators. Architectures for functions are therefore motivated, see 1.6.4.2.

73
 Introduction

1.6.4.1 Subsystems
The architectures are dependent on the business model for how to purchase and integrate subsystems
to a vehicle. Hence, it is relevant to define the subsystems. For vehicle motion functionality, the rele-
vant subsystems (or “motion support devices”) are typically:
• Propulsion system
• Brake system
• Wheel suspension
• Wheels and tyres
• Steering system
• Environment sensing system
Each of these can typically be purchased as one subsystem. Each typically have mechanical and signal
interface to the remaining vehicle. Different vehicle manufacturers can have different strategies for
signal interface and how much the subsystems are allowed to be dependent of each other.

1.6.4.2 Vehicle Function Architecture

As the number of electronically controlled functions increase, an architecture for vehicle functionality,
or “Function Architecture”, becomes necessary to meet fast introduction of new functionality and to
manage different variety of vehicle configurations. A function architecture is a set of design rules for
how functions interact with each other (e.g. signalling between control functions). Also, the decompo-
sition in subsystems should be reflected in this architecture, see 1.6.4.1. An older expression which is
related to function architecture is cybernetics. Examples (from Vehicle Dynamics/Motion function do-
main, often including energy) of modern expressions which are related are Integrated Chassis Control
(GM), Integrated Vehicle Dynamics Control (Ford), Complete Vehicle Control (Volvo) and Vehicle Dy-
namics Integrated Management (Toyota). There is no exact and generally well accepted definition of
such architecture. However, it becomes more and more essential, driven by increasing content of elec-
tronic control in vehicles. One way to visualize a reference architecture is given in Figure 1-55 and Fig-
ure 1-56.
Physical connection
Vehicle Control Information
Driver sensor 1 Request
Function 1
Information
sensor 2
Fusion Function 2
Prediction
sensor 3
Arbitrator
Momentary & Capability
Request
Coordinator
Momentary & Capability

Environment Actuated actuator 1 … … actuator N

(road, obstacles, …) vehicle

Figure 1-55: Concepts of a vehicle motion (and energy) function architecture. Arbitrators,
Coordinators and Actuators are the most important architectural objects for vehicle dynamics
(vehicle motion).

74
 Introduction

Vehicle Transport Mission, Route, Route Segment Layer Predictive Energy Route HMI Human
Environment Management Machine
Sensing Longitudinal Speed, Energy Split Interface
Traffic Situation Layer Longitudinal
ACC
Traffic
LaneSteering Driver ACCbtn
Interpretation
V2V / V2I Collision Interpretation Dist btn
Maps Avoidaice Arbitration
…
FwdSens Longitudinal Acceleration,
Vehicle Capability Actual Power Split SteWhl
AnSens
Vehicle Motion Layer
Driver APed
Vehicle Interpretation
Vehicle Control Arbitration
State BPed
Estimator Momentary irp Accelerations,
Energy Power Split
Management Accel to Force
Vehicle Stability
Capability Control Coordination
Estimator

Actual Device Capability Actual Wheel Torques, Steering Angles

Motion Support Device Layer Physical connection to

⋯ SteAct FrntBrkAct RearBrkAct
EngAct brake only on passenger
TransmAct cars. Heavy trucks have
Vehicle
Motion brake-by-wire.
Sens

requests
actual values
capabilities
physical connections

Figure 1-56: One example of reference architecture of vehicle motion functionality. Red arrows:
Requests, Blue arrow: Information, Black lines with dot-ends: physical connection.
In order to be able to formulate design rules in reference architecture of functionality the following are
relevant questions:
• Which physical quantities should be represented on the interface between Devices (Sensors and
Actuators) and Vehicle Level Functionality?
• Partitioning within a reference architecture of vehicle motion functionality could be realised as
shown in Figure 1-56. Different Layers/Domains are defined:
• Human Machine Interface Domain: This includes the sensors/buttons which the driver
uses to request services from the vehicle’s embedded motion functionality.
• Vehicle Environment Domain: Includes surrounding sensors mounted on vehicle but also
communication with other vehicles (V2V) and infrastructure (V2I) and map information.
• Route Layer: Planning the whole transport mission, horizon 10..1000 km. These functions
exist mainly for commercial traffic and might be done outside the subject vehicle.
• Traffic Situation Layer: Interpret the immediate surrounding traffic which the vehicle is
in, road/lanes and other road users, horizon ≈100..300 m. Example of functions: adaptive
cruise control, collision avoidance, and lane steer support.
• Vehicle Motion Layer: This includes the Energy management, powertrain coordination,
brake distribution, and vehicle stability such as ESC, ABS. Horizon ≈10..30 m This layer
also estimates the vehicle states e.g. . In addition, this layer would be able to give
vehicle level capability of max/min acceleration and their derivative.
• Motion Support Device Layer: This includes the devices/actuators that can generate ve-
hicle motion. This layer is also consisting sensors which could include the capability and
status of each device e.g. max/min wheel torque.
• Formalisation of different types of:
• Blocks, e.g. Controller, Information Fusion, Arbitrator and Coordinator.

75
 Introduction

• Both arbitrators and coordinators have inputs and outputs as requests, typically ex-
pressed in same physical quantity. An arbitrator has more requests in that out and a
coordinator has more requests out than in.
• Signals, e.g. Request, Actual (or Status) and Capability.
• Parameters used in Functional blocks. One can differ between Physical parameters (or
Model parameters) and Tuning parameters. Some parameters can be common across the
whole vehicle, which enables a kind of communication between blocks without normal
signals, but instead exchanging values during start-up of the system.
Each signal should have a definition of which physical quantity and unit it refers to. For signals of Ac-
tual type, there also needs a concept to handle how accurate they are; e.g. as a tolerance or sending an
upper and lower value between the physical quantity shall be. The definition of physical quantity is
very important, not the least for Request signals. An example is, if using tyre longitudinal slip as quan-
tity, sender and receiver of signal have to agree on slip definition. Since there are many slip definitions
around, it might suggest using wheel rotational speed instead, which is less ambiguous.
The functionality is then allocated to ECUs, and signals allocated to network communication. The ref-
erence architecture can be used for reasoning where the allocation should be done. Which functional-
ity is sensitive for e.g. time delay and thus should be allocated in the same ECU? Detailed control algo-
rithm design is not stipulated by a reference architecture. Instead the reference architecture assists
how detailed control algorithms be managed in the complete problem of controlling the vehicle mo-
tion. Whether solutions of Functional Safety (ISO 26262, 2011-2012) is represented in a reference ar-
chitecture of functionality can vary.
Vehicles consisting of several units add special challenges, especially if the units are actuated. A heavy
truck trailer is always actuated with at least brakes.

1.6.4.3 Virtual Vehicle Architecture

Virtual vehicle architecture, VVA, here refers to how to organize complete vehicle models, which can
replace some real vehicle pre-series. VVA is a set of rules how a vehicle model should be modularized,
such as variable/signal interfaces, parameterization, format- and toolchains. The vision is that a whole
project organisation, within an OEM and its suppliers, could deliver modules to virtual pre-series and
that these modules would fit together. The FMI format, see 0, is one example of what could facilitate
this.
1.6.4.3.1 § Product Liability
Contribution from Mats Jonasson, Volvo Cars and Vehicle Dynamics at Chalmers

It becomes increasingly important for vehicle manufacturers to take responsibility of their products.
The main reason is that numbers and complexity of sensors, actuators and control functions increases.
The risks are often connected to how vehicle moves, i.e. to vehicle dynamics/vehicle motion. The num-
ber of risks increase both because the systems becomes more complex and because more information
is available on-board. The availability of information can mean that it is ethically wrong to not act on
it, e.g. to not automatically brake if camera detects a possible pedestrian ahead. An acceptable excuse
to not brake could however, e.g., be that braking would mean a more severe risk due to causing a colli-
sion from rear-coming traffic.
Failure Mode Effect Analysis (FMEA) has been around for a long time and it is applicable on any design
in any product, and not only safety aspects. Basically, one identifies things that can go wrong with a
certain conceptual design early in development and decide how to follow it up during the develop-
ment. The follow-up normally includes to formulate requirements on the design. ISO26262 (ISO
26262, 2011-2012) describes a similar way of working, but stricter and more specialized on traffic
safety hazards and vehicles as products.
A very brief description of (ISO 26262, 2011-2012) follows here, see also figure below. A certain Item
(part, subsystem or function) is identified for Hazard Analysis early in product development. Hazards
are defined and assessed to a certain Automotive Safety Integrity Level (ASIL) in a Risk Assessment. The
ASIL levels, listed with increasing severity, are QM, ASIL-A, ASIL-B, ASIL-C, ASIL-D. The levels ASIL-A to

76
 Introduction

ASIL-D require Safety Requirements managed under an ISO26262 process during the continued devel-
opment, while the QM does not. A Safety Requirement is a requirement on the design which reduces
the ASIL to a tolerable level.
Conceptually, one can calculate ASIL for a certain hazard: 𝐴𝑆𝐼𝐿 𝑅 𝑠𝑘 𝑆𝑒 𝑒𝑟 𝑡𝑦 ( 𝑥𝑝𝑜𝑠𝑢𝑟𝑒
𝑜𝑛𝑡𝑟𝑜 𝑡𝑦) . For instance, loss of acceleration has generally lower Severity than loss of decelera-
tion. For instance, the Exposure becomes lower if the hazards only appears in sharp turns or if there is
redundancy in the solution. For instance, a hazard during automated driving increases the ASIL via the
Controllability factor, since human driver is then less alert and less likely to control the situation.
ISO26262 describes how the work should be documented. The documentation can be asked for in a
legal court case.

Figure 1-57: § Overview of the safety life cycle, from (ISO 26262, 2011-2012).

1.6.5 Verification Methods with Real Vehicle

1.6.5.1 In Traffic
Driving on public roads in real traffic is the most realistic way to verify how well a vehicle fulfils the
requirements during real use. It can be used for completely new vehicle models; or new systems,
mounted on old models. The drivers can be either ordinary drivers (FOT=Field Operational Test) or
test drivers (expeditions). A general existing vehicle population can also be studied by collecting data,
e.g. as Accident Statistics Databases.

1.6.5.2 On Test Track

For vehicles and systems which are not yet allowed on public roads, or tests which are very severe or
need a high degree of repeatability, test are carried out at test tracks. There are specialized test tracks
for certain conditions, such as hot climate or slippery surfaces.

77
 Introduction

1.6.5.3 Rapid Prototyping, RP

Rapid prototyping here refers to rapid prototyping of control algorithms. The method is to have a real
vehicle but especially equipped with a real-time computer with I/O to the vehicle. Changes of control
algorithms can be done in a laptop on-board and then new versions can be downloaded to the real-
time computer and tested a minute later. This enables a really quick development loop.

1.6.5.4 Augmented Reality, AR or Manipulated Environment

This is a new method. A typical example is: A real driver drives a real vehicle on a real road/test track.
Some additional (virtual/simulated) traffic objects are presented to driver, e.g. on a head-up display.
The same objects can be fed into the control functions, as if they were detected by the vehicle’s cam-
era/radar, which enables functions such as automatic braking to be triggered.

1.6.5.5 Objective Measures and Subjective Assessment

The vehicle can be instrumented so
that measurements can be logged and
later compared with requirements.
Also, the tests can be performed with
driving robot, so that the manoeuvres
are well repeated.
However, driving also with human
driver has its own value; the subjective
assessments can be registered via in-
terviews and debriefing after per-
formed tests. A typical scale used for
subjective rating or assessment is
shown in Figure 1-58.

Figure 1-58: Subjective rating scale (SAE J1441, 2016.)

1.6.6 Verification Methods with Virtual Vehicle

This section introduces different methods called “XIL=X in the Loop”. The loop referred to is then the
representation of the whole system, except for the X. The loop is typically a simulation model, but
there can also be hardware.

1.6.6.1 Testing with Real Driver

1.6.6.1.1 DIL =Driver in the Loop
Verification with a driving simulator (or DIL) is when a real human, not a driver model, uses real
driver devices (pedals, steering wheel) to influence a simulation model of the remaining system (vehi-
cle and environment). The loop is closed by giving the human feedback through display of what would
be visible from driver seat, including views outside windscreen.

78
 Introduction

Feedback can be further Alternative verification methods

improved by adding a mo- Office Driving Real vehicle
tion platform to the simulation simulator test
driver’s seat, sound, vibra- Test new technology before it is built
(save time and money) + - --
tions in seat, steering
wheel torque, etc. The ve- Design parameter
sweep/optimization +! - --
hicle model run in the sim-

Requirements
ulator can utilize HIL, SIL
Repeatable and Parallelized testing + + -
and MIL, from 0, for some Safe testing + + -
part of the vehicle model. Representative integration in vehicle - +! ++
Driving simulator can be w.r.t. to vehicle - - +!
Represen-
compared to other verifi- tative
w.r.t. driver - +! +
cation methods, see Figure behaviour w.r.t. to surroundings -- -
(road & traffic) (env. sens) (env. sens) +!
1-59. Subjective scale from
Figure 1-58 can be used. Figure 1-59: Comparison of verification methods for complete vehicle
functions.

1.6.6.2 Testing using Driver Model

Generally, this means that driver, vehicle and environment are modelled and simulated. To perform a
serious test there is also need to put effort on test scripts so that the model is run in the intended test
manoeuvre and result is pre-processed. For verification of functions involving with algorithms, the
HIL, SIL and MIL below are different ways to represent the algorithm in the vehicle model.
1.6.6.2.1 HIL = Hardware in the Loop Simulation
The hardware is often one or several ECUs (Electronic Control Units). If several ECUs are tested, the
hardware can also contain the communication channel between them, e.g. a CAN bus. The hardware is
run with real-time I/O to simulation model of the remaining system (vehicle, driver and environment).
In some cases, there is also mechanical hardware involved, such as if the ECU is the brake system ECU,
the actual hydraulic part of the brake system can also be included in the HIL set-up, a so called “wet
brake ECU HIL”.
For HIL, there can be two types of hardware involved: The hardware under test (the H in HIL) and
hardware that helps representing the rest of the system, i.e. a part of the loop (the L in HIL).
1.6.6.2.2 SIL = Software in the Loop Simulation
The software is often one or several computer programs (intended for download in electronic control
units). The software is run with synchronized time discrete I/O to a simulation model of the remaining
system (vehicle, driver and environment).
The software is often used in compiled format (black box format) so that the supplier of the software
can retain his intellectual property.
1.6.6.2.3 MIL = Model in the Loop Simulation
The model, or more correctly, a control algorithm, is a conceptual form of the computer programs (in-
tended for download in electronic control units). The control algorithm is run with I/O to a simulation
model of the remaining system (vehicle, driver and environment).
The control algorithms can appear in compiled format so that the supplier of the control algorithms
can retain his intellectual property. Then it is hard to tell the difference between MIL and SIL.

1.7 Heavy Trucks

The following section describes heavy trucks, mainly as compared to passenger vehicles.

79
 Introduction

1.7.1 General Differences

Trucks are normally bought and owned by companies, not private persons. Each truck is bought for a
specialized transport task. Life, counted in covered distance, for trucks is typically 10 times passenger
cars. The lifetime cost of fuel is normally 5 times the vehicle cost, compared to passenger car where
these costs are about equal. The cost for driver salary is a part of mileage cost, typically same magni-
tude as fuel cost. If investment cost for vehicle and repairs are distributed over travelled distance,
these are typically also of same magnitude. So, the cost for a transport typically comes from one third
fuel, one third driver salary and one third vehicle investment and repairs.

1.7.2 Vehicle Dynamics Differences

A truck has 5..10 times less power installed per vehicle weight. Trucks have their centre of gravity
much higher, meaning that roll-over occurs at typically 4 m/s2 lateral acceleration, as compared to
around 10 m/s2 for passenger cars. Trucks have centre of gravity far behind mid-point between axles,
where passenger cars have it approximately symmetrical between the axles. Trucks are often driven
with more units after, see Figure 1-60. The weight of the load in a truck can be up to 2..4 times the
weight of the empty vehicle, while the maximum payload in passenger cars normally are significantly
lower than the empty car weight. Trucks often have many steered axles, while passenger cars normally
are only steered at front axle. The view for the driver (and environment sensors) is from higher up and
without inner rear mirror. The lateral margins in lane are much smaller.
Another difference is that it is not as easy as for two-axle vehicles, to find the vertical loads under each
axle. This is both with and without longitudinal load transfer. The load varies a lot in mass and posi-
tion, there is often multiple axles in each axle group and there can be a vertical force exchange be-
tween the units. See 4.5.2.2.2.2.

1.7.3 Definitions
In Figure 1-60, we can find the following units:
• Towing units: Tractor or Rigid (Truck)
• Towed units: Full trailer (FT), Semi-trailer (ST), Centre-axle trailer (CAT), (Converter) Dolly
• The couplings between the units can of 2 types:
o Fifth-wheel coupling (e.g. between Tractor and Semi-trailer). Designed to take force in all
3 directions. Furthermore, it is designed to be roll-rigid, but free in pitch and yaw.
o A Turntable is similar to a fifth wheel but has a rolling bearing instead of a pitch-hinged
greased surface, which leads to less yaw friction and no pitch degree of freedom. In con-
verter dollies, one often sees both, fifth wheel on top of turntable.
o Hitch coupling or Drawbar coupling (e.g. between Rigid and Full trailer). Designed to
take forces in longitudinal and lateral directions, but only minor in vertical direction. Also,
it is rotationally free in all 3 directions. A full-trailer has pitch-moment-free rear end of
drawbar. A converter dolly or centre-axle-trailers have pitch-rigid rear end of drawbar.
• There is often a need for several axles close to each other, to manage a high vertical force. Such
group of axles is called “axle group”, “axle arrangement” or “running gear”.

80
 Introduction

Conventional CVs 16.5 m, ST Tra

Tractor-ST Tractor-SemiTrailer
40 t
18.75 m, CAT Tru
Truck CAT Truck-CentreAxleTrailer
40 t
18.75 m, FT
Tru
Truck FT Truck-FullTrailer
40 t
B-double or Tractor-LinkTrailer- 25.25 m, ST Link Tra
Tractor B-
double SemiTrailer 60 t
Existing LCVs

Tractor-ST- Tractor-SemiTrailer- 25.25 m, CAT ST Tra

CAT CentreAxleTrailer 60 t
25.25 m, ST Tru
Truck-Dolly-SemiTrailer Dy
60 t
Nordic CV FT
24 m, Tru
Truck-FullTrailer
60 t
Tractor-SemiTrailer-Dolly- 31.5 m, ST ST Tra
Dy
SemiTrailer 80 t
Prospective LCVs

A-double FT
Tractor-SemiTrailer- ≈30 m, ST Tra
FullSemiTrailer 80 t
Truck Truck-CentreAxleTrailer- 27.5 m, CAT CAT Tru
DuoCAT CentreAxleTrailer 66 t
AB-double Truck-Dolly-LinkTrailer- 34 m, ST Link Tru
or Truck B- Dy
double SemiTrailer 90 t

vertical-free, pitch-free, roll-free (hitch coupling on pitch-free drawbar)

vertical-rigid, pitch-free , roll-free (hitch coupling on pitch-rigid drawbar)
vertical-rigid, pitch-free, roll-rigid (fifth-wheel coupling)
vertical-rigid, pitch-rigid, roll-rigid (turn-table coupling)
(all above couplings are longitudinal-rigid, lateral-rigid, yaw-free)
Figure 1-60: An overview of combination vehicles (CVs). Based on (Kharrazi , 2012). The
FullTrailer is registered as one unit, but it can be modelled as two units with a coupling between.

1.8 Smaller Vehicles

This section is about smaller vehicles, meaning bicycles, electric bicycles, motorcycles and car-like ve-
hicles for 1..2 persons. The last vehicle type refers to vehicles which are rare today, but there are rea-
sons why they could become more common: Increasing focus on energy consumption and congestion
in cities can be partly solved with such small car-like vehicles, of which the Twizy in Figure 1-61 is one
example. All vehicles in Figure 1-61 can be referred to as Urban Personal Vehicle (UPVs), because they
enable personalised transport in urban environments. The transport can be done with low energy con-
sumption per travelled person and distance, compared to today’s passenger cars. The transport will
also give higher levels of flexibility and privacy for the travelling persons, compared to today’s public
transportation. If such vehicles also could be shared between users, this concept will even more help
congestion in cities but might require automation of booking and maybe even automation of driving.
Beside UPVs there could be considered smaller Urban Freight Vehicles (UFVs), but driver salary costs
would then be very high per transported mass (or volume) of goods. So, UFVs might need to wait for
automation of driving.
Figure 1-61 show some UPVs. UPVs/UFVs may require some categorizations:
• Climate and user type: Sheltered or open.

81
 Introduction

• Transport and user type: Short travels (typically urban, 5-10 km, 50 km/h) or long travels (typi-
cally inter-urban, 10-30 km, 100 km/h).
• Chassis concept:
o Narrow (e.g. normal bicycles and motorcycles) or wide (at least one axle with 2 wheels, re-
sulting in 3-4 wheels on the vehicle). Note that UPVs in both categories are typically still less
wide than passenger cars.
o Roll moment during cornering carried by suspension roll stiffness or roll moment during cor-
nering avoided by vehicle cambering. The first concept can be called “Roll-stiff vehicle”. The
second concept can be called “Cambering vehicle” or “Leaning vehicle”. One-tracked vehicles
are always Cambering vehicles. two-tracked are normally Roll-stiff, but there are examples of
Cambering such (see upper right in Figure 1-61). See Figure 1-62.
o (This compendium does only consider vehicles which are “Pitch-stiff”, i.e. such that can take
the pitch moment during acceleration and braking. Examples of vehicles not considered,
“Pitching vehicles”, are: one-wheeled vehicles as used at circuses and two-wheeled vehicles
with one axle, such as Segways.)
• Note that also Roll-stiff Vehicles camber while cornering, but outwards in curve and only
slightly, while Cambering Vehicles cambers inwards in curve and with a significant angle.
Cambering Vehicles is more intricate to understand when it comes to how wheel steering is
used. In a Roll-stiff Vehicle, the wheel suspension takes the roll-moment (maintains the roll
equilibrium), which means that driver can use wheel steering solely for making the vehicle
steer (follow an intended path). In Cambering Vehicles, the driver has to use the wheel steering
for both maintaining the roll equilibrium and following the intended path. A model is given in
4.5.2.3. Figure 1-62 shows some possible conceptual design of a cambering vehicle. It is not
possible to do a partly roll-stiff and partly cambering vehicle, unless one uses suspension
springs with negative stiffness, i.e. the suspension would need to consume energy to tilt the
vehicle inward in curve.
3 wheeled cambering Piaggio MP3
concept vehicle from BMW

Toyota
iRoad
Twizy from Renault

ZeeBee from Clean Motion Eco Electric Bicycle, Monark

Figure 1-61: Examples of Urban Personal Vehicles. Left: “Roll-stiff”. Right: “Cambering“.
From www.motorstown.com, www.cleanmotion.se, www.monarkexercise.se,
www.nycscootering.com.

82
 Introduction

“Roll-stiff concept” … …as opposed to “Cambering concept”

Translate Lean sprung body Lean whole vehicle, Lean whole vehicle,
2 wheel axle single wheel axle

Figure 1-62: How Cambering vehicles avoid taking the roll moment while cornering. The vehicles
are viewed from rear and turns to the left. The dash-dotted arrows mark the resulting forces
which are equal between the 5 concepts.

1.9 Notation List

Figure 1-63 to Figure 1-65 show notation of parameters, variables and subscripts used in this compen-
dium. The intention of this compendium is to follow International Standards (ISO 8855), but devia-
tions from this is sometimes used, especially motivated by conflicts with areas not covered by (ISO
8855). Some alternative notations are also shown, to prepare the reader for other frequently used no-
tation in other literature.

83
 Introduction

Notation
Categorization Recommended Alternatives
Subject for notation in code by hand Unit Description / Note
Basic (Shaft) Torque T T M Nm
Basic Coefficient of friction mu μ mue, muh 1=N/N
N/(m/s) or
Basic Damping coefficient d d c,k,D
Nm/(rad/s)
Basic Density roh ρ
Ratio between
Basic Efficiency eta hη 1=W/W useful/output power
and used/input power
Basic Energy E E W Nm=J
Basic Force F F f N
9.80665 average on
Basic Gravity g g m/s2
Earth
Basic Height h h H
Basic Imaginary unit j j i - j*j=-1
Basic Mass m m M kg
Subscript can be given
Basic Mass moment of inertia J J I kg*m2 to denote around
which axis
Basic Moment (of forces) M M T Nm Not for shaft torque
Basic Power P P W=Nm/s=J/s
Ratio between input
Nm/(Nm) or and output rotational
u,
Basic Ratio r r (rad/s)/(rad/s speed, or output and
i (ISO8855)
) input torque, in a
transmission
Basic Rotational speed w ω w rad/s
Basic Shear stress tau τ N/m2
1. General

N/m or
Basic Stiffness coefficient c c C, k, K
Nm/rad
Basic Strain eps ε m/m
Basic Translational speed v v V m/s
Basic Wave length lambda λ m
Can be track width,
Basic Width W W w m vehicle width, tyre
width, …
Dynamics (Time) Frequency f f 1/s=periods/s
Dynamics Angular (time) frequency w ω w rad/s
Dynamics Angular spatial frequency W Ω rad/m
Dependent variables in a dynamic Both state variables
Dynamics z z <various>
system and output variables
Input variables in a dynamic
Dynamics u u <various>
system
Dynamics Mean Square value MS MS
Output variables in a dynamic
Dynamics y y <various>
system
Dynamics Power spectral density G Φ PSD
Dynamics Root Mean Square value RMS RMS
Spacial frequency as radians per
Dynamics W Ω rad/m
travelled distance
1/m=periods/
Dynamics Spatial frequency fs fs
m
States variables in a dynamic
Dynamics x x <various>
system
The independent
Dynamics Time t t t, T s variable in a dynamic
system
Operators Transfer function H H
1. General Air resistance coefficient cd cd Cd 1
2. Vehicle

Cornering stiffness or lateral tyre dF y /da = dF y /ds y at

1. General Cy Cy Ca N/rad or N/1
stiffness a=s y =0
1. General Longitudinal tyre stiffness Cx Cx N/1 df yw /ds x at s x =0

Figure 1-63: Notations

84
 Introduction

Notation
Categorization Recommended Alternatives
Subject for notation in code by hand Unit Description / Note
1. General Mass m m M kg
1. General Rolling resistance coefficient f f 1=N/N
1. General Track width W W m
1. General Tyre stiffness C C c N/1 dF/ds at s=0
Kus, kus, N/(N/rad) =
1. General Understeer gradient Ku Ku
U (ISO8855) rad/(m/s2)
If no subscript, undefined
1. General Vehicle side slip angle b β b rad
or CoG
1. General Wheel base L L l, lf+lr, WB m
1. General Wheel radius R R r, Rw m
2. Road Curvature kappa κ roh, r rad/m=1/m road or path
2. Road Curve radius R R r, roh, r m road or path
Positive when right side
2. Road Road bank angle pxr φ xr of ground is lower than
left side ground
2. Road Road inclination angle pyr φ yr rad Positive downhill
2. Road Vertical position of road zr zr z, Z m
Vehicle acceleration, in interial ax, ay, decomposed in vehicle
3. Motion a x, a y, a z m/(s2)
system az coordinates direction
dervx, der(vx),
Time derivatives of each 2 decomposed in vehicle
3. Motion dervy, der(vy), m/(s )
component v x , v y , v z coordinates direction
dervz der(vz)
r=[rx,ry,rz] often position of Center
3. Motion Vehicle position x,y,z x, y, z m
or [X,Y,Z] of Gravity, CoG
decomposed in vehicle
3. Motion Vehicle velocity, in interial system vx,vy,vz v x, v y, v z u,v,w m/s
coordinates direction
From ground, air, towed
Fx, Fy, units, colliding objects,
F x, F y, F z, [N, N, N, Nm,
4. Forces Forces and moments on vehicle Fz, Mx, etc. May appear also for
M x, M y, M z Nm, Nm]
2. Vehicle

My, Mz in road plane: F=[Fx, Fy,

Mz]
Subscript "i" is
placeholder for particular
Forces on one wheel, axle or side Fix,Fiy,F wheel, axle or side. May
4. Forces F ix , F iy , F iz lowercase f N
from ground iz appear also extended
with moments: [Fx, Fy, Fz,
Mx, My, Mz]
Forces on one wheel, axle or side Fixv, F ixv , F iyv , in vehicle coordinate
4. Forces lowercase f N
from ground Fiyv,Fizv F izv system

Forces on one wheel, axle or side Fixw,Fiyw F ixw , F iyw , in wheel coordinate
4. Forces lowercase f N
from ground ,Fizw F izw system
Sign convention positive
when either of "rotated
5. Angles Camber angle g γ g rad as wheel roll angle" or
"top leaning outward vs
vehicle body".
The angle required for a
Low speed or Ackermann speed
5. Angles dswA δswA rad certain vehicle path
steering wheel angle
curvature at low speeds
5. Angles Euler rotations ψ, θ, φ rad Order: yaw, pitch, roll
px, py, Angles, not rotations. Roll
5. Angles Roll, pitch, yaw angle φx, φy, φz j, q, y rad
pz and Pitch normally small

decomposed in vehicle
Vehicle angular velocity, in interial wx, wy,
5. Angles ωx , ω y , ω z rad/s coordinates direction
system wz
(roll, pitch, yaw)
May refer to steering
5. Angles Steering angle d δ delta rad
wheel, road wheel or axle

Normally an average of
5. Angles Steering angle of road wheels drw δrw RWA rad
angles on front axle

Figure 1-64: Notations

85
 Introduction

Notation
Categorization Recommended Alternatives
Subject for notation in code by hand Unit Description / Note
5. Angles Steering angle of steering wheel dsw δsw SWA rad
6. Slip Tyre (lateral) slip angle a α alpha rad a=arctan(sy)
6. Slip Tyre lateral slip sy sy sy=tan(a)
1=
6. Slip Tyre longitudinal slip sx sx k, -k
(m/s)/(m/s)
6. Slip Tyre slip s s
used e.g. as Ca=cornering
7. Subscript axle a a
stiffness for any axle

COG, cog,
7. Subscript centre of gravity CoG CoG
CG, cg
Often used as double
7. Subscript front f f F, 1 subscript, e.g. fr=front
right
Often means with respect
7. Subscript inner i i
to curve
Often used as double
7. Subscript left l l L
subscript, e.g. fl=front left
2. Vehicle

Often means with respect

7. Subscript outer o o
to curve
Often used as double
7. Subscript rear r r R, 2 subscript, e.g. fr=rear
right
7. Subscript road r r
E.g. drw=road wheel
7. Subscript road wheel rw rw RW
steering angle
7. Subscript sprung mass s s
SW, E.g. dsw=steering wheel
7. Subscript steering wheel sw sw
H (ISO8855) angle
7. Subscript unsprung mass u u us
used e.g. as F1v to denote
7. Subscript vehicle v v wheel force on wheel 1 in
vehicle coordinates
used e.g. as F1w to
denote wheel force on
7. Subscript wheel w w
wheel 1 in wheel
coordinates
1. Propulsion Engine torque Te Te Nm
3. Subsystem in vehicles

1. Propulsion Propulsion torque Tprop T prop Nm often expressed at wheel

2. Brake Brake torque Tbrk T brk Nm
3. Steering Steering wheel torque Tsw T sw SWT Nm
4. Suspension Caster angle CA rad
4. Suspension Caster offset c m
4. Suspension Kingpin inclination angle KPI rad
4. Suspension Pneumatic trail t m
4. Suspension (Wheel) Camber angle φCamber rad

Figure 1-65: Notations

1.10 Typical Numerical Data

The purpose of the tables in this section is to give approximate numerical values of parameters. Vehi-
cle parameters are often dependent on each other; changing one leads to that it is suitable to change
others. The parameters are given with the intention to be consistent with each other, for each vehicle
example. The tables balance between being generic and specific, which is difficult. Therefore, please
consider the table as very approximate.

86
 Introduction

Parameter Notation Typical Value Unit

Passenger car Heavy truck
Air density roh 1 kg/(m^3)
Earth gravity g 9.80665 m/(s*s)
Road friction, at dry asphalt 1.0
Road friction, at wet asphalt 0.6
mu N/N
Road friction, at snow 0.3
Road friction, at wet ice 0.1
Tyre Cornering Coefficient (lateral slip
(N/rad)/N
coefficient), i.e. slip stiffness CCy 10..15 5..10
= (N/1)/N
normalized with vertical load
Ratio of tyre longitudinal and lateral
CCx/CCy 1.5..2 1
slip coeffcient stiffness

Figure 1-66: Typical values of parameters, common for typical passenger cars and heavy trucks.

1.10.1 For Passenger Vehicles

Group/Type Parameter Notation Typical Unit Note / Typical / Range /
Value Relation to other
Vehicle Length, bumper to bumper 5.00 m
Longitudinal distance from 40-50% of wheel base:
Vehicle lf 1.3 m
CoG to front axle lf=0.55*L;
Vehicle Mass m 1700 kg
Radius of gyration is sligthly
Moment of inertia for yaw
Vehicle Jzz 2900 kg*m*m less (0.9) than half wheel base:
rotation
=m*(0.9*L/2)^2
Sum of 4 wheels with
Vehicle Unsprung mass mus 200 kg
suspensions
Vehicle Track width W 1.70 m
Vehicle Wheel base L 2.90 m
For calculation of air
Projected area in a resistance. Examples: Volvo
Vehicle Afront 2 m^2
transversal view XC90 has 2.78 m^2, Volvo S60
has 2.27 m^2
For calculation of longitudinal
cd, clf, 0.31,
Vehicle Aerodynamic coefficients 1 resistance and lift force over front
clr 0.10, 0.07 and rear axle
Wheel and Tyre Rolling resistance 0.01 N/N
Wheel and Tyre Wheel radius 0.30 m
Wheel and Tyre Wheel rotational inertia 0.5kg*m*m For one wheel
Wheel and Tyre Tyre vertical stiffness ct 250 000 N/m For one wheel
Propulsion Engine inertia 0.5kg*m*m
Transmission ratio, highest rad/rad= Engine to wheel. In magnitude
Propulsion 5.00
gear Nm/Nm of 5.
Transmission ratio, lowest rad/rad= Engine to wheel. In magnitude
Propulsion 10.00
gear Nm/Nm of 10.
Brake proportioning, N/N or Often tuned so that braking
Brake 70/30
front/rear Nm/Nm both axles when braking 0.8*g
Steering Steering ratio 16 rad/rad Steering wheel to Road wheel
Vertical stiffness at wheel. Sum of
4 wheels. So that bounce frequency
Suspension heave stiffness f is between 1 and 2 Hz:
Suspension cs 100 000 N/m
(without roll) sqrt(c/m)=f*2*pi;
Bump stop at (3..4)*g when fully
loaded.
Suspension roll stiffness, Vertical stiffness at wheel. Sum of
Suspension carb 14 000 N/m
only from anti-roll-bars both axles.
Wheel rate. Sum of 4 wheels. Some
40..60% of critical damping:
Suspension Suspension damping ds 13 000 N/(m/s) d = (0.4..0.6)*2*sqrt(c*m);
2..3 times softer in compression
than rebound.

Figure 1-67: Typical data for a passenger vehicle.

87
 Introduction

1.10.2 For Heavy Vehicles

Compared to passenger cars, trucks differ much more in size and configuration, see Figure 1-60. Also, a
certain individual vehicle also differs much more between empty and fully loaded. Globally, “Tractor
with Semitrailer” is the most common heavy vehicle, which is why it is used as example.
load length=13.6
front overhang
wheel base=8 =1.5

overall height =
overall height

3.7 m
=4 m

rear rear forward

overhang axle spread offset =4 front
=1.5 overhang overhang
=2.6 =1 =1.5
wheel base=5
vehicle overall length=16.5 m

Figure 1-68: Typical data for a heavy vehicle, exemplified with “Tractor with Semi-trailer”

Typical Value Note / Typical / Range /

Group/Type Parameter Empty With Unit Relation to other
max
payload
Trackwidth, centre-to-centre for
Vehicle 2.25 m Outer tyre edge 2.55
single tyres
Projected area in a transversal For calculation of air
Vehicle 10.0 m^2
view resistance
c_d coefficient in aero dynamic
Vehicle Aerodynamic coefficient 0.4..0.6 1 resistance formula (normal truck
0.4, long combination truck 0.6)

Tractor CoG heigth 1.00 m

Tractor Mass (total), tractor 7500 kg
Tractor Moment of yaw inertia 22500 kg*m*m Around unit CoG
Tractor Unsprung mass 1700 kg Sum of all axles at unit
Semitrailer CoG heigth 1.00 2.00 m
Semitrailer Mass (total), semi-trailer 10000 32500 kg
Semitrailer Moment of yaw inertia 150000 500000 kg*m*m Around unit CoG
Semitrailer Unsprung mass 2400 kg Sum of all axles at unit
Wheel and Tyre Wheel radius 0.50 m 0.4-0.5
Wheel and Tyre Wheel rotational inertia 10.0 kg*m*m For one wheel, single tyre
Wheel and Tyre Rolling resistance 0.005 N/N Or less, 0.003
kNm/s=k
Propulsion Engine max power 370
W
Propulsion Engine inertia 5 kg*m*m
Propulsion Transmission ratio, highest gear 3 1=rad/rad Engine to wheel
=Nm/Nm
Propulsion Transmission ratio, lowest gear 30 Engine to wheel
Steering wheel to Road
Steering Steering ratio 20 rad/rad
wheel
Steering wheel to Road
Steering Steering wheel rotational inertia 0.04 kg*m^2
wheel

Figure 1-69: Typical data for a heavy vehicle

88
 Vehicle Interactions and Subsystems

2 VEHICLE INTERACTIONS AND

SUBSYSTEMS
2.1 Introduction
The study of vehicle dynamics starts with the interfaces between the vehicle and its environment, see
Figure 2-1. The vehicle tyres are the primary force interface for all motion (in road plane, i.e. accelera-
tion, steering, braking) but also undesired disturbances (out-of- road plane, i.e. road unevenness, road
bumps, etc.). Additionally, the aerodynamic loads on the vehicle will create forces that are often unde-
sirable (wind resistance, side gusts, etc.) but can sometimes be exploited for better contact with the
road (down-force). An example of extreme interactions to the vehicle is the impact forces from a crash.
An interaction of another kind is the driver. One often divides the whole system into 3 parts: vehicle,
driver and environment, see Figure 2-1. Driver and environment can often be clustered as 2.9 Driving
and Transport Application, which includes all outside subject vehicle. Some aerodynamic models are
given in 2.8. Driver models are discussed with some selected example models, see 2.9.2.
Driver
(via Driver) Traffic, Infrastructure
Driver/Rider,
User or
Customer Steering Wheel,
Pedals, Buttons,
Air, Weather Seat Accelerations,
Visual Instruments,
Sounds, …
Aerodynamic Relative Motion,
Collision Forces, (Object) vehicles, Other road users,
Air density, Weather, Forces, Vision, …
Information, … Traffic signs, Road markings,
Light/dark, …
Infrastructure, …
(Subject)

Freight
Vehicle Road surface construction, Covered
Weight with water/snow, Topography,
Payload, Curvature, Unevenness, …
Occupants, … and Location Tyre Forces

kg
Ground, Road
Environment
Figure 2-1: How the vehicle interacts with driver and environment.
The chapter also contains descriptions of the subsystems which are most relevant for Vehicle Dynam-
ics, with focus on Wheels and Tyres, see Figure 2-2.
The Environment Sensing System is described as a subsystem, see 2.7. There are also a lot of other
sensors in the vehicle, sensing the conditions in the subject vehicle, see 3.5, 4.6 and 5.10. These sen-
sors are often more established and therefore often treated as included in a certain subsystem, such as
wheel (rotational) speed sensors are often a part of the brake system.

2.1.1 References for this Chapter

• Tyres: (Pacejka, 2005), (Rill, 2006), and (Michelin, 2003)
• Vehicle Aerodynamics: Reference (Barnard R. , 2010)
• Driver models: “Chapter 38 Driver Models in Automobile Dynamics Application” in Reference
(Ploechl, 2013)

89
 Vehicle Interactions and Subsystems

2.2 Wheels and Tyres

2.2.1 Introduction
The wheels and tyres of a vehicle have the following tasks:
o Carry the vertical load
o Transmit longitudinal forces (propulsion > 0 and brake < 0)
o Transmit lateral forces (steering)
To minimize negative effects, we can also list these:
o Roll with minimum energy loss, minimum tyre wear, minimum particle emissions and noise
o Isolate disturbances, mainly vertical. This includes also to not cause disturbances, which over-
all means that the wheel and tyre should be round and balanced.
Acceleration & Brake Pedal Environment Sensing

Ctrl steering Steering

wheel

(vehicle)
Propulsion body

Suspension

brake disk
Propulsion, Brake and Steering are
knuckle, bearing Brake “active” in the meaning that they are
“controllable through algorithms run
Wheel & Tyre on an electronic control unit ECU”.

Figure 2-2: Vehicle with the 6 Vehicle Dynamics relevant subsystems: Wheels and Tyres,
Suspension, Propulsion, Brake, Steering, and Environment Sensing. Towed units, with similar
subsystems, are not shown.

2.2.1.1 Wheel Integration in Suspension

To understand the “force flows” of a wheel, it is good to understand the conceptual integration, regard-
ing shaft, brake disks and wheel bearings. Cars typically have their inner bearing rings rotating with
the wheel, while trucks instead have outer rings rotating, see Figure 2-3.
Car, view from top or rear Truck, view from top or rear
brake calliper brake calliper
𝑁 𝑝 𝑝 brake disk
𝑀𝑤 𝑙
𝑀𝑤 hub 𝑁
𝑙 𝑝
brake disk

𝑇 𝑓 Wheel
shaft bolted axle end (non-driven) 𝑝
onto here
𝑤 𝑙 Non-driven: shaft ends
Wheel here, 𝑇 𝑓 0.
from http://19526-presscdn-0- bolted onto Driven: shaft continue to 𝑤 𝑙 axle end (driven) 𝑇
66.pagely.netdna-cdn.com/wp- propulsion system, via
𝑓
brake disk
content/uploads/2014/12/wheel-bearings.jpg drive-joint if steered.

⃗
Figure 2-3: Concepts of wheels are typically integrated. All vectors are 3D, while moment 𝑁 is 2D
in the wheels xz-plane and moment 𝑇 𝑓 is perpendicular to the same plane.

2.2.1.2 Tyre Design

Before discussing the mechanics of tyre and road interactions, the physical structure of the wheel as-
sembly should be understood. Consisting of a steel rim and an inflated rubber toroid, pneumatic tyres
were invented and patented by Robert William Thomson in 1845 and are essentially the only type of
tyre found on motor vehicles today.
The physical construction of the tyre carcass affects the response of the tyre to different road loadings.
The carcass is a network of fabric and wire reinforcement that gives the tyre the mechanical strength.

90
 Vehicle Interactions and Subsystems

The structure of the carcass can be different: Bias-ply, Bias-ply Belted, and Radial-ply, see Figure 2-4.
Bias-ply tyres were the first types of pneumatic tyres to be used on motor vehicles. Radial ply tyres fol-
lowed 1946 and became the standard for passenger cars and is today also dominating also on trucks.
Note how the bias-ply constructions have textile structures oriented at an angle to the tyre centreline
along the x-z plane. This angle is referred to as the crown angle and is further illustrated in Figure 2-4.
Note the textile orientation for the bias-ply and radial tyres. Also note the difference in crown angles
between the two tyre constructions. This difference plays an important part in the rolling resistance
characteristics of the tyre which is 2.2.1.6.
The tyre components in Figure 2-4 have been constructed to provide the best tyre performance for dif-
ferent loading directions. Trade-offs are necessary between handling performance and comfort, be-
tween acceleration and wear, as well as between rolling resistance and desired friction for generating
forces in ground plane. The rubber components and patterns incorporated in the tread are critical to
the friction developed between the tyre and road under all road conditions (wet, dry, snow, etc.). Fric-
tion is most relevant in longitudinal and lateral vehicle dynamics. The belts define the circumferential
strength of the tyre and thus braking and acceleration performance. The sidewall and plies define the
lateral strength of the tyre and thus influence the lateral (cornering) performance of the vehicle. The
sidewall as well as the inflation pressure are also significant contributors to the vertical stiffness prop-
erties of the tyre and affect how the tyre transmits road irregularities to the remainder of the vehicle.
A tyre that has strong sidewalls will support vertical load well, but at the cost of vertical compliance.
bias-ply:

belt

radial-ply (most common today):

sidewall

Figure 2-4: Left: Carcass Construction, (Wong, 2001). Left top: Bias-ply construction. Left bottom:
Radial construction. Right: Radial Tyre Structure, (Cooper Tire & Rubber Co., 2007).

marking: S/[mm] / 100*H/S construction Dr/[inch]

example, car: 165 / 65 R 14
example, truck: 315 / 80 R 22.5

Unladen radius R0 = Dr*25.4/2+(H/S)*S

for car example = 14*25.4/2+0.65*165 = 285 mm
for truck example = 22.5*25.4/2+0.8*315 = 538 mm

From https://putneys.ca/winter-tires-time-
take-off-now/
Figure 2-5: Left: Tyre marking (radial tyre). Right: Typical summer and winter tread patterns.

91
 Vehicle Interactions and Subsystems

2.2.1.3 Wheel Angles

A wheel or a tyre is best described in its own coordinate system, see Figure 2-6. The 3 forces and 3 mo-
ments are acting on the wheel from the ground. The other quantities, such as slip angle and camber
angle 𝜀𝑤 , in Figure 2-6 are examples of variables which have influence on the forces and moments in
the coming chapters.
perpendicular
to road plane

Plane perpendicular to road

𝑀
𝑀
𝑀

Figure 2-6: Tyre coordinate system. Forces and moments on tyre from ground. From (ISO 8855).
2.2.1.3.1 Steer Angle
A wheel on a vehicle has a Steer angle, . The Steer angle is the angle from vehicle longitudinal axis to
the wheel plane about the vehicle vertical axis (ISO 8855). Assuming vehicle 𝑥𝑦-plane is parallell to
road plane, this angle is same as angle from vehicle longitudinal axis to the intersection between wheel
plane and road plane.
2.2.1.3.2 Camber Angle
A wheel on a vehicle on a road surface has a Camber angle. The Camber angle is the angle from vehicle
longitudinal axis to the wheel plane about the vehicle longitudinal axis (ISO 8855). Assuming vehicle
𝑥𝑦-plane is parallell to road plane, this angle is same as the deviation from right angle between wheel
plane and road plane. A symmetry within an axle means that camber angle typically has opposite signs
on left and right wheel. However, for two-track vehicles, one can often see the sign convention that a
wheel top leaning outward from vehicle body is negative, regardless of on left or right side.
2.2.1.3.3 Steer and Camber
If a wheel has both significant Steer angle and significant Camber angle at the same time, these angles
alone does not define the orientation, since steering-before-cambering gives another orientation than
cambering-before- steering. One way is to express the orientation in a rotational angle around the
steering axis, see 2.2.1.3.5. An alternative is to use the concept of Euler rotations.
2.2.1.3.4 Castor Angle
A wheel, suspended to be steered, has a Castor angle (or Caster angle). This is the angle between the
vehicle vertical axis and the projection of the steering axis (2.2.1.3.5) on the vehicle 𝑥 -plane (ISO
8855). Castor angle is positive when top of steering axis is inclined rearward. Castor angle provides an
additional aligning torque, see 2.2.4.7 and 0, and changes the Camber angle when the wheel is steered.
2.2.1.3.5 Steering Axis Inclination
A wheel, suspended to be steered, has a Steering axis inclination (or Kingpin inclination).

92
 Vehicle Interactions and Subsystems

This is the angle between the vehicle vertical axis and the projection of the steering axis (2.2.1.3.5) on
the vehicle 𝑦 -plane (ISO 8855). Steering axis inclination is positive when top of steering axis is in-
clined inward. See also 0.
2.2.1.3.6 Static Toe Angle
A wheel on an axle has a Static toe angle (a.k.a. Toe-in). The Static toe angle is the angle between vehi-
cle longitudinal axis and the wheel plane about the vehicle vertical axis, with the vehicle at rest and
steering in the straight-ahead position (ISO 8855). Positive Static toe angle (a.k.a. Toe-in or “ Toe-
out”) is when forward portion of the wheel is closer to the vehicle centreline than the wheel centre.
Typically, left and right wheel on same axle has same Static toe angle. Hence it measures an axle prop-
erty rather than a wheel property.
One could define an “instantaneous toe angle”, for an axle during arbitrary vehicle operation (non-rest
and non-zero steering), as ( + ( 𝑙 𝑓 ))⁄ . This will have a different value than Static toe angle
due to suspension linkage geometry and elasticity in suspension bushings.
Toe (regardless of toe-in or toe-out) generally generates opposing lateral forces on left and right side
leading to propulsion energy loss and tyre wear. Toe-out on front axle and toe-in on rear axle makes
the vehicle more yaw stable (less over-steered). Tone-in on front axle makes vehicle more yaw agile
but it also improves on-centre steering feel. Normal design choice for a passenger car is positive Static
toe angle (Toe-in) on both axles, and more on front axle.

2.2.1.4 How Tyres Carries Vertical Load

Figure 2-7 shows how a pneumatic tyre carries the vertical load. In a traditional bicycle or motorbike
wheel, the pre-tensioned spokes have a similar role as the air and rubber parts of a pneumatic tyre.

Upper Sidewall
Air, inflated
Belt

Shaft
Lower Sidewall
(Front and Rear Sidewall are neglected here, but
could have been modelled as shear springs.)

Inflated, but not loaded Loaded

slightly increased push,

increased

Upper
pull

𝑭𝒛
pull
decreased pull

Sidewall
but same
push
pull
Belt,

Shaft Shaft Shaft

decreased

Lower
pull

pull

These 3 all change in

Sidewall direction to carry .
The largest of these
Air changes is in (lower)
𝑭𝒛 sidewall!

Figure 2-7: A pneumatic tyre is pre-tensioned by inflation pressure. This figure explains how the
wheel and tyre takes (vertical) load without (significant) increase of inflation pressure.

2.2.1.5 Tyre Model Architecture

This section describes how a tyre model can be instantiated once per wheel and integrated in a model
together with models of vehicle, driver, and environment, see Figure 2-8. It does not define a variable
flow (causality) between the wheel and the tyre model; such causality depends on how the wheel iner-
tias and elasticity is modelled. This sectionFigure 2-8 differs between parameters and variables, in
context of dynamic study, i.e. in time domain. Within each simulation, parameters are constant, but
variables can vary. If design optimization, as opposed to simulation, the design parameters also has a

93
 Vehicle Interactions and Subsystems

role of variables (i.e. they are varied) in the optimization, but not in time domain. This section follows
approximately the STI tyre model interface standard (Besselink, 2011), but not in all details.
Operational parameters

Driver
(Wheel and
Tyre) Design Behavioural Physical
Parameters Rest of
Tyre response Wheel response
variables variables vehicle Environment

Wheel and Tyre model

Operational
variables

Figure 2-8: How wheel and tyre models can come into a model context.
2.2.1.5.1 (Wheel and Tyre) Design Parameters
We limit ourselves to today’s traditional pneumatic tyres. Then the design is captured by:
• Carcass/Material: Rubber quality and plies arrangement.
• Tread/Grooves: Groove pattern, Groove depth, Tread depth, Spikes pattern (if spikes)
• Main dimensions: Outer radius, Width, Aspect ratio
• Installation parameters: Inflation pressure
2.2.1.5.2 Operational Parameters
These are operating conditions which vary slowly, and in this description assumed to be constant dur-
ing one manoeuvre/driving cycle. These are:
• Road surface (dry/wet, asphalt/gravel/snow/ice, …)
• Road compliance and damping (hard/soft, …)
• Wear state of tyre
• Age of tyre
• Temperature
2.2.1.5.3 Operational Variables
These are operating conditions which vary quickly, and in this description assumed to be variable in
time during one manoeuvre/driving cycle. These are:
• Tyre velocities ( ) or Tyre slip (𝑠 and 𝑠 (or ) and sign( ))
• Vertical tyre force, ( )
• Wheel torque (𝑇 𝐴 𝑛 𝑇 𝑝 𝑙 𝑛 +𝑇 𝑘 )
• Camber angle
The wheel’s rotational velocity, , can be contained in Wheel and Tyre model instead of inside Model of
rest of the vehicle, which would mean that will not be part of the Operational variables interface.
Tyre forces in road plane ( ) can be given instead of tyre slip. Another alternative is to give the
corresponding actuation, e.g. wheel shaft torque and wheel steer angle relative to wheel course angle.
With those setups, the response variables in 2.2.1.5.4.
Vertical tyre force can be modelled as arbitrarily varying in time or as an offset amplitude for different
frequencies where offset is from a mean value. The mean value would then be a parameter instead of a
variable. The latter alternative can be more efficient if simulating a longer driving cycle, where follow-
ing each road wave would be very computational inefficient.
2.2.1.5.4 Response Variables
Response refers to response to Operational variables changes.
Physical Response Variables
The variables from tyre model to model of rest of the vehicle is essential the forces and moments, see
Figure 2-6:

94
 Vehicle Interactions and Subsystems

• Longitudinal and lateral forces ( )

• Roll moment or Overturning moment (𝑀 )
• Rolling resistance moment (𝑀 𝑙𝑙 )
• Spin moment or Aligning moment (𝑀 )
Other responses are:
• Wear rate [worn rubber mass or tread depth per time unit]
• Loaded radius (𝑅𝑙 )
Behavioural Response Variables
The Responses can be modelled as Behavioural variables:
• Slip characteristics (slip stiffnesses [𝑁/ ], peak friction coefficient, … or other parame-
ters in physical or empirical tyre slip model, as in 2.2.3.4 and 2.2.4.3)
• Rolling resistance coefficient and Loaded radius (𝑅𝑙 )
• Vertical stiffness and damping coefficients
• Wear coefficient, Relaxation lengths or deformation stiffnesses [𝑁/ ], …
One can select to not introduce Behavioural response variables but only use Physical response varia-
bles. If Behavioural response variables are introduced, one should consider them as special from DAE
point of view, e.g. they can be difficult to differentiate. Also, they require an agreed definition to be in-
terpreted, which is the reason to why they are not proposed as interface variables to the vehicle
model.

2.2.1.6 The Wheel as a Transmission

The whole wheel, rim and tyre, can be seen as a transmission from rotational mechanical power to
translational mechanical power. The shaft is on the rotating side; where we find rotational speed, ,
and the torque, 𝑇. The torque 𝑇 is then the sum of torque on the propulsion shaft and torque on the
brake disk or drum. The wheel hub is on the translating side of the transmission; were we find trans-
lating speed, , and force, . See Figure 2-9. Normally, we approximate and denote 𝑛
and 𝑛 .
If we neglect deformation of tyre/ground and sliding, we find the intended function of the wheel:
𝑟 𝑢𝑠 and 𝑇⁄𝑟 𝑢𝑠 . There will be force losses which makes smaller, see rollnig re-
sistance in 2.2.2, and speed losses which makes smaller, see longitudinal force and slip in 2.2.3.
x-direction =
= vehicle longitudinal forward

vehicle, wheel 𝑭𝒉 𝒃𝒛
including torque source, excluding wheel
𝑻
𝝎
𝑭𝒉 𝒃𝒙
𝒗𝒙
𝑻 𝒗𝒙 𝑭𝒉 𝒃𝒙
𝑭𝒉 𝒃𝒛
𝑒
𝑅𝑙
𝑛

𝑛
ground

Figure 2-9: A wheel as a transmission from rotational [ 𝑇] to translational [ ].

The longitudinal force in ground plane , depends on the design of the tyre, the ground surface and
the operational variables, such as velocities [ ] and vertical force . The tyre can be under braking,
free rolling, pure rolling or propulsion/traction, listed in order of increasing and increasing 𝑇.

95
 Vehicle Interactions and Subsystems

2.2.1.7 Tyre Rolling and Radii

Ideal rolling is shown in Figure 2-10 a). This relationship does not hold when the tyre is deflected as in
Figure 2-10 b). Not even the deflected radius can be assumed to be a proportionality constant between
angular and translational velocity, since the tyre contact surface slides, or slips, versus the ground.
Then, an even truer picture of a rolling tyre looks like Figure 2-10 c), where the deformation at the
leading edge also is drawn. This means that is only ≈ 𝑅 ∙ and only for limited slip levels.
a) Ideal rolling: b) With vertical load ≠ 0 c) With vertical load ≠ 0
and torque x ≈ 𝑇 ≈ 0: and (significant) torque 𝑇 > 0 :
𝑭𝒛 𝑭𝒛
w 𝑻
vx vx 𝑭𝒙 vx

w w

𝑅𝑙2
𝑅𝑙
𝑅 inflated unloaded radius 𝑅 free radius when loaded 𝑅2 free radius when loaded and propelled
𝑅 𝑅𝑙 lever when loaded 𝑅𝑙2 lever when loaded and propelled
𝑅 > 𝑅 and 𝑅2 < 𝑅 and
≈ 𝑅 ≈ 𝑅𝑙 ≠ 𝑅2 ≠ 𝑅𝑙2
Figure 2-10: Radius and speed relations of a tyre. 𝑅 and 𝑅𝑙 are not the same across a), b) and c).
There is a relative speed between tyre and the road surface. The ratio between this relative speed and
a reference speed is defined as the “tyre slip”. The reference speed can be the circumferential speed or
the translational speed of the tyre, depending on the application. For a driven wheel, the slip definition
in Eq [2.1] is often used and for braked wheels Eq [2.2] is often used. This is to avoid division by small
numbers in take-off and brake tests, respectively. The physical model in 2.2.3.1 reveals that Eq [2.1] is
the physically most motivated.
𝑅∙ 𝑅∙
𝑠 [2.1] 𝑠 [2.2]
|𝑅 ∙ | | |
With the definition in Equation [2.1] and a model for how tyre longitudinal force varies with 𝑅 and
(Eq [2.16] taken as example) we get Figure 2-11, which shows cuts in Figure 2-31.
+
𝑠 /

-40 40 +
/𝑁

-40 40 +
𝑅 / /𝑠
Figure 2-11: Slip 𝑠 and force as function of 𝑅 for constant . 50 [𝑘𝑁⁄ ] and
5000 [𝑁]. Semi-inverted scales around 𝑅 | | 0 and 𝑠| | . Dashed shows where definition
of 𝑠 is not relevant, since is not dependent of 𝑠 , since < 0.
It is not obvious which 𝑅 to use in Equations [2.1]..[2.2], e.g. 𝑅 𝑅 𝑅2 𝑅𝑙 or 𝑅𝑙2 . However, this com-
pendium recommends the free radius (𝑅 or 𝑅2 ), rather than the loaded radii (𝑅𝑙 or 𝑅𝑙2 ) because the
free radii are better average value of the radius around the tyre and the tyre’s circumference is tangen-
tially stiff, so speed has to be same around the circumference.

96
 Vehicle Interactions and Subsystems

Sometimes one defines the Rolling radius 𝑅 ( ⁄ )| 𝑇= , i.e. a speed ratio with dimension length,
between translational and rotational speeds, measured when the wheel is undriven (𝑇 0). This ra-
dius can be used for relating vehicle longitudinal speed to wheel rotational speed sensors, e.g. for
speedometer or as reference speed for ABS and ESC algorithms.
Yet another approach is to use the radius ( ⁄ )| = , i.e. the ratio when the wheel is pure rolling. Us-
ing 𝑅 ( ⁄ )| 𝑇= or 𝑅 ( ⁄ )| = in the slip definitions shifts (𝑠 ) curve and 𝑇 (𝑠 ) curve, see
more in 2.2.3.5.
The variable 𝑠 is the longitudinal slip value, sometimes also denoted as 𝜅 or 𝜆. When studying brak-
ing, one sometimes uses the opposite sign definition, so that the numerical values of slip becomes posi-
tive.

2.2.1.8 Tyre Contact Length

The contact length 𝐿 represents the deformation shape. The 𝐿 is not a design parameter, but an inter-
mediate parameter or variable, explaining both rolling resistance and tyre forces from slip. One way to
model 𝐿 in a physical way is shown in Figure 2-12. The tyre belt is modelled as circumferentially rigid
(constant length), which gives:
𝜋 𝛾
𝜋 𝑅𝑓 𝜋 𝑅 𝑧 +𝐿
𝜋
2
(𝑅𝑓 Δ ) + (𝐿⁄ )2 𝑅2𝑧 [2.3]
𝐿⁄
tan(𝛾)
𝑅𝑓 Δ
Additionally, we need a model for the vertical deformation. Simplest possible is a linear spring:
Δ [2.4]
tyre.L tyre.Dz
0.20 0.01

𝛾 𝑅𝑧 0.15 0.00

0 4000 8000
0.10
𝑅𝑓 tyre.Fz [N]
𝑅𝑓

0.05

𝑭𝒛 𝐿/ 20
tyre.gamma

𝑅 𝑧
0.00
0
-0.05
0 2000 4000 6000 8000 0 4000 8000
tyre.Fz [N] tyre.Fz [N]

𝑅𝑙 𝑅𝑓 tyre.Fsw z tyre.Fbeltz tyre.L

𝛾 6000 0.2

ground 0.0
level 4000
0 4000 8000
tyre.Fz [N]

𝑭𝒛 2000

𝐿
tyre.Rfree tyre.RFz
0.3004
0
0.3000

0 2000 4000 6000 8000 0 4000 8000

tyre.Fz tyre.Fz [N]

Figure 2-12: Left: Physical model. Right: Example result of model defined by Eqs [2.3] and [2.5].
A typical usage of this model is to know variable and parameters 𝑅𝑓 and . This leaves 4 un-
known variables (Δ 𝛾 𝐿 𝑅 𝑧 ) and equally many equations. An explicit solution is easy for Δ , but diffi-
cult for the remaining 3 variables [𝛾 𝐿 𝑅 𝑧 ]. For a single operation conditions, one can solve by itera-
tion. For whole simulations, it is more computational efficient to pre-process numerically in a table for
how [Δ 𝐿 𝑅 𝑧 ] varies with 𝛾 (for certain 𝑅𝑓 , which typically not varies during a simulation). For
each time instant during simulation, interpolation in this table gives [𝛾 𝐿 𝑅 𝑧 ] from known Δ .
The stiffness can be found for an existing tyre with a certain inflation pressure, but it is far from a
design parameter. Also, the linear model is often too simplified. To improve this, alternative vertical
deformation models can be used. If studying only single operation conditions, Finite Element models
are suitable. But for simulations, a more computational efficient solution is almost a necessity. The

97
 Vehicle Interactions and Subsystems

model in Eq [2.5] can replace Eq [2.4]. This is simple but indicates the mechanisms for how sidewall
stiffness and inflation pressure come into play: Assume that the vertical load is carried by sidewalls
( 𝑤 ) and the belt area between the sidewalls ( 𝑙 ). Assume belt has no bending stiffness and side-
walls follow Hertz contact theory:
𝑤 + 𝑙

𝑙 𝑝 𝑛𝑓𝑙 𝑊 𝐿 (𝑊 tyre width 𝑝 𝑛𝑓𝑙 tyre inflation pressure) [2.5]

2
𝑤 𝑘 𝑤 𝐿
So, Eq [2.5] can be solved for [𝐿 𝑤 𝑙 ]. Then [2.3] can be solved for [𝛾 Δ 𝑅 𝑧 ], using iterations
or preprocessing with swept 𝛾 and interpolation, very similar as described above. The model is plotted
for varying in Figure 2-12.
The 𝑘 𝑤 is a radial stiffness of the sidewall. The set [𝑊 𝑘 𝑤 𝑝 𝑛𝑓𝑙 ] is closer to design parameters than
the parameter [ ].
When driving on soft ground, the deformation of the ground adds variables, beside 𝐿, that influences
rolling resistance and forces in ground plane. Driving on soft ground is not well covered in this com-
pendium, except shortly in 2.2.2.4.4. Present section assumes rigid ground.

2.2.2 Rolling Resistance of Tyres

2.2.2.1 Definition of Rolling Resistance
Rolling resistance on flat and hard ground can be defined in two ways:
𝑦 •𝑣 𝑛𝑜 𝑡𝑒-
{𝑠𝑡𝑒 𝑦}
𝐸𝑛 𝐿 ∫|𝑇 𝜔 ⃗ 𝑦| ∫|𝑇 𝜔 𝑣 |
• Energy definition: 𝑅𝑅 𝐸 {
𝑟 𝑠 𝑝
}
𝑧 |𝐷 𝑛 | ⃗ 𝑦|
𝑧 ∫|𝑣 𝑧 ∫|𝑣 | 𝑠𝑡 𝑡𝑒
𝜔 𝑣 𝑓
𝑇𝜔 𝑣 𝑠 𝑇 𝑇
| | { | 𝜔| } | | { 𝑠 } | 𝐹 |≈
𝑧𝑣 ( ) ( )
𝑛 𝑓 >0 𝑧 𝑧
𝑠
𝑧
𝐶𝐶 𝐹𝑧
𝑧

𝑓 𝑇⁄
≈{ }≈ | | This definition includes both force and speed losses.
≈0 𝑧
𝑇 𝐿 𝑇 𝑇⁄
• Torque (or Force) definition: 𝑅𝑅 sign( ) sign( ) This def-
𝑧 𝑧 𝑧
inition only includes force losses relative to a nominal force 𝑇/𝑅; not velocity or speed
losses.
The 𝑅𝑅 (or 𝑓 ), is Rolling Resistance Coefficient.
Energy definition 𝑅𝑅 𝐸 has the advantage that it does not require the radius 𝑅. But it is not useful in
most vehicle dynamics models, since it does not resolve into force and velocity. And, the 𝑅𝑅 𝐸 gener-
ally varies more than 𝑅𝑅 . So, the compendium uses the force definition. It should be mentioned that
(ISO28580, 2018) uses the Energy definition.
§ The 𝑅𝑅 is dimensionless. With the energy definition, we can understand it as a normalized energy
loss and with force definition we can understand it as a normalized torque loss. In both cases
[𝑁 ⁄(𝑁 )], but [𝑁 ] is energy and torque, respectively.
The speed losses are governed by the following equation, which is explained more in 2.2.3:
𝑛𝑜 𝑡𝑒- 𝑅
• {
𝑟 𝑠 𝑝} 𝑥( 𝑥 ) { 𝑥 ≪ } 𝑥 𝑠𝑥 { ≈ 𝑁𝑜 } 𝑥 𝑠𝑥 |𝑅 |
𝑥

2.2.2.2 Physical Explanation of Rolling Resistance

The rolling resistance is difficult to model physically. In the following, two possible explanations are
given, see Figure 2-13 and Figure 2-14. They focuses tyres on hard ground and moderate speeds, both
positive and negative. On soft ground, there are mechanisms as in 2.2.2.4.4. And, at very high speeds,
there is inertial impact mechanism which causes energy loss. The two explanation models below
mainly takes on the challenge to explain why a rolling resistance moment appears opposite to roll di-
rection, also when speed is close to zero.

98
 Vehicle Interactions and Subsystems

The overall explanation of rolling resistance for pneumatic tyres on hard flat surfaces is that the pres-
sure distribution is offset in rolling direction. In Figure 2-13, radial damping and friction is the cause
for the offset. Figure 2-13 also shows another off-set effect; that the whole contact patch is moved for-
ward due to longitudinal force and shearing of the sidewalls.
The rolling resistance coefficient is almost the same for very low speeds; even when the wheel starts
rolling from zero speed, after gradually increasing the torque up to the rolling resistance moment. The
radial friction in Figure 2-13 can explain that, but there is also another explanation, see Figure 2-14. It
explains an additional reason for why the contact patch is moved ahead of wheel hub. The belt is cir-
cumferentially stiff and takes a short-cut along the chord, through the contact. This builds up shear
stresses in the sidewall, 𝜏 𝑤 . The belt is flexible for bending, so belt radius is proportional to belt ten-
sion force, . This is because same effect as for tension in rope around a cylinder: 𝑓𝑜𝑟 𝑒 𝑝𝑟𝑒𝑠𝑠𝑢𝑟𝑒
𝑟 𝑢𝑠 𝑡ℎ . In our case, the pressure corresponds to the summed effects of inflation pressure 𝑝 𝑛𝑓𝑙
and radial stress in sidewalls 𝜎 𝑤 . So, the radius becomes smaller than original radius in inlet and
larger in outlet. Assumption of constant contact length and geometric constraints from tensional rigid
belt requires that contact patch is offset towards the inlet, so 𝑒 sign( ) |𝑒|;.
Pressure distribution components Resulting pressure and Equivalent forces

𝑇 𝑇

Rl Rl
Pressure from elastic
radial deformation e
damping Pressure from radial
deformation speed
friction
pressure
Contact patch offset, due to shear of
tyre sidewalls (drawn for > 0) equivalent
Figure 2-13: Normal force distribution on a tyre. The measure e is the force offset. In steady state,
the forces in hub and contact patch are the same: and .

99
 Vehicle Interactions and Subsystems

1. wheel 2. wheel, except belt 2. belt

𝑟 𝑢𝑠
w 𝑇𝑛 sidewall
𝑛 𝑙
𝑇 𝑇 𝑇 𝑛 𝑛 𝑙
v 𝑛 𝑛 𝑙 𝑛 𝑛 𝑙

undeformed
Rl deformed 𝑤
𝜏 𝑤

4. wheel 𝑒 𝑒 3. belt in contact

𝜏 𝑤

𝑟 𝑢𝑠 2 > 𝑟 𝑢𝑠 < 𝑤+ 𝑤+
+
𝑤

Figure 2-14: 1. Wheel at low speed ( small and negative). 2. Free body diagrams with belt
separate. 3. How belt tension force changes before and after contact. 4. Influence of radii variation
on contact patch position.

2.2.2.3 Mathematical Representation of Rolling Resistance

The rolling resistance force is defined as the loss of longitudinal force on the vehicle body, as com-
pared to the longitudinal force which would have been transferred with an ideal wheel. Assuming
force equilibria in longitudinal and vertical direction, 𝑛 and 𝑛
, see Figure 2-9.
𝑇⁄𝑅
𝑅𝑅 | | [2.6]

denotes the longitudinal force on the wheel, 𝑇 denotes the applied torque and 𝑅 denotes the tyre
radius. For a free rolling tyre, where 𝑇 0, 𝑓 becomes simply ⁄ . One often sees definitions of
𝑅𝑅 which assumes free rolling tyre; but Eq [2.6] is also valid when 𝑇 ≠ 0, which is useful.
A free body diagram of the forces on the wheel can be used to explain the rolling resistance. Consider
Figure 2-13 which represents a free rolling wheel under steady state conditions. The inertia of the
wheel is neglected.
Longitudinal and vertical force equilibria are already satisfied, due to assumptions above. However,
moment equilibrium around wheel hub requires:
𝑇 𝑒
𝑇 ∙ 𝑅𝑙 ∙𝑒 0 ⇒ [2.7]
𝑅𝑙 𝑅𝑙
This result suggests that the force , which pushes the vehicle body forward, is the term 𝑇⁄𝑅𝑙 (arising
from the applied torque T) minus the term ∙ 𝑒⁄𝑅𝑙 . The term can be seen as a force 𝑙𝑙 and referred
to as the rolling resistance force. We seldom know neither 𝑅𝑙 or 𝑒, but they are rather constant and the
form of Eq [2.7] is same as Eq [2.6] if:
𝑅𝑅 |𝑒⁄𝑅| [2.8]
Eq [2.8] is a definition of rolling resistance coefficient based on physical mechanisms internally in the
tyre with road contact, while Eq [2.6] is based on quantities which are measurable externally, see Eq
[2.6]. Sometimes one sees 𝑅𝑅 ⁄ as a definition, but that is not suitable since it assumes ab-
sence of torque.

100
 Vehicle Interactions and Subsystems

It is not obvious which radius to use. 𝑅𝑙 is can be motivated because 𝑅𝑙 is the lever for . However, in
2.2.1.7, we found arguments for using other radii. If same radius is used, for slip and rolling resistance,
we can fully see the tyre as a transmission with the nominal ratio 𝑅 and zero energy loss if 𝑅𝑅 0
and 𝑠 0.
It is important to refer to this phenomenon as rolling resistance as opposed to rolling friction. It is
not friction in the basic sense of friction, because ≠ 𝑅𝑅 ∙ except for the special case when un-
driven wheel (𝑇 0). Figure 2-15 shows an un-driven and pure rolling.
Rolling resistance is a torque loss. Other torque losses, which can be included or not in tyre rolling re-
sistance, are:
• losses associated with friction in gear meshes,
• drag losses from oil in the transmission,
• wheel bearing (and bearing sealings) torque losses,
• drag from brake discs,
• drag losses from aerodynamic around the wheel, and
• uneven road in combination with suspension damping that dissipates energy.
These should, as rolling resistance, be subtracted from propulsion/brake torque. However, sometimes
they are included as part of the tyres rolling resistance coefficient, which can be misleading. The wheel
bearing torque loss have two torque terms: one is proportional to vertical load on the wheel (adds typ-
ically 0.000 to rolling resistance coefficient), and the other is of constant magnitude but counter-di-
rected to rotation speed. The former term can be included in rolling resistance coefficient. The aerody-
namic losses due to wheel rotation are special since they vary with wheel rotational speed, meaning
that they (for constant vertical load) are relevant to include when studying the variation of rolling re-
sistance coefficient with vehicle speed. A summarizing comment is that one has to be careful with
where to include different torque losses, so that they are included once and only once.

2.2.2.4 Variation of Rolling Resistance

Several parameters will affect the rolling resistance moment (or Rolling resistance coefficient). Design
parameters, see Figure 2-8:
• Tyre material. Natural rubber often gives lower rolling resistance.
• Radial tyres have more flexible sides, giving lower rolling resistance also bias ply have a
greater crown angle causing more internal friction within the tyre during deflection.
• Geometry:
o Diameter. Large wheels often have lower coefficient of rolling resistance
o Width
o Groove depth
o Tread depth
• Higher inflation pressure gives lower rolling resistance on hard ground but higher rolling re-
sistance on soft ground (and vice versa), see Figure 2-16 and Figure 2-17.

101
 Vehicle Interactions and Subsystems

General case: Special case: Special case:

Driven w 𝑇 Undriven or Pure-rolling
any 𝑇 Free-rolling 0
𝑇 0 𝑇 𝑒
𝑅𝑛 𝑚 ; | |
𝑜 | | 𝑇

e e
e

(k is stiffness for how much

contact patch moves
longitudinally due to )

Figure 2-15: Driven wheel with rolling resistance. Special cases “Free-rolling” and “Pure rolling”.
Operational parameters, see Figure 2-8:
• Elevated temperatures give low rolling resistance, via increased inflation pressure. Tyres need
to roll approximately 30 km before the rolling resistance drop to their lowest values.
• Road/ground, sometimes covered: Clean asphalt, Asphalt with water/leaves/sand/…,
Loose/hard gravel, Snow/ice. Soft ground or covered hard ground increases rolling resistance.
• Wear state. Worn tyres have lower rolling resistance than new ones (less rubber to deform).
Operational variables, see Figure 2-8:
• Vertical force.
• Speed. Rolling resistance increases with vehicle speed due to rubber hysteresis and air drag.
• Tyre loads (propulsion/braking and lateral forces)
Approximately for modern tyres (written year 2020): Passenger car tyres have 𝑅𝑅 0.005. .0.0 ,
Truck tyres have 𝑅𝑅 0.00 . .0.005. The lower values are “eco-tyres” and can have some trade-off
with vertical comfort and road grip.
2.2.2.4.1 Variation of Tyre Type
Trucks tyres have a much lower rolling resistance coefficient than passenger vehicle tyres, approxi-
mately half. Tyres have developed in that way for trucks, because their fuel economy is so critical.
2.2.2.4.2 Variation of Vertical Force
In a first approximation, the rolling resistance force is proportional to vertical force, i.e. RRC is con-
stant. But, typically, the RRC decreases slightly with vertical force. This, and parasitic bearing losses,
𝑡𝑜𝑟𝑞𝑢𝑒 sign( ) 𝑜𝑛𝑠𝑡 𝑛𝑡, explains why commercial vehicles lift axles when driven with low pay-
load.
2.2.2.4.3 Variation of Speed
As an example, left part of Figure 2-16 shows the influence of speed and tyre construction on rolling
resistance. The sudden increase in rolling resistance at high speed is important to note since this can
lead to catastrophic failure in tyres. The source of this increase in rolling resistance is a high energy
standing wave that forms at the trailing edge of the tyre/road contact.

102
 Vehicle Interactions and Subsystems

There are some empirical relationships derived for the tyre's rolling resistance. It is advisable to refer
to the tyre manufacturer's technical specifications when exact information is required. This type of in-
formation is usually very confidential and not readily available. Some general relationships have been
developed, from (Wong, 2001):
Radial-ply passenger car tyres: 𝑅𝑅 0.0 6 + 0.04 ∙ 0 6 ∙ 2
Bias-ply passenger car tyres: 𝑅𝑅 0.0 69 + 0. 9 ∙ 0 6 ∙ 2
Radial-ply truck tyres: 𝑅𝑅 0.006 + 0. ∙ 0 6 ∙ 2
Bias-ply truck tyres: 𝑅𝑅 0.007 + 0.45 ∙ 0 6 ∙ 2
As seen in Figure 2-16, a rule of thumb is that rolling resistance coefficient is constant up to around
100 km/h.
.05
0.4

Coefficient of rolling resistance, f

Coeff. of Rolling Resistance

.04
Sand

.03
0.2

.02 Bias Ply

Bias-Belted Mediumhard
Medium HardsoilSoil
Radial
Concrete
.01 0
0 25 50 75 100 125 150 175 0 20 40
Speed [km/h] Inflation Pressure [psi]
Figure 2-16 : Left: Rolling resistance coefficient variation with speed for different tyre types
(Gillespie, 1992). Right: Range of Coefficient of Rolling Resistance for different road/ground
material.
2.2.2.4.4 Variation of Road Surface
The road or ground can mainly vary in two ways: how rigid and how slippery it is. It is not solely a
question of the road, because such rigidness should be judge relative to tyre inflation pressure and
such slipperiness should be judged relative to tyre tread pattern, including whether the tyre has spikes
or not.
Right part of Figure 2-16 shows that the rolling resistance changes a lot due to different road/ground
material and inflation pressures. As can be expected, a range of values exist depending on the specific
tyre and surface materials investigated. On hard ground, the rolling resistance decreases with in-
creased inflation pressure, which is in-line with the explanation model used above, since higher pres-
sure intuitively reduces the contact surface and hence reduces e. On soft ground the situation is re-
versed, which requires a slightly different explanation model, see Figure 2-17. On soft ground, the
ground is deformed so that the wheel rolls in a “wheel-local up-grade” with inclination angle . Intui-
tively, a higher inflation pressure will lead to more deformation of the ground, leading to a steeper .
The phenomena in Figure 2-17 is a plastic deformation of the ground, so if a vehicle has several tyres
which takes the same path on ground, the rolling resistance will be different due to a memory (a state
variable) in the ground.
The “wheel-local up-grade” in Figure 2-17 follows each tyre so that it is constant in time for each tyre.
This is a difference to simply uneven (rigid) road, where the wheels experience alternating wheel-local
up- and ground-grade. In average, the alternating up and down does not generate rolling resistance in
tyre, but it can cause energy-loss in wheel suspension and consequently this contributes to energy
consumption for the vehicle.

103
 Vehicle Interactions and Subsystems

Wheel rolling on hard ground: Wheel rolling on soft ground:

𝑇 𝑣 𝑇 𝑣 𝑇
horizontal=
=vehicle-
𝑣 𝑣
longitudinal
𝑣

𝑒
𝑣

Figure 2-17 : Rolling resistance explanation for hard and soft ground. Subscripts: c=contact,
v=vehicle.
2.2.2.4.5 Variation of Longitudinal Force, Propulsion and Braking
Figure 2-13 shows an offset of contact patch due to longitudinal force . This makes 𝑅𝑅 dependent
of . The phenomena is not well studied, since 𝑅𝑅 is often given for free-rolling tyres, but according
to Figure 2-18 RRC increases with positive and decreases with negative . The wheel radius also
decreases with magnitude of force. For negative forces, these two effects have opposite influence, so
the change is less for negative force, as seen in Figure 2-18.
Rolling resistance coefficient,
𝑇⁄
𝑅𝑅
𝑧
0.08

Normalised
0.04
Longitudinal
Force,
𝑧
-0.4 -0.2 0 +0.2 +0.4
Figure 2-18 : Rolling resistance dependency of longitudinal tyre force. Inspired by (Wong, 2001)
2.2.2.4.6 § Separation of Actuated Wheel Torque and Rolling Resistance
Contribution from Toheed Ghandriz, Vehicle Dynamics

The force play in FBD numbered 3 in figure below is primarily proposed for understanding. But there
are sometimes reasons to use FBD numbered 4, where the torques 𝑇 and 𝑛 𝑒 from Figure 2-9
are treated separately and converted to one force each. Such reason is that 𝑇 𝑇𝐴 is possible to con-
trol (actuatable) while 𝑛 𝑒 𝑇 is not.
1 𝑇𝐴 2 3 4
𝑇 𝑝 +𝑇 𝑘 𝑇𝐴 𝑇𝐴 𝑇 𝑇𝐴

𝑇 𝑅
𝑅𝑅 𝑅 𝑇 𝑅 𝑅
𝑒 𝑅𝑅 𝑅 𝐴 𝑇𝐴 /𝑅

Figure 2-19: § Force equivalent FBDs with both rolling resistance (RR) and actuation (Act). Wheel
is modelled inertia free in both translation and rotation. 𝐴 and 𝑅 is a virtual decomposition
of the real force .
The decomposition of in 𝐴 and 𝑅 means a deviations from the advice in 1.5.2.1 to not draw
two forces acting at the same point of part and having same direction. Then note:

104
 Vehicle Interactions and Subsystems

• The tyre models where force depends on slip (2.2.3..2.2.6) applies to , not to 𝐴 .
• The 𝑅 is not a separate physical force acting on the vehicle. It is 𝐴 𝑅 that
physically acts on the vehicle and should appear in vehicle equilibrium.
So, 𝑅 is very different from aero dynamic resistance force and grade resistance forces, which are
physical forces acting on the vehicle.

2.2.2.5 § Energy Losses on Uneven or Deformable Ground

Previous treats flat and rigid ground, except that uneven or deformable ground is briefly mentioned in
2.2.2.4.4. If ground is rigid but uneven, there will be “local up- and down-hills”. Such will also appear if
the ground is deformable, but they will be more of “local uphills which follows each wheel”. If ground
is uneven, it will additionally cause energy losses in the suspension damper. Uneven ground, if wheel
takes longitudinal or lateral forces, also increase the slip losses due to large variations in normal force.
Figure below introduces the directions global (e.g. and 𝑣 ), vehicle or road (e.g. 𝑣 ) and contact
e.g. ). The rolling resistance coefficients in this section will be “effective rolling resistance coeffi-
cients”, which means that they are different, typically larger, than the contact-local coefficient 𝑅𝑅
from the force offset 𝑒.

+ 𝑙

in vehicle (or in contact

road) directions directions

global horizontal

Figure 2-20: § Explanation of rolling resistance as wheel rolling in a “local uphill”. The “global
uphill” or “road” is marked to distinguish between and .
2.2.2.5.1 Uneven and Rigid Ground
Force Losses
Moment equilibrium of the wheel gives:
0 𝑇 𝑅 𝑒
If we model the “local”, or contact-local, rolling resistance as on rigid ground, 𝑒 𝑅𝑅 𝑅, we get:
𝑓𝑜𝑟 𝑒
0 𝑇 𝑅 𝑅𝑅 𝑅 ⇒{ }⇒
𝑒𝑞𝑢 𝑒𝑛 𝑒
⇒ 0 𝑇 ⁄𝑅 ( 𝑣 cos( ) + 𝑣 sin( )) ( 𝑣 cos( ) 𝑣 sin( )) 𝑅𝑅 ⇒
⇒ 0 𝑇 ⁄𝑅 + 𝑣 ( cos( ) + sin( ) 𝑅𝑅 ) + 𝑣 ( sin( ) cos( ) 𝑅𝑅 ) ⇒
𝑇𝑖 ⁄ 𝑖𝑧𝑣 ( in( 𝜑𝑖 𝑦𝑟 )+co ( 𝜑𝑖 𝑦𝑟 ) ) 𝑠 𝑇𝑖 ⁄ 𝑖𝑧𝑣 ( 𝜑𝑖 𝑦𝑟 + )
⇒ 𝑣 co ( 𝜑𝑖 𝑦𝑟 ) in( 𝜑𝑖 𝑦𝑟 )
≈ {| |} ≈ +𝜑𝑖 𝑦𝑟
≈
𝑠 || 𝑛
{ } ≈ 𝑇 ⁄𝑅 𝑣 ( + 𝑅𝑅 )
𝑠 |𝑅𝑅 |
One can define a rolling resistance in the vehicle direction, 𝑅𝑅 𝑣:

105
 Vehicle Interactions and Subsystems

𝐿 𝑇𝑖 ⁄ 𝑖 𝑣
𝑇𝑖 ⁄ (𝑇𝑖 ⁄ 𝑖𝑧𝑣 ( 𝜑𝑖 𝑦𝑟 + ))
𝑅𝑅 𝑣 ≈ ≈ 𝑅𝑅 +( )
𝑉 𝑙 𝑖𝑧𝑣 𝑖𝑧𝑣

Note that negative is a “local downhill”, so 𝑅𝑅 𝑣 is the sum of two positive values if “local uphill”,
which should be intuitive correct. If the ground is uneven and rigid, there are equally much local
downhill as local uphill on average; also if the road has a “global grade”. Then, on average, there is no
influence of , so 𝑅𝑅 𝑣 ≈ 𝑅𝑅 .
Speed Losses
It should be intuitive to use the longitudinal slip model, see 2.2.3, in the contact directions:
𝑠 𝑠 . Inserting the slip definition 𝑠 (𝑅 )⁄|𝑅 | gives:
𝑅 𝑓𝑜𝑟 𝑒
⇒{ }⇒
|𝑅 | 𝑒𝑞𝑢 𝑒𝑛 𝑒
⇒( 𝑣 cos( )+ 𝑣 sin( ))
𝑅 ( cos( ))
( 𝑣 cos( ) 𝑣 sin( )) ⇒
|𝑅 |
𝑠 𝑅 ( )
⇒ {| |} ⇒ ( 𝑣 + 𝑣 ( )) ≈ ( 𝑣 𝑣 ( )) ⇒
|𝑅 |
⇒ 𝑣 𝑣 ≈ ( 𝑣+ 𝑣 ) 𝑠 ⇒
⇒ 𝑣 𝑣 𝑠 ≈ 𝑣 𝑠 + 𝑣 ⇒
𝑠 + 𝑠 | 𝑠 |
⇒ 𝑣 ≈ 𝑣 { } 𝑣 ( 𝑠 + )
𝑠 𝑛 𝑠 | |
Note that negative is a “local downhill”, so helps in creating 𝑣 which reduces the re-
quired 𝑣 and, hence the required 𝑇 . If the ground is uneven and rigid, there are equally much local
downhill as local uphill on average; also if the road has a “global grade”. Then, on average, there is no
influence of , so 𝑣 ≈ 𝑣 𝑠 .
The same modelling concept can be used with saturated sign(𝑠 ) min( |𝑠 | ) and
then one can study traction problems due to uneven ground; i.e. when the vehicle gets stuck.
2.2.2.5.2 Losses in Suspension Damping
Above concluded that uneven and rigid does not cause energy losses on average. However, unevenness
can still cause energy loss due to increased motion in vehicle suspension dampers.
2.2.2.5.3 Deformable Ground
If the ground is deformable (soft) it deforms due to the force from the tyre. The ground is generally
more plastic than elastic, so a simplification is that deformation is completely plastic.

2.2.3 Longitudinal Force

The longitudinal force, , between the tyre and ground influences the vehicle propulsion and braking
performance. We can see it as a depending on the sliding between the tyre and the ground.
First, compare friction characteristics for a translating block of rubber with a rolling wheel of rubber.
Figure 2-21 shows the basic differences between classical dry-friction, or Coulomb friction, of such
sliding block and the basic performance of a rolling tyre. Experiments have shown that the relative
speed between the tyre and the road produces a frictional force that has an initial linear region that
builds to a peak value. After this peak is achieved, no further increase in the tangential (friction) force
is possible. There is not always a peak value, which is shown by the dashed curve in the figure. The
slope in the right diagram will be explained in 2.2.3.1, using the so called “tyre brush model”.

2.2.3.1 Brush Model for Longitudinal Slip

The brush model is frequently used to explain how tyre develop forces in ground plane, see e.g. Refs
(Pacejka, 2005) and (Svendenius, 2007). The brush model is a physically based model that uses shear
stress and dry friction at a local level, i.e. for each contact point in the contact patch. Figure 2-22 shows
the starting point for understanding the brush model.

106
 Vehicle Interactions and Subsystems

Sliding block (or wheel with ω=0) Rolling wheel of rubber

𝑁
𝑁
T
𝑅

𝑁 𝑁
𝑁 𝑁 (𝑁 )

𝑁 𝑝 𝑘 𝑁
slide 𝑁
𝑛 𝑁 𝑛
stick

𝑠 𝑝 ⁄ 𝑓
𝑅 ⁄ 𝑓

slide

• A stick mode where Δ 0. • No stick mode (if rolling, i.e. > 0).
• In slide mode, the friction force, , • The friction force, , is depending on
is only depending on sign Δ . relative sliding speed, ⁄ 𝑓 .

Figure 2-21: Friction characteristics.

w*R

vx

bristle or block

Figure 2-22: Tyre ground contact for braked tyre. Origin to the “Brush model”. Picture from
Michelin.
2.2.3.1.1 Uniform Pressure Distribution and Known Contact Length
A first simple variant of the brush model, uses the following assumptions:
o Sliding and shear stress only in longitudinal direction (as opposed to combined longitudinal
and lateral)
o Uniform and known pressure distribution over a constant and known contact length (as op-
posed to using a contact mechanics-based approach, which can calculate pressure distribution
and contact length 𝐿.)

107
 Vehicle Interactions and Subsystems

o Contact length is known (as opposed to function of vertical force).

o No difference between static and dynamic coefficient of friction
o Only studying the steady state conditions, as opposed to including the transition between op-
erating conditions. Here, “steady state” refers to the distributions of tyre tread shear defor-
mation 𝛾(𝜉 ) and shear stress 𝜏(𝜉 ), which are assumed to not vary with time.
At first, we also assume that the tread is built up of 𝑁 bristles of rubber. Each bristle is assumed as a
sheared piece of rubber without force interaction between neighbouring bristles. If there is no sliding
between rubber and ground, the shear stress of the element develops as in Hooke’s law: 𝜏
𝐴𝑟𝑒 𝛾 𝐴𝑟𝑒 . The 𝐴𝑟𝑒 𝑊 𝐿⁄𝑁, but we let 𝑁 → , we can use a continuous model:
𝜏 𝐴𝑟𝑒 𝜏 𝑊 𝜉 , where 𝑊 is width of tyre and 𝜉 a coordinate along the belt in the contact, see
Figure 2-23. Figure 2-23 also shows that we assume friction contact between bristle and ground. (The
bristles can alternatively be seen as bending beams fixed in the belt end or rigid beams connect with
rotational springs in the belt end.)
rolling direction
trailing end leading edge
of contact L=length of contact patch of contact View from rear,
(=rubber (=rubber at one certain 𝜉 :
View from outlet) tyre inlet)
side: 𝜉 𝐿 𝜉 𝜉 𝑅 𝜉 0
W=width
𝜉
H=height
𝜉
road
(relative to wheel hub)
𝛾
bristle
Velocities of bristle
increasing deformation ends in stick zone:
Contact patch,
𝑅
view from above:
same deformation Drawn for 𝑅 >

𝜉
pressure 𝑅
stick zone stick
𝜉
slip
slip zone
“mode”
𝜉
shear stress 𝜏
(or shear deformation angle, 𝛾)

Figure 2-23: Physical model for simple brush model for longitudinal slip. Drawn for propelled tyre.
The bristles represent the rubber tread, not including elasticity in sidewall. Drawn for 𝑅 ∙ > .
If we assume no sliding:
𝜏 ∙𝛾 ℎ𝑒𝑟𝑒 𝜏 𝑠ℎ𝑒 𝑟 𝑠𝑡𝑟𝑒𝑠𝑠 𝑠ℎ𝑒 𝑟 𝑜 𝑢 𝑢𝑠 𝛾 𝑠ℎ𝑒 𝑟 𝑒𝑓𝑜𝑟 𝑡 𝑜𝑛 𝑛 𝑒
When a bristle enters the contact patch, it lands un-deformed, i.e. with 𝛾 0. The further into contact,
along coordinate 𝜉, we follow the element, the more sheared will it become. Since the ground end of
the element sticks to ground, the increase becomes proportional to the speed difference and the
transport time 𝑡 𝑛 𝑝 , which is the time it takes for a bristle to reach the coordinate 𝜉:
= 𝑟 𝑝
∫= ∙ 𝑡 Δ 𝑅∙ 𝑟 𝑝 𝜉 𝑛 𝑝 𝑡 𝑛𝑝
𝛾 { } ∫ 𝑡 { }
𝐻 𝑅∙ 𝐻 = 𝑛 𝑝 |𝑅 ∙ |
𝜉 ⁄| 𝜔| 𝜉 ⁄| 𝜔|
𝑅∙ 𝑅∙ 𝑅∙ 𝜉 𝑅∙
∫ 𝑡 ∫ 𝑡 ∙ ∙𝜉
𝐻 = 𝐻 = 𝐻 |𝑅 | 𝐻 ⏟|𝑅 ∙ |

The force is the and integral of 𝜏 from 𝜉 0 to 𝜉 𝐿:

108
 Vehicle Interactions and Subsystems

𝐿 𝐿 𝐿
𝑊
∫𝜏∙ 𝐴 𝑊∙ ∫ 𝛾∙ 𝜉 𝑊∙G ∫ ∙𝑠 ∙𝜉∙ 𝜉 ∙ 𝑠 ∙ ∫𝜉 ∙ 𝜉
𝐻 𝐻
𝑊𝐿 ξ= ξ=
𝑊 𝐿2
∙𝑠 ∙𝑠 (𝑁𝑜𝑡𝑒: 𝑇ℎ 𝑠 𝑠𝑠𝑢 𝑒𝑠 𝒏𝒐 𝒍𝒊𝒅𝒊𝒏𝒈 𝑛 𝑜𝑛𝑡 𝑡. )
⏟ 𝐻

Note that this model simplifies from a function of 2 variables ( and ) to 1 variable (𝑠 ). With this
definition of slip, 𝑠 , we automatically handle and both positive and and both negative. The
case when and have different signs ( < 0) is special and will be handled below.
(It is possible to reach Eq [2.9] via integration over 𝜉 instead. Then, one uses 𝑛 𝑝 | |. After a
more complex derivation, the resulting slip definition 𝑠 (𝑅 ∙ )⁄|𝑅 ∙ | falls out in the final ex-
pression for , i.e. Eq [2.9]. Despite this, one often sees 𝑠 (𝑅 ∙ )⁄| |.)
As long as friction limit is not reached (|𝜏| < 𝑝) within the whole contact, this model is valid. The
rsik for sliding is highest at 𝜉 𝐿 so the validity can be formulated |𝜏(𝐿)| < 𝑝 𝑊𝐿
𝑧
⇒ 𝐻∙
|𝑠 | ∙ 𝐿 < 𝑧
⇒ |𝑠 | < 𝑧
⇔ | |< 𝑧
.
𝑊𝐿 2 2
So, if that condition is not fulfilled, the friction limit is reached within the contact, i.e. at a break-away
point at 𝜉 𝜉 < 𝐿. Then, we have to split the integral in two. The point 𝜉 is defined by 𝜏(𝜉 ) 𝐻 𝑠
2 𝑝 𝑊 𝐿2 𝐿
𝜉 𝑊 𝐿2
𝑠 𝜉 𝑝 ⇒ 𝜉 2 2
𝑧
. For 𝜉 > 𝜉 , the rubber element will slide with a
constant 𝜏 ∙ 𝑝.
𝐿 𝜉 𝐿

𝑊 ∙∫𝜏 ∙ 𝜉 𝑊∙∫ ∙𝑠 ∙𝜉∙ 𝜉+𝑊∙ ∫ ∙𝑝∙ 𝜉

𝑊 𝐿2
𝜉
2
𝜉 𝐿
∙𝑠 ∙ + ∙ 𝑝 ∙ 𝑊 ∙ (𝐿 𝜉 ) {𝑝 𝑛 𝜉 }
𝐿2 𝑊 𝐿 𝑠
2 2
+ ∙ ∙( ) ∙ ∙( ) 𝑓𝑜𝑟 𝑠 >
4
⏟ 𝑠 ⏟ 𝑠 4 𝑠
𝑖 𝑘 𝑖𝑝

The force terms, 𝑘 and 𝑙 𝑝 , from each of stick and slip regions are identified. These two terms
are separately shown plotted in Figure 2-25.
The case when and have different signs leads to that 𝜉 0 , since the bristles will deform in the
opposite direction; slip over the whole contact or macro slip. So, the whole contact has 𝜏 ∙ 𝑝, which
leads to . Hence, the total expression for becomes as in Eq [2.9]. We also add subscript 𝑥
on and 𝐻, to prepare for a corresponding model for lateral forces, in 2.2.4.
sign(𝑅 )∙ ∙ sign(𝑠 ) ∙ ∙ 𝑓 <0
∙ ∙
∙𝑠 𝑒 𝑠𝑒 𝑓 |𝑠 | ≤ ⇔| |≤
∙
∙ [2.9]
sign(𝑠 ) ∙ ∙ ∙( ∙ ) 𝑒 𝑠𝑒
{ 4∙ |𝑠 |
∙ 𝑊 ∙ 𝐿2 𝑅
ℎ𝑒𝑟𝑒 𝑒𝑛𝑠 𝑜𝑛: [𝑓𝑜𝑟 𝑒] 𝑛 𝑠
∙𝐻 |𝑅 |
Eq [2.9] is plotted in Figure 2-25. Eq [2.9] is also called Dugoff tyre model. It exists also as lateral slip
and combined slip tyre model, see (Dugoff, Fancher, & Segel, 1969).
It is important to reflect over which of the physical quantities that reasonably has to be modelled as
varying. This will of course depend on the driving manoeuvre studied, but here is a typical situation:
The slip 𝑠 and normal load are typical varying and defined by the vehicle model. The quantities

109
 Vehicle Interactions and Subsystems

𝑊 𝐻 are often reasonably constant, so they can be parameters. However, it is often not reason-
able to assume that the contact length 𝐿 is constant. 𝐿 is rather a function of : 𝐿( ) .
The case when < 0 is unusual and can only occur when |𝑠 | > . An example is when vehicle
moves rearward with /𝑠, would be that the wheel spins forward, e.g. with 𝑅 /𝑠.
Then 𝑠 + and ∙ . Also, if increase to 𝑅 + /𝑠, we get same ∙ , but
with 𝑠 + . One can also note that there is another 𝑅 , for same , which gives 𝑠 + . This is
𝑅 ⁄ /𝑠 and then ∙ ∙( ∙ ⁄(8 ∙ )), which is < . So, is uniquely de-
fined for any ( ), but has double solutions for some 𝑠 , when |𝑠 | > .
If we instead hold a certain forward vehicle speed, e.g. /𝑠, and study how varies with , we
can identify 4 specific levels of :
• Full rearward traction: 𝑅 ⁄𝑠. This gives 𝑠 , with different signs on and ,
and
• Locked wheel: 𝑅 0 ⁄𝑠. This also gives 𝑠 and
• Pure rolling wheel: 𝑅 ⁄𝑠. This gives 𝑠 0 and 0;
• Full forward traction: 𝑅 + ⁄𝑠. This gives 𝑠 + and ∙ ∙(
∙ ⁄(4 ∙ )) ≈ {𝑡𝑦𝑝 𝑦} ≈ (0.95. .0.98) ∙ , which is <
The first case is achievable, with an electric motor braking but not with friction brakes, where and
have different signs. The last case shows that we cannot reach in the direction the vehicle
moves, since there will always be a small part on the inlet side of the contact where the shear stress
has not reached 𝑝. In practice, we can see it as ≈ , but when using the model mathematically, it
can be good to note such small phenomena.
2.2.3.1.2 Longitudinal Tyre Slip Stiffness
In summary for many models, the following is a good approximation (compared to tests) for small lon-
gitudinal slip, and certain normal load and certain friction coefficient:
∙𝑠 [2.10]
For the brush model, or any other model which describes (𝑠 … ), one can define the “Longitu-
dinal tyre (slip) stiffness” , which have the unit 𝑁 𝑁⁄ 𝑁/ (( /𝑠)⁄( /𝑠)). It is the derivative of
force with respect to slip when at 𝑠 𝑠 0:

𝜕
(𝜕 (𝑠 … ))| ( …) [2.11]
= 𝑦=

Note that is not a stiffness in the conventional sense, force/deformation. The tyre also has such a
deformation stiffness. Often, it is obvious which stiffness is relevant, but to be unambiguous one can
use the wording: “slip stiffness” [N/1=N/((m/s)/(m/s))] and “deformation stiffness” [N/m].
With the brush models with both pressure distributions, Eq [2.9] and Eq [2.16], we get the
∙ 𝑊 ∙ 𝐿2 ⁄( 𝐻). With 0.5 [𝑀 𝑁⁄ 2 ] (typical shear modulus in rubber), 𝑊 𝐿 0. . .0. [ ]
(typical sizes of contact patch for passenger car) and 𝐻 0.0 [ ] (approximate tyre tread depth) one
gets ≈ 5[𝑘𝑁] < ∙ 𝑊 ∙ 𝐿2 ⁄( 𝐻) <≈ 40[𝑘𝑁/ ]. Empirically, we can measure for passenger
car tyres around 5. .50 [𝑘𝑁/ ]. This indicates that the brush model models the essential physical phe-
nomena and that the sheared part (the bristles) is rather only the tread than the whole elastic part
sidewall and tread together.
2.2.3.1.3 Influence of Vertical Load and Friction in Brush Model
The vertical load on the tyre affects the force generation, . For large slip our dry friction model moti-
vates that is proportional to . This is also a good approximation for small slip, via . Then,
one can define the Longitudinal Slip Stiffness Coefficient, :
𝑠 𝑠 [2.12]

110
 Vehicle Interactions and Subsystems

The contact length 𝐿 will reasonably vary according to some deformation model. Hertz’s contact the-
ory for line contact motivates that 𝐿 2√ . This is implemented as 𝐿2 /𝑘, where 𝑘 is a material
modulus with dimension force/area, gives:
∙ 𝑊 ∙ 𝐿2 ∙ 𝑊 ∙ /𝑘 ∙𝑊
( )
∙𝐻 ∙𝐻 ⏟∙ 𝐻 𝑘

So, the assumption 𝐿 2√ explains why ≈ . This can be verified with experiments,
see Figure 2-55. A small tendency for degressive increase (𝜕 2 ⁄𝜕 2 < 0) can be found. A definition of
from a general ( 𝑠 ) will include , which is the vertical force around which the model
should be valid:

( ( ))| ( ) [2.13]
𝑧 = 𝑧0

The result is summarised in Figure 2-24. Increasing only increases the saturation level, while leaving
the slope at lower 𝑠 constant, or possibly slightly increased. Variation of involves the contact length
model. The assumption “𝐿 2√ ” simply scales the curve in force direction. Overall, it can be con-
cluded that there are arguments for two conceptual ways how the tyre characteristics changes, varied
vertical force and varied friction coefficient, see Figure 2-24.

Air
Asphalt temperature:
-15 to -10 C
=

⁄
Normalized force,

Snow
=

Reference 𝑠
Doubled vertical force (simplified) Ice
Doubled vertical force (tendency)
Doubled friction coefficient (simplified)
Doubled friction coefficient (tendency)
slip, 𝑠

Figure 2-24: How tyre characteristics typically vary due to varying vertical force and road friction.
Left: Theory. Right: Measurements with varying friction, i.e. varying surfaces, from PhD course by
Ari Tuononen, Aalto university, Finland, Saariselkä, Finland 2014-03-15..22.
The influence of vertical force on (and on lateral slip stiffness ) is further discussed in 2.2.4.3 and
2.2.5.4.
2.2.3.1.4 Influence of Different Static and Dynamic Friction
A common model for friction is that coefficient of friction to remain sticking, 𝑘 , is higher than the
coefficient of friction when slipping has started, 𝑙 𝑝 . This is sometimes called “stiction”, 𝑘 𝑛
𝑘 ⁄ 𝑙 𝑝 . If this is implemented in the model, the brush model can explain why the overall (𝑠 )
often has a peak, as indicated already in Figure 2-21. In the derivation in Eq Error! Reference source
𝐿
not found., the differing between 𝑘 and 𝑙 𝑝 , affects like this: 𝑙𝑝 𝑊 ∙ ∫𝜉 𝑙 𝑝 ∙ 𝑝 ∙ 𝜉 and
𝐿
𝜉 2
𝑖 𝑘 𝑧
.

111
 Vehicle Interactions and Subsystems

sign(𝑠 ) ∙ 𝑙𝑝 ∙ 𝑓 <0 ⇔ < |𝑠 | <

𝑘 ∙ 𝑘 ∙
∙𝑠 𝑒 𝑠𝑒 𝑓 |𝑠 | ≤ ⇔| |≤
∙
𝑘 [2.14]
sign(𝑠 ) ∙ 𝑙𝑝 ∙ ( 𝑘 ( )) 𝑒 𝑠𝑒
{ |𝑠 | 𝑙𝑝
2
∙𝑊∙𝐿 𝑅
ℎ𝑒𝑟𝑒 𝑒𝑛𝑠 𝑜𝑛: [𝑓𝑜𝑟 𝑒] 𝑛 𝑠
∙𝐻 |𝑅 |
This model only gives a peak if 𝑘 𝑛 > , see Figure 2-26. The model explains that a peak can occur,
but the shape of the curve does not correlate well with tyre measurements. The model can work to ex-
plain some phenomena on vehicle level, but a better model can be found if assuming parabolic pres-
sure distribution, see 2.2.3.1.5.
2.2.3.1.5 Brush Model with Parabolic Pressure Distribution
Hertz’s contact theory for line contact motivate an elliptical pressure distribution. A parabolic pres-
sure distribution approximates an elliptical and it gives an alternative brush model, as compared to
the one appearing from uniform pressure. The coefficients in the parabolic pressure function have to
𝐿
be chosen such that ∫ 𝑝(𝜉 ) 𝑊 𝜉 and 𝑝(0) 𝑝(𝐿) 0 :
6∙ 𝜉 𝜉
𝑝(𝜉 ) ∙ ( ) [2.15]
𝑊 𝐿 𝐿 𝐿
If we do the corresponding derivation as for the uniform pressure distribution, e.g. assuming a stick
and slip zones, the location of where slip starts, 𝜉 , becomes:
6∙ 𝜉 𝜉 𝑊 𝐿3 𝑠
𝜏(𝜉 ) 𝑠 𝜉 𝑝(𝜉 ) ∙ ( ) ⇒ 𝜉 𝐿
𝐻 𝑊 𝐿 𝐿 𝐿 6 𝐻
The slip where 𝜉 appears outside 0 < 𝜉 < 𝐿 is when 𝜉 becomes < 0, which is when the whole con-
tact slips:
𝑊 𝐿3 𝑠 𝑠
𝜉 𝐿 𝐿 ( )<0 ⇒𝑠 >
6 𝐻
Total longitudinal force, , becomes:
∙ 𝑓 <0
( ∙ |𝑠 |)2 ( ∙ |𝑠 |)3 ∙ ∙
∙ |𝑠 | + 𝑒 𝑠𝑒 𝑓 |𝑠 | < ⇔
∙ ∙ 7∙( )2
sign(𝑠 )
[2.16]

⇔| |≤ ∙

{ ∙ 𝑒 𝑠𝑒
2
∙𝑊∙𝐿 𝑅
ℎ𝑒𝑟𝑒 𝑒𝑛𝑠 𝑜𝑛: [𝑓𝑜𝑟 𝑒] 𝑛 𝑠
∙𝐻 |𝑅 |
The shape of this curve becomes as shown in Figure 2-25. The uniform pressure distribution model is
drawn as reference. Note that the parabolic pressure distribution does not give any linear part, but the
derivative at 𝑠 0 is same, ∙ 𝑊 ∙ 𝐿2 ⁄( 𝐻).

112
 Vehicle Interactions and Subsystems

Figure 2-25: Shape of force-to-slip relation derived with brush model with different pressure
distributions. Also, the force terms from stick- and slip-regions are shown. The arctan and tanhyp
curves are examples of curve-fitting to same / 𝑠 in 𝑠 0.
Figure 2-25 indicates that tanhyp gives a better curve fit than arctan. However, that conclusion is
drawn without introducing different static and dynamic friction coefficients, see 2.2.3.1.6. Figure 2-26
shows that the curve shape before peak is influenced by how different 𝑙 𝑝 and 𝑘 are.

2.2.3.1.6 Different Static and Dynamic Friction

We now introduce different 𝑘 and 𝑙 𝑝 . We also identify each force term, for stick and slip regions,
as follows:
𝜉 𝜉
𝑠
𝑘 𝑊 ∙ ∫ 𝜏(𝜉 ) ∙ 𝜉 𝑊∙∫ ∙𝑠 ∙𝜉 ∙ 𝜉 {𝜉 𝐿 ( ) } ⋯
𝑊 𝐿2 𝑘
2
∙( ) 𝑠
3 𝑖 𝑘 𝑧
𝐿 𝐿
6∙ 𝜉 𝜉
𝑝 ∙ ( )
𝑊 𝐿 𝐿 𝐿
𝑙𝑝 𝑊∙ ∫ 𝑙𝑝 ∙𝑝∙ 𝜉 𝑙𝑝 ∙𝑊 ∙ ∫𝑝 ∙ 𝜉 𝑠 ⋯
𝜉 𝜉 𝜉 𝐿 ( )
{ 𝑘 }
2
𝑠 𝑠
𝑙𝑝 ∙ ∙( ) ( )
𝑘 𝑘
∙ 𝑘 ∙
𝐵𝑜𝑡ℎ 𝑒𝑥𝑝𝑟𝑒𝑠𝑠 𝑜𝑛𝑠 𝑜𝑛 𝑦 𝑓𝑜𝑟 𝑠 <
The force terms, 𝑘 and 𝑙 𝑝 , from each of stick and slip regions are identified for 𝑠 <
∙ 𝑘 ∙ ⁄ . When slip is larger, 𝑠 > ∙ 𝑘∙ ⁄ , the whole contact slips, 𝑘 0 and
𝑙𝑝 𝑙𝑝 . If we sum to 𝑘 + 𝑙𝑝 we get:
∙ 𝑘 ∙
𝑙𝑝 ∙ 𝑓 < 0 𝑜𝑟 |𝑠 | >
sign(𝑠 ) 𝑙𝑝 ( ∙ |𝑠 |)2 𝑙𝑝 ( ∙ |𝑠 |)3
[2.17]

∙ |𝑠 | ( ) +( )
𝑘 ∙ 𝑘∙ 𝑘 7∙( 𝑘 )2
{ 𝑒 𝑠𝑒
∙ 𝑊 ∙ 𝐿2 𝑅
ℎ𝑒𝑟𝑒 𝑒𝑛𝑠 𝑜𝑛: [𝑓𝑜𝑟 𝑒] 𝑛 𝑠
∙𝐻 |𝑅 |
Typical values of friction coefficients that give good resemblance with measurements are 𝑘 𝑛
. . . .0, see experiment with tyre tread sample in (Ludwig & Kim, 2017). Note that the peak of the

113
 Vehicle Interactions and Subsystems

curve neither appears at ⁄ 𝑙 𝑝 nor ⁄ 𝑘 . For example, the curve with [ 𝑙 𝑝 𝑘]

[ ] peaks at ⁄ ≈ 6 00 ⁄5000 ≈ . 6, which is far from 𝑘 .
Eq [2.17] gives same peak 𝑝 𝑘 and asymptotic 𝑙 𝑝 independent of . The peak slip 𝑠 𝑝 𝑘 be-
comes independent of ⁄ , which is often constant:
𝜇 𝑖𝑝
4 3
𝜇 𝑖 𝑘 ±3 𝑧 ±3
𝑝 𝑘 ± 𝑘 𝜇 𝑖𝑝 2
and 𝑠 𝑝 𝑘
𝑖 𝑘
𝜇 𝑖𝑝
𝑖 𝑘
𝜇 𝑖𝑝
(3 2 ) 3 2 3 2
𝜇 𝑖 𝑘 𝜇 𝑖 𝑘
⏟ 𝜇 𝑖 𝑘

𝑝 𝑘

The model and analysis can be transferred to lateral slip, except that there is often a tendency that
⁄ ≠ , but rather that tyre_smallStiction.Fx
tyre_noStiction.Fx
⁄ decreases with . Hence,tyre2_noStiction.Fx
tyre_criticalStiction.Fx
𝑠 𝑝 𝑘 increases
tyre_largeStiction.Fx
tyre2_noStiction.Fx
with .
tyre2_smallStiction.Fx tyre2_criticalStiction.Fx
tyre2_smallStiction.Fx tyre2_criticalStiction.Fx tyre2_largeStiction.Fx
tyre2_largeStiction.Fx tyre2_noS
9000 9000
9000
𝑙𝑝 𝑘 .65 06 (tuned in)
8000 𝑙 𝑝 8000
8000
𝑘
7000 7000
7000
𝑘 𝑘
6000 6000
6000
𝑘 .5
𝑘
5000 5000
5000
𝑘

4000 4000
4000
𝑘 .5

3000 𝑘 3000
3000

2000 2000
2000

1000 1000
1000

0 0
0.0 0.2 0.4 0.6 0.8 0.01.0
0.0 0.2
0.2 0.4
0.4 0.6
0.6 0.80.8 1.01.0
sx sx [rad-1] sx

Figure 2-26: Brush model with uniform (left) and parabolic (right) pressure. Varying 𝑘/ 𝑙 𝑝.

2.2.3.1.7 § Different Slip Definitions Influence on Tyre Characteristics Estimation

It matters for the curve’s transition shape to saturation, at which part of slip curve does it matter
which “transportation speed” to use, i.e. 𝑛 𝑝 𝑅 or 𝑛 𝑝 in 𝑠 (𝑅 )/ 𝑛 𝑝 . If
we estimate tyre characteristics, e.g. , using samples on the way towards saturation, it has a big
influence how the shape is expressed in . See figure below.

Tyre model: “Brush model

with parabolic pressure
distribution”, but with two
slip definitions:
𝑅
𝑠 𝑝
𝑅
𝑅
𝑠 𝑝
The difference is in the
denominator, i.e. the
transport velocity 𝑛 𝑝.

Ideal estimation assumed.

Conclusion: Different tyre

behaviour is estimated.

Figure 2-27: § Effect of slip definition on friction estimation

114
 Vehicle Interactions and Subsystems

2.2.3.2 Measurements
Some examples of measurements are given Figure 2-28. It is important to understand that there is a
big spread between different tyres (e.g. studded or not, as shown in the figure) and that the physical
phenomena we try to measure and model is not only the tyre, but the contact between one certain tyre
and one certain ground surface (e.g. ice, as shown in figure). Additionally, there is one certain wheel
suspension which can cause different high frequency oscillations which affect the averaged measured
signals. The figure shows both longitudinal and lateral grip, so both relevant for 2.2.3 and 2.2.4, but not
the combined situation in 2.2.5.
Straight-line braking Steering without wheel torque
𝑥/

/
utilized friction

utilized friction
0.05 Studded tyre
Studded tyre
Nordic non-studded winter tyre Nordic non-studded winter tyre
European non-studded winter tyre European non-studded winter tyre

longitudinal slip [percent] side slip angle [degrees]

Figure 2-28: Typical measurements for passenger car tyres on ice. From Hjort and Eriksson, VTI
report 875, 2015. https://www.vti.se/en/Publications .

2.2.3.3 Parameter Fitting in Physical Tyre Models

The brush model variants presented above are physical tyre models. They can to some extent represent
changing tyre design parameters and changing ground properties. It is almost impossible to know the
correct numerical values of the real design parameters, such as 𝑊 𝐻 in 𝑊 𝐿2 ⁄( 𝐻 )
or . But if one has experimental data, one can at least fit and , to one experiment. Changing or
then makes sense. However, 𝐿 is no design parameter, but a variable dependent of . So, if also chang-
ing , one would need more experimental data with variation in or a model for 𝐿 𝐿( ). The com-
pendium argues for using slip definition from Eq [2.1] instead of Eq [2.2] when fitting parameters in
physical tyre models.

2.2.3.4 Curve Fit Tyre Models

The brush model is a physical tyre model. However, if a model is required which is numerically accu-
rate for one specific tyre of which one have experimental data, and there is no need for changing tyre
design parameters and ground properties, one can use a curve fit tyre model instead.
Those “Curve Fit Tyre Models” often uses a mathematical curve approximation, such as trigonometric
and exponential forms instead of models as in Eq [2.9]. Most curve fit tyre models use the "collapse
from 3 to 2 independent variables", by which is meant that the fitted curves have 2 independent varia-
bles: [ ] 𝑓𝑢𝑛 (𝑠 (𝑅 ) 𝑠 (𝑅 )) instead of [ ] 𝑓𝑢𝑛 ( ) . This selection
of number of independent variables is typically the only very physical assumption in the curve fit mod-
els. The argument of physics to use slip definition from Eq [2.1] instead of Eq [2.2] are typically ne-
glected for curve fit tyre models, and it is usual that Eq [2.2] is selected. Regardless which slip defini-
tion is used, it is important that same slip definition is used when numerical parameter values are fit-
ted as when the tyre model is used, e.g. for simulation.
2.2.3.4.1 Magic Formula Tyre Model
The most well-known curve fit tyre model is probably the “Magic Formula”. It was proposed by Profes-
sor Hans Pacejka, 1934-2017. It is described, e.g., in (Bakker, 1987). The curve fit has the general form:

115
 Vehicle Interactions and Subsystems

𝑜𝑟 𝑒 𝑦(𝑥) 𝐷 ∙ sin( ∙ arctan(𝐵 ∙ 𝑥 ∙ (𝐵 ∙ 𝑥 arctan(𝐵 ∙ 𝑥)))) + 𝑆𝑉

[2.18]
𝑆 𝑝 𝑠 𝑥 + 𝑆𝐻
The variable Slip is the tyre slip, either longitudinal slip 𝑠 or lateral slip 𝑠 . The parameters are: 𝐷 [𝑁]
is peak force, 𝐵 [𝑁⁄ ] is slip stiffness, [ ] is shape and [ ] is curvature. The parameters 𝑆𝑉 and 𝑆𝐻
simply shifts the curve so that it passes through the origin, which might not be the case for measure-
ment data, since there can be errors in tyre radius and correction for rolling resistance. The influences
of these parameters are shown in Figure 2-29.

Figure 2-29: Magic Formula Tyre Parameters, (Pacejka, 2005)

2.2.3.4.2 TMsimple and TMeasy Tyre Models
TMsimple and TMeasy are two other curve fit models. TMsimple is shown in Eq [2.19] (without influ-
ence of vertical force). Both are shown in Figure 2-30. TMsimple is a simplified variant of TMeasy. For
example, in TMsimple, it is not possible to set the maximum force value to a specific slip. TMeasy is de-
scribed in Ref (Hirschberg, Rill, & Weinfurter, 2007).
(𝑠) 𝑚 sin(𝐵 ( 𝑒 | |⁄𝐴 ) sign(𝑠))
ℎ𝑒𝑟𝑒 𝐵 𝜋 arcsin( ∞ ⁄ 𝑚 ) 𝑛 𝐴 𝑚 𝐵⁄arctan( ) [2.19]
𝑡ℎ 𝑚 ∞ 𝑛 𝑠 𝑛 𝑞 [ . 5]
Tyre model TMeasy
max

Figure 2-30: Left: TMsimple (Lex, 2015). Right: TM-Easy Tyre Model, (Hirschberg, Rill, &
Weinfurter, 2007).
2.2.3.4.3 More Advanced Models
There are numerous of more advanced tyre models, such as Swift and FTire. They mix physical and
curve fit parameters. FTire is almost a finite element model and demands very many parameters.
2.2.3.4.4 Very Simple Tyre Models
There are many more models with different degree of curve fitting to experimental data. However, one
can often have use for very simple curve fits, such as:

116
 Vehicle Interactions and Subsystems

• Linearized: ∙𝑠
• Linearized and saturated: sign(𝑠 ) ∙ 𝑛( ∙ |𝑠 | ∙ ). This means approximating the
curve _𝑥 𝑠 with 3 spliced lines with the 2 borders 𝑠
( ) ± ∙ ⁄ . One can also splice in
more lines, so that the model stays piecewise linear.
• Stiff: 𝑠 0 ⇔𝑅 (as if linear with → )
• Stiff and saturated: 𝑓 𝑠𝑡 𝑘 𝑠 0 𝑒 𝑠𝑒 Discrete state switching: ℎ𝑒𝑛 𝑠 <
0 𝑡ℎ𝑒𝑛 𝑠𝑡 𝑘 ← 𝑡𝑟𝑢𝑒 𝑒 𝑠𝑒 ℎ𝑒𝑛 > 𝑡ℎ𝑒𝑛 𝑠𝑡 𝑘 ← 𝑓 𝑠𝑒 (Approximately described.)

2.2.3.5 Transients or Relaxation in Contact Patch

Both the physical and empirical tyre models assume of steady state condition in the contact patch,
meaning steady state deformation pattern and steady state sliding speed distribution. Transients be-
tween different steady state conditions take some time which is why the phenomena is called relaxa-
tion. Often in Vehicle Dynamics, the relaxation is such a quick process that it can be assumed to take
place instantaneously, i.e. the algebraic relation (𝑠 ) can be used. But sometimes it is relevant
to model the transients more carefully. Transients are triggered by variations; variations in slip (
and/or ), vertical force ( ) and surface conditions (e.g. expressed in varying ). Transients can also
appear for constant conditions (constant ) if the tyre and its suspension constitute a system
that comes into stick-slip oscillations.
Note that also start from stand-still and deceleration to stand-still are examples of strong transients in
slip. We can see this in Figure 2-11 where we see that the smaller | |, the steeper the curve ( ) be-
comes during sign change. So, if 0, the ( ) will have same step-form as a dry friction contact.
Similarly, we see infinitely steep gradients near 0 in Figure 2-31. If using slip, we experience
this as a mathematical singularity; slip approaches ± when approaches zero. Modelling transients
often solve both the mentioned transients in slip, vertical force and surface conditions as well as driv-
ing situations involving/close to stand-still.
The physical phenomena to be modelled is elasticity; in tyre sidewalls and/or in contact patch.
Parabolic pressure distribution, 0 5000 𝑁 𝑙𝑝 𝑘

If wheel locked (e.g. by If vehicle stand-still (e.g. held If vehicle has constant non-zero velocity (e.g.
friction brake), i.e. 0: by other wheels), i.e. 0: controlled by other wheels), i.e. 𝑜𝑛𝑠𝑡 𝑛𝑡 ≠ 0:

>0 𝑅
𝑅
( < < 𝑅
𝑙𝑝 𝑝 𝑘 𝑘)

<0

Figure 2-31: Eq [2.17] plotted over the [𝑅 ]-plane.

117
 Vehicle Interactions and Subsystems

2.2.3.5.1 Transients due to Relaxation in Contact Patch Modelled as Filter

The elasticity in the side walls, modelled in 2.2.3.5.1, explains delay; e.g. when changing 𝑚 stepwise,
the force will not follow 𝑓 directly, but with a delay. Another phenomenon that causes delay is that
the bristles in the contact patch needs to adopt to a new deformation pattern and sliding speed distri-
bution. Following the brush model, the physically correct way would be to formulate the equations as
a partial differential equation (PDE), with derivatives with respect to both time and position along the
contact patch, 𝜉. A “quasi-physical” way to model this is to apply a 1 st order time delay of force:
( /𝜏) ∙ (𝑓 (𝑠 ⋯) )
where 𝑓 (𝑠 ⋯ ) is the force according to a steady state model, e.g.Eq [2.9]
𝐿 𝐿𝑟 [2.20]
and the time delay, 𝜏 𝑣 𝑟 | ∙𝜔 |
and 𝐿 is the relaxation length, which
𝑇𝑟 𝑝𝑜𝑟 𝑟𝑖
often is given as a fraction (≈ 5. .50%) of tyre circumference.
Alternatively, one can also express the delay as a 1st order time delay of the slip, as follows:
𝑓 (𝑠 𝑙 ⋯)
𝑠 𝑙 ( /𝜏) ∙ (𝑠 𝑠 𝑙 )
[2.21]
where 𝑓 (𝑠 𝑙 ⋯ ) is the force according to a steady state model, e.g. Eq
[2.9] and 𝜏 is as defined in Eq [2.20].
Eq [2.20] is similar to “spring in series”, as in 2.2.3.5.1, if a linear tyre-to-slip-model is used. This can
motivate that delaying force (Eq [2.20]) is more physical than delaying the slip (Eq [2.21]). The delay
in slip rather proposes that relative speed as state variable, which is not physical in this context. How-
ever, the delayed force has the non-physical effect that sometimes can become > in cases
when wheel is off-loaded quickly, i.e. when is a large negative value. So, an extension to
max(𝑓 (𝑠 𝑙 …) ) can be motivated.
It would make sense from physical point of view, if the relaxation length was approximately same
magnitude as the contact length, or possibly the length of the sticking zone. However, commonly given
size of relaxation length is 5. .50% of tyre circumference, which is normally several times larger than
the contact length. This can be because one measures delay due to sidewall elasticity also, but then in-
terpreted as a relaxation length.
With 𝑓 𝑠 , and 𝑚 > 0 and > 0 , we can simplify Eq [2.20] to Eq [2.22]:
𝑅 𝑚
∙𝑅 𝑚 ∙ ∙ ⇒ {𝑠 𝑠𝑠 }⇒
𝐿 𝐿 𝐿
[2.22]
⇒ ≈ ∙ (𝑅 𝑚 )
𝐿
2.2.3.5.2 Transients due to Elasticity of Sidewalls
Figure 2-32 shows a physical model which can model how the force change is delayed during quick
changes in slip. The model can also handle vehicle stand-still. This type of model is often called a rigid
ring tyre model, because the belt is modelled as a rigid ring. The ring is here massless but has mass and
rotational inertia. The longitudinal and vertical support are here rigid, but they can be modelled as
compliant. If no significant inertia in wheel hub, the driveshaft compliance will be series coupled with
the rotational compliance, 𝑤 . The torque 𝑇 𝑚 is the sum of torque from propulsion system and brake
system. Damper elements can be added beside the compliances.
The mathematical model becomes as follows, where 𝑓 denotes a steady state tyre model, e.g. from
2.2.3.1:
𝑤 ∙ (𝑅 𝑚 𝑅 𝑙 )
𝑅 𝑙 [2.23]
𝑓 (𝑠 …) ℎ𝑒𝑟𝑒 𝑠
|𝑅 𝑙 |

118
 Vehicle Interactions and Subsystems

Wheel rotational
elasticity of sidewalls Massless
ring/belt

Rim &
Rim &
wheel shaft Same dynamic wheel shaft
behaviour between
rim and road

Rigid longitudinal
and vertical supports

𝑓 𝑤
Road

𝑅 𝑙 𝑅 𝑚

Brush model or similar:

𝑓 𝑠
𝜔 𝑣 Wheel rotational elasticity
𝑠 of sidewalls transformed to
𝜔 𝑓
Note: This can be seen as a non-linear damper: tangential elasticity with
𝑓 𝑅 𝑙 stiffness 𝑤 𝑁/ .

Figure 2-32: Tyre model including the rotational elasticity of tyre sidewalls.
Damping in parallel with the elasticity is often motivated also:
+
𝑤 ∙ (𝑅 𝑚 𝑅 𝑙 )
𝑤 ∙ (𝑅 𝑚 𝑅 𝑙 ) [2.24]
𝑅 𝑙
𝑓 (𝑠 …) ℎ𝑒𝑟𝑒 𝑠
|𝑅 𝑙 |
If used in a system where 𝑚 and are input variables to tyre, the force will become a state vari-
able. It is then not a problem that slip is undefined for 0, because the explicit form of equations
will become as follows. Note that we simplify by only considering the case when 𝑙 > 0. And the
model validity is limited to … such that uniquely defines 𝑠 .
𝑠 ← ( …)

𝑙 ← (𝑜𝑛 𝑦 𝑓𝑜𝑟 𝑙 > 0)

𝑅 ( 𝑠 sign( )) [2.25]
← 𝑤 ∙ (𝑅 𝑚 𝑅 𝑙 )
ℎ𝑒𝑟𝑒 𝑠 𝑡ℎ𝑒 𝑛 𝑒𝑟𝑠𝑒 𝑜𝑓 𝑓𝑢𝑛 𝑡 𝑜𝑛 𝑓 𝑠𝑢 ℎ 𝑡ℎ 𝑡 𝑠 ( )
With 𝑓 𝑠 , and 𝑚 > 0 and > 0 , we can simplify Eq [2.25]to Eq [2.26]:
𝑤 ∙
𝑤 ∙𝑅 𝑚 ⇒ {𝑠 𝑠𝑠 }⇒
[2.26]
⇒ ≈ 𝑤 ∙𝑅 𝑚 𝑤 ∙
Eq [2.25] means that we read the function 𝑓 from force to slip 𝑠 . When knowing slip, we can cal-
culate the rotational speed 𝑙 . Then, the state derivative can be calculated, so that the state can
be updated in each time step.
With any reasonable tyre function 𝑓 , there is a maximum magnitude of force, | | , above
which there is no slip 𝑠 that gives that . In most problems, one never ends up there in the simula-
tions, since when approaching | | , the velocity 𝑙 changes quickly in the direction that

119
 Vehicle Interactions and Subsystems

makes | | stays < . But, if decreases stepwise, one might end up there anyway for short time
intervals. In that case, it often gives physically acceptable solutions on vehicle level, to simply saturate
𝑠 so that | | is saturated at a certain level, e.g. 0.95 . For the brush model with uniform pres-
sure distribution, Eq [2.9], the inverted function becomes as follows, including such saturation:
∙
𝑓𝑜𝑟 | | ≤
∙ sign( )
( ) ∙ 𝑓𝑜𝑟 | | < 0.95 ∙
𝑠 4∙ | |
∙ [2.27]
∙ sign( )
∙ 𝑒 𝑠𝑒
{ 4∙ 0.95
∙ 𝑊 ∙ 𝐿2
ℎ𝑒𝑟𝑒 𝑒𝑛𝑠 𝑜𝑛: [𝑓𝑜𝑟 𝑒]
∙𝐻
2.2.3.5.3 Relation between the two Transients Models
The models in 2.2.3.5.1 and 2.2.3.5.1 describe two different phenomena which both are present in the
real world. Since it is difficult to distinguish between the two delay-creating parameters, 𝑤 and
⁄𝐿 , one often models only one phenomenon. But then, one adjusts the numerical value of the used
parameter so that the model captures about the same delay as a real-world test. If 𝑤 ⁄𝐿 , the
two models coincide approximately.
This compendium does not give any recommendation of which of the two models is best.
Note that, with unsaturated in Eq [2.20], one has to limit the integration of to | | ≤ ∙ .
2.2.3.5.4 § Expansion of Brush Model to Partial Differential Equation
Contribution from Luigi Romano, Vehicle Dynamics at Chalmers

The brush model derived in 2.2.3.1 was derived for steady state deformation pattern, i.e. the defor-
mation was only 𝛾(𝜉 ), not 𝛾 (𝜉 𝑡). Some call such tyre models “(single-)point contact models”. The
models in 2.2.3.5.1 and 2.2.3.5.2 uses a point contact approach, but add in time derivatives. These
models fit well as parts in larger dynamic models of the whole vehicle and one can use ordinary simu-
lation methods for dynamic system, 1.5.1.1.5.
If assuming 𝛾(𝜉 𝑡), or “line contact model”, one actually models the transients in a physical way. The
Mathematical model then becomes a partial differential equation, PDE, as opposed to ODE. To imple-
ment model in a typical dynamic vehicle model (ODE) we can discretize the PDE to ODE. One can also
discretize already in the Physical model, see Figure below.

120
 Vehicle Interactions and Subsystems

”Control volume” 𝛾 𝜉 𝑡 ”Follow each bristle” 𝛾 𝑡 𝜉 𝑡

Discretization
tyre in 𝜉, made in
𝜉 𝐿 𝑅 𝜉 𝜉 0 physical model 𝜉 𝐿 𝑅 𝜉 (𝑡) 𝜉 0
𝜉 𝜉
𝑡 𝑒 H
road
𝛾
bristle at 𝜉 𝛾(𝜉 𝑡) bristle at time 𝑡
(not same bristle over time) 𝑅 𝑡
𝜕𝛾
𝛾(𝜉 𝑡) 𝜕𝜉 𝑡
𝜕𝜉

𝜉 𝐿 𝑅 𝜉 𝜉 0 𝜉 𝐿 𝑅 𝜉 (𝑡+ 𝑡) 𝜉 0
𝜉 𝜉
𝑡 𝑒 𝑡 + 𝜕𝑡
bristle at 𝜉
(not same bristle over time) 𝜕𝛾 𝛾 (𝑡 + 𝑡)
𝛾 𝜉 𝑡+𝜕𝑡 𝛾+ 𝜕𝑡 bristle at time 𝑡 + 𝑡
𝜕𝑡
𝜕𝛾 𝜕𝛾 𝑅
𝐻 𝑅 𝐻 𝜉 𝑅 𝛾
𝜕𝑡 𝜕𝜉 𝐻
𝛾 0𝑡 0 𝑛 𝛾 𝜉0 𝛾 𝜉 𝛾 0 ℎ𝑒𝑛 𝜉 > 0 𝑛 𝛾 𝛾 𝑓𝑜𝑟 ..𝑁

𝜕 𝜕 𝜉𝑖+1 𝜉𝑖 1
𝐻 𝛾 ≈𝑅 𝐻 ≈
𝜕𝜉 𝜕𝜉 2 𝐿/𝑁
𝛾 0 𝑛 𝛾 𝛾 𝑓𝑜𝑟 ..𝑁

Figure 2-33: § Two views of how bristle deformation varies in space and time. Left: Using a
“control volume” in which new bristles comes in and moves out. Right: Following a certain bristle.
Note that the figure only shows model and equation for the stick zone. With the model derived from
“following each bristle”, it is straight-forward to implement also stick, slip and transitions for each
bristle. With the model derived from the control volume, it is less easy to do this, but for many driving
cases one can assume that there is exactly one break-away point 𝜉 as in the steady state model. Refer-
ence (Romano, Bruzelius, & Jacobson, 2020) does this for a start from stand-still and brake to stand-
still, showing that the model in 2.2.3.5.2 can be motivated starting from the PDE from the “control vol-
ume”.

2.2.3.6 Simultaneous Rolling Resistance and Longitudinal Force

So far, the models in 2.2.3 have not included rolling resistance. As we regard rolling resistance as a
torque, not a force, it does not affect the (𝑠 ) curve. But the rolling resistance does move the 𝑇 (𝑠 )
curve vertical. The curve moves upwards if the wheel is rolling forward and downwards if rolling rear-
wards, see Figure 2-34. Utilizing 𝑠 ( ) 𝑠 𝑅𝑅 gives 𝑠 𝑅𝑅 ⁄ ≈
0.0 ⁄ 0 0.00 .
The (𝑠 ) curve is normally the suitable view for vehicle level studies, while the 𝑇 (𝑠 ) curve is some-
times needed for involving a model of the propulsion and brake systems. As mentioned in 2.2.1.7, one
can select different 𝑅 in the slip definition. Figure 2-34 uses an 𝑅 for slip definition such that 0
for 𝑠 0, i.e. for pure rolling. Often, for wheel slip control (ABS, TC, …), one instead uses a radius for
slip definition such that 𝑇 0 for 𝑠 0, i.e. for free-rolling. That is because it is much easier to know
when 𝑇 0 than when 0. When 𝑇 0, the motion can make it look as if the wheel was rolling at a
slightly different radius ⁄ 𝑅 ( + sign( ) 𝑅𝑅 ⁄ ) ≈ (0.999 𝑜𝑟 .00 ) 𝑅 which is the
Rolling radius 𝑅 in 2.2.1.7.

121
 Vehicle Interactions and Subsystems

force
+ + 𝑅𝑅
+ 𝑇/𝑅𝑙 (if > 0)
+ 𝑅𝑅
𝑇/𝑅𝑙 (if < 0)
𝑅𝑅
𝑅
𝑠
𝑅
Operating point when free- 𝑅𝑅
rolling ( 0) forward ( > 0)

𝑅𝑅

𝑅𝑅

Figure 2-34: Longitudinal tyre force ( ) and normalized wheel torque (𝑇/𝑅𝑙 ). Slip 𝑠 defined so
that curve (𝑠 ) passes through diagram origin, which means that 𝑇(𝑠 )⁄𝑅𝑙 does not.

2.2.4 Lateral Force of Tyre

After a vehicle starts moving, controlling the direction of travel becomes a high priority for the opera-
tor. For wheeled vehicles, the primary mode to control travel direction is to change the orientation of
the tyre, i.e. to apply a steer angle. Tyres generate a lateral force when they are oriented at an angle
different to the direction of the vehicle motion. The tyre typically deforms as in Figure 2-35.
a) yz plane, view from rear b) xy plane, view from above
left non-steered wheel rest of vehicle left non-steered wheel rest of vehicle

𝑤
𝑀
𝑀
𝑀
𝑡𝑝
𝑀 𝑀

≈
𝑀 𝑀 ≈𝑀
Figure 2-35: Deformation and forces of a Cornering Tyre. Wheel side slip angle is .
It is essential to distinguish between the steer angle and wheel (lateral or side) slip angle of the tyre.
Lower right part of Figure 1-43 shows this difference. The steer angle, or d, is the angle between ve-
hicle longitudinal direction and tyre longitudinal direction. The wheel side slip angle, is the angle be-
tween tyre longitudinal direction and the tyre translational velocity (=wheel hub velocity).
Assuming no longitudinal tyre slip, 𝑅 , the relation between the lateral force of a tyre and the
wheel side slip angle is typically as shown in Figure 2-39. The behaviour of the curve is similar to that
exhibited for longitudinal forces Figure 2-29 and Figure 2-30. It becomes even more similar if lateral
slip angle is replaced by lateral wheel slip, 𝑠 , which is tan( ) ≈ for small lateral slip.
𝑓𝑠 0
𝑠 sign(𝑅 ∙ ) {
. 𝑒. 𝑅
} tan( ) [2.28]
|𝑅 ∙ | 𝑅∙
Using magnitude in denominator |𝑅 ∙ | gives same sign of 𝑠 and for all combinations of signs of
and , which leads to easier formulas.

122
 Vehicle Interactions and Subsystems

2.2.4.1 Tyre brush Model for Lateral Slip

With corresponding simplification as in 2.2.1.6, we now use the brush model to also explain the lateral
properties.
2.2.4.1.1 Model with Independent Bristles
Figure 2-36 shows the model for pure lateral slip (no longitudinal slip) and should be compared to Fig-
ure 2-23. The difference is that the model for lateral slip has the deformation of the bristles perpendic-
ular to drawing in the upper left view in the figure.
rolling direction
L=length of contact patch view from rear,
at one certain x
tyre
𝑅
outlet inlet rim
x sidewall
x=L x=xc x=0 and tread
H=height
in series
road (relative
𝑅 to wheel hub)
(relative to wheel hub)
Contact patch, W=width
view from above:
Velocities of bristle
line where upper ends
ends in stick zone:
of bristles are attached
line where lower ends of 𝑅
bristles appear when sliding

Drawn for 𝑅
x and >0

pressure
x
stick
slip
“mode”
x
shear stress
(lateral, i.e. perpendicular
to top drawing)

Figure 2-36: The brush model’s physical model for lateral slip. The bristles are to be thought of as
the tread in series with sidewall lateral elasticity.
Each bristle in Figure 2-36 is thought of as a part of the tread in series with a part of the sidewall. Fur-
ther, the bristles are considered as independent of each other, which is debatable. However, it is
enough for a quantitative explanation of the brush model for lateral slip. The derivation of model equa-
tions becomes similar as for the longitudinal model:
∙ 𝑡(𝜉 ) ∙ 𝑡(𝜉 ) 𝑡 (𝜉 ) 𝜉 ⁄ 𝑇 𝑛 𝑝
𝜏 ∙𝛾 ∙ ∙ { }
𝐻 𝐻 𝑇 𝑛 𝑝 ≈ |𝑅 ∙ |
∙𝜉 𝑊 𝐿2
∙ ∙ ∙𝜉 { 𝑠 } ∙𝑠 ∙𝜉
𝐻 |𝑅 ∙ | 𝐻 |𝑅 ∙ | 𝐻 |𝑅 ∙ | 𝑊 𝐿2
Note that subscript 𝑦 has been introduced where we need to differ towards the longitudinal brush
model. Correspondingly, subscript 𝑥 should be used in longitudinal model. As for longitudinal model,
we have to express the force differently for when friction limit is not reached within the contact and
when it is.

123
 Vehicle Interactions and Subsystems

∙ ∙
∙𝑠 𝑓 |𝑠 | ≤ ⇔| |≤
∙
∙
sign(𝑠 ) ∙ ∙ ∙( ∙ ) 𝑒 𝑠𝑒
{ 4∙ |𝑠 | [2.29]
2
∙𝑊∙𝐿
ℎ𝑒𝑟𝑒 𝑒𝑛𝑠 𝑜𝑛: [𝑓𝑜𝑟 𝑒] 𝑛 𝑠
∙𝐻 |𝑅 ∙ |
(Only valid for pure lateral slip, i.e.: 𝑅 )
Note that the lateral tyre slip 𝑠 is the sliding speed of tyre over ground in lateral direction, divided by
the same “transport speed” as for longitudinal slip, i.e. the longitudinal transport speed 𝑅𝑤 . Note
also that the lateral force and lateral tyre slip are counter-directed, which is logical since it is of fric-
tion nature.
Tyre Lateral Slip vs Wheel Slip Angle
The lateral tyre slip 𝑠 ⁄|𝑅 ∙ | in Eq [2.29] and Eq [2.31] can be compared with lateral wheel slip
angle arctan( ⁄ ), mentioned in context of Figure 2-35:
• “Lateral wheel slip” 𝑠 𝑤 𝑤⁄ 𝑤 tan( ), is how wheel (hub and tyre) moves over
ground and independent of wheel rotational speed.
• “Lateral tyre slip” 𝑠 ⁄|𝑅 ∙ |, is the slip used in the constitutive relation (𝑠 ).
• If no longitudinal tyre slip 𝑠 0, i.e. if 𝑅 , we have 𝑠 𝑤 𝑠 . Then 𝑠 𝑤 can be used in
the constitutive relation.
For a linearization, the most correct way is that lateral force is 𝑠 , as opposed to α. Often one
finds α as starting point in the literature, but this compendium uses 𝑠 . (In (Pacejka, 2005),
pp184-185, there is also a note that 𝑠 is more appropriate than .)
A difference is how one linearizes a vehicle model in and . A non-steered axle modelled with
α , needs to be approximated with “ 𝑟 𝑡 𝑛( 𝑣 ⁄ 𝑣 ) ≈ 𝑣⁄ 𝑣 ” to make the vehicle
model linear, see derivation of Eq [4.49]. However, with 𝑠 ; the vehicle model becomes linear
without further approximations. For a steered axle it is less obvious, but it does not help the lineariza-
tion to use α.
2.2.4.1.2 Model with Dependent Bristles, String Model
Opposed to the assumption in 2.2.4.1.1, the lateral deformations of the bristles are dependent on each
other, especially since the tread is mounted on the belt and the belt is rather like a string. So, we as-
sume a certain deformation of the sidewall, expressed in 𝜀 𝑛 and 𝜀 for the belt=”string” in Figure
2-37. This gives a slightly different model compared to 2.2.4.1.1. Models with such belt deformation
are called “tyre string models”. The shape of the string is dependent on the sidewall elasticity, e.g. tyre
profile height, but also of the side force itself. So, the model is intrinsically implicit; the string shape
influences the side force and the side force influences the string shape.
The derivation of model equations becomes similar as for the longitudinal model. Here is an interme-
diate result, an expression for the shear stress 𝜏 :
∙ 𝑡 (𝜉 ) (𝑅 tan(𝜀 𝑛 ) ) ∙ 𝑡(𝜉 )
𝜏 ∙𝛾 ∙ ∙
𝐻 𝐻
𝑡(𝜉 ) 𝜉⁄ 𝑇 𝑛 𝑝 (𝑅 tan(𝜀 𝑛 ) )∙𝜉
{ } ∙
𝑇 𝑛 𝑝 ≈ |𝑅 ∙ | 𝐻 |𝑅 ∙ |

𝑊 𝐿2
∙ tan(𝜀 𝑛 ) sign( ) ∙𝜉 { }
𝐻 |⏟
𝑅∙ | 𝐻
( 𝑦 )

124
 Vehicle Interactions and Subsystems

(𝑠
∙⏟ tan(𝜀 𝑛 ) sign( )) ∙ 𝜉
𝑊 𝐿2
𝑦𝜀

view from rear,

rolling direction at one certain x
rim
L=length of contact patch sidewalls
𝑅 tan 𝜀 𝑛
tyre (relative to wheel hub)
outlet 𝑅 belt
inlet
x 𝑅 tan 𝜀 𝑛
x=L x=xc x=0 tread
H=height
road 𝑅
(relative to wheel hub)

Contact patch, view W=width

from above: rim
𝜀 𝜀𝑛
Belt deformation is
drawn simplified as
piecewise linear.
Velocities of bristle ends in stick zone:
𝑅 tan 𝜀 𝑅 𝑅 tan 𝜀 𝑛

𝑅
Drawn for 𝑅 and >0

𝜉
pressure
𝜉 stick
slip
“mode”
𝜉 shear
stress
(lateral)
Figure 2-37: Tyre sidewall deformation and tread deformation with the belt (the “string”) in
between. The drawn bristles are here assumed to represent only tread parts, while the sidewall is
treated as an elastic structure between rim and belt.
Comparing to 2.2.4.1.1, we can note that subscript 𝑡𝑟 is added to underline that and /𝐻
now means only the tread, not including the sidewall. Sidewall elasticity is instead handled with 𝜀 𝑛 .
The variable 𝑠 𝜀 is an auxiliary mathematical variable introduced only to make the expressions more
manageable; it will be eliminated later.
∙
∙𝑠 𝜀 𝑓| |≤
∙
sign(𝑠 𝜀) ∙ ∙ ∙( ∙ ) 𝑒 𝑠𝑒
4∙ |𝑠 𝜀 |
{
∙ 𝑊 ∙ 𝐿2
ℎ𝑒𝑟𝑒 𝑒𝑛𝑠 𝑜𝑛: [𝑓𝑜𝑟 𝑒] 𝑛 𝑠 𝑛 𝑠 𝜀 𝑠 tan(𝜀 𝑛 ) sign( )
∙𝐻 |𝑅 ∙ |
Now, this model is still implicit because 𝜀 𝑛 depends on 𝜏(𝜉 ). Introducing simplest possible (linear)
constitutive equation for this dependency as in Eq [2.30]:

125
 Vehicle Interactions and Subsystems

∫𝜏 𝜉 𝑤𝜀 𝑛 tan(𝜀 𝑛 ) [2.30]

Eq [2.30] makes it possible to derive the following explicit model:

𝑤𝜀 𝑛 ∙
∙𝑠 𝑓| |≤
𝑤𝜀 𝑛 +
2
𝑤𝜀 𝑛 |𝑠 | 𝑤𝜀 𝑛 𝑠 𝑤𝜀 𝑛
+ √( ) +
∙ ∙
𝑠 𝑛(𝑠 ) ∙ ∙ 𝑒 𝑠𝑒 [2.31]
{
∙ 𝑊 ∙ 𝐿2
ℎ𝑒𝑟𝑒 𝑠 𝑛 𝑒𝑛𝑠 𝑜𝑛: [𝑓𝑜𝑟 𝑒]
|𝑅 ∙ | ∙𝐻
(Only valid for pure lateral slip, i.e.: 𝑅 )
It should be noted that the constitutive relation in Eq [2.30] only states how the string is angled. It is
still physically consistent to separately add, outside the tyre model, a constitutive equation for the lat-
eral translational deformation of the sidewall; something as 𝑤 𝑛 𝑙 𝑛 , where
would be the lateral deformation between wheel rim and contact patch. Such sidewall elasticity would
appear in series with the lateral tyre slip force model, in a similar way as the torsional sidewall elastic-
ity in 2.2.3.5.2 appeared in series with the longitudinal tyre slip force model.
2.2.4.1.3 Comparison of the Bristle Models
Eq [2.31] (dependant bristles) is to be compared to Eq [2.29] (independent bristles). Using 𝑤
𝑤𝜀 𝑛 ⁄( 𝑤 𝜀 𝑛 + ) the models are identical up to / and have the same asymptote
for 𝑠 → . See Figure 2-38.

Figure 2-38: Comparison of model with independent and dependent bristles (String model).
The influence of vertical load was discussed in 2.2.3.1.2 but is better explained with dependent bris-
tles. Assumes that only the tread stiffness (and not 𝑤 𝜀 𝑛 ) varies with contact length, and that
this variation is proportional, ( ) , as we found in 2.2.3.1.2. This indicates a degressive char-
acteristics of ( ), which is also observed in measurements.
The model with dependent bristles is probably more correct. Anyway, we will use the other most in
this compendium, since it is much easier to combine with the longitudinal model (Eq [2.9]) to model
combined (longitudinal and lateral) slip.

2.2.4.2 Lateral Tyre Slip Stiffness

In summary for many models (and tests!) the following is a good approximation for small lateral slip
(and negligible longitudinal slip and constant normal load):

126
 Vehicle Interactions and Subsystems

∙𝑠 or 𝛼 ∙ [2.32]
For the brush model, or any other model which describes (𝑠 … ) or (α … ), one can de-
fine the “Lateral tyre slip stiffness” or “Tyre Cornering Stiffness”, or 𝛼 , which have the unit N/1 or
N/rad. It is the derivative of force with respect to slip or slip angle. Reference (ISO 8855) defines the
cornering stiffness as 𝛼 for slip angle 0. It is implicit that also 𝑠 0. And 0 when 𝑠 0:
𝜕 𝜕
𝛼 ( )| (𝜕 )| [2.33]
𝜕𝛼 𝛼= = 𝑦
𝑦= =

When using only small slip, it does not matter if the cornering stiffness is defined as the slope in an
versus diagram or versus 𝑠 𝑡 𝑛( ) diagram. Therefore, the notation for cornering stiffness
varies between 𝛼 and . Cornering stiffness has the unit 𝑁 which can be interpreted differently: as
[𝑁 𝑁⁄ 𝑁/(( /𝑠)⁄( /𝑠)) ] or as [𝑁 𝑁 ⁄𝑟 ].
The longitudinal tyre slip stiffness, , is normally larger than the lateral tyre slip stiffness, , which
can be explained with that the tyre is less stiff in lateral direction. Since it is the same rubber one could
argue that both and 𝐻 should be the same, but both due to longitudinal grooves in the tread and due
to lateral deformable sidewall, it is motivated to introduce different subscripts: ( ⁄𝐻) > ( ⁄𝐻) .
One could elaborate with different friction coefficients and , but in this compendium it is claimed
that friction is well modelled as isotropic. More about this in 2.2.4.7.5.
The cornering tyre forces initially exhibit a linear relation with the slip angle. A non-linear region is
then exhibited up to a maximum value. In Figure 2-39, the maximum slip angle is only 16 degrees (or
𝑠 tan( 6 𝑒 ) 0. 9) and one can expect that the tyre forces will drop as the slip angle ap-
proaches 90 degrees.
Adhesion Limit
Adhesion Limit
|Cornering Force| [kN]

|Slip Angle| [deg]

0 4 8 12 16

Figure 2-39: Left: Influence of tyre design. Right: Influence of inflation pressure, (Gillespie, 1992).

2.2.4.3 Influence of Vertical Load

As discussed for longitudinal slip and the rolling resistance behaviour of tyres, the vertical load on the
tyre affects the force generation. The general behaviour of the tyre’s cornering performance as the ver-
tical load changes is presented in Figure 2-39. These figures show that the cornering stiffness is influ-
enced by vertical load. A first approximation is that with proportionality coefficient
𝛼 , the Cornering Coefficient (or Lateral Slip Stiffness Coefficient):
𝑠 𝑠 or 𝛼 𝛼 [2.34]
We have the linearized in two ways: with respect to and 𝑠 . A slightly better linearization is ≈
𝑁 𝑚 ( 𝑁 𝑚 ) 𝑠 where subscript Nom is the tyre’s nominal load. Some examples are
shown in Figure 2-40 and Figure 2-41. More about influence of vertical force in 2.2.5.3.2.1.

127
 Vehicle Interactions and Subsystems

Figure 2-40: Example of cornering stiffness versus vertical load for a truck tyre 295/80R22.5.

Figure 2-41: Cornering stiffness versus vertical load for some passenger car tyres. From flat track
tests.

2.2.4.4 § Influence of Tyre Wear

Worn tyres get smaller groove depth and bristle height 𝐻. So, worn tyres should be more slip stiff.

128
 Vehicle Interactions and Subsystems

Figure 2-42: § Preliminary measurements of lateral slip stiffness for twin-mounted trailer tyres for
heavy vehicles. Two tyres, “1” and “2”, are measured. “W” means worn to a small groove depth but
still legal. From Mattias Hjort and Sogol Kharrazi, VTI.

2.2.4.5 Curve Fit Tyre Models for Lateral Slip

The general form of the lateral force versus lateral slip curve is also suitable for the Magic Formula,
TM-Easy, or similar curve fitting approach when sufficient test data is available.

2.2.4.6 Transients and Relaxation in Contact Patch

As for longitudinal, there is a delay in how fast the steady state conditions can be reached in contact
patch, which is sometimes important to consider. A similar model, as for relaxation in longitudinal di-
rection 2.2.3.5.1, is to add a first order delay of the force:
𝐴 ∙ (𝑓(𝑠 …) )
where 𝑓(𝑠 … ) is the force according to a steady state model and 𝐴
𝑣𝑇𝑟 𝑝𝑜𝑟 | ∙𝜔| |𝑣 | [2.35]
𝐿𝑟 𝐿𝑟
or 𝐴 𝐿𝑟
and 𝐿 is the relaxation length, which is a frac-
tion (≈ 5. .50%) of tyre circumference.

2.2.4.7 Other Forces and Moments in Lateral Use of Tyre

(This section has strong connection with 2.6.2.)
The deformations of the tyre during cornering are quite complex when compared to the case of pure
longitudinal motion, see Figure 2-35. Hence, there are more effects than simply a lateral force. Some of
these will be discussed in the following.
2.2.4.7.1 Tyre Aligning Moment due to Lateral Shear
In the lowest diagram in Figure 2-36, one can see that shear stress is concentrated to the outlet side of
the contact patch for small slip angles. So, the equivalent lateral force acts behind the centre of wheel
rotation for small slip angles. As seen in Figure 2-35 b) it acts at a position 𝑡𝑝 behind the wheel’s 𝑦 axis.
The distance 𝑡𝑝 is called the pneumatic trail, see also Figure 2-43, and the resulting yawing moment
around a vertical axis through centre of the contact patch will be 𝑡𝑝 , which is often called the
tyre aligning moment, 𝑀 . The moment is named after that is normally acts to align the wheel in direc-
tion of zero side slip. Figure 2-43 shows 𝑀 𝑡𝑝 , but also a similar moment, (𝑡𝑝 + ). Addi-
tional effects from 𝜏 are not drawn.
If the tyre is on a steered axle, the aligning moment influence on steering wheel torque is important.
When finding that influence, the moment around the steering axis intersection with ground is the im-
portant moment (𝑡𝑝 + ), which can be called the steering moment. The distance is the me-
chanical trail, which is built into the suspension linkage design. One typically designs the suspension
so that > 0, which makes the whole steering moment act in the same direction as the aligning mo-
ment. If driver takes his hands from steering wheel in a curve, and/or if steering power assistance is

129
 Vehicle Interactions and Subsystems

lost, the steering will tend to steer in the direction of body motion above the steered axle, which is nor-
mally relatively smooth and safe.
Figure 2-43 shows the combined response of lateral force and slip angle. It is interesting to note that
the steering torque reaches a peak before the maximum lateral force capacity of the tyre is reached. It
can be used by drivers to find, via steering wheel torque, a suitable steer angle which gives a large lat-
eral force but still does not pass the peak in lateral force. The reason why pneumatic trail can become
slightly negative is because pressure centre is in front of wheel centre, see Figure 2-13.
Tyre Aligning Moment in due to Lateral Shear in Brush Model
A model for (yawing) aligning moment around a vertical axle through centre of contact point, 𝑀 , will
now be derived. Any model for lateral shear stress can be used, but we will here only use the uniform
pressure distribution and independent bristles in 2.2.4.1.1. A corresponding expression as Eq [2.29] is
derived, but for 𝑀 instead of .
𝐿
𝐿
𝑀 𝑊 ∙ ∫ 𝜏 ∙ (𝜉 ) 𝜉 ⋯

𝐿
C ∙𝑠 𝑓𝑜𝑟 < ⇔𝑠 <
6 [2.36]
2 2
∙𝐿
( ) 𝑒 𝑠𝑒
{ 8∙ ∙𝑠 ∙ ∙𝑠
𝑊 𝐿2
ℎ𝑒𝑟𝑒
𝐻

steering axis
shear stress,

intersection
𝜏

with ground 𝑡𝑝
contact
patch
𝑡𝑝 𝑡𝑝 +
tyre
steering moment peaks at lower slip than lateral tyre force

lateral tyre force (for a tyre with peak)

total trail = caster trail + pneumatic trail + 𝑡𝑝

steering moment + 𝑡𝑝

side slip angle

Figure 2-43: General Response of Steering torque to Side slip angle. Tyre aligning moment = ∙ 𝑡𝑝
is one part of the steering moment= ∙ (𝑡𝑝 + ).
The curve of Eq [2.36] is plotted in Figure 2-44.

130
 Vehicle Interactions and Subsystems

Small and increasing ,

and constant non-zero
lever

Large , but lever

approaches zero

Figure 2-44: Aligning moment (𝑀 ) around contact patch center for uniform pressure distribution.
The lateral force and the aligning torque can be used to calculate the steering forces. If also the steer-
ing assistance is known, the steering wheel torque can be calculated. It can be noted that the model
does not include the moment from steering rotation itself, i.e. the torque counteracting .
2.2.4.7.2 Influence from Longitudinal Tyre Force
Figure 2-43 does not show effects from 𝜏 , but this is shown in Figure 2-45.
2.2.4.7.3 Tyre Spin Torque
The none-symmetry in shear stress around wheel centre in Figure 2-45 can appear as steady state;
non-symmetry in 𝜏 due to brush model and in 𝜏 due to non-symmetric vertical pressure. But there is
one additional reason to yaw moment in tyre contact patch. That is the friction yaw velocity of the
wheel 𝑤 𝑙 . The additional moment from this effect is often called (tyre) spin torque. One way to
model it is, are conceptually an elastic torsional spring in series with friction. The influence from spin
torque is only important at low speed, e.g. steering in low speed, where 𝑤 𝑙 ≈ 𝑤 𝑙 .
moment moment around
shear around wheel steering axis
stresses force center intersection

shear stress, 𝜏 𝑠
steering axis
intersection 𝑠 𝑠+
with ground
contact
patch
shear stress, 𝜏

tyre
𝑡𝑝

𝑡𝑝 𝑡𝑝 +

Figure 2-45: Influence from both lateral and longitudinal shear stress on aligning moment.
2.2.4.7.4 Camber Force
Camber force (also called Camber thrust) is the lateral force caused by the cambering of a wheel. Cam-
ber thrust 𝑚 is approximately linearly proportional to camber angle 𝑤 for small angles:

131
 Vehicle Interactions and Subsystems

Camber thrust 𝑚 𝑚 ∙ 𝑤 . The camber stiffness, 𝑚 , is typically 5..10 % of the

cornering stiffness. The tyre lateral forces due to lateral slip 𝑙𝑝 and due to camber 𝑚 are su-
perimposable when they are small:
𝑙𝑝 + 𝑚 ∙𝑠 ∙ 𝑤 [2.37]
𝑙𝑝 and 𝑚 use same friction contact to ground so the sum will be saturated by .
A physical model for camber thrust can be derived very similarly to how models in 2.2.3.1 and 2.2.4.1,
see Figure 2-46: Brush modes for Camber force (or Camber thrust) 𝑚 . There action line of
𝑚 is often ahead of contact patch centre, which is called Camber Lead.
Subscripts: vertical section, through
Conditions: wheel centre, seen from rear
: 𝑟𝑜𝑢𝑛
No slip (𝑅𝑤 𝑤 and 0)
: ℎ𝑒𝑒
𝑤 𝑚
cos 𝑚 ≈
≈
≈
forward same relative
motion vs
𝑚

ground, so
≈ 𝑅𝑤

outlet
𝑅𝑤 𝑤
inlet same forces
𝑅𝑤 cos

𝜉
𝜉 𝐿 𝜉 0 𝜉 𝐿⁄
𝜉 0
Moving
ground 𝑚 Fixed ground
𝑤 𝑤 𝑚
Contact
ground rotates Upper bristle ends follows 𝑤 sin 𝑚 ≈
patch
this straight line, so ≈ 𝑤 𝑚
𝑝𝑝 𝜉 0
Lower bristle follows the rotating ground, so
Deformation 𝜉 𝜉 𝐿⁄
𝑙 𝑤 𝑤 𝑚
and 𝜏 𝜉 have
𝑚
this shape Conditions with low and/or very stiff bristles
Conditions where 𝒚 is saturated by 𝒑: (as a “metal-to-metal rolling contact”):
stick slip stick 𝜏
𝜏 + 𝑝
𝜏 0 𝑚 0
𝜏 0 𝑀 𝑚
𝜏 𝑝
𝑚 𝜏 𝑝 𝜏

Figure 2-46: Brush modes for Camber force (or Camber thrust) 𝑚 . For intuitive
understanding, note that the ground is thought of as rotating opposite to the vertical component
of wheel rotation ⃗ 𝑤 .
𝑡 𝑡 𝑘𝑒𝑠 𝑡 𝑒 𝑡𝜉 𝜉 𝜉
∫0 (𝑣 𝑜𝑤 𝑟 𝑦 𝑣𝑢𝑝𝑝 𝑟 𝑦(𝜉))∙ 𝜉 ∫0 ((𝜉 𝐿 ⁄2) 𝜔𝑧𝑔 )∙ 𝜉
𝜏 (𝜉 ) ∙ 𝛾 (𝜉 ) { 𝑓𝑜𝑟 𝑟 𝑠𝑡 𝑒 } ∙ 𝐻𝑦
∙ 𝐻𝑦
𝑓𝑟𝑜 𝑛 𝑒𝑡 𝑡𝑜 𝜉
𝜉 𝑛 𝑝 𝑡𝜉 𝑦 𝑦
2
𝜉 𝐿
𝜉
𝑦
{ } ∙ ∫ 𝜉(𝑅𝑤 𝑤 𝑡𝜉 𝐿⁄ ) ∙ 𝑡𝜉 ∙ [𝑅𝑤 𝑤 𝑡𝜉 ] ∙
𝑅𝑤 𝑤 𝑡𝜉 𝐻𝑦 𝐻𝑦 2 2 2 𝐻𝑦
2 2
𝜉 𝜉 𝑦 𝜉
(𝑅𝑤 𝑤 𝐿 𝑡𝜉 ) {𝑡𝜉 ≈ 𝑤 𝑚 } 𝑤 𝑚 ∙ (𝑅𝑤 𝑤 ( ) 𝐿
2 𝑤 𝜔𝑤 2 𝐻𝑦 𝑤 𝜔𝑤
𝜉 𝑦
) 𝑦
∙ (𝜉 2 𝐿 𝜉) ≈ ∙ 𝑚 (𝜉 2 𝐿 𝜉)
𝑤 𝜔𝑤 2 𝐻𝑦 𝑤 2 𝐻𝑦 𝑤
𝐿
Integration 𝑚 ∫ 𝜏 (𝜉 ) ∙ 𝜉 gives:
𝑊 𝐿3
𝑚 𝑚 ∙ 𝑚 ℎ𝑒𝑟𝑒 𝑚
𝐻 𝑅𝑤 [2.38]
Note: Only valid when no part of the contact patch is saturated by friction.

132
 Vehicle Interactions and Subsystems

Checking with typical values for a passenger car tyre gives: 𝑚 ⁄

( 3
𝑊 𝐿 ⁄( 2
𝐻 𝑅𝑤 ))⁄( ∙ 𝑊 ∙ 𝐿 ⁄ ∙ 𝐻 ) 𝐿 6 𝑅𝑤 ≈ 0. ⁄(6 0. ) ≈ 0.05
⁄ ( ) 5% which is
of same magnitud as 5..10%, which was mentioned in beginning of 2.2.4.7.4.
2.2.4.7.5 Overturning Moment
The contact patch is deflected laterally from the centre of the carcass. This creates an overturning mo-
ment 𝑀 due to the offset position of the normal force. The lateral force also contributes to the over-
turning moment.

2.2.5 Combined Longitudinal and Lateral Slip

Operation of vehicles often involves a combination of steering/cornering and braking/propulsion.
Generally, one can experience two causal effects of combined slip:
• Loss of sidegrip due to increased wheel torque or longitudinal tyre slip. Typically, this is when
propulsion or braking directly cause undesired lateral vehicle motion (reduced steerability or
reduced yaw stability).
• Loss of longitudinal grip due to increased lateral tyre force or tyre side slip. Typically, this is
when steering cause undesired wheel rotation, and only indirectly causes undesired longitu-
dinal vehicle motion (reduced acceleration or reduced deceleration).
The first effect is more directly affecting vehicle motion than the latter. So, one can say one have to be
more careful when changing wheel torques in a curve than when changing steering angle while accel-
erating or braking. Engineering for these manoeuvres requires models of vehicle and wheels/shafts
but also tyres. These tyre models have to represent combined slip, i.e. how [ ] varies with (𝑠 𝑠 ).
If the tyre has isotropic adhesion properties in the lateral and longitudinal direction, one can assume
that the maximum force magnitude is determined by the maximum resultant friction force, ∙ .
2 2
2 2 2
+ ≤( ∙ )2 ⇒ ( ) + ( 𝑦 ) ≤ 2 [2.39]
𝑧 𝑧

Equation [2.39] can be plotted as a circle, called the “Friction Circle”. Since the lateral and longitudinal
properties are not isotropic (due to carcass deflection, tread patterns, camber, etc) the shape may be
better described as a “Friction Ellipse” or simply “Friction limit”.
When not cornering, the tyre forces are de- 1
scribed by a position between -1 (braking) and maximum turning left
+1 (acceleration) along the Y-axis. Note that without propulsion and
braking
the scales of both axes are normalized to the 0.5 turning left while
maximum value for friction. braking
partial propulsion
Fy/(mu*Fz)

The “actuation” of the wheel means the pro-

pulsion, braking and steering (and sometimes 0
max propulsion
suspension control) of the wheel. An ideal ac-
tuation allows all conditions within the bound-
-0.5
aries of the friction circle to be achieved any-
time during a vehicle manoeuvre. An example
of limitation in actuation is a wheel on a non- -1
steered rear axle. They cannot access any of
-1 -0.5 0 0.5 1
the lateral parts of the circle; unless the vehi- Fx/(mu*Fz)
cle slides laterally.
Figure 2-47: Friction Circle with some
examples of utilization. View from above,
forces on tyre.
At the boundary of the friction circle, tyres become more sensitive to changes in slip. It is therefore ex-
tra important to model the direction of the force in relation to shear deformation and relative slip mo-
tion in the tyre contact patch. Here, isotropic shear and friction properties are assumed:

133
 Vehicle Interactions and Subsystems

𝑡𝑒𝑟 𝑠 𝑛 𝑠𝑝𝑒𝑒
[2.40]
𝑜𝑛 𝑡𝑢 𝑛 𝑠 𝑛 𝑠𝑝𝑒𝑒 𝑅∙
When considering a combined slip model, one can establish the “total slip” 𝑠 𝑠, through the defini-
tion 𝑠 2
𝑠 2
𝑠 + 𝑠 . Then, it can be tempting to look for a function
2 2
𝑓(𝑠 ) and then decom-
pose into and through [ ] [+𝑠 ⁄𝑠 𝑠 ⁄𝑠 ] . This is not fully physical, which is
easiest understood by looking at the brush model for uniform pressure distribution: When 𝑠 is small
enough, there will be no slip in the contact, because 𝜏 < 𝑝 in whole contact. For such conditions and
isotropic linear deformation model of the bristles, the longitudinal and lateral models from before are
valid. So, is independent of 𝑠 and is independent of 𝑠 . So, using 𝑠 gives a non-physical depend-
ence. Anyway, such approximate models can be useful for conditions with larger slip, see 2.2.5.3.1.

2.2.5.1 Anisotropic Brush Model with Independent Bristles

The longitudinal model Eq [2.9] and the lateral model with independent bristles in 2.2.4.1.1 will now
be used with the anisotropy ≠ . This is motivated by that sidewall contributes more to lateral
bristle elasticity than to longitudinal bristle elasticity. With significantly different tread grooves in lon-
gitudinal and lateral direction, one could motivate also difference in friction coefficient , but here we
assume same in both directions. The derivation of mathematical model comes from the physical
model inFigure 2-48. It becomes very similar to the derivation for the pure longitudinal and lateral
models.

rolling direction view from rear,

showing rubber element
L=length of contact patch on one certain x

tyre 𝑅 W=width
outlet inlet
𝜉
𝜉 𝐿 𝜉 𝜉 𝜉 0
H=height
road (relative
(relative
to wheel
to wheel hub) hub)
Contact patch, Velocities of bristle ends in
view from above: stick zone:
𝑅
force on bristle
𝐻 𝑠 ⁄
𝑠 𝑠 ⁄𝐻 Drawn for 𝑅 > and >0
𝑠 𝑠 𝑠 ⁄𝐻 Positions of bristle ends in
𝐻 𝑠 ⁄ 𝑠 stick zone, for a bristle which
entered contact time 𝑡 ago:
x
𝑅 𝑡
pressure
x stick arctan ⁄ 𝑡
slip 𝑡
“mode” Drawn for 𝑅 > and >0
x 𝜏 𝑙𝑝 longitudinal
𝜏 shear stress Continuous
2
𝜏 across break-away
2
point (𝜏 𝑙 𝑝 + 𝜏 𝑙 𝑝
x 𝜏 𝑙𝑝 lateral 𝜏 2 + 𝜏 2 ), but redistribution
𝜏 towards sliding direction in slip zone.
shear stress
Figure 2-48: Physical model for deriving brush model for combined slip.

134
 Vehicle Interactions and Subsystems

For the stick zone:

𝜏 ∙𝑠 ∙𝜉 𝑛 𝜏 ∙𝑠 ∙𝜉
𝑊 𝐿2 𝑊 𝐿2
If no slip in contact:
𝐿 𝐿 𝐿
𝐿2
𝑊 ∙∫𝜏 ∙ 𝜉 𝑊∙∫ ∙𝑠 ∙𝜉∙ 𝜉 ∙ 𝑠 ∙ ∫𝜉 ∙ 𝜉 ∙𝑠 ∙ ∙𝑠
𝑊 𝐿2 𝐿2 𝐿2
𝐿 𝐿 𝐿
𝐿2
𝑊 ∙ ∫𝜏 ∙ 𝜉 𝑊∙∫ ∙𝑠 ∙𝜉∙ 𝜉 ∙ 𝑠 ∙ ∫𝜉 ∙ 𝜉 ∙𝑠 ∙ ∙𝑠
𝑊 𝐿2 𝐿2 𝐿2
Now, if there is a break-away point (𝜉 ) where slip starts, we can find it from 𝜏(𝜉 ) 𝑝 ⇒ 𝜏2 +
𝜏 2 2
𝑝 . Introducing an auxiliary parameter, 𝑘
2 ⁄ , and an auxiliary variable, 𝑠𝑘

√(𝑘 ∙ 𝑠 )2 + 𝑠 2 , gives the following break-away point:

𝐿
𝜏(𝜉 ) 𝑝 ⇒ 𝜏2 + 𝜏2 2
𝑝2 ⇒ ⋯ ⇒ 𝜉 𝑧
2 𝑦 𝑘

The forces, when 0 < 𝜉 < 𝐿 and > 0 becomes:

𝐿 𝜉 2 𝐿
𝑊∙∫ 𝜏 ∙ 𝜉 𝑊∙∫ ∙ 𝑠 ∙ 𝜉 ∙ 𝜉 + 𝑊 ∙ ∫𝜉 𝜏 𝑙𝑝 𝜉
𝑊 𝐿2
𝜉 2 𝐿 ∙ ∙𝜉 2 ∙ 𝑧
𝑊∙∫ ∙ 𝑠 ∙ 𝜉 ∙ 𝜉 + 𝑊 ∙ ∫𝜉 ∙𝑝 𝜉 ⋯ + (𝐿 𝜉 ) and
𝑊 𝐿2 𝐿2 𝐿
𝐿 𝑛 𝑜 𝑜𝑢𝑠 𝑦 𝑦∙ 𝑦 ∙𝜉 2 ∙ 𝑧 𝑦
𝑊∙∫ 𝜏 ∙ 𝜉 {
} ⋯ + (𝐿 𝜉)
𝑡ℎ 𝑥 𝑟𝑒 𝑡 𝑜𝑛 𝐿2 𝐿
Arranging for all combinations of ( ), we find Eq [2.41].

+ ∙
[ 𝑦
] 𝑓 <0
∙
+ 𝑠 ∙ 𝑧 2 2 ∙ 𝑧
[ 𝑠 ] 𝑒 𝑠𝑒 𝑓 𝑠𝑘 ≤ 2∙ ⇔ √ + ≤
[ ] 𝑦 2
𝜉 2 𝐿 𝜉
+ (( 𝐿 ) ∙𝑠 + ∙ 𝐿
)
[ ] 𝑒 𝑠𝑒 [2.41]
𝜉 2 𝐿 𝜉 𝑦
{ (( 𝐿 ) ∙𝑠 + ∙ 𝐿
)
𝑅 2 2
ℎ𝑒𝑟𝑒 𝑠 𝑠 𝑠 √𝑠 +𝑠
|𝑅 ∙ | |𝑅 ∙ |
𝐿 ∙ 𝑊 ∙ 𝐿2
𝑘 𝑠𝑘 √(𝑘 𝑠 )2 + 𝑠 2 𝜉
𝑠𝑘 ∙𝐻
The terms with (𝜉 ⁄𝐿)2 comes from shear stress in the stick zone, and the direction is in proportion to
the stiffnesses and (anisotrop elasticity). The terms with (𝐿 𝜉 )⁄𝐿 comes from the shear stress
in the slip zone and it is counter-directed to the relative velocity between the surfaces (isotropic fric-
tion).
Figure 2-49 shows results from the model. It can be observed that is independent of 𝑠 and is in-
dependent of 𝑠 for ≤ ∙ ⁄𝑠. This is a reasonable consequence of that no sliding occurs so that
forces are purely defined by the elasticity, not the friction. At 𝑠 ≈ 0.0 . .0.04, we see that increases
with utilization of | | at some areas. This is a redistribution of force from longitudinal to lateral, due
to that the tyre stiffer in longitudinal than lateral.

135
 Vehicle Interactions and Subsystems

𝑠 sign 𝑅

leaning
Anti-symmetric,

𝑠
quadrant
in the 3rd

𝑠
Symmetric, in
quadrant
the 2nd

Figure 2-49: Tyre (𝑠) with iso-curves for longitudinal slip 𝑠 and lateral tyre slip 𝑠
⁄|𝑅𝑤 | (5 left diagrams) and for 𝑠 (𝑅𝑤 )/|𝑅𝑤 | and slip angle
𝑟 𝑡 𝑛 ( ⁄ ) (5 right diagrams). Tyre model from Eq [2.41] used. Levels used for slip and
𝑡 𝑛(𝑠 𝑝 𝑛 𝑒): 0 ±0.0 ±0.0 ±0.05 ±0. ±0. ±0. ±0.4.
∙ 𝑧
Note that, for √ 2
+ 2
≤ 2
, is independent of 𝑠 and of 𝑠 . This is because for such low
slip and uniform pressure distribution, there is no slipping part in the contact patch. It is the slipping
part that creates the ”cross-dependence” that depends on 𝑠 and on 𝑠 . The independence ap-
pears as the perpendicular iso-𝑠 and iso-𝑠 curves for < in the upper diagrams in Figure
2-49.

136
 Vehicle Interactions and Subsystems

For small 𝑠 , it is still a good model that (𝑠 ) 𝑠 and (𝑠 ) 𝑠 . Then: 𝑠

⁄(𝑅𝑤 ) {𝑠 (𝑅𝑤 )⁄(𝑅𝑤 )} ( 𝑠 ) ⁄ ( 𝑠 ) 𝑠 𝑤.
Increasing 𝑠 from 0 to small positive 𝑠 means reduction of | |, which should be intuitive since utili-
zation of friction in one direction reduce force in the other. However, decreasing 𝑠 from 0 to small
negative 𝑠 means increase in | |, which can be counter-intuitive. However, the explanation is that the
“transport speed” 𝑅 decreases, which means a “slip stiffer” tyre. The increase in for small brak-
ing is seen in Figure 2-50: slight brake torque on a wheel improves lateral grip!

Figure 2-50: Combined Longitudinal and Lateral Slip.

2.2.5.2 Anisotropy, Parabolic Pressure, Stick and Slip Friction,

Independent Bristles
Section 2.2.5.1 is for uniform pressure distribution and same friction coefficient for stick and slip. We
will now change to the more realistic parabolic distribution and replace with 𝑘 and 𝑙 𝑝 . The
model gets one more parameter and therefore more tunable to experiments, see Eq [2.42].
+ 𝑙𝑝 ∙ 3 𝑖 𝑘∙ 𝑧
[ ] 𝑓 < 0 𝑜𝑟 𝑠𝑘 >
𝑦 𝑦
𝑙𝑝 ∙
[ ] 𝜉 2 𝜉 2 𝜉 3
+( ∙𝑠 ∙( ) + 𝑙𝑝 ∙ ( ( ) + ( ) ) )
𝐿 𝐿 𝐿
[ 2 2 3 ] 𝑒 𝑠𝑒
𝜉 𝜉 𝜉
( ∙ 𝑠 ∙ (𝐿) + 𝑙𝑝∙ ( ( ) + ( ) )
𝑦
) [2.42]
{ 𝐿 𝐿
𝑅 2 2
ℎ𝑒𝑟𝑒 𝑠 𝑠 𝑠 √𝑠 +𝑠 𝑘
|𝑅 ∙ | |𝑅 ∙ |
𝜉 𝑠𝑘 ∙ 𝑊 ∙ 𝐿2
𝑠𝑘 √(𝑘 𝑠 )2 + 𝑠 2
𝐿 𝑘 ∙𝐻
Eq [2.42] is plotted for steady state conditions in Figure 2-51.
§ However, using Eq [2.42] in a vehicle simulation, often becomes singular. This is either when 𝑅 ∙
0 (locked wheel, which means that 𝑠 and 𝑠 becomes not physically interpretable) or when
(𝑅 )2 + 2 0 (direction of relative velocity becomes not physically interpretable). A way to
avoid such computational problems is to introduce a small positive velocity parameter, 𝜀 , and divde
by max(|𝑅 ∙ | 𝜀 ) instead of |𝑅 ∙ | and max(√(𝑅 )2 + 2 𝜀 ) instead of √(𝑅 )2 + 2.
However, one should keep in mind that this gives a tyre model which cannot fully stop if the road has a
slope, longitudinally or laterally. It also does not model that the break-away force for a non-rotating
tyre on a stand-still vehicle is 𝑘 . If one need to model such, e.g. for start in uphill or
banked road, the tyre model needs to be developed to a dynamic model with discrete states.

137
 Vehicle Interactions and Subsystems

leaning
𝑠 sign 𝑅
Anti-symmetric,

𝑠
quadrant
in the 3rd

𝑠
sign 𝑅
sign 𝑅
Symmetric, in

quadrant
the 2nd

𝑠
𝑠

Figure 2-51: Tyre (𝑠) with iso-curves for longitudinal slip 𝑠 and lateral tyre slip 𝑠
⁄|𝑅𝑤 | (5 left diagrams) and for 𝑠 (𝑅𝑤 )/|𝑅𝑤 | and slip angle
𝑟 𝑡 𝑛 ( ⁄ ) (5 right diagrams). Tyre model from Eq [2.42] and Eq [2.47] used. Levels used for
slip and 𝑡 𝑛(𝑠 𝑝 𝑛 𝑒): 0 ±0.0 ±0.0 ±0.05 ±0. ±0. ±0. ±0.4.
If 𝑠 is varied while 𝑠 0 in Eq [2.42], leads exactly to the longitudinal slip model with peak (Eq
[2.17]). Similar peak in is found for varying 𝑠 while 𝑠 0. The peaks in and are equally large,
but the peak in occurs at larger slip: 𝑠 𝑝 𝑘 > 𝑠 𝑝 𝑘 . Also for combined slip, a maximum in oc-
curs, but on a “2-dimensional ridge” in the “cake plot” in Figure 2-52.

138
 Vehicle Interactions and Subsystems

Figure 2-52: Plots of the vector field [ ] 𝑓([𝑠 𝑠 ]) or 𝑓 (𝑠), using the results from
Figure 2-51. “Cake-plot” is an expression from (Weber, 1981).
2.2.5.2.1 § Some Specific Operating Conditions

𝑣
𝑅 . 3
≈ 0.9709
𝑅
0.
tan 0. ≈
≈ 0. 00

0
0. 𝑟

+
Figure 2-53: § Part from Figure 2-51. Shows specific conditions: (rotationally) locked ( 0),
infinite spin ( ± ), and 𝑥| |. The 𝑥| | appears at slight negative ≈ 500 𝑁.

2.2.5.3 Approximate Combined Slip Models

2.2.5.3.1 Using Scalar Force Function of Combined Slip
A simple but not fully physically consistent combined slip model can be expressed using “total slip,
𝑠 ”, as shown in Eq[2.43].
+𝑠 𝑅 ∙
[ ] [ 𝑠 ] [ 𝑤 ]
𝑠 2
√(𝑅𝑤 ∙ )2 + ( )
𝑅𝑤 ∙
ℎ𝑒𝑟𝑒 𝑠 2 𝑠2 𝑠2 + 𝑠2 ℎ𝑒𝑟𝑒 𝑠 𝑛 𝑠
|𝑅𝑤 ∙ | |𝑅𝑤 ∙ |
𝑓 <0 [2.43]
𝑛 {
𝑓( 𝑠 ) 𝑒 𝑠𝑒
ℎ𝑒𝑟𝑒 𝑒. . 𝑓 {Eq [2.9], [2.14], [2.16] or [2.17]} or
𝑓 min( ∙𝑠 ) or 𝑓 tanhyp (( ⁄ )∙𝑠 )
ℎ𝑒𝑟𝑒 ≈ {𝑡𝑦𝑝 𝑦} ≈ 5. . 0 [𝑁⁄𝑁]

139
 Vehicle Interactions and Subsystems

A problem with Eq[2.43] is that anisotropy cannot be represented. Inspired by Eq [2.41] and Eq [2.42],
we can adjust it to Eq [2.44], which has only one example of function 𝑓. Eq [2.44] is not fully physically
motivated. However, it can be tuned to experiment data which has different and .
+𝑠 𝑅 ∙
[ ] [ 𝑠 ] [ 𝑤 ]
𝑠 2
√(𝑅𝑤 ∙ )2 + ( )
𝑅𝑤 ∙
ℎ𝑒𝑟𝑒 𝑠 2 𝑠2 𝑠2 + 𝑠2 ℎ𝑒𝑟𝑒 𝑠 𝑛 𝑠
|𝑅𝑤 ∙ | |𝑅𝑤 ∙ |
𝑓 <0 [2.44]
𝑛 {
𝑓( 𝑠𝑘 ) 𝑒 𝑠𝑒
ℎ𝑒𝑟𝑒 𝑒. . 𝑓 min( 𝑠𝑘 )

ℎ𝑒𝑟𝑒 𝑠𝑘 √(𝑘 𝑠 )2 + 𝑠 2 𝑘 ⁄

ℎ𝑒𝑟𝑒 𝑡𝑦𝑝 𝑦 ≈ 5. . 5 [𝑁⁄𝑁] 𝑛 𝑘 ≈ .5

§ Combined Slip Model on Inverted Form
A common usage of combined slip is that ( ) are known and we look for [ ]
𝑓𝑢𝑛 ( ) . We will try to find that function.
𝑅𝑤
𝑠 𝑠 𝑠2 𝑠2 + 𝑠2
|𝑅𝑤 | |𝑅𝑤 |
𝑠 𝑠
tanhyp ( 𝑠 ) +
𝑠 𝑠
So, 7 equations and 8 unknowns (𝑠 𝑠 𝑠 ). This is because there is no unique solu-
tion, but the solution is a curve in space. To simplify the example, we assume > 0. Then we
replace 𝜔 with ⁄|𝑅 | and 𝜔 with ⁄|𝑅 |. By this, disappears.
𝜔
𝑠 𝜔 𝑠 𝜔 𝑠2 𝑠2 + 𝑠2
𝜔
𝑠 𝑠
tanhyp ( 𝑠 ) +
𝑠 𝑠
So, 7 equations and 7 unknowns (𝑠 𝑠 𝑠 𝜔 𝜔 ). However, the equation system is trans-
cendent so a symbolic solution can not be found, cf. sin(𝑥) 𝑥 can not be symbolically solved.
2.2.5.3.2 Friction Circle Inspired Combined Slip Model
A combined slip model for cases when one knows without involving 𝑠 is shown in Eq [2.45]. It is
useful when we know the wheel torque, e.g. by prescribed propulsion or braking and do not consider
the wheel rotational inertia, then 𝑇/𝑅𝑤 . Eq [2.45] also shows the corresponding, less usual, case
when is known without involving 𝑠 . These models are not fully physically motivated, but can in-
stead be considered “a mathematical scaling, inspired by the friction ellipse-like curves in Figure
2-51”. It is a rather good approximation when ≈ . But for ≈ 0, it is less good to scale with
since the forces and are then more dependent of the shear elasticity than of friction and
consequently rather independent of each other.
2 2
√ ( ) ∙ | 𝑛 √ ( 𝑦
) ∙ | 𝑦= 𝑦=
[2.45]
∙ 𝑧 = = ∙ 𝑧

If this concept of scaling is applied on 𝜕 ⁄𝜕𝑠 , we get:

𝜕 𝑦 2 𝜕 2
𝑦
√ ( ) ∙ | 𝑛 √ ( ) ∙ | 𝑦= 𝑦=
𝜕 𝑦 𝑧 = = 𝜕 𝑧
𝜕 𝑦
[2.46]
𝜕
ℎ𝑒𝑟𝑒 | 𝑛 |
𝜕 𝑦 𝜕 =
𝑦=

140
 Vehicle Interactions and Subsystems
|
§ Comparison with Physically Motivated Model

𝑅 𝑅 𝑅
𝑠 𝑠 𝑠

;
𝑅 𝑠 𝑅 𝑠 𝑅 𝑠

𝑠 𝑠 𝑠
𝑠 𝑠 𝑠 ≈
𝑅 𝑅

Figure 2-54: § Left and middle shows that causality can be changed for a physically motivated
model, such as in Eq [2.41] and Eq [2.42], even if the middle generates an algebraic loop. The right
shows how Eq [2.45] and Eq [2.46] “cheats”, by using an 𝑠 definition which assumes 𝑅 ≈ ,
which is only motivated without combined slip.

2.2.5.4 Influence of Vertical Force

Contribution from Niklas Fröjd, Volvo Trucks

Figure 2-55 shows experiment data on how slip stiffness varies with vertical force.
Contact lenght [mm]

Lateral,

Figure 2-55: Measurements for varying vertical force. Left: Slip stiffnesses. Right: Contact length.
2.2.5.4.1 Explanation of Higher Slip Stiffness Longitudinal than Lateral
The model in 2.2.4.1.1.1 explains why tyre is more slip stiff in longitudinal than lateral direction, i.e.
why > . We can then assume iso-tropic brush bristles in both friction and shear stiffness, i.e.
⇒ and 𝑤𝜀 𝑛 ⁄( 𝑤 𝜀 𝑛 + ) . From experiments, we typi-
cally find ⁄ { 𝑡 𝑛𝑜 𝑛 𝑁 𝑚 } 𝑁 𝑚 ⁄ 𝑁 𝑚 𝑁 𝑚⁄ 𝑁 𝑚 𝑘𝑁 𝑚 .5 . . .
This could be explained with the model if 𝑤 𝜀 𝑛 ⁄ ⁄(𝑘𝑁 𝑚 ) . . . This shows that neither
of 𝑤 𝜀 𝑛 nor is neglectable.
2.2.5.4.2 Model for how Lateral Slip Stiffness is Degressive with Vertical Force
The model in 2.2.4.1.1.1 can also explain why lateral slip stiffness is degressive with , as indicated
already in 2.2.3.1.3. The was found to be proportional to , due to that contact length increase pro-
portional to √ . A very conceptional reasoning of how 𝑤 𝜀 𝑛 could vary with follows now: We de-
fined 𝑤 𝜀 𝑛 𝜀 𝑛 ⁄ . For a fix lateral deformation 𝜀 𝑛 ≈ ⁄𝐿 and ≈ 𝐿. So, 𝑤 𝜀 𝑛 ⁄𝐿2
⁄ . With and 𝑤 𝜀 𝑛 ⁄ , and given 𝑁 𝑚 and 𝑁 𝑚 at given nominal
𝑁 𝑚 , the model gives Eq [2.47] and Figure 2-56. So, by knowing Nom and measuring 𝑁 𝑚 and
𝑁 𝑚 one gets 𝑘𝑁 𝑚 as 𝑘𝑁 𝑚 𝑁 𝑚 ⁄ 𝑁 𝑚 and thereby a quantified model of ( ) and ( ).

141
 Vehicle Interactions and Subsystems

𝑛
𝑁 𝑚
𝑁 𝑚 𝑘𝑁 𝑚 2
𝑁 𝑚 [2.47]
𝑁 𝑚 +
𝑁 𝑚

Figure 2-56: Model with Dependent Bristles explains that Lateral Slip Stiffness is degressive with
. The is (direction independent) slip stiffness due to tread (𝑡𝑟).

2.2.5.5 Transients and Relaxation in Contact Patch

The models above for combined slip assume steady state conditions. For separate longitudinal and lat-
eral slip, there is a delay in how fast the steady state conditions can be reached, which is sometimes
important to consider. A similar model, as for relaxation in longitudinal direction 2.2.3.5.1, is to add a
first order delay of the force:
[ ] [𝐴 ∙ (𝑓 (𝑠 𝑠 …) ) 𝐴 ∙ (𝑓 (𝑠 𝑠 …) )]
where [𝑓 𝑓 ] are the forces according to steady state models and [2.48]
[𝐴 𝐴 ] are relaxation lengths, as defined in Eq [2.20] and Eq [2.35].
Most reasoning in 2.2.3.5 is applicable also for combined slip relaxation.

2.2.6 Summary of Tyre Force vs Slip Models

Categorization of tyre models for longitudinal force and slip:
• No lateral slip:
o Ideally rolling (𝑅 leaving to be defined by other than the tyre)
o Linear + 𝑠 + 𝑠
o Saturated due to road friction:
▪ Simplest, saturated linear: + sign(𝑠 ) min( |𝑠 | )
▪ General: 𝑓( 𝑠 )
• Influence from lateral slip: 𝑓(𝑠 )
Categorization of tyre models for lateral force and slip:
• No longitudinal slip:
o Ideally tracking ( 0 leaving to be defined by other than the tyre)
o Linear 𝑠 𝑠
o Saturated due to road friction:
▪ Simplest, saturated linear: sign(𝑠 ) min( |𝑠 | )
▪ General: 𝑓( 𝑠 )
• Influence from longitudinal slip: 𝑓(𝑠 )
A general Combined force and slip model: [ ] 𝑓(𝑠 𝑠 ) or [ ] 𝑓( ).

142
 Vehicle Interactions and Subsystems

2.2.7 Vertical Properties of Tyres

The most important vertical property of a tyre is probably the stiffness. It mainly influences the verti-
cal dynamics, see Chapter 5. For normal operation, the vertical force of the tyre can be assumed to
vary linearly with vertical deflection. If comparing a tyre with different pressures, the stiffness in-
creases approximately linear with pressure. See Figure 2-57.

60000

60
0
0

Figure 2-57: Vertical properties of a truck tyre.

Figure 2-58: Different tyre models which will filter road irregularities differently. Picture from
Peter Zegelaar, Ford Aachen.

2.2.8 Tyre Wear

There are many other aspects of tyres, for instance the wear. Wear models are often based around the
Archard’s (or Reye’s) wear hypothesis worn material is proportional to work done by friction, i.e. fric-
tion force times sliding distance. Wear rate (worn material per time) is therefore friction force times
sliding speed. Different approaches to apply this to tyres and expanding to temperature dependency
etc. is found for instance in Reference (Grosch & Schallamach, 1961). A generalization of 𝑊𝑒 𝑟𝑅 𝑡𝑒 [in
mass/s or mm tread depth/s], for one certain tyre at certain temperature, becomes as follows:

143
 Vehicle Interactions and Subsystems

𝑊𝑒 𝑟𝑅 𝑡𝑒 𝑟 𝑡 𝑜𝑛 𝑜𝑟 𝑒 ∙ 𝑆 𝑛 𝑆𝑝𝑒𝑒 ⇒ 𝑊𝑒 𝑟𝑅 𝑡𝑒 𝑘𝑤 𝑣 ∙ ∙ ≈

)2 + 2 )2 + 2
≈𝑘∙ ∙ √(𝑅 ≈ 𝑘𝑤 𝑣 ∙ ( ∙ 𝑠) ∙ √(𝑅 ≈

≈ 𝑘𝑤 𝑣 ∙ ( ∙ 𝑠) ∙ (𝑠 ∙ 𝑇 𝑛 𝑝 ) ⇒ 𝑊𝑒 𝑟𝑅 𝑡𝑒 ≈ 𝑘𝑤 𝑣 ∙ ∙ 𝑠2 ∙
[2.49]
2 2
where √ + 𝑠 √𝑠 2 + 𝑠 2; 𝑇 𝑛 𝑝 defined as in Eq Error!
Reference source not found. and 𝑘𝑤 𝑣 is a constant for a certain tyre with a
certain temperature, rolling on a certain road surface, which characterises
the wear averaged over the contact patch.

2.3 Suspension System

Suspension can mean suspension of wheels (or axles), suspension of sub-frame and drivetrain and
suspension of cabin (for heavy trucks). In this compendium, only wheel and axle suspension are con-
sidered. Suspension design is explained in 2.3, but also used in complete vehicle models in 3.4.5-3.4.7,
4.3.10, and 4.5.3.
Suspension influences road grip and ride comfort, so merely all vehicle motion, Chapters 3..5. The in-
fluence is through how vertical forces and camber and steer angles on the wheels change with body
motion (heave, roll, pitch), road unevenness (bumps, potholes, waviness) and wheel forces in
ground plane (from Propulsion, Braking and Steering subsystems). Important is also that suspension
influences the material stresses (extreme values and fatigue), both in the suspension itself and other in
other parts of the vehicle body. Figure 2-59 shows one way to see the suspension systems role.

motion of sprung body

driver, above each wheel
environment
(except road surface)
(sprung)
WhlTorques, suspension body
Wheel (linkage,
prop, brk, ste AxleSteAngles & Tyre elasticities,
systems (HW&SW) dampers)

WhlRotSpeeds, forces on vehicle from each

vertical Fy Fy tyre to ground contact
AxleSteForces
displacement
Fz Fx Fz Fx
under each wheel

road surface
Figure 2-59: Wheel/axle suspension described as modular sub-model per axle. It may be noted that
both wheel model (main geometry such as wheel radius) and tyre model (how and vary with
tyre slip and ) is a part of each Wheel & Tyre sub-model.
The simplest view we can have of a suspension system is that it is an individual suspension between
the vehicle body and each wheel, consisting of one linear spring and one linear damper in parallel.
Chapter 5 uses this simple view for analysis models, because it facilitates understanding and it is
enough for a first order evaluation of the functions studied (comfort, road grip and fatigue load) dur-
ing normal driving on normal roads.
The full 3D aspect of suspension is not covered here in 2.3. Instead, a division into 2D is done in 2.3.3
Suspension -- Heave and Pitch and 2.3.4 Suspension -- Heave and Roll, aiming at Longitudinal and Lat-
eral dynamics, respectively. The full 3D aspects are briefly addressed in 4.5.3.2.5 and 0.

144
 Vehicle Interactions and Subsystems

2.3.1 Components in Suspension

Each wheel can rotate in its hub. Each hub can be individually suspended to the body or left and right
hub can be mounted on a rigid beam which is suspended to the body. The suspension parts are below
grouped in: Linkage, Elasticities and Dampers. One might count in additional parts in the suspension,
such as bearings, shafts, brake parts, etc.

2.3.1.1 Linkage
Linkage, which has the purpose to constrain the relative motion between wheel and body via kinemat-
ics to one dof (approximately vertical translation), or, for a steered axle, also allow one more dof (ap-
proximately yaw rotation) per axle. The linkage defines how longitudinal and lateral tyre forces are
brought to the body (sprung mass).
The linkage consists of links (or members) and joints; mainly ball joints, but sometimes others, such as
hinge joints. The coordinates (“hard-points”) of these joints are the real design parameters, but the dy-
namic behaviour of a complete vehicle model can be expressed in much fewer parameters, namely the
“effective pivot points”. These effective points are used in 2.3.3 and 2.3.4.

2.3.1.2 Elasticities or Compliances

Springs are examples of elasticities or compliances. The springs develop forces when the wheels are
vertically displaced relative to the body. There is often one spring per wheel but also an anti-roll bar
per axle. The anti-roll bar connects left and right wheel to each other to reduce body roll.
Springs often have a rather linear relation between the vertical displacement and force of each wheel,
but there are exceptions:
• Anti-roll bars make two wheels dependent of each other (still linear). Anti-roll bars can be
used on both individual wheel suspensions and rigid axle suspensions.
• The springs are intentionally designed to be non-linear in the compressed end of their stroke
with bump stops. Bump stops at passenger cars are typically designed at ( .5. .4) vertical
acceleration when vehicle is fully loaded. (A somewhat opposite non-linearity appears in the
rebound end of the stroke, due to wheel lift. Here it is the contact force with ground that is sat-
urated to zero, not the spring force. The difference is the damping force.)
• In context of bump stop, it is relevant to mention an almost opposite phenomena in the other
end of the deformation range; a wheel cannot take vertical pulling forces to the ground. This is
called wheel lift, and it is the saturation of force under the tyre to zero, as opposed to becom-
ing negative.
• The compliances can be non-linear during the whole stroke, e.g. air-springs and leaf-springs.
Air-springs are non-linear due to the nature of compressing gas, e.g. assuming ideal gas: 𝑝 𝑉
𝑛 𝑅 𝑇 ⇒ 𝑜𝑟 𝑒 𝑛 𝑅 𝑇 ⁄(𝐿 𝑜 𝑝𝑟𝑒𝑠𝑠 𝑜𝑛) .
• The compliances can be controllable during operation of the vehicle. This can be to change the
pre-load level to adjust for varying roads or varying weight of vehicle cargo or to be controlla-
ble in a shorter time scale for compensating in each oscillation cycle. The latter is very energy
consuming and no such “active suspension” is available on market.
The springs are the main compliance, but also other smaller compliances are present and makes the
effective stiffness lower: the links themselves, the bushings in the joints between the links and the
brackets where the links are connected to the body. For the complete vehicle model, the tyres vertical
compliance adds to the suspension compliance.

2.3.1.3 Dampers
Dampers have the purpose to dissipate energy from any oscillations of the vertical displacement of the
wheel relative to the body. High damping damps oscillations, but high damping also increases the
shock force transmittance (with this reasoning, the name “shock-absorber” is misleading). The most
common design is the hydraulic piston type. Simplest understanding or modelling is a linear relation

145
 Vehicle Interactions and Subsystems

between the vertical deformation speed and force. But the following non-linearities should be men-
tioned:
• The dampers are normally intentionally designed to be different in different deformation di-
rection. Typical values for passenger cars are about 3 times more damping in rebound (exten-
sion) than compression (bump). For heavy vehicles, the difference is often even larger, maybe
a factor 5 or 10.
An intuitive explanation to the asymmetry can be found by comparing driving over different
transient unevenness, a bump or a hole. In both cases, large spring stiffness and large damping
cooperate to transfer the road unevenness and worsen comfort. But average damping cannot
be too low, since damping damps oscillations. When driving over a bump, the force can be very
large due to the bump stop in the spring. When driving over a hole, the force is limited by
wheel lift from ground. The spring, including bump stop and wheel lift, is therefore generally
stiffer for a bump than for a hole. So, to compensate for this, the damping is made opposite:
more damping when driving over a hole, i.e. rebound (extension).
• Damping in leaf springs is non-linear since they work with dry friction.
• The dampers can be designed to be controllable during operation of the vehicle. This can be
used to change the damping characteristics to adjust for varying roads or varying weight of ve-
hicle cargo or to be controllable in a shorter time scale for compensating in each oscillation cy-
cle. The latter is called “semi-active suspension” and is available on some high-end vehicles on
market.

2.3.2 Axle and Wheel Rates

All compliances (springs, bushings, etc) contribute to the stiffness between the body and the wheels.
The wheels are not independent of each other; we have especially a connection between left and right
due to the anti-roll bars. Therefore, we define both axle (compliance) rate and wheel (compliance)
rate, see Figure 2-60. We can also call these Effective stiffnesses of the axle and wheel, respectively.
Test to define and/or determine
axle (stiffness) rate, 𝒄𝒂𝒙𝒍 :
• Fix the sprung body and the wheels on
sprung the other axle(s)
• Lift both wheel on the axle with
body 𝑧
𝑙
• 𝑙
𝑟

suspension Test to define and/or determine

wheel (stiffness) rate, 𝒄𝒘𝒉𝒍 :
• Fix the sprung body and all other wheels
• Lift the wheel with 𝑤 𝑙 (slowly
to not engage damping)
left wheel right wheel • 𝑤 𝑙 𝑤 𝑧

on axle i on axle i 𝑟 𝑤 𝑙

Note: Lift slowly to avoid engaging (speed dependent) damping.

Check also lowering, to detect if dry friction damping.
Figure 2-60: Definition of axle and wheel (stiffness) rates.
Alternatively to lifting one wheel, the axle can be rolled (without lifting the midpoint of the axle):
𝑙 𝑓 𝑊⁄ and 𝑊⁄ . It is then natural to find the measure axle
roll stiffness 𝑙𝑙 [𝑁 ⁄𝑟 ] 𝑤 𝑙 𝑊/ , instead of 𝑤 𝑙 [𝑁⁄ ]. If the vehicle is symmetric and
track width is known, these measures carry the same information and 𝑙𝑙 𝑤 𝑙 𝑊 ⁄ .
2

With these definitions, the vertical wheel forces for a two-axle vehicle will be as follows. It is assumed
that the vehicle is symmetrical and that there are connections between the wheels only as anti-roll
bars on each axle. The body is fixed and road under the wheels are displaced with . Notation 𝑙 is

146
 Vehicle Interactions and Subsystems

axle rate and 𝑤 𝑙 is wheel rate for axle . The time derivative of spring force and vertical veloci-
ties are used to avoid involving pre-tension in the springs.
𝑙 𝑙 𝑙
𝑪 𝟎 𝑤 𝑙 ( 𝑤 𝑙 )
[ 𝒘𝟏 ] [ ] ℎ𝑒𝑟𝑒 𝑪𝒘𝒊 [ ]
𝟎 𝑪𝒘𝟐 2𝑙 𝑙
2𝑙 ( 𝑤 𝑙 ) 𝑤 𝑙
[ 2 ] 2

If both body and road under the wheels is moving, we simply exchange [ 𝑙 2𝑙 2 ]𝑇 with
[ 𝑙 2𝑙 2 ] 𝑇 [ 𝑙 2𝑙 2 ]𝑇
. The axle roll stiffness becomes 𝑙𝑙
( ℎ 𝑥 ⁄ 4) 𝑊 [Nm/rad] for 𝑀 𝑙 .
Similarly, for axle , we can define axle (damping) rate 𝑙 , wheel (damping) rate 𝑤 𝑙 and axle roll
damping 𝑙𝑙 . However, 𝑙 is often simply 𝑤 𝑙 since there are typically no dampers connect-
ing left and right wheel. Corresponding 4 × 4 damping matrix becomes a diagonal matrix, since 𝑙
normally is 𝑤 𝑙.
If we know the design parameters (stiffness and location of the actual spring) we can calculate the
rates. This will be exemplified in 2D and briefly discussed for general (3D) below.

2.3.2.1 Explanation in 2D
A very simplified suspension is assumed in Figure 3-30. The stiffnesses 𝑓 and are the axle spring
rates (or effective stiffnesses of each axle). The real spring may have another stiffness 𝑙 , but its effect
on vertical force is captured in the effective stiffness. An example of how the effective stiffness is found
for a 2D model (heave and pitch) from a real suspension design is given in Figure 2-61.
Transverse (Panhard) rod Strut
Shock absorber rod 𝑝 𝑛 𝑝
Coil 2 3
1 spring
𝑝

Drive
axle a

Drum Control arm

physical spring
4 “effective suspension”
5
virtual spring
with stiffness 𝑝 𝑛 with stiffness
𝑝 𝑛
𝑝

“trivial
suspension”

6a 6b
Moment equilibrium of arm:
𝑝 𝑛 𝑝 ⇒ 𝑝 𝑛 𝑝
𝑝
Compatibility: 𝑟𝑟
𝑟
Equivalence in stiffness:
2
⇒ 𝑝 𝑛 ⇒ 𝑝 𝑛
𝑟 𝑟 𝑟
Figure 2-61: From suspension design spring stiffness to effective stiffness. 6a: Final Mathematical
model. 6b: Interpretation (“Reverse modelling”) back from Mathematical to Physical model,
showing “Trivial (linkage) suspension”.

147
 Vehicle Interactions and Subsystems

Note that the factor ( / )2 is not the only difference between effective and real stiffness, but the ef-
fective can also include compliance from other parts than just the spring, such as bushings and tyre.
There will also be a need for a corresponding effective damping coefficient, see 3.4.5.2.2, or axle
(damping) rate. How forces in road plane is transferred is not well captured in this model, see in
Figure 2-61. Compared to suspension models later in compendium, Figure 2-61 ends with a “trivial
(linkage) suspension model”. In 3.4.5.2, the suspension linkage is better modelled, which allows valid-
ity for (propulsion and braking). Corresponding for is modelled in 4.3.10.3.

2.3.2.2 § Expansion to 3D
It should be noted that above is one of 2 possible 2D views of explaining effective stiffness. A similar
reasoning could be done in view from rear. Then we would see the same physical spring but other lev-
ers ( and ). So, that view would give another value of effective stiffness. A realistic model for both
and has to use pivot axis in 3D instead. If a prototype is available, one can also measure effective
stiffness, because it is the actual stiffness that one would feel if one tries to lift the wheel by grabbing
the tyre-to-ground contact surface. In a first approach one can explain a 3D model as moment equilib-
rium around a moment-free pivot axis. However, for a general 3D motion, there should be a screw
joint along the pivot axis, see mathematical formulation of equilibrium in 4.5.3.2.5. The lead of the
screw is often assumed to be zero and it is no examples below where a non-zero lead is identified in an
example suspension.

2.3.3 Suspension -- Heave and Pitch

2.3.3.1 Examples of Suspension Designs
In Figure 3-32, a “trailing arm” is drawn both for front and rear axle. For rear axle, that is a realistic de-
sign even if other designs are equally common. However, for front axle a so-called McPherson suspen-
sion is much more common, see Figure 2-62. Figure 2-63 shows a suspension for a heavy vehicle.

2.3.4 Suspension -- Heave and Roll

2.3.4.1 Examples of Suspension Designs
There are axles with dependent wheel suspensions, which basically look as the roll centre axle model
in Figure 4-37 , i.e. that left right wheel are rigidly connected to each other. Then, there are axles with
dependent wheel suspensions, which look more like the model with wheel pivot points in Figure 4-37.
For these, there are no (rigid) connections between left and right wheel.
Many axles have a so-called anti-roll bar, which is an elastic connection between left and right side. It
is connected such that if the wheel on one side is lifted, it lifts also the wheel on the other side. Note
that, if an anti-roll bar is added to an independent wheel suspension it is still called independent, be-
cause the connection is not rigid.

gf
90 deg
ef

ef

gf
Figure 2-62: Example of typical front axle suspension, and how pivot point is found. The example
shows a McPherson suspension. From Gunnar Olsson, LeanNova.

148
 Vehicle Interactions and Subsystems

Non reactive Level sensor Reactive

Air spring

V-stay

Stabilizer bar

Bump stop

Air spring member

V-stay Spring link
bracket
U-bolts
Anti-roll bar link Spring link bearing
Shock absorber
Reaction rod Spring hanger bracket
Reaction rod bracket Anti-roll bar

=0
𝑇 𝑇

Figure 2-63: Axle suspensions/installations for double rear axle heavy vehicles.
Figure 2-64 and Figure 2-65 show design of two axles with independent wheel suspensions. Figure
2-66 shows an axle with dependent wheel suspension. These figures show how to find wheel pivot
points and roll centre. In the McPherson suspension in Figure 2-65, one should mention that the strut
is designed to take bending moments. For the rigid axle in Figure 2-66, one should mention that the
leaf spring itself takes the lateral forces. Symmetry between left and right wheel suspension is a rea-
sonable assumption and it places the roll centre symmetrically between the wheels, which is assumed
in the previous models and equations regarding roll centre.
Pivot point
motion for points for wheel
90 deg
moving with hub

motion for a point

moving with hub, where From the other
the wheel which is in 90 deg wheel on same
contact with ground Roll centre
for axle axle

90 deg
Figure 2-64: Example of how to appoint the pivot point for one wheel, and roll centre height, for an
axle with double wishbone suspension.

149
 Vehicle Interactions and Subsystems

motion for
90 deg
points moving
with hub Pivot point
for wheel
motion for a
point moving
with hub, where
the wheel which
is in contact
with ground
From the other
90 deg Roll centre wheel on same
for axle axle

90 deg

Figure 2-65: Example of how to appoint the pivot point for one wheel, and roll centre height, for
axle with double McPherson suspension.
Generally, a “rigid axle” gives roll centre height on approximately the same magnitude as wheel radius,
see Figure 2-66. With individual wheel suspension one has much larger flexibility, and typical chosen
designs are 30..90 mm front and 90..120 mm rear.

Figure 2-66: Example of how to appoint the pivot point for one wheel and roll centre for axle with
rigid axle suspended in leaf springs. From (Gillespie, 1992).
The target for roll centre height is a trade-off. On one side, high roll centre is good because it reduces
roll in steady state cornering. On the other side, low roll centre height is good because it gives small
track width variations due to vehicle heave. Track width variations are undesired, e.g. because it
makes the left and right tyre lateral force fight against each other, leaving less available friction for lon-
gitudinal and lateral grip. Roll centre is normally higher rear than front. One reason for that is that the
main inertia axis leans forward, and parallelism between roll axis and main inertia axis is desired.

2.4 Propulsion System

A generalised propulsion system is shown in Figure 2-67, along with a specific example of a conven-
tional one. There are 1+2=3 control degrees of freedom marked for the generalized one (e.g. engine
power, transmission ratio and storage power) while there is only 1+1=2 for the conventional. The
larger number of control degrees of freedom comes from that there are 2 drivetrains, e.g. an ICE as
prime mover and an electric machine inside the transmission. Most of 2.4 assumes a single drivetrain,
except Figure 2-67 and 2.4.7, which assumes multiple drivetrains. This compendium distinguishes be-
tween propulsion systems and drivetrains, where there is exactly one propulsion system for a whole
vehicle, but it can consist of one or more drivetrains.
Note that the approach in Figure 2-67 uses a one-wheel 1D longitudinal model, see 1.6.2.2.1.

150
 Vehicle Interactions and Subsystems

Conventional propulsion system

Driver

e-throttle control other systems, e.g. brake

(1)

𝑇 𝑇
(on one
driven axle)
Body
Prime mover
(ICE, internal Wheels
combustion
engine)
Transmission
(Clutch & Gearbox & Final gear)

Transportation Human/Automated Generalized propulsion system

task Driver
Acceleration
Request (1)

Propulsion control other actuation systems, e.g. brake

(1) (2)
Prime 𝑇 𝑇 Wheels Road /
Transmission (incl. road Body
mover (as many as
contact)
𝑠 … Traffic
driven wheels)
P (position along
road, grade, …)
Energy storage (e.g. battery
in electric or hybrid vehicle)

Figure 2-67: Propulsion system models, as appearing in a vehicle model

2.4.1 Modelling Concepts

It is often suitable to model propulsion systems as in “1.6.1.1.1 Single cut force notation”, see Figure
2-68.
input signal output signal • Nodes only on connections (shafts).
• A node has a positive direction, from left to
Component Component Component
right, in which 𝑁 and 𝑁 are positive.
A Node 1 B Node 2 C • A system model with natural causality has to
have, for each component, equally many
equations as connected shafts, plus number
Component
𝑇𝑁 Node N 𝑇𝑁 of output signals.
D In example: Components [A, B, C, D] has to
Node 3 𝑁 have [1, 3+1, 1, 1] equations.
𝑁
Figure 2-68: Proposed notation and sign conventions for Propulsion system models, following
“1.6.1.1.1 Single cut force notation”.
The alternative would be as shown in Figure 2-69. It might be better if the model should resolve mo-
ments in different directions in 3D (not shown in Figure 2-69), such as can be relevant e.g. for conical
gear transmissions.

151
 Vehicle Interactions and Subsystems

• Each Connection is between 2 (or more!) Connectors.

Component A Component B • The Connection has no direction.

Connector Connector 𝑇 Connector Connector 𝑇

𝑇𝐴 𝑇𝐴2 Connection
1 Component B 2
2
1 Component A 2
𝑇𝐴2 + 𝑇 0
𝐴 𝐴2 2
𝐴2

𝐴 𝑇𝐴 𝐴 𝐴2 𝑇𝐴2 𝐴2
𝑇 2 𝑇 2 2

Figure 2-69: Alternative notation and sign conventions for Propulsion system models, following
”1.6.1.1.2 Multiple cut forces notation”.

2.4.2 Prime movers

The conversion of stored energy to power occurs in the prime mover, see Figure 2-67. Details of the
conversion processes and transmission of power to the tyres are not covered in this compendium.
Some basic background is still necessary to describe the longitudinal performance of the vehicle. The
main information that is required is a description of the torque applied to the wheels over time and/or
as a function of speed. Sketches of how the maximum torque varies with speed for different prime
movers (internal combustion engine (ICE), electric machine/motor (EM) or similar) are shown in Fig-
ure 2-70. The torque speed characteristics vary dramatically between electric and internal combustion
engines. Also, gasoline and diesel engines characteristics vary.
The curve for an EM in Figure 2-70 shows that the main speciality, compared to ICEs, is that their op-
eration range is nearly symmetrical for negative speeds and torques. However, the curve should be
taken as very approximate, since EMs can work at higher torque for short periods of time. The strong
time duration dependency makes electric motors very different to ICEs from a vehicle dynamics point
of view. Other properties that makes them special are quick and accurate response, well known actual
torque and that it is much more realistic to divide them into several smaller motors, which can operate
on different wheels/axles.

Consumes power 𝑇 Generates power

𝑇
𝑇

Generates power Consumes power

Internal Combustion Engine. ICE Electric Motor, EM Human (on bicycle)

𝑇 𝑇
𝑇
/[𝑊/𝑘 ]

𝑢𝑟 𝑡 𝑜𝑛 [𝑠]
0 0 00 03 04 05

Figure 2-70: Torque Characteristics of 3 Prime Movers: ICE, EM and Human

A bicycle uses the human body as prime mover. It reminds of an EM in that it has torque from zero
speed. The duration dependency is high, with maximum power ≈ 0 W/kg up to a minute duration
and ≈ W/kg for an hour duration.

152
 Vehicle Interactions and Subsystems

2.4.2.1 Efficiency and Consumption

The efficiency (𝜂) is the output power per input power. The specific consumption can be the inverted
value, but often the input power is measured in fuel rate [ /𝑠] or [ 𝑡𝑟𝑒/𝑠], which makes specific con-
sumption inversely proportional to efficiency and not the exact inverse value.
𝑜𝑢
𝑓𝑓 𝑒𝑛 𝑦 𝜂
𝑖
𝑙 𝑖 𝑙
𝑆𝑝𝑒 𝑓 𝑜𝑛𝑠𝑢 𝑝𝑡 𝑜𝑛 𝑆𝑝𝑒 𝑓 𝑜𝑛𝑠𝑢 𝑝𝑡 𝑜𝑛2 [2.50]
𝑜𝑢 𝑜𝑢 𝑜𝑢

ℎ𝑒𝑟𝑒 𝑢𝑒 𝑅 𝑡𝑒 𝑠 𝑛 [𝑘 /𝑠] 𝑜𝑟 [ 𝑡𝑟𝑒/𝑠]

𝑛 𝑠𝑝𝑒 𝑓 𝑒𝑛𝑒𝑟 𝑦 𝑜𝑛𝑡𝑒𝑛𝑡 𝑛 [ ⁄𝑘 ] 𝑜𝑟 [ ⁄ 𝑡𝑟𝑒]
The efficiency is dependent on the operating point in the speed vs torque diagram, or map, for the
prime mover. An example of a specific consumption map for an ICE is given in Figure 2-71. Maps with
similar function can be found for other types of prime movers, such as the efficiency map for an EM,
see Figure 2-72.
Figure 2-71 and Figure 2-72 also show that the efficiencies can be transformed to the traction diagram.
The maps for different gears partly overlap each other, which show that an operating point of the vehi-
cle can be reached using different gears. The most fuel or energy efficient way to select gear is to select
the gear which gives the lowest specific fuel consumption, or highest efficiency. Such a gear selection
principle is one way of avoiding specifying the gear selection as a function over time in the driving cy-
cle. For vehicles with automatic transmission, that principle can be programmed into the control algo-
rithms for the transmission. However, the gear selection is often a trade-off with acceleration reserve,
see 3.3.4.7, which argues for lower gear. Assuming very tightly stepped transmission, or CVT, gives
that a high efficiency can be kept down to around 0.3 of maximum power.
Efficiency 𝜂 With gear ratio for lowest
Lowest specific
consumption
specific consumption
for that power
𝜂𝑚
force, F
Propulsion

gear 1 0
( typically between 0.1 and 0.5.)

gear 2

0
0 Speed, v

Figure 2-71: Left: Fuel consumption map. Curves with constant specific fuel consumption
[ ⁄(𝑘𝑊 ℎ )], which is ⁄efficiency. Middle: Specific fuel consumption curves transformed to
Traction Diagram, for different gears. Right: How efficiency with efficiency-optimal gear ratio
drops when < 𝑚 .

153
 Vehicle Interactions and Subsystems

Propulsion
force, F
gear 1

gear 2

0
0 Speed, v
Figure 2-72: Left: Efficiency map for a typical brushless DC motor, from (Boerboom, 2012). Elliptic
curves show where efficiency, in [1/W/W], is constant. Right: The efficiency curves can also be
transformed in Traction Diagram, see 3.2.1, for a given gear.

2.4.2.2 § Efficiency and Losses in Electric Machines

When treating more than one quadrant, it is often easier to do maps for 𝑙 ( 𝑇) than for 𝜂( 𝑇),
because definition of 𝜂 is reasonably varying between quadrants, see figure below. Note that quadrant
1 and 3 are very symmetric, while quadrant 1 and 4 is only rather symmetric. This is because the ma-
chine operates in two main way: motor and generator.

Quadrant 1 and 3:
Motor operation,
converting electric
power to mechanical:
𝜂 𝐼⁄ 𝑇

Quadrant 2 and 4:
Generator operation,
converting mechanical
power to electrical:
𝜂 𝑇 ⁄ 𝐼

Figure 2-73: § 4 quadrant efficiency map for electric motor, from Ref (Brian Bole, 2012).
2.4.2.2.1 § Efficiency Model for Electric Machine
A model 𝜂 𝜂(𝑇 ) is useful when studying selection of gears and combination of multiple electric
drivetrains. For this, interpolation in a map from empiric measurements can be used. However, a sym-
bolic expression based on physics can have advantages in computation performance and validity
range. If introducing 2 linear resistors ( 𝑅 𝐼) one can explain some of the losses, see following fig-
ure.

154
 Vehicle Interactions and Subsystems

Physical model: Mathematical model:

𝐼 𝑚 𝐼𝑚
𝑇 𝑚 𝐼𝑚
𝑅 𝑚+𝑅 𝐼
𝑇 𝐼 𝐼𝑚 + 𝑚 ⁄𝑅𝑝
𝑇𝜔

𝑅𝑝
𝜂
0 [𝑉] (Assuming 𝑇 has same sign as 𝐼)

Numerical solution:
with 00 𝑅 0. 𝑅𝑝 5000

Figure 2-74: § Model of power efficiency of an electric machine based on 2 resistors.

The symbolic expressions are found, using Matlab Symbolic toolbox, as:
>> %syms I U Im Um Rs Rp T w eta; sol=solve([T*w==Um*Im, U==Um+Rs*I,
I==Im+Um/Rp, eta==T*w/(I*U)], [eta, I, Um, Im]), sol.eta=

𝑇 𝑅𝑠 𝑅𝑠 ( + 𝑅𝑠 ⁄𝑅 𝑝 )
𝜂 𝑝 ( + +√ 𝑇 ) 𝜂 𝑝 (𝑇 𝑅𝑠 𝑅𝑝 )
⁄𝑅𝑝 + 𝑇 𝑅𝑝

Already in the mathematical model in figure above and from expression for 𝜂 𝑝,
we see that 𝑇 and
only apears in the combination 𝑇 , which is the mechanical power. So, the model with 2
resistors gives constant 𝜂 along curves with 𝑇 𝑜𝑛𝑠𝑡 𝑛𝑡. So, the single 𝜂 peak at a cer-
tain ( 𝑇) cannot be reflected by this model. The compendium does not go further in physical
modelling of EMs, so a non-physical factor is added with 2 tuning parameters, :
𝜂 𝜂 𝑝 (𝑇 𝑅𝑠 𝑅𝑝 ) ( 𝑘 ) ( 𝑘𝑇 𝑇 )
An example of parameterisation of this model is plotted in the following figure.

155
 Vehicle Interactions and Subsystems

Figure 2-75: § Model of power efficiency of an electric machine based on 2 resistors and 2 tuning
parameters. Parameters: 00 [𝑉] 𝑅 0.06 [𝛺] 𝑅𝑝 500 [𝛺] 𝑘𝜔 9.88
8 [ ⁄(
0 𝑟 ⁄𝑠) ] 𝑘 𝑇 4.50 0 6 [ ⁄(𝑁 )2 ]
2

2.4.3 Transmissions
In some contexts, “transmission” means the 1-dimensional transmission of rotational mechanical
power from an input shaft to one output shaft. Such are called “Main transmissions”. In other contexts,
“transmission” means the system that distributes the energy to/from an energy buffer and to/from
multiple axles and/or wheels. Such are called “Distribution transmissions”.

2.4.3.1 Main Transmissions

Main transmission can be either stepped transmissions or continuously variable transmissions, CVTs.
Among stepped transmissions, there are manual an automatic. Among automatic, there are those with
power transmission interruption during shifting and other with powershifting, see 2.4.3.1.1. Clutches
and torque converters can also be part of models of main transmissions, see 2.4.4 and 2.4.4.3. A
stepped transmission, can be modelled e.g. as:
𝑇 𝑟 ∙ 𝑇𝑛 ∙ 𝜂 𝑀 Δ𝑇 ∙ sign( )
𝑛
[2.51]
𝑟
ℎ𝑒𝑟𝑒 𝑟 𝑟 𝑟2 ⋯ 𝑟𝑁 𝑟≠0
Δ𝑇 is the “parasitic” or “load independent” losses, arising from oil, sealings and bearings. Eq [2.51] is
not valid for neutral gear, because then there is no speed equation, but instead two torque equations:
𝑇 Δ𝑇 ∙ sign( ) and 𝑇 𝑛 Δ𝑇 𝑛 ∙ sign( 𝑛 ) .
For any 1-dimensional transmission of rotational mechanical power between two rotating shafts, the
total efficiency, 𝜂 𝑙 ⁄ 𝑛 𝑇 ∙ ⁄(𝑇 𝑛 ∙ 𝑛 ) , is depending on operating condition. If as-
suming a nominal ratio, 𝑟, the total efficiency can be decomposed in 𝜂 𝑙 𝜂 𝑇 ∙ 𝜂𝜔 , where 𝜂 𝑇
𝑇 ⁄(𝑟 ∙ 𝑇 𝑛 ) and 𝜂𝜔 ⁄( 𝑛 ⁄𝑟 ) .
2.4.3.1.1 Powershifting Main Transmissions
The most common powershifting transmission is using planetary gears and torque converter, see Fig-
ure 2-84 and (Jacobson B. , 1993). During 2000-2010, developments in mechatronics has enabled to
leave out the torque converter and the planetary gears resulting in what often meant with powershift-
ing transmissions, see concept in Figure 2-76. In practice, one can often manage with 2 clutches, and
instead select different paths through gear wheels with synchronisers. An advanced design of

156
 Vehicle Interactions and Subsystems

powershifting transmission for a hybrid propulsion system is seen in Figure 2-85. A dummy sequence
of shifting is simulated in Figure 2-86.
Interpret pedals.
Request engine torque Decides how to shift. Engages
and when to shift gear. and disengages the clutches
C3

e in C2
out Driven wheel
Vehicle body
Prime
mover C1 w x

Figure 2-76: Conceptual design of powershifting transmission with 3 gears.

2.4.3.1.2 § Neutral Mode
Propulsion system can have the possibility to be in Neutral mode (or Neutral gear). This means that
there is a clutch and that it is disengaged. One cannot find a value of ratio 𝑟 such that Eq [2.51] repre-
sents Neutral mode in a good way. Instead there are 2 new equations:
𝑇 Δ𝑇 ∙ sign( )≈0
𝑇 𝑛 +Δ𝑇 𝑛 ∙ sign( 𝑛 ) ≈ 0
(and 𝑛 and are not stipulated by any equation from the transmission)

2.4.3.2 Distribution Transmissions

The distribution between prime mover and energy buffer, as well as between multiple axles and/or
wheels, can conceptually be done in two mechanical ways, exemplified in the following with distribu-
tion between left and right:
• Distribute in certain fractions of (rotational) speed. A (rotationally) rigid shaft between left
and right wheel is one example of this. We find this in special vehicles, such as go-carts, and in
other vehicles as when a differential lock is engaged. There are 3 shafts in such an axle: input
shafts and two output shafts (to left and right wheel). The equations will be:
𝑛 𝑙 𝑓
𝑛 [2.52]
𝑇𝑛 𝑇𝑙 𝑓 +𝑇
• Distribute in certain fraction of torque. This requires some type of planetary gear arrangement.
A conventional (open) differential gear is one example of this, where the equations will be:
𝑙 𝑓 +
𝑛
𝑇𝑙 𝑓 𝑇 [2.53]
𝑇𝑛 𝑇𝑙 𝑓 +𝑇
The case of multiple (wheel-individual) drivetrains, we can see it as a 3 rd generalized way. There are
then 2 input shafts, so the common input connector (instead of input shaft with 𝑛 and 𝑇 𝑛 ) can be an
electric connection with voltage 𝑛 and current 𝐼 𝑛 . Compared to eqs [2.52] and [2.53], we find only 1
equation (not 3): 𝑛 𝐼 𝑛 𝑇𝑙 𝑓 𝑙 𝑓 +𝑇 . So, there are 2 control dofs, e.g. 𝑇𝑙 𝑓 and
𝑇 , which are known from (requested by) a control algorithm, i.e. via an input port to each machine.
Generally speaking, the open differential is rather straight-forward to use in most vehicle dynamics
analyses: The speeds are given by vehicle motion (e.g. curve-outer wheel runs faster than curve-inner
wheel, defined by vehicle yaw velocity and track width). The torques are defined by the differential, as
half of the propulsion torque at each side.

157
 Vehicle Interactions and Subsystems

A locked differential, it is generally more difficult to analyse and understand in a vehicle manoeuvre.
Here, the wheels are forced to have same rotational speed, and, in a curve, that involves the tyre longi-
tudinal slip characteristics, so the explicit form solution involves more equations.
So, open/locked differential is the basic concept choice. But there are additions to those: One can build
in a friction clutch which is either operated automatically with mechanical wedges or similar or oper-
ated by control functions. Another was is to build in one-way clutches which can be coupled in dynam-
ically, see (Lidberg & Alfredsson, 2009). One can also build in an EM which moves torque from one
wheel to the other. However, the compendium does not go further into these designs.
Shafts are also parts of the distribution transmissions. If oscillations are to be studied, these has to be
modelled with energy storing components:
• Rotating inertias or Flywheels ( 𝑇 𝑛 𝑇 ) and
• Elasticities, compliances or springs: (𝑇 ∙( 𝑛 )).

2.4.4 Clutches and Brakes in Transmission

The modelling techniques shown in this section are also applicable also for tyres (2.2) and brakes (2.4.7).
Seen as machine elements, brakes are special cases of clutches, specialized by that one of the clutch
halves is fixed. There are brakes on the wheels, but there are also brakes inside the transmission. A
tyre operates as a transmission clutch (2.2.1.6) which can stick and slip (0). Also, when operating be-
tween forward and rearward rolling, the rolling resistance of the tyre acts as a brake, with stick and
slip.
Clutches with dry friction are difficult to model in dynamic systems since they introduce “discrete dy-
namics”, sticking and slipping, see 1.5.1.4. The conceptual mathematical model of a controlled clutch is
given in Figure 2-77 and Eq[2.54]. Here, the interface variables are 𝑇 2 and . Capacity [𝑁 ]
is an input signal representing the magnitude of torque when the clutch slips. In the model below, a
discrete state variable, 𝑥 0 𝑜𝑟 + , is introduced to model the discrete dynamics.
𝑡 =clutch torque capacity 2 variants of (Discrete) State Transition diagrams:

𝑇 𝑇 𝑙 >0
𝑥 1: 𝑇 < 𝑥 =0: 𝑥 =+1:
2 𝑇 𝑙 0 𝑇>+ 𝑇 +
𝑙 2 𝑙 <0

Discrete states: 𝑇 𝑇>+

𝑥 +
0

𝑙 𝑙
𝑥

𝑥
𝑇<
Figure 2-77: Model of a clutch. The 𝑥 is a discrete state, declared as Integer.
𝑓𝑥 𝑡ℎ𝑒𝑛 𝑇 +
(Continuous) Equations
{𝑒 𝑠𝑒 𝑓 𝑥 0 𝑡ℎ𝑒𝑛 𝑙 0
using the discrete state 𝑥 :
𝑒 𝑠𝑒 𝑓 𝑥 + 𝑡ℎ𝑒𝑛 𝑇 𝑒𝑛 𝑓
(
ℎ𝑒𝑛 𝑥 𝑛 )
𝑙 > 0 𝑡ℎ𝑒𝑛 𝑥 0 [2.54]
(Discrete State) 𝑒 𝑠𝑒 ℎ𝑒𝑛 (𝑥 0 𝑛 𝑇< ) 𝑡ℎ𝑒𝑛 𝑥
ransition Equation: 𝑒 𝑠𝑒 ℎ𝑒𝑛 (𝑥 0 𝑛 𝑇 > + ) 𝑡ℎ𝑒𝑛 𝑥 +
{𝑒 𝑠𝑒 ℎ𝑒𝑛 (𝑥 + 𝑛 𝑙 < 0 ) 𝑡ℎ𝑒𝑛 𝑥 0 𝑒𝑛 ℎ𝑒𝑛

158
 Vehicle Interactions and Subsystems

2.4.4.1 Modelling Format and Tool Aspects

A tool with discrete dynamic support should log at least two values for an event: the value before and
after the event. So, there will be 2 identical time values in the time vector. Several transitions can hap-
pen during on event, e.g. switching from 𝑥 to 𝑥 + without 𝑥 0 in between in Figure
2-77. For this reason, the simulation tool needs to do “event iteration”, which means that the whole
model is evaluated repeatedly during the event until no further changes happens. It can be discussed
whether intermediate values during the event iteration should be logged or not. Modelica is designed
to not log those.

2.4.4.2 Different Surroundings for the Clutch

Different implementations can be needed for different surroundings; energy dissipating or generating
components (dampers or power sources), kinetic energy storing components (flywheels) or potential
energy storing components (elasticities), see examples in Figure 2-78. The challenge is to handle that
the set of state variables can change between the different discrete states.
Connected to control algorithm, which

…
calculates “clutch torque capacity”, 𝑡 .
… …
…

Further
components
… J J …Further
components elasticity clutch elasticity
flywheel clutch flywheel

…
…

… …
… J …
damper clutch damper
(e.g. 𝑇( ) prime (e.g. (𝑠 )
flywheel clutch elasticity mover model) tyre model)
Figure 2-78: Examples of differently modelled surroundings of a clutch.
For complete main transmissions, as automatic transmissions, there are several clutches involved, the
implementation of the ideal model in Eq [2.54] can be very demanding. It is modelled in Modelica. The
discrete dynamics is modelled with the discrete state (xd declared Integer) and the operator
“pre(z)”, which holds the value from last time instant or, in an event, from the last event iteration.
2.4.4.2.1 Clutch and Elasticity in Series
The easiest surrounding to a clutch is in series with elasticity, assuming velocities can be input. Figure
2-79 shows such. It also adds 𝑠𝑡 𝑡 𝑜𝑛 𝑠𝑡 > ; a different static and dynamic friction, 𝑘 𝑠𝑡
𝑙𝑝 .

159
 Vehicle Interactions and Subsystems

12 3 temporary elasticity
𝑡 (rigid shaft when
(permanent) 𝑡 clutch is slipping)
elasticity 2
2
𝑇 𝑇 𝑇 𝑇
𝑇 𝑇
2 𝑇
clutch 2 𝑇
clutch
//Clutch:
if slip==0 then 1 //Clutch and Elasticity:
if slip == 0 then
3
w1 = wx; w1 = wx;
else der(state) = cs*(wx - w2); T = state;
der(T) = sign(slip)*der(cc); else // if abs(slip) == 1 then
end if; T = +sign(slip)*cc;
0 = (wx - w2); //rigid elasticity
der(state) = void; //state not interpreted as T
end if;
when pre(slip)== 0 and T >+st*cc then slip=+1; when pre(slip)== 0 and T >+st*cc then slip=+1;
reinit(T, +cc); reinit(state, +cc);
elsewhen pre(slip)== 0 and T <-st*cc then slip=-1; elsewhen pre(slip)== 0 and T <-st*cc then slip=-1;
reinit(T, -cc); reinit(state, -cc);
elsewhen pre(slip)==+1 and w1<wx then slip= 0; elsewhen pre(slip)==+1 and w1<wx then slip= 0;
elsewhen pre(slip)==-1 and w1>wx then slip= 0; end when; elsewhen pre(slip)==-1 and w1>wx then slip= 0; end when;
//Elasticity:
der(T) = cs*(wx-w2);

T cc -cc +st*cc -st*cc T cc -cc +st*cc -st*cc

1.5 1.5

1.0
1 3
1.0

0.5 0.5

0.0 0.0

-0.5 -0.5

-1.0 -1.0

-1.5 -1.5
0 5 10 15 0 5 10 15

w1 w2 wx w1 w2 wx

2 1 2 3
0 0

-2 -2

0 5 10 15 0 5 10 15

//Clutch
5if
slip and Elasticity:
T > +st*cc and w1 > w2 then der(T) = +der(cc); 2 slip

0elseif T < -st*cc and w1 < w2 then der(T) = -der(cc);

0
-5else der(T) = cs*(w1 - w2); end if;
0 5 10 15 0 5 10 15

Figure 2-79: Model of clutch and elasticity in series, modelled for speed input from both sides.
Implemented in Modelica. Implementation 1,2 and 3. Without discrete state and without when (2)
only works well (without chattering) for 𝑠𝑡 .0. Switched physical model (3) has a state
(state) depending on the discrete state 𝑠 𝑝; elasticity is modelled only when slip=0.
2.4.4.2.2 Clutch between Inertias
When connecting 2 inertias or 2 elasticities with a clutch is more complicated. For clutch between in-
ertias, it is proposed to use 𝑙 as state variable, because it makes it easy to keep 𝑙 0 during
stick.be modelled, see Figure 2-80. The values 𝑥 ± are temporary during a state event; it enables a
total state transition during one event between 𝑥 and 𝑥 + , without (wrongly) logging an
intermediate 𝑥 0. With 𝑠𝑡 < , there is risk for chattering solutions.

160
 Vehicle Interactions and Subsystems

Discrete states: (Discrete) State Transition diagrams:

𝑇 𝑇 𝑇 > +𝑠𝑡
𝑥 +
𝑙 >0 𝑥 =+1: 𝑇 +𝑠𝑡 𝑙 >0
>0

<0
𝑙
𝑙 <0
𝑥 +

𝑙
𝑥 2: 𝑇 < 𝑠𝑡 𝑥 =0: 𝑥 =+2:
0 𝑇 0 𝑇 > +𝑠𝑡 𝑇 +
𝑙

𝑙 𝑙 𝑙 >0
𝑥

>0
𝑥 𝑙 <0 𝑥 1: 𝑇 𝑠𝑡 𝑙 <0

𝑙
𝑥 𝑙 <0
(𝑥 and + are only temporary during the event.)
𝑇 < 𝑠𝑡
Figure 2-80: Modelica of a clutch which works to be connected between inertias.
2.4.4.2.3 Functional or Inverse Clutch Model
In some cases, it can be suitable to prescribe how the clutch is operates in terms of 𝑙 (𝑡) instead of
(𝑡). This means that we rather assume a successful engagement as that 𝑙 → 0 during an assumed
engagement time, than assuming a (𝑡) and find out how long time it takes to engage (or fail to en-
gage. The 𝑙 (𝑡) is then declared as input and the (required) torque 𝑇(𝑡) and the torque capacity
(𝑡) becomes an output. This can simplify the modelling. This is not further discussed in this compen-
dium but see http://blog.xogeny.com/blog/part-2-kinematic/.

2.4.4.3 Implementation of Clutches as Stiff Dampers

A way around model implementation problems is to use an approximate clutch model as in Eq [2.57].
The advantage is that it can handle any surrounding without leading to changing set of state variables.
The drawback is that it has a trade-off between modelling the intended physics (Eq [2.57]) and the
computational efficiency in simulation; the more → , the closer to the intended model we come,
but the simulation time will increase towards infinity. The approximation can be seen as putting a stiff
damper in direct series with an ideal clutch. There are also other approximations, which could be seen
as putting a stiff spring with stiffness in series with an ideal clutch. This will work if clutch is sur-
rounded by inertias:
𝑇𝑙 𝑓 𝑇 𝑛 𝑙 𝑙 𝑓
𝑇𝑙 𝑓
sign( 𝑙 )
𝜀 | 𝑙| 𝜀 | 𝑙| | 𝑙|
min ( + ( )) 𝑓 <
𝜀 𝑛 𝑚 𝜀 𝑛 𝑚 𝑛 𝑚 [2.55]

𝜀 | 𝑙|
max ( ) 𝑒 𝑠𝑒
𝜀 𝑛 𝑚 𝜀 𝑛 𝑚
+ 𝜀 (| )
{ 𝑙|

161
 Vehicle Interactions and Subsystems

(Note: “Semi-invert diagram axes”, i.e. that

value varies linearly below 1, but 1/value
varies linearly above 1.)

𝑇
𝑇

𝑓
𝑓

𝑇𝑙
𝑇𝑙

𝑙 𝑙
𝑛 𝑚 𝑛 𝑚

Figure 2-81: Example of approximation of clutch model. Strategies: 𝑛 𝑚 = typical slip speed of
clutch, 𝜀 ≪ . Dashed curve shows before approximation.
Figure 2-82 shows an example that the ideal and approximate models can give comparable results
with respect to torques and speeds. About computational efficiency, the ideal needs around 5 𝑠 time
step with Euler forward integration, while the approximate needs 100 times smaller time step. If bet-
ter agreement than in Figure 2-82 is needed, 𝜀 needs to be reduced, which slows down the approxima-
tion even more. Note also that, for energy dissipation, the approximate model of course calculates a
higher energy dissipation, since it assumes the clutch has to slip to transfer torque.
approxClutch.Aped approxClutch.c0
approxClutch.Aped
2 approxClutch.c0
2 𝐴 𝑒 𝑒𝑟 𝑡𝑜𝑟 𝑝𝑒
1
1 𝑢𝑡 ℎ 𝑒𝑛 𝑒 𝑒𝑛𝑡
0
0 0 2 4 6 8 10 12 14 16
0 2 4 6 8 10 12 14 16
approxClutch.w eng approxClutch.w gin idealClutch.w eng idealClutch.w gin
approxClutch.w
600eng approxClutch.w gin idealClutch.w eng idealClutch.w gin
600
550 /[𝑟 /𝑠]
𝑛 𝑛
550
500
500
450
450 /[𝑟 /𝑠]
𝑛
400
400
350
350
300
300
250
250
200
200 0 2 4 6 8 10 12 14 16
0 2 4 6 8 10 12 14 16
Modelled in two ways: approxClutch.vx idealClutch.vx
approxClutch.vx idealClutch.vx
• Approximation, 𝜀 0.05 20
/[ /𝑠]
• Ideal 20
0
0 0 2 4 6 8 10 12 14 16

Figure 2-82: Example

0
with
2
clutch,
4
modelled
6 8
in two
10
ways
12 14 16

There are clutch models also in the standard Modelica (see 1.5.4.5) library, see Figure 2-83. Note that
the library is built such that the “small inertia” is needed, which forces down computational efficiency
during clutch slip.

162
 Vehicle Interactions and Subsystems

idealGear
engine engine_flywheel small_inertia body
spring damper

tau ratio=7.5 c=1000e3

clutch d=500
J=0.5 J=0.001 J=1500*0.3*0.3

(Inputs to "eninge" and "clutch" governed by equations)

Figure 2-83: Model example with clutch, using standard Modelica library
“Modelica.Mechanics.Rotational”
Brakes, one-way clutches and backlashes often causes similar difficulties and can be modelled simi-
larly as clutches.

2.4.5 Hydrodynamic Torque Converters

Hydrodynamic torque converters serve almost same purpose as a clutch, but it is much less complex to
model as a member of a dynamic system. Such converters have a pump in input side and turbine on
output side. They can operate with substantial slip, and when slip, there is a torque amplification,
which leads to that a vehicle with converter have typically good acceleration performance and drivea-
bility also without the corresponding lowest gear needed on same vehicle without converter.
The following model gives the steady state characteristics of a hydrodynamic torque converter. Steady
state characteristics are often enough but combined with that hydrodynamic torque converters often
are possible to lock-up with a clutch, mounted in parallel to the impellers.
𝑇 𝑇𝑛
(𝜈) 2 𝜆(𝜈) ℎ𝑒𝑟𝑒 𝑛 𝜆 𝑟𝑒 𝑓𝑢𝑛 𝑡 𝑜𝑛𝑠 𝑛 𝜈
𝑇𝑛 𝑛 𝑛
𝑇𝑜𝑢 𝜔𝑜𝑢
But if locked-up: 𝑇𝑖
𝑛 𝜔𝑖
[2.56]
The moment capacity, 𝜆 is often given on a dimensionless form
𝜆 𝑇 𝑛 ⁄( 2𝑛 𝐷𝑒𝑛𝑠 𝑡𝑦 𝑢𝑡𝑒𝑟𝑅 𝑢𝑠 5 ).

𝑇 𝑇𝑛
𝑇𝑛 𝜆 2
𝑛

stepped 1
e in out gearbox
(with planetary
lock-up gears) 1 1
clutch ⁄
𝜈 ⁄ 𝑛 𝜈 𝑛

Figure 2-84: Left: Traditional automatic transmission. Right: Conceptual curves of the torque
converter.

2.4.6 Energy Storages and Energy Buffers

Contribution from Jelena Andric and Majid Astaneh, Chalmers

Fuel tank for fossil fuel or hydrogen, battery, flywheel, hydro static accumulator and super capacitor
are examples of energy storages. An energy buffer often refers to an energy storage that can not only be
emptied (during propulsion), but also refilled by regenerating energy from the vehicle during deceler-
ation. With that nomenclature, a fuel tank is an energy storage, but not an energy buffer. Also, a battery
which can only be charged from the grid, and not from regenerating deceleration energy, is not an en-
ergy buffer. Today, the most common energy storage is fossil fuel tank and the most common energy
buffer is battery.
Eq [2.57] gives a simple model of an energy buffer. Also the energy converters to mechanical rotation
(electric machines, if energy buffer is a battery) is modelled.

163
 Vehicle Interactions and Subsystems

( 𝑛 )∙𝜂 𝑓𝑜𝑟 𝑛 >

{
( 𝑛 )/𝜂 𝑒 𝑠𝑒 [2.57]
ℎ𝑒𝑟𝑒 𝑛 𝑇𝑛 ∙ 𝑛 𝑛 𝑇 ∙
Including how the buffer is connected, one more equation can be found: Typically, 𝑛 or
𝑇𝑛 𝑇 .
The model uses stored energy, . For batteries, one often uses state of charge, 𝑆𝑜 , instead. Conceptu-
ally, 𝑆𝑜 / 𝑚 , where 𝑚 is a nominal maximum charge level.
A first approximation of the efficiencies of modern batteries, can be 𝜂 𝜂 𝑜𝑛𝑠𝑡 𝑛𝑡
0.85. .0.95, but typically the efficiency is dependent of many things, such as 𝑛 . The model
above does not consider any leakage when buffer is “resting”, i.e. when 𝑛 . Seen over a whole
transport operation, the energy is typically both stored in and taken out from the battery, so a “round-
trip efficiency” becomes smaller, only 𝜂 𝑛 𝑝 ≈ 𝜂 𝜂 .

2.4.6.1 Fuel Cells

Contribution from Leo Laine, Volvo Trucks and Chalmers
From vehicle motion point of view, a fuel cell with hydrogen storage is an energy storage with electric
energy output. It has limited change rate of output power, typically a time constant of 10 s, so it is typi-
cally combined with a battery. Together with a battery, regeneration and a higher change rate of out-
put power is possible.

2.4.6.2 § Size of Energy Buffers

For vehicle dynamics engineers, it can be difficult to get an intuitive understanding of the size of an en-
ergy buffer like a battery. Since the storage is often limiting range, speed changes and “road grade flat-
tening”, is can help to express energy in the following measures with more vehicle dynamics related
units, instead of [ ] or [𝑘𝑊ℎ]:
• Range: 𝑛 [ ] [ ]⁄ 𝑛 [𝑁] ⁄(𝑅𝑅 + 0.5 𝜌 𝐴 2)
. Note that en-
ergy lost in friction brakes are not included.
• Speed change: 𝑝 𝑛 [ ⁄𝑠 ] √ [ ]⁄ . Note that driving resistance in not included.
• Road grade flattening: 𝑙 𝑛𝑛 [ ] [ ]⁄( ). Note that driving resistance in not
included.

2.4.7 Special Topology Propulsion Systems

There are two propulsion systems which are topologically special: Hybrid Propulsion Systems and
Multiple Electric Drivetrains. Simply speaking, the first has multiple sources (e.g. an ICE and one
electrical motor) while latter has multiple sinks (e.g. propulsion on two axles). In both cases, it often
appears something like different driving modes, which can reduce the need of many gears and invoke
a need for modes selection, similar/related to gear selecton.

2.4.7.1 § Hybrid Propulsion Systems

Hybrid propulsion systems can be categorized in series and parallel. The definition is based on that
one of the propulsion sources is identified as prime mover, which is often the ICE. If the prime mover
is connected with a 1- -dof transmission to the vehicle, it is a parallel hybrid. The secondary propul-
sion source then adds force. If there is a 2- -dof transmission between prime mover and vehicle, it is a
series hybrid. The secondary propulsion source then adds force.

164
 Vehicle Interactions and Subsystems

2.4.7.1.1 § Example of Hybrid Propulsion System

89
82
85

44
41
37

Electric
motor

m 5 2
𝑚
9
C1
4

3
ICE 8
e 1 C0 C2 6 B1 x 2
𝑅𝑤
7

Figure 2-85: § Hybrid propulsion system with powershifting, designed using planetary gears.
Upper left: Design. Upper right: Gear/Clutch schedule. Lower half: Dynamic model.
C0.c C0.Tleft wm Tm
400 1500

200
1000
0
500
-200
0 10 20 30 40 50 60 70
0
C1.c C1.Tleft
400 -500
0 10 20 30 40 50 60 70
200
0 we Te

-200
0 10 20 30 40 50 60 70 400

C2.c C2.Tleft 0

200 0 10 20 30 40 50 60 70

0 vx Fx_kN
20
0 10 20 30 40 50 60 70
15
B1.c B1.Tleft
10
400
200 5

0 0

-200 -5
0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70
Figure 2-86: § Simulation of the transmission in Figure 2-85 (with approximate clutch models
from Eq [2.55]). Example sequence of shifts: Simply shift each 10th second, in order as in table in
Figure 2-85.

165
 Vehicle Interactions and Subsystems

2.4.7.2 § Multiple Electric Drivetrains

Electric machines (EMs) requires less complex transmissions than ICEs, since they have a broader
speed range, 𝑚 / 𝑚 𝑛 (in principle since 𝑚 𝑛 0), and therefore require fewer gears. EMs are
also less complex than ICEs. They also allow less complex transmissions, since they have a broader
speed range, 𝑚 / 𝑚 𝑛 (in principle since 𝑚 𝑛 0), which requires fewer gears. If an EM can
manage with 1 gear, there is no need for gearboxes, which further contribute to reduce complexity. So,
it makes sense to design electric vehicles without gearboxes but allow multiple EMs, i.e. multiple
drivetrains. The following is an example where a single electric drivetrain with a 2-gearbox is chal-
lenged by a multiple electric drivetrain without gearboxes.
The EM model from 2.4.2.2.1 is used with scaling in torque dimension.
2.4.7.2.1 Single Drivetrain Vehicle
The single drivetrain vehicle is designed as follows: EM for single drivetrain scale with 1 or no scaling.
The intersection of ⁄ and the driving resistance defines the top speed, see 3.2.3.1. Lowest ratio
(gear 2) is selected for the top speed. We now have design point 1. Highest ratio (gear 1) is then se-
lected so that the constant power hyperbolas are spliced without gap or overlap. This defines design
points 2 (startability, 3.2.7.1) and 3 (maximum speed at startability road grade). The efficiency map for
each gear can be used to identify gear shift curves for maximum efficiency, as was indicated in Figure
2-72.

Design point 3

Design point 2

Gear shift curve for

maximum efficiency

Design point 1

Figure 2-87: § Efficiency maps for a Single Drivetrain: 1 electric motor (20 kW), gearbox with 2
gears.

166
 Vehicle Interactions and Subsystems

2.4.7.2.2 Multiple Drivetrain Vehicle

Propulsion system components to the multiple drivetrain vehicle can be designed in many ways. We
exemplify with 3 drivetrains, each with an EM scaled with 0.5. The ratios of drive train 1 and 2 are
same as for gear 2 in single drivetrain vehicle. The ratio of drivetrain 3 is such that all 3 drivetrains to-
gether can reach design point 2. Design point 3 is not fully reached, but other advantages are reached,
see figure below.

Figure 2-88: § Efficiency maps for a Multiple Drivetrains: 3 electric motors (3×10 kW), no
gearboxes. Efficiency maps for 3 Modes are shown.
The figure shows 3 “Modes”. A Modes is a selection of which drivetrains (which EMs) to use, and with
which distribution, to generate the required force on the vehicle. The 3 modes in the figure are de-
fined as:
• Mode 1: Use only Drivetrain 1
• Mode 2: Use Drivetrain 1 and 2, with equal distribution of force and power
• Mode 3: Use Drivetrain 1 only, if it is enough for the required . If more is needed, add as much
as needed from Drivetrain 2. If even more is needed, add as much as needed from Drivetrain 3.
Note that one can define modes in more advanced way, such as in Ref (Utbult, Jonasson, Yang, &
Jacobson, accepted 2020). Also note, that anyway the modes are defined, they constitute discrete se-
lections which reminds very much of gear selection.

2.5 (Wheel) Braking System

Braking can refer to either wheel braking which means adding wheel torques from (wheel) braking sys-
tem or vehicle braking which can be achieved with negative wheel torques from propulsion system. Sel-
dom used, it would also be possible to apply unusual wheel steering angles, e.g. steer left and right wheel
in different directions. Other ways are possible such as raise brake shields on the vehicle body to increase
longitudinal aerodynamic resistance.
There are several systems that can brake a vehicle:
• Service brake system (brake pedal and ABS/ESC controller, which together applies brake pads
to brake discs/drums)
• Parking brake (lever/button that applied brake pads to brake discs/drums, normally on rear
axle on cars but all axles on heavy vehicles)
• Prime mover brakes:
o Engine braking (ICE operates at “engine brake” as marked in Figure 2-70)

167
 Vehicle Interactions and Subsystems

o Electric machines (machines can be used symmetrical, i.e. both for positive and nega-
tive torques, see Figure 2-70)
• Heavy vehicles often have Retarders. They normally use hydraulic or Eddy current to dissipate
engine, as opposed to dry friction. So, they cannot brake at low speeds or stand-still.
• Large steer angles will decelerate the vehicle, see 3.2.2.2.
This section is about Friction brakes, meaning Service brakes and Parking brake. In vehicle dynamics
perspective, these have the following special characteristics:
• Friction brakes are almost unlimited in force for a limited time since they can lock the wheels
for most driving situations and road friction (ICE and electric motors are often limited by their
maximum power, since it is often smaller than available road friction.) However, if the friction
brakes are used for a long time, the brake lining will start to fade. This means friction coeffi-
cient is lowered due to high temperature (oxidation and melting of pad/lining material).
• Friction brakes can only give torque in opposite direction to wheel rotation. (Electric motors
can brake so much that wheel spins rearwards.)
• Friction brakes can hold the vehicle at standstill. (Using electric machines for stand-still in a
slope, a closed loop control would be necessary, resulting in that vehicle “floats” a little.)
The basic design of a passenger car brake system is a hydraulic system is show in Figure 2-89. Here,
the brake pedal pushes a piston, which causes a hydraulic pressure (pressure = pedal force/piston
area). The hydraulic pressure is then connected to brake callipers at each wheel, so that a piston at
each wheel pushes a brake pad towards a brake disc ( Disc orce Pressure PistonArea). The brake
torque on each wheel is then simply: 𝑇 NumberOf rictionSurfaces DiscCoefficientOf riction
Disc adius Disc orce. (Normally, there are 2 friction surfaces, since double-acting brake calipers.) By
selecting different piston area and disc radii at front and rear, there is a basic hydro mechanical brake
distribution ratio between front and rear axle. There are normally two circuits for redundancy. It
should be mentioned that DiscCoefficientOf riction varies a lot; during one strong brake event, it can
typically drop 10..25% due to temperature rise and sliding velocity decrease.
Brake systems for heavy trucks are generally based on pneumatics, as opposed to hydraulics, see Fig-
ure 2-91. Ref (Tagesson K. , 2017), has a good descriptive chapter about brake systems for heavy vehi-
cles.

Driver pushes Master cylinder Front Front Front

Slave cylinder
on pedal

II HI
Hydraulics

X (common)
Pistons with
brake pad Front Front

Brake
disc HH
LL
Hydraulic pump for
ABS/ESC and hydraulic Rear
valves are not drawn.

Figure 2-89: Layouts of a hydraulically applied brake system, which is conventional on passenger
cars.
Brake systems for modern road vehicles are almost always mechatronic systems, i.e. they contain both
mechanical parts and control algorithms. As minimum, one can include the wheel slip control, see
4.6.2.1.4, or ABS/EBD, see 3.5.2.3/3.5.2.4.

168
 Vehicle Interactions and Subsystems

ABS intervention 1.
2.
Master Cylinder
Inlet Valves
ESC intervention
(Pedal braking. One wheel 3. Wheel Cylinders/Callipers (No pedal braking. One wheel
brake controlled=released) 4. Brake Discs brake controlled=applied)
5. Outlet Valves
6. Tank/Low Pressure
Accumulator
1. 7. Pump 1.
8. Check Valve
9. 9. Isolation Valve/Traction 9.
Control Valve

8. 8.
7. 7.

2. 5. 5.

5. 2.
2. 5.
6. 6.
2.
3. 3. 3. 3.
controlled = controlled =
= released 4. … 4.
= applied 4. … 4.
(several wheels, only 2 drawn) (several wheels, only 2 drawn)
Figure 2-90: Concept of hydraulically applied brake system for ABS and ESC functions.

Figure 2-91: Pneumatically applied brake system for heavy vehicles. Electronics Brake System,
EBS, from Volvo GTT, Mats Sabelström.

169
 Vehicle Interactions and Subsystems

2.6 (Wheel) Steering System

Steering can refer to either wheel steering which means adding yaw angles to the wheels relative to the
vehicle body or vehicle steering which can be achieved with wheel steering but also other actuators, such
as propelling or braking the wheels on side.
The steering system is here referred to the arrangements between steering wheel and the road
wheel’s steering on the steered axle. It is normally the front axle that is steered. Driver’s interaction is
two-folded, both steering wheel angle and torque, which is introduced in 2.9. In present section, we
will focus on how wheel steer angles are distributed between the wheels.

2.6.1 Chassis Steering Geometry

The most basic intuitive relation between the wheels steer angles is probably that all wheels’ rotation
axes always intersect in one point. This is called Ackermann geometry and is shown in Figure 2-92.
The condition for having Ackermann geometry is, for the front axle steered vehicle that:
𝑅 ⁄
tan( ) 𝐿
⇒ + [2.58]
𝑅 + ⁄ tan( ) tan( ) 𝐿
tan( ) 𝐿 }
The alternative to Ackermann steering geometry is parallel steering geometry, which is simply that
. Note that Ackermann geometry is defined for a vehicle, while parallel steering is defined for
an axle. This means that, for a vehicle with 2 axles, each axle can be parallel steered, which means that
the vehicle is non-Ackermann steered. However, the vehicle can still be seen as Ackermann steered
with respect to mean steer angles at each axle.
For low-speed, Ackermann gives best manoeuvrability and lowest tyre wear. For high-speed, Parallel
is better in both aspects. This is because vehicles generally corner with drift outwards in curves, which
means that the instantaneous centre is further away than Ackermann geometry assumes, i.e. more to-
wards optimal for parallel. Hence, geometry is typically chosen between Ackermann and parallel.
Practical arrangement to design the steering geometry is shown in Figure 2-93. The design of linkage
will also make the transmission from steering wheel angle to road wheel steer angle non-linear. This
can lead to different degrees of Ackerman steering for small and large steering wheel angles.

di do 1 turn centre
a Ackermann error, front.
b Ackermann error, rear.

1 tan(d o ) = 1 tan(d i ) + w L
L

Common intersection of all

wheels’ axes of rotation
Rr w

Figure 2-92: Ackermann steering geometry. Left: One axle steered. Right: Both axles steered and
including “Ackermann errors”. From (ISO 8855).
In traditional steering systems, the steering wheel angle has a monotonically increasing function of the
steer angle of the two front axle road wheels. This relation is approximately linear with a typical ratio

170
 Vehicle Interactions and Subsystems

of 15..17 for passenger cars. For trucks the steering ratio is typically 18..22. In some advanced solu-
tions, steering on other axles is also influenced (multiple-axle steering, often rear axle steering). There
are also solutions for dynamically adding steer angle through a planetary gear and electric angle-con-
trolled motor on the steering shaft, so called Active Front Steering (AFS). In reference (Tagesson K. ,
2017), there is a good descriptive chapter about steering systems for heavy vehicles.
Ackermann 𝑤

steering
𝑓 𝑟𝑝 𝑛
(trapezoidal Rack 𝑙
𝑓 𝑟𝑝 𝑛
𝑤
𝑤
geometry)
(Erasmus Darwin
1758, Rudolph
Ackermann 1810)
Tie Rod

Steering
Knuckle
Steering
View from top Axis
and rear:
𝑙

0
symmetry
when 𝑤
𝑅𝑆𝐴

𝑙
𝛾 𝛾

𝑅𝑆𝐴 𝑟𝑝 𝑛 𝑤 𝑟𝑝 𝑛 𝑤

Figure 2-93: Left: Ackermann (Trapezoidal) Steering. Right: Rack Steering, common on passenger
cars.
Eq [2.59] shows the relation between steering angles for rack steering, with 𝑇𝑅𝐿=Tie Rod Length,
𝑆𝐴𝐿=Steering Arm Length, 𝑅𝑆𝐴=Rack to Steering Axis lengths.
2
𝑇𝑅𝐿2 (𝑅𝑆𝐴 𝑆𝐴𝐿 cos(𝛾 + 2
𝑙 )) + (𝑅𝑆𝐴 𝑆𝐴𝐿 sin(𝛾 + 𝑙) + 𝑟𝑝 𝑛 ∙ 𝑤)
[2.59]
𝑇𝑅𝐿2 (𝑅𝑆𝐴 𝑆𝐴𝐿 cos(𝛾 ))2 (𝑅𝑆𝐴
+ 𝑆𝐴𝐿 sin(𝛾 ) 𝑟𝑝 𝑛 ∙ 𝑤)
2

The 𝑇𝑅𝐿 determines a toe-in. To get toe-in , design (oradjust) the 𝑇𝑅𝐿 to:
𝑇𝑅𝐿 √(𝑅𝑆𝐴 𝑆𝐴𝐿 cos(𝛾 2
toe-in )) + (𝑅𝑆𝐴 𝑆𝐴𝐿 sin(𝛾 toe-in ))
2 .

2.6.2 Steering System Forces

(This section has large connection with 2.2.4.7Other Forces and Moments in Lateral.)
The steering wheel torque, 𝑇 𝑤 , should basically be a function of the tyre/road forces, mainly the
wheel-lateral forces. This gives the driver a haptic feedback of what state the vehicle is in. The
torque/force transmission involves a servo actuator, which helps the driver to turn the steering sys-
tem, typically that assists the steering wheel torque with a factor varying between 1 and 10, but less
for small 𝑇 𝑤 (highway driving) than large 𝑇 𝑤 (parking), see Figure 2-94. Here, the variation in assis-
tance is assumed to be hydraulic and follows a so-called boost curve. At 𝑇 𝑤 0, the assistance is
≈0.45/0.55≈1 and for 𝑇 𝑤 4 Nm, it is ≈0.9/0.1≈10.

171
 Vehicle Interactions and Subsystems

Rack
Force Assisted

Assistance

UnAssisted

SteWhlTq

Figure 2-94: Left: Boost Curve with different working areas depending on the driving envelope.
Middle: Torque distribution between manual torque, FM, and assisting torque, FA, depending on
applied steering wheel torque. From Reference (Rösth, 2007). Right: Unassisted and assisted
steering wheel torque.

2.6.2.1 § Torque Steer and Brake Steer

For vehicle dynamics, one important effect of a steered axle, is that the lateral force on the axle tries to
align the steering in the direction that the body (over the steered axle) moves without tyre side slip.
This is designed in via the sign of the castor trail, see Figure 2-95. Also, asymmetry in longitudinal tyre
forces (wheel shaft torques and/or brake torques) affects the steering wheel torque. This is analysed
in the following.

Steering-axis inclination angle, Wheel 1 with

KPI Caster (friction)
angle, CA brake brake
View from View from
torque, Tb:
rear: right: (Tb<0 if w>0) Tb
bearing
w

forward

Fx1=Tb/R
Castor offset at
wheel center
(drawn positive)
Wheel 1 with
Fx1 shaft torque, Ts:

Castor offset at ground, c, R

(drawn positive)
Steering axis offset
at ground, s Ts
(drawn positive)
(sometimes called Scrub radius) Fy1
Ts*sin(KPI)
Fy1 = resulting
lateral force
Pneumatic trail, t Fx1=Ts/R
(positive if “behind”) lateral force
distribution

Figure 2-95: § Steering geometry and offsets.

On a steered axle, there is a Steering axis (or Kingpin axis) around which the wheel is rotated when it is
steered. The steering axis intersects with the ground plane at a point which normally has an offset

172
 Vehicle Interactions and Subsystems

from the contact patch centre, both longitudinally and laterally. The offsets are called Castor offset at
ground and Steering axis offset at ground, respectively. (Steering axis offset at ground is sometimes
called Scrub radius, but that is also used for the resulting distance of both offsets, so it is an ambiguous
name.) This figure also defines Normal steering axis offset at wheel centre. We will use the 3 latter
measures to explain why steering is affected by differences in shaft torque left/right, differences in
brake torque left/right, and lateral wheel forces. The steering axis distance from wheel centre in a side
view, is called Castor offset at wheel centre.
Often, the actual forces between tyre and ground are not in the exact centre of the nominal contact
patch, which also creates “effective” version of castor offset and scrub radius. For instance, the Pneu-
matic trail adds to the Castor offset at ground due to the lateral force distribution being longitudinally
offset from nominal contact point; so that the lever becomes not only , but + 𝑡. The moment, 𝑇 ,
on steering system (around Steering axis, turning towards increasing steer angle) is affected by tyre
forces on a steered axle as in Equation below.
Castor offset at ground is, on passenger vehicles, 15-20 mm (at motorcycles approximately 100 mm).
On rear wheel driven passenger vehicles, it is typically 5 mm due to higher Castor Angle which gives a
beneficial higher Camber angle gain at cornering. On a front wheel driven passenger vehicle, a non-
zero castor offset is not chosen due to drive axle lateral displacement. Castor offset gives a self-aligning
steering moment, which generally improves the steering feel.
(𝑇 +𝑇 ) 𝑇
𝑇 ∙ 𝑠 cos(𝐾 𝐼) + 𝑅 sin(𝐾 𝐼) +
𝑅 𝑅
(𝑇 2 + 𝑇 2 ) 𝑇2
+ 𝑠 cos(𝐾 𝐼) + 𝑅 sin(𝐾 𝐼) +
𝑅 𝑅 [2.60]
+( ) ( ( + 𝑡 ) + 2 ( + 𝑡2 )) cos( 𝐴)
𝑇2 𝑇 𝑇2 𝑇
𝑠 cos(𝐾 𝐼) + 𝑘 ( ( +𝑡 )+ 2 ( + 𝑡2 )) cos( 𝐴)
𝑅 𝑅
The equation shows that difference in both brake torque and shaft torque affects steering and so does
the sum of lateral forces. For reducing torque steer and disturbances from one-sided longitudinal
forces due to road irregularities, kingpin offset, scrub radius different road friction should be as small
as possible, but it is limited by geometrical conflicts between brake disc, bearing, damper, etc.
Positive scrub radius contributes to self-centring, thanks to lifting the car body, see below. Negative
scrub radius compensates for split- braking, or failure in one of the brake circuits. Hence, the scrub is
a balance between these two objectives. Scrub radius is often slightly negative on modern passenger
cars. Scrub radius is often positive on trucks, maybe 10 cm, due to packaging.

2.6.2.2 § Effects from Lifting

The geometry in figure shows one part of the lifting effect. This is that if steer angle is changed from
zero, it lifts the vehicle slightly which requires a steering torque. One can see this as a kind of “return
spring effect”.
shows how KPI and scrub radius causes the vehicle body to lift a distance 𝑠 ′′′ 𝑠 ∙ cos(𝐾 𝐼) ∙
( cos(| |)) ∙ sin(𝐾 𝐼). This will require a work 𝑇 ∙ ∙ 𝑠 ′′′ . This leads to an 𝑇
(additional to Eq above) as follows:
( 𝑡 𝑜𝑛 ) 𝑇
∙ 𝑠 ′′′ 𝑠 ∙ cos(𝐾 𝐼) ∙ ( cos(| |)) ∙ sin(𝐾 𝐼) [2.61]
∑ ∙ 𝑙
=𝑙 𝑓

It should be noted that the Eq is not complete with respect to all “returning effects”. There are also ef-
fects from Castor angle and Castor trail as well as that the tyre has a width and radius. However, in to-
tal, these give rise to a returning steering torque which is depending on the steer angle.
It can also be noted that steering effort for low speed or stand-still is largely influenced by whether
brakes are applied or not, due to the magnitude (not sign) of scrub radius.

173
 Vehicle Interactions and Subsystems

Contact point Contact point

𝑇 tyre/ground for 𝒕𝒆𝒆𝒓 ≠ 𝟎

𝑠 ′′ sin 𝐾 𝐼
for 𝒕𝒆𝒆𝒓 𝟎
View from
rear: s
KPI

𝑠 ′′′
Fx1

Steering axis offset

at ground, s
Figure 2-96: § Lift effect due to steer angle and positive Steering axis offset at ground.

2.7 Environment Sensing System

This subsystem has to be mentioned since it is maybe the most important new enabler for today’s de-
velopment of automated driving. The technology to sense (radar, camera, lidar, GPS, etc) is not typi-
cally part of vehicle dynamics, but many vehicle dynamics control functions can be invented or im-
proved through usage of the information from the subsystem. Some typically available information is
listed in 3.5.1 and 4.6.1. Another vehicle dynamics aspect is that some sensor fusion, but primarily
some predictions, can be made using vehicle dynamics models.

2.8 Vehicle Aerodynamics

Contribution from Alexey Vdodin, Vehicle AeroDynamics at Chalmers and Adam Brandt, Vehicle AeroDynamics
at Chalmers and CEVT
The flow of air around the vehicle body produces different external forces and moments acting on the
vehicle. The fluid mechanics will not be covered in this course. However, practical first order models
for aerodynamic forces have been established and are presented here.

2.8.1 Longitudinal Relative Wind Velocity

𝑓 𝑓
𝑀

or 𝑅 or 𝑅
ℎ ≈ℎ ℎ 0
Aerodynamic reference point or 𝑅
Centre of gravity, CoG
Figure 2-97: Force-equivalent ways to model longitudinal-wind aerodynamic forces in 𝑥 -plane.
𝑓 and differs, but is the same for all.
From vehicle motion point of view, variables for aerodynamic forces and moments are naturally de-
fined as positive when acting on vehicle in directions defined positive by (ISO 8855). They are denoted
and 𝑀 in 2.8. So, > 0 when the wind pushes the vehicle forward. Consequently, <0
when the wind is resisting forward motion of the vehicle. The longitudinal force play is special in the
sense that it often appears in discussions where it is natural to name it “resistance force” (resisting for-
ward motion), which makes it natural to use the opposite sign definition. So, we use a double notation
on the longitudinal aerodynamic force: 𝑅 , see Eq [2.62].
∙ 𝜌 ∙ 𝐴𝑓 𝑛
𝑅 ∙ 2
𝑙 sign( 𝑙) [2.62]

174
 Vehicle Interactions and Subsystems

The 𝑙 is the longitudinal component of the wind velocity relative to the vehicle. The parameters
𝜌 and 𝐴𝑓 𝑛 represent the drag coefficient, the air density and a reference area of the vehicle, re-
spectively. The 𝐴𝑓 𝑛 is the area of the vehicle projected on a vehicle transversal plane. The most com-
∙𝜌∙𝐴𝑓𝑟𝑜
mon case is relative headwind, which makes sign( 𝑙) , which makes 𝑅 2
∙
2
𝑙 > 0.
Typical values of drag coefficients ( ) for cars can be found from sources such as: (Robert Bosch
GmbH, 2004), (Barnard R. , 2010), (Hucho, 1998), and (Schuetz, 2015). These coefficients are derived
from coast down tests, wind tunnel tests or CFD (Computational Fluid Dynamics) calculations. The air
resistance can often be neglected for city speeds, but not at highway speeds.
Since a car structure moving through the air is not unlike an aircraft wing, there are also an aerody-
namic lift force and pitch moment. This affects the vertical forces on front and rear axle, and conse-
quently the tyre to road grip. Hence, it affects the lateral stability.
𝑙 ∙ 𝜌 ∙ 𝐴𝑓 𝑛
∙ 2 𝑙
[2.63]
𝑝𝑚 ∙ 𝐿 ∙ 𝜌 ∙ 𝐴𝑓 𝑛
𝑀 ∙ 2 𝑙 sign( 𝑙 )
The coefficient 𝑙 represents the lift characteristics of the vehicle. For extreme vehicle, such as racing
cars, one can achieve negative 𝑙 , but often by sacrificing with higher . The forces and are
assumed to act through the same reference point, often centre of gravity (CoG), which defines 𝑀 .
One can replace [ 𝑀 ] with equivalent [ 𝑓 ] or
[ 𝑓 ], Figure 2-97 and Eq [2.64].
𝑙𝑓 ∙ 𝜌 ∙ 𝐴𝑓 𝑛 2
𝑓 ∙ 𝑙
𝑙 ∙ 𝜌 ∙ 𝐴𝑓 𝑛 2 [2.64]
∙ 𝑙
For a reference height ℎ , often ℎ 0

2.8.2 Lateral Relative Wind Velocity

When the wind comes from the side, there can be direct influences on the vehicle lateral dynamics. Es-
pecially sensitive are long but light vehicles (such as buses or vehicles with unloaded trailers). The
problem can be emphasized by sudden winds (e.g. on bridges or exiting a forested area). Besides direct
effects on the vehicle lateral motion, side-winds can also disturb the driver through disturbances in the
steering wheel feel. Similar expressions to the longitudinal loads are derived for lateral forces and
from side-winds.
∙𝜌∙𝐴 2
∙ 𝑙 sign( 𝑙)
[2.65]
𝑚 ∙𝜌∙𝐴∙𝐿 2
𝑀 ∙ 𝑙 sign( 𝑙)

The speed 𝑙 is the lateral component of the vehicle velocity relative to the wind. Note that 𝐴 and
𝐿 may now have other interpretations and values than in Equations [2.62]-[2.64], e.g. 𝐴𝑓 𝑛 or 𝐴 .

2.8.3 § Variation of Relative Wind

2.8.3.1 § Sign Change of Relative Wind Components
The above coefficients 𝑙 𝑝𝑚 𝑙𝑓 𝑙 are normally measured or computed for negative 𝑙 , i.e.
negative relative headwind ( 𝑙 < 0). For positive relative tailwind ( 𝑙 > 0), these coeffi-
cients get other numerical values due to absence of symmetry in vehicle body shape across an 𝑥𝑦-
plane. So, these coefficients are functions of sign( 𝑙 ).

175
 Vehicle Interactions and Subsystems

The coefficients 𝑚 are typically more independent of sign( 𝑙 ) > 0 due to approximate sym-
metriy in vehicle body shape across an 𝑥 -plane.
Some of the equations above in 2.8 are selected to include “sign(⋯ )”. The corresponding coeffi-
ciencts typically, thereby, retain their (positive) sign. In the other equations, the corresponding coef-
ficiencts may or may not change sign.

2.8.3.2 § Varying Relative Wind Yaw Angle

The coefficients 𝑙 𝑝𝑚 𝑙𝑓 𝑙 (for horizontal relative wind 𝑙 ) vary with lateral relative wind
𝑙 . So, the coefficients 𝑙 𝑝𝑚 𝑙𝑓 𝑙 are not only depending of sign( 𝑙 ), as mentioned in
2.8.3, but also 𝑙 or relative wind yaw angle 𝑙 arctan ( 𝑙 𝑙 ).

The coefficients 𝑚 are almost independent of sign( 𝑙) but they still vary with relative wind
yaw angle 𝑙.

2.8.3.3 § Transient Wind Variations

Note that the equations still assume steady-state conditions for the aerodynamics; if wind direction
relative to vehicle changes quickly (by changing wind direction or yawing vehicle vs ground), there
can be transients, which is warned for, but not covered, in this compendium.

2.9 Driving and Transport Application

The driver drives and experiences the vehicle in the short time scale through pedals, steering wheel
and seat. But the load (cargo weight and position) and choice of route is important on the longer time
scale. One can differ between, Ref (Pettersson, 2019):
• Transport Application is how the vehicle is used by one user/owner during the vehicle’s life-
time. It can be commuting × 5 km/day, 5 days/week (e.g. for passenger car) or loading and
transporting timber on 10 km forest road + 300 km highway, twice per day (e.g. for a truck).
• Transport Operation is how the vehicle is used along a specific route. It is typically 10 min to
10 h driving. Driving Cycles, (𝑡), mentioned in 3.3.1.1 have the purpose to describe approxi-
mately the same.
• Transport Mission is the purpose of one Transport Operation, such as where and when to stop
and load/unload a certain payload.

2.9.1 Mission, Road and Traffic

This section is kept very short, but it is included for completeness, beside 2.9.2. See more in 3.3.1.

2.9.2 Driver
To study how different vehicle designs work in a vehicle operation, a driver model is needed. In its eas-
iest form, a driver model can be steering wheel angle 𝑤 0 or 𝑤 ̂ sin( 𝑡) . Another ex-
treme interpretation of what can be called a driver model is an implicit/inverse statement, like “driver
will push accelerator pedal so that speed 0 [ ⁄𝑠] during the manouvre”, which leads to that ac-
celerator pedal position becomes an output, as opposed to input, to the vehicle model. Beyond those
very extreme driver models, there is often need for a driver model which react on vehicle states in re-
lation to an environment or traffic. In this section, driver models are primarily thought of as models
of the driver of the subject vehicle, but when modelling surrounding traffic carefully, each object vehi-
cle can also use a driver model.
The driver interacts with the vehicle mainly through steering wheel, accelerator pedal and brake pe-
dal. In addition to these, there are clutch pedal, gear stick/gear selector, and various buttons, etc., see
Figure 2-98, but we focus here on the first 3 mentioned.

176
 Vehicle Interactions and Subsystems

• Accelerator Pedal Position, Other:

• Brake Pedal Force, • Clutch Pedal and Gear
• Steering Wheel Angle Stick or Gear Selector
• Parking Brake
• Direction indicators
• HMI (buttons, lamps,
• Accelerator Pedal Force, text, sounds, …)
• Brake Pedal Position, • …
• Steering Wheel Torque,
• Seat motion,
• View of environment relative to vehicle

Figure 2-98: Interface of driver, and some commonly assumed causality.

Driver’s control of vehicle dynamics, or vehicle motion including position, can be discussed in longitu-
dinal (mainly pedals) and lateral (mainly steering wheel).
Driver reacts on several stimuli, such as motion (mainly through seat), sounds and optical. Among mo-
tion, it is primarily the translational accelerations (and their time derivative, jerk) that is sensed by the
driver, but also rotational velocities can be sensed by humans. Among the optical, looming (optical ex-
pansion of an object in the driver’s field of view [ 𝑒 /𝑠]) is often used as a cause for how driver uses
the pedals. The optical flow (the pattern of apparent motion of objects, surfaces, and edges in the
driver’s field of view) is often used as a cause for highest comfortable speed and yaw velocity.
Driver models are here discussed as models of the “human driver” for use in vehicle verification simu-
lations. However, driver models can also be understood as models of “virtual driver”, and then they are
actually implemented as algorithms in the vehicle product, e.g. as prediction algorithms or automated
driving controllers. In the first context, it is often important to vary the driver model (at least its pa-
rameters, maybe even its equations) for robustness, as mentioned in Figure 1-7. In the latter context,
the driver model of the subject vehicle is rather varied for optimization, see Figure 1-7 again.
An important aspect of driver modelling is how the user (driver or occupant) experiences the vehicle.
This is often referred to as subjective rating, but for some cases one can establish methods to objec-
tively calculate a measure of how good or bad the experience is. The measure can sometimes be a
physical quantity but often it has to be a rating or grading without unit. Examples are “driveability
[rating 0-10, high is good]”, “steering effort [deg/s, low is good]” and “ride comfort [m/s 2, low is
good]”, see Figure 1-58.

2.9.2.1 Driver Modelling

As in all modelling, it is important to model, or select model, after what the model should be used for.
Driver modelling for “verification of vehicle functions” and for implementation in “driving automa-
tion functions” are similar in that they should react on the vehicle’s environment, but there are also
differences. Driver models for verification of vehicle functions should be as human-like as possible.
They should also judge feedback to driver, such as assessing steering effort. Driver models for use in
driving automation functions should also be human-like to facilitate cooperation between human and
automated driving, such as hand-over/take-back or simultaneous control. However, there are also rea-
sons to not mimic all aspects from a human driver, such as the human’s inability to watch in several
directions simultaneously.
A categorization of modelling concept is whether the model uses equations that intend to reflect the
biological processes human’s perception, cognition and neuro-muscular or equations from a vehicle
model. The first concept (exemplified in 2.9.2.3.3) would rather use angle to obstacle as opposed to
distance to obstacle, since humans rather see angles than distances. The latter concept (exemplified in
2.9.2.2.1) assumes that driver has adapted to the specific vehicle and (subconsciously) operates the
vehicle in a kind of inverse model thinking. Overall, both concepts can reflect approximately the same
driving, but they are differently parameterized; typically, in biological parameters and vehicle parame-
ters, respectively.
Driver can be modelled in 2 parts: Strategic and Operative. A division in Longitudinal and Lateral is
also relevant. One can think of different ways of arranging these dimensions on each other; one possi-
ble way is shown in Figure 2-99.

177
 Vehicle Interactions and Subsystems

Longitudinal motion relative Driver

to other road users/obstacles Pedals,
Strategic Requested longitudinal motion Operative
and to road grades and of subject vehicle, e.g. ForwardOrReverse,
curves, road unevennesses, Longitudinal Longitudinal CruiseButtons
lane widths, legal speed, ...
Actual longitudinal
Requested lateral motion of subject motion of subject vehicle
Lateral motion relative to
Strategic vehicle, e.g. or 𝜅 Operative
other road users/obstacles SteeringWheel
and to road edges. Lateral (𝜅 curvature ⁄ 2+ 2 ≈ ⁄ ) Lateral
Actual lateral motion
of subject vehicle.
Figure 2-99: One possible arrangement of Strategic vs Operative and Longitudinal and Lateral.

2.9.2.2 § Longitudinal Driver Models

Each pedal has both pedal position and pedal force as communication channels to the driver.
In broad terms, the accelerator pedal position is interpreted by a control system as a certain requested
engine torque, which means a certain wheel shaft torque on the propelled axle/axles. In some ad-
vanced solutions, one could think of using accelerator pedal force as a feedback channel from vehicle
to driver.
In broad terms, the brake pedal force gives a certain friction brake torque application on each axle. In
some advanced solutions, one could think of changing brake pedal position versus force characteristics
as a feedback channel from vehicle to driver. But already with traditional (hydraulic) brake systems
for passenger cars, the brake pedal gives a feedback to driver in that ABS interventions are felt.
For electric and hybrid vehicles, the brake pedal can typically influence the propulsion system as well,
in that at low levels of pedal braking, the electric propulsion motors give negative wheel shaft torque.
This is for regenerating energy. Regenerative braking can even be engaged at low levels of accelerator
pedal position.
A model of how the driver operates the pedals is called a longitudinal driver model. But different ma-
noeuvres call for different driver models. Examples of manoeuvres/situations are:
• Full acceleration or deceleration, including driver delays
• Keep constant speed
• Keep distance to lead vehicle
• When a cut-in happens in front of object vehicle: Expand to comfortable/safe gap
• Approach slower object vehicle from behind: Reduce to comfortable/safe gap
• Longitudinal speed control during overtaking
• Longitudinal speed control to stand-still
2.9.2.2.1 § Example of Longitudinal Driver Model
An example for approaching a slower vehicle from behind is defined through simple kinematic rela-
tions and some assumptions, which indicates how to calculate the required (constant) acceleration to
reach a certain safe gap. The safe gap can be defined by “gap needed to exactly avoid collision if brak-
ing as object vehicle but with a certain reaction time”. In traffic schools, one is often taught that safe
time gap 𝑆𝑇 is 𝑆𝑇 [𝑠], as a rule of thumb. In Eq, it is shown how such a model is derived, sub-
scripts: s=subject, o=object, 0=present, p=predicted. The requested acceleration just have to be con-
verted to pedal force or position.

178
 Vehicle Interactions and Subsystems

Prediction, of object and subject:

𝑠𝑝 𝑠 + 𝑡𝑝 (object vehicle assumed to not accelerate)
𝑝 + 𝑡𝑝
𝑠𝑝 𝑠 + 𝑡𝑝 + 𝑡𝑝2 ⁄
Requirement on prediction at 𝑡𝑝 :
𝑠 𝑝 𝑠 𝑝 (𝑠 𝑠) 𝑓 𝑆 𝑓𝑒𝑇 𝑒 𝑝 𝑆𝑇
𝑝
[2.66]
Solve for and 𝑡𝑝 gives the driver model on explicit form:
( )2
(𝑠 𝑠 𝑆𝑇 )
(𝑠 𝑠 + 𝑆𝑇 )
𝑡𝑝

Use only if 𝑡𝑝 > 0, else “prediction” happens in passed time.

Examples of applying the driver model in above Eq is shown in Figure below, where also a general
PVA-time-diagrams (PVA=position, velocity, acceleration) is shown. PVA-time-diagrams are often use-
ful when studying vehicle operations with other moving objects. Different scenarios can be tested, and
driver models can be derived using simple formulas, such as constant speed and constant acceleration.
predicted collision (if
𝑜𝑛 𝑡𝑢 𝑛 no speed changes)
𝒑𝒐 𝒊𝒕𝒊𝒐𝒏 𝑠 predicted collision (if
certain deceleration
of object vehicle)
known (sensed/estimated) unknown (predicted, several scenarios)

safe speed adaptation

prediction (if object keeps
speed and subject decelerates
safe time gap with constant )

prediction time 𝑡𝑝 for “safe speed adaptation”

TTC=Time To Collision
eTTC=enhanced TTC

𝑝𝑟𝑒𝑠𝑒𝑛𝑡 𝑡 𝑒
𝑡 𝑒
𝑜𝑛 𝑡𝑢 𝑛
𝒗𝒆𝒍𝒐𝒄𝒊𝒕𝒚
subject vehicle

object vehicle
𝑡 𝑒
𝑜𝑛 𝑡𝑢 𝑛
𝒂𝒄𝒄𝒆𝒍𝒆𝒓𝒂𝒕𝒊𝒐𝒏
𝑡 𝑒

Figure 2-100: § Left: PVA-time-diagrams (PVA=position, velocity, acceleration) for studying

longitudinal interaction with an object vehicle ahead. Right: PVA diagram for the driver model in
Eq above using 𝑆 𝑓𝑒𝑇 𝑒 𝑝 [𝑠] and some randomly selected 𝑠 and 𝑣 .

2.9.2.3 § Lateral Driver Models

2.9.2.3.1 § Causality of Steering Wheel Angle and Torque
The driver interacts with steering wheel through two channels, steering wheel angle and steering
wheel torque. The angles of the road wheels are functions of steering wheel angle, almost independent
of forces and torques in the steering system. Further, the steering wheel torque is a function of the

179
 Vehicle Interactions and Subsystems

lateral force on the front axle, reduced with assistance from the power steering or steering servo, but
almost independent of the angles.
A frequently used model is that the driver decides a steering wheel angle and expects a certain steer-
ing wheel torque, 𝑇 𝑤 , feedback. Only in very driver-active steering situations, the causality is modelled
the other way, e.g. studying what happens if driver takes hands-off, i.e. steering wheel torque =0. It can
be that driver uses the effect that steering system returns to 0 without driver torque after a quick
step steering by driver.
2.9.2.3.2 § Steering Wheel Angle Driver Models
How driver operates the steering wheel can also be called a lateral driver model. But different ma-
noeuvres call for different driver models. Examples of manoeuvres/situations are:
• Going into curve from straight and vice versa
• Turning in intersection
• Avoidance manoeuvre
• Stabilizing vehicle when sudden lose yaw stability, e.g. in curve
• Compensating road unevenness which cause lateral disturbance
• Lateral speed control during overtaking
2.9.2.3.3 § Example of Steering Wheel Angle Driver Model Based on Biology
An example of a driver model is the Salvucci and Gray model in Figure and Eq below. It is motivated
from human’s biological cognition. The steering wheel angle is denoted 𝑤.
nearpoint

𝑊/ 𝑓
𝑛
farpoint

𝑊/ 𝑊 is some width (such as

vehicle, road or lane width)
Figure 2-101: § Salvucci and Gray model for how driver steers.

𝑤 𝑘𝑓 𝑓 + 𝑘𝑛 𝑛 +𝑘 𝑛 ⇔
[2.67]
⇔ 𝑤 (𝑡) 𝑤 (𝑡 ) + 𝑘𝑓 𝑓 (𝑡) + 𝑘𝑛 𝑛 (𝑡) +𝑘 ∫
0
𝑛 (𝑡 . . 𝑡) 𝑡

It is not well defined in figure above what it means that vehicle “points”. It can be either vehicle head-
ing or vehicle course angle. Probably, the course angle, is most motivated to use. The course angle is
different for different points in the vehicle, so one might need to define also a “mind-point” in the vehi-
cle. In many cases the mind-point is the position of the driver.
The 3 𝑘 parameters that have to be tuned to certain driver, vehicle and operation. Additionally, the dis-
tance to aim-points needs to be tuned, such as if there should be extra margins on top of 𝑊/ and
whether 𝑊 is vehicle, road or lane width. Further on, all these assumptions can change during a simu-
lation, e.g. as driving on a windy country road and suddenly a meeting vehicle shows up might make
the far point jump. So, even if the model equation appears simple, it is far from obvious how to tune
and use it. Hence, there are variants of it which has only one aim-point. We will refer to the reduced
model, with only one aim-point:

𝑤 𝑘 +𝑘 ⇔ 𝑤 (𝑡) 𝑤 (𝑡 )+𝑘 (𝑡) + 𝑘 ∫ 𝑛 (𝑡 . . 𝑡) 𝑡 [2.68]

0

This kind of model is fed with variables from the vehicle and environment states. A drawback is that
the dynamic vehicle behaviour is built into the 𝑘 parameters, meaning that if vehicle dynamics is
changed, e.g. adding a trailer or reducing road friction, the 𝑘 parameters reasonable have to be re-

180
 Vehicle Interactions and Subsystems

tuned. If identifying the 𝑘 parameters as physical parameters, using a dynamic vehicle model, one can
reduce the need for such re-tuning.
Above driver model can be categorized as having “look ahead”. Simpler driver model can be based on
only where the vehicle is at present time, such as present lateral position in lane. More advanced
driver models use a dynamic model for prediction. The vehicle dynamics model is especially needed in
predicting models.

181
Vehicle Interactions and Subsystems

182
 Longitudinal Dynamics

3 LONGITUDINAL DYNAMICS
3.1 Introduction
The primary purpose of a vehicle is transportation, which requires longitudinal dynamics. The chapter
is organised with one group of functions in each section as follows:
• 3.2 Steady State Function
• 3.3 Functions Over (Long)
• 3.4 Functions in (Short) Events
• 3.5 Control Functions

3.1.1 References for This Chapter

• 2.4 Propulsion System and “Chapter 23. Driveline” in Ref (Ploechl, 2013)
• “Chapter 24. Brake System Dynamics” in Ref (Ploechl, 2013)
• “Chapter 27 Basics of Longitudinal and Lateral Vehicle Dynamics” in Ref (Ploechl, 2013)
• “Chapter 6: Adaptive Cruise Control” in Ref (Rajamani, 2012)

3.2 Steady State Functions

Functions as top speed and grade-ability are relevant without defining a certain time period. For such
functions, it is suitable to observe the vehicle in steady state, i.e. independent of time. Those functions
are therefore called steady state functions, in this compendium. The main subsystems that influences
here are the propulsion system, see 2.4, and the (Friction) Brake system, see 2.4.7.

3.2.1 Traction Diagram

The force generated in the prime mover is transmitted through a mechanical transmission to the
wheel which then generates the propulsive forces in the contact patch between tyre and road. In an
electric in-wheel motor, the transmission can be as simple as a single-step gear. In a conventional vehi-
cle, it is a stepped transmission with several gear ratios (i.e. a gearbox). Then, the drivetrain can be
drawn as in Figure 3-1. The torque and rotational speed of the engine is transformed into force and
velocity curves via the mechanical drivetrain and driven wheel. The result is a Traction diagram. The
transformation follows the following formula, if losses are neglected:
𝑇
𝑟 𝑡 𝑜∙ 𝑛 𝑊ℎ𝑒𝑒 𝑅 𝑢𝑠 ∙ [3.1]
𝑊ℎ𝑒𝑒 𝑅 𝑢𝑠 𝑟 𝑡𝑜

prime mover driven wheel 𝒓𝒂𝒕𝒊𝒐

𝑻 𝑭 𝑻
transmission 𝒓𝒂𝒕𝒊𝒐 𝑻 𝒓𝒂𝒅𝒊
(vehicle) body
𝝎 𝑻 𝒓𝒂𝒅𝒊
𝝎⁄𝒓𝒂𝒕𝒊𝒐 𝒗 𝝎 𝒗
𝝎 𝒓𝒂𝒕𝒊𝒐
𝒓𝒂𝒕𝒊𝒐 𝑻 𝑭

𝝎⁄𝒓𝒂𝒕𝒊𝒐
prime mover characteristics:
Traction Diagram
𝑻 𝑭
𝒓𝒂𝒕𝒊𝒐
Multiply torque by
𝒓𝒂𝒅𝒊
𝒓𝒂𝒅𝒊
Multiply rotational speed by
𝝎
𝒓𝒂𝒕𝒊𝒐 𝒗

Figure 3-1: Construction of Traction Diagram.

183
 Longitudinal Dynamics

A traction diagram for a truck is given in Figure 3-2, which also shows that there will be one curve for
each gear.

lowest gear

highest gear

Figure 3-2: Example of engine map and corresponding traction diagram map from a truck.
(D13C540 is an I6 diesel engine of 12.8-litre and 540 hp for heavy trucks.)

Figure 3-3: Relations between diagrams. The circle markers show alternative gears for one
example of downhill operating point, where neutral gear is possible.

Losses in transmission can be included by loss models for transmission, such as:
𝑇
𝜂𝑇 ∙ 𝑟 𝑡 𝑜 ∙ ℎ𝑒𝑟𝑒 𝜂 𝑇 ≤
𝑟 𝑢𝑠
𝜂𝜔 ∙ 𝑟 𝑢𝑠 ∙ ℎ𝑒𝑟𝑒 𝜂𝜔 ≤ [3.2]
𝑟 𝑡𝑜
𝑣 𝑙 ∙
ℎ𝑒𝑟𝑒 𝜂 𝑇 ∙ 𝜂𝜔 𝜂 𝑙 ≤
𝑛 𝑛 𝑇∙
This will move the curves in the first quadrant downwards due to ηT < and to the left due to 𝜂𝜔 < .
Tyre rolling friction is a torque loss mechanism, which on its own yields 𝜂𝜔 and ηT < . Tyre lon-
gitudinal slip is a speed loss mechanism, which on its own yields 𝜂𝜔 < and ηT . See 2.2.1.6. The
multiplication with 𝜂 is only demonstrative and should be seen more generic: in many cases the losses
are additional instead of multiplicative, e.g. for rolling resistance: 𝑟 𝑡 𝑜 (𝑇 𝑇)⁄𝑟 𝑢𝑠
𝑟 𝑡 𝑜 (𝑇 𝑓 )⁄𝑟 𝑢𝑠 . Which wheels’ to use is discussed in 3.2.2.

184
 Longitudinal Dynamics

A traction diagram is a kind of “one degree of freedom graphical model”. The traction diagram is on
complete vehicle level, so the force axis represents the sum of forces from all wheels. This can include
more than one propulsion system and also brakes.

3.2.1.1 § Gear Selection

Gear selection is at which vehicle conditions the gear should be shifted. If we call that “when” to shift,
there is also a question “how” to shift. The “when” is discussed in the following section, while “how” is
not covered well in this compendium, but it is briefly discussed in 2.4.3.1 and exemplified in 2.4.7.1.
3.2.1.1.1 Gear Selection for Steady-state Operating Conditions
In Figure 3-3, there gear selection was discussed for an operating point defined by 3 quantities: ,
and . To be noted is that it was discussed also assuming steady state (constant over time). To
simplify somewhat, we will now discuss optimal gear selection for only a 2-dimensional diagram, con-
ceptually as in the following figure. The regions are typically selected for optimality: For VehicleFor-
ceRequest> small, one typically selects gear for lowest fuel consumption. For VehicleForceRequest<
small, one selects gear optimizes to allow engine and retarder braking, to avoid overheating the fric-
tion brakes.

VehicleForceRequest
VehicleForceRequest

Gear=1 Gear=2 Gear=3 Gear=1 Gear=2 Gear=3

VehicleSpeed VehicleSpeed
0 0
Gear=Neutral Gear=Neutral
-small -small
Gear=1 Gear=2 Gear=3 Gear=1 Gear=2 Gear=3

Figure 3-4: § Gear selection diagram. Right: with hysteresis to avoid non-stationary and too
frequent shifts
The curves in the diagram are also depending on other conditions, such as road grade and vehicle
gross weight. The diagram only indicates sequential gears, while, especially heavy vehicle, often jump
over gears, e.g. shifting directly from gear 1 to 3. The energy optimality of neutral gear is explained in
Figure 3-3. In essence, the neutral gear eliminate torque losses in propulsion system which otherwise
would remain even in the operating condition of 𝑉𝑒ℎ 𝑒 𝑜𝑟 𝑒𝑅𝑒𝑞𝑢𝑒𝑠𝑡 ≈ 0.
3.2.1.1.2 Gear Selection for Transient Operating Conditions
The previous description was made for a steady-state operating condition. Transients, i.e. variations
over time, add the problem to avoid non-stable gear shifting, and especially that gear is shifted too fre-
quent. This is because each shift causes energy losses, but also other negative consequences such as
reduced driveability of the vehicle and lifetime of the propulsion system components.
Too frequent shifting can be avoided by hysteresis in different ways. One way is to define separate up-
and down-shift curves, see previous figure. The hysteresis is achieved by putting these curves much
enough apart. This can be characterized as using a memory.
Another way to avoid too frequent shifting, is to use a timer which explicitly forbids shifting within a
certain time after the last shift. Using the time partly overcome the problem that the area in the dia-
gram where neutral gear is optimal is very narrow.
So, there is a conflict between selecting neutral gear for low energy consumption and avoiding too fre-
quent gear shifts.

185
 Longitudinal Dynamics

With a predictive cruise, 3.5.2.2.1, it is much easier to utilize the neutral gear, since the algorithm can
simply look in the prediction and find out if a too frequent gear shift is predicted or not. In this way
one can allow neutral gear much more often without the drawbacks of too frequent shifting.

3.2.2 Power and Energy Losses

There are power losses 𝑙 (and energy losses 𝑙 ∫ 𝑙 𝑡) which causes an energy consump-
tion 𝑛 for a transport operation. If the operation starts and stops at same speed (often zero) and
same altitude, 𝑛 𝑙 . And 𝑙 ∑ 𝑙 ∑( ∫ 𝑙 𝑡 ) ∫ (∑ 𝑙 ) 𝑡 ∫ 𝑙 𝑡
where denotes different losses. One can count the energy consumption per distance 𝐷
𝑛 ⁄𝑥 (∫ 𝑙 𝑡)⁄∫ 𝑡 [⁄ 𝑁 ⁄ 𝑁]. The 𝐷 can be seen as a time-averaged re-
sistance force, summed over all “parts”, ∀ , where there are losses. If the operation has same character
(hilliness, speed, etc) lover a long distance, ∑ 𝐷 → 𝐷 when 𝑥 → and 𝑡 → . See also 3.3.4.1.

3.2.2.1 Driving Resistance Force

Some losses can be identified as true forces, visible directly in a free body diagram, i.e. we don’t need
to go via a “Energy loss per distance, 𝐷”. Such are forces from gravity due to road grade and aerody-
namic resistance.
Another way to approach this is to study Figure 3-2 and extrapolate that a very low transmission ratio,
i.e., a very high gear, we would enable infinite speed, which of course is not realistic. This is because
“driving resistance force” is missing in Figure 3-2. Such force can be added to traction diagram as a
curve that typically increases with speed. The top speed is found as the intersection between propul-
sion curve and driving resistance force curve, see Figure 3-5.
One part of the driving resistance force comes from driving uphill: the grade or gravitational load on
the vehicle. This is negative when driving down-hill. There is also aerodynamic driving resistance
force, see Eq [2.62]. Grade and aero-dynamic resistance are (vehicle) body forces.
Also rolling resistance is, often, counted as a resistance force. However, it is not a body force, but in-
stead it acts as a torque on the wheel, not a longitudinal force, on each wheel : 𝑇 𝑓 . But
for a non-driven wheel, the rotational equilibrium of the wheel leads to a small negative longitudinal
force in tyre-to-road contact. So, rolling resistance can be included in traction diagram in either the
“supply” or resistance curve. If one want to use the traction diagram to show the friction limit where a
wheel start to spin, > as in 3.2.6, it can be suitable to include rolling resistance in supply
curve for that wheel, but in the resistance curve for the other wheels. In the following, rolling re-
sistance is represented as a torque on driven wheels (which lowers the supply curve) and a longitudi-
nal force on non-driven wheels (which lifts the resistance curve). The traction diagram can then also
host a curve for friction limit for spinning driven wheels, as in Figure 3-10.
Traction diagram
Supplied (from propulsion system):
𝑣 𝑇 𝑓 𝑣 𝑅𝑤

Available for acceleration: 𝑣

Driving resistance: 𝑅
𝑅 + 𝑅 𝑙𝑙 𝑛 𝑛 𝑣 +𝑅
(downhill positive,
air resistance: after ISO8855)
𝑅 0.5 𝜌 𝐴 2

rolling resistance, drv: Note: The rolling resistance on driven axle is

𝑅 𝑙𝑙 𝑛 𝑛 𝑣 𝑓 𝑅 𝑛 𝑛 𝑣
included in the transformation from the prime
mover to 𝑣 . But the rolling resistance on
road grade resistance: 𝑅 non-driven axle leads to a force, 𝑅 𝑙𝑙 𝑛 𝑛 𝑣 , in
topspeed sin the diagram.
Figure 3-5: Traction diagram. Head wind speed, 𝑤 𝑛 , is assumed to be zero. See also Figure
3-10.

186
 Longitudinal Dynamics

2
𝑅 𝑅 𝑙𝑙 + ∙ ∙ sin( )+ ∙ ∙𝜌∙𝐴∙( 𝑤𝑛 )

𝑅 𝑙𝑙 ∑ 𝑓 𝑤 𝑙 [3.3]
non-driven wheels
𝑓 ℎ𝑒𝑒 𝑠 𝑟𝑒 𝑢𝑛 𝑟 𝑒𝑛: 𝑅 𝑙𝑙 𝑓∙ ∙ ∙ cos( )
As seen in Figure 3-5, the supply and resistance curves are drawn in same diagram. The resulting in-
tersection identifies the top speed of the vehicle. This is a stable point in the diagram, so the vehicle
condition at top speed is for steady state (no acceleration).
Figure 3-5 also shows that the acceleration can be identified as a vertical measure in the traction dia-
gram, divided by the mass. However, one should be careful when using the traction diagram for more
than steady state driving. We will come back to acceleration performance later, after introducing the
two effects “Load transfer” and “Rotating inertia effect”.

3.2.2.2 Losses due to Longitudinal Tyre Slip

Consider a vehicle with 𝑁 ≥ wheels. Assume a certain vehicle speed and a certain desired propul-
sion force ∑ = ..𝑁(𝑇 ⁄𝑅𝑤 ). Also assume that wheel torques 𝑇 can be distributed according to 𝑁
equations, e.g. 𝑇 𝑇2 𝑇3 𝑇4 0 for a conventional front axle driven 4-wheeled car. Also assume
same longitudinal stiffness coefficient on all wheels and that out-of-road-plane equilibria and sus-
pension equations defines the vertical forces . . 𝑁 . The, the power loss due to longitudinal tyre slip
can be calculated as:

𝑙 ∑ (𝑇 (𝑅 )) ∑ (𝑇 𝑠 |𝑅 |) ∑ ( |𝑅 |)
𝑅
= :𝑁 = :𝑁 = :𝑁
2 2
∑ ( |𝑅 |) ≈ ∑
𝑅 𝑅
= :𝑁 = :𝑁
An example with a fore-aft-symmetric 2-axle vehicle, propelled on one axle gives:
02 2 2
𝑙 ≈ ( + )
𝑅 ⁄ ⁄ 𝑅
The same vehicle, but propelled equally much on both axles gives:
( ⁄ )2 ( ⁄ )2 2
𝑙 ≈ ( + )
𝑅 ⁄ ⁄ 𝑅
So, twice as much energy is lost due to longitudinal tyre slip if propelling on 1 instead of 2 axles.
When negative wheel torque, one can brake with friction brakes and then there is no energy loss. How-
ever, if braking with electric propulsion, the loss can be negative, meaning that energy is regenerated
to electric energy storage. If braking so much with electric propulsion that wheel rotates rearwards,
there would again be an energy loss, 𝑙 𝑇 𝑛𝑒 𝑡 𝑒 𝑛𝑒 𝑡 𝑒 > 0.

3.2.2.3 Losses due to Lateral Tyre Slip

(This section might require some studying of Chapter 4 for full understanding.)
There are more driving-resistance effects than covered in Equation [3.3]. One example is that none-
Ackermann steering geometry (toe or parallel steering on an axle, or two non-steered axles).
Another effect, which appears also for Ackermann steering geometry, is that power is lost due to lat-
eral axle slip. Now, we use the same simple model as in Figure 4-21, but additionally use ⁄
and define power losses 𝑙 as sliding velocity counterdirected to force:
𝑙 𝑓 𝑓 𝑓 𝑠𝑓 𝑠 .
We also define a Cornering Resistance Coefficient, 𝑅 :

187
 Longitudinal Dynamics

⁄ 2 2
𝑙 𝑓 𝑟
𝑅 (( ) +( ) )≈ ( ) [3.4]
𝐿 𝑟 𝐿 𝑓 𝑦

𝑅 is such that the additional propulsion force due to cornering is ≈ 𝑅 or the additional
power is ≈ 𝑅 . During a transport operation, the cornering in each time instant is typi-
cally described by two variables, e.g. ( 𝑅𝑝 ), but only the combined scalar measure 2⁄
𝑅𝑝 influ-
ences 𝑅 . Hence, we can plot the following graph:
𝑅

Figure 3-6: Left: Cornering Resistance Coefficient. Right: Required steer angle.
Vehicle data: 500 𝑘 𝐿 𝑓 . 5 𝑓 60 [𝑘𝑁/ ] 80 [𝑘𝑁/ ].
Notes:
• The model used above is not advanced enough to differ between which axle is driven. For such
purpose, one would need e.g. the model in Figure 4-17.
• Normal driving is often below 2 or 3 ⁄𝑠 2 , so the coefficient typically stays below 0.01. So, the
influence on energy consumption, during such “maximum normal” negotiation of corners, is
still of the same magnitude as rolling resistance coefficient 𝑅𝑅 ≈ 0.005. .0.0 0.
• For ideally tracking axles, see 2.2.6, 𝑓 → and → , which gives that 𝑓 → 0 and conse-
quently no power loss and no required propulsion force. Therefore, high cornering stiffness is
fuel efficient when cornering.
• When driving extreme cornering, such as driving as fast as possible in a circle on a test-track,
one will experience that the top speed is much lower than driving straight ahead. That is NOT
explained by [3.4]. It would require inclusion of a combined tyre slip model.

3.2.3 Functions After Start

Figure 3-7 shows how the functions can be found in a traction diagram.

3.2.3.1 Top Speed *

Function definition: Top speed is the maximum longitudinal forward speed the vehicle can reach and maintain
on level and rigid ground without headwind.
Top speed is the speed where the sum of all driving resistance terms is equal to the available propul-
sion forces.

3.2.3.2 Gradeability *
Function definition: Grade-ability is the maximum grade that a vehicle is capable to maintain the forward mo-
tion on an uphill road at a certain constant speed, at a certain road friction level and with a certain load. (from
Reference (Kati, 2013))
For vehicles with high installed propulsion power per weight, the road friction can be limiting, but this
is not visualised in Figure 3-7. Since the speeds are higher than for start-ability, the air resistance can-
not be neglected.

188
 Longitudinal Dynamics

Road slope of this curve Road slope of this curve (Grade-ability here limited by
( 𝑟 𝑒2 ) is the Grade- ( 𝑟 𝑒3 ) is the Grade-
ability for speed 2 ability for speed 3 propulsion, not road friction)

Longitudinal
force
𝑅 for
varying
road grades
Road slope of this curve
( 𝑟 𝑒 ) is the Down-
grade holding capability
for speed

0
0 3
Top Speed,
2 speed

maximum engine brake

Figure 3-7: How Top speed, Gradeability and Down grade holding capability is read-out from
Traction diagram.

3.2.3.3 Down-grade Holding Capability *

Function definition: Down-grade holding capability is defined as the maximum down-grade in which the
vehicle with certain weight is able to maintain a certain speed without using friction brakes.
The function is typically of interest for heavy trucks and certain typical certain weight and certain
speed is payload corresponding to maximum allowed weight and 80 km/h downhill.
The function is defined assuming there are clear friction brakes and other brakes, where the other
brakes are typically engine brake and retarders. For newer vehicle concepts having electric propul-
sion, also regenerative braking via reversed electric propulsion motors can be discussed to be allowed.
However, because sometimes the energy storage will be full so that regenerative braking cannot be
applied. Also, a small energy storage will have a limited downhill length, which might call for also pre-
scribing a certain downhill distance.

3.2.4 Starting with Slipping Clutch

As seen in previous traction diagrams, there is no available positive propulsion force at zero speed.
This means that the diagram can still not explain how we can start a vehicle from stand-still.
The concepts in Figure 3-1 were used to create the force-velocity diagram in
Figure 3-8. It shows the smooth curve of a Continuously Variable (ratio) Transmission (CVT) in com-
parison to the stepped transmission. The CVT is the ideal situation for the engine since it can always
let the engine work at a maximum power or minimum fuel consumption (minimum for the momen-
tarily required power). If the CVT has unlimited high ratio, it can actually have a non-zero propulsion
force at zero vehicle speed. Without losses, this force would be infinite, but in reality, it is limited, but
still positive, so the vehicle can start from stand-still.
A stepped transmission, as well as a CVT with limited ratio range, instead needs a clutch to enable
starting from vehicle stand-still. This is shown in Figure 3-8. The highest force level on each curve can
be reached at all lower vehicle speeds, because the clutch can slip. It requires the clutch to be engaged
carefully to the torque level just below the maximum the engine can produce. In traditional automatic
transmissions, the slipping clutch is replaced with a hydrodynamic torque converter, to enable start
from stand-still.

189
 Longitudinal Dynamics

Longitudinal
force, 𝑭𝒙
ICE on continuously variable ratio
(can keep power at maximum ICE power, 𝑇 𝑚 / )

ICE on lowest gear (highest ratio)

slipping clutch ICE on highest gear (lowest ratio)

fully engaged clutch

(start from v=0 is not possible)
0
0 Speed, 𝒗𝒙
Figure 3-8: Force/Speed Curves for a Multiple Gear Transmission and for CVT.

3.2.5 Steady State Vertical Force Distribution be-

tween Axles
The vehicle performance discussed previously does not rely on knowing the distribution of (vertical)
load between the axles. To be able to introduce limitations due to road friction, this distribution must
be known. Hence, we set up the free-body diagram in Figure 3-9.

m*g

Figure 3-9: Free Body Diagram for steady state vehicle. With ISO coordinate system, the road
gradient is positive when downhill. (Rolling resistance force on non-driven axles is included in .)
From the free-body diagram we can set up the equilibrium equations as follows and derive the formula
for load on front and rear axle:
Moment equilibrium, around rear contact with ground ():
∙ ∙ ( ∙ cos( ) + ℎ ∙ sin( )) 𝑅 ∙ ℎ 𝑓 ∙(𝑓+ ) 0 ⇒
∙ cos( ) + ℎ ∙ sin( ) ℎ
⇒ 𝑓 ∙ ∙ 𝑅 ∙
𝑓+ 𝑓+
[3.5]
Moment equilibrium, around front contact with ground ():
∙(𝑓+ ) ∙ ∙ ( 𝑓 ∙ cos( ) ℎ ∙ sin( )) 𝑅 ∙ ℎ 0 ⇒
𝑓 ∙ cos( ) ℎ ∙ sin( ) ℎ
⇒ ∙ ∙ +𝑅 ∙
𝑓 + 𝑓 +

For most vehicles and reasonable gradients, one can neglect ℎ ∙ sin( ) since it is ≪ | 𝑓 ∙ cos( )| ≈
| ∙ cos( )|.

3.2.6 Friction Limit

With a high-powered propulsion system, there is a limitation to how much the vehicle can be pro-
pelled, due to the road friction limit. It is the normal load and coefficient of friction, which limits this.

190
 Longitudinal Dynamics

For a vehicle which is driven only on one axle, it is only the normal load on the driven axle, 𝑣 𝑛,
that is the limiting factor:
min( 𝑝 ∙ 𝑣 𝑛) [3.6]
One realises, from Figure 2-15, that the rolling resistance on the driven axle works as a torque loss and
that the road friction limitation will be limiting 𝑇 𝑛 ∙ 𝑟 𝑡 𝑜 𝑒 ∙ 𝑣 𝑛 rather than limiting 𝑇 𝑛 ∙
𝑟 𝑡 𝑜. Expressed using the rolling resistance coefficient, 𝑓 𝑙𝑙 , gives:
𝑇𝑛 ∙𝑟 𝑡 𝑜
min( 𝑝 ∙ 𝑣 𝑛) min ( 𝑓 𝑙𝑙 ∙ 𝑣 𝑛 ∙ 𝑣 𝑛) [3.7]
𝑟 𝑢𝑠
This is shown in the traction diagram in Figure 3-10, where it should also be noted that the rolling re-
sistance curve only consists of the rolling resistance on the non-driven axles. See also Figure 2-15.

3.2.7 Start Functions

3.2.7.1 Start-ability *
Function definition: Start-ability is the maximum grade that a vehicle is capable to start in and maintain the
forward motion at a certain road friction level and a certain load. (Reference (Kati, 2013))

𝑇𝑛 𝑟 𝑡𝑜 𝑒
force, Fx
Longitudinal

𝑣 𝑛
𝑝
𝑟 𝑢𝑠
𝑇𝑛 𝑟 𝑡𝑜
𝑓 𝑙𝑙 𝑣 𝑛
𝑟 𝑢𝑠

Friction limit: 𝑣 𝑛

𝑅
Maximum 𝑅 0.5 𝜌 𝐴 2

available 𝑅 𝑙𝑙 𝑓 𝑙𝑙 𝑛 𝑣 𝑛

𝑅 𝑙 𝑝 sin
0
0 Speed, 𝒗
Figure 3-10: Traction diagram with Road Friction limitation and Driving Resistance curves.
Figure 3-11 shows how we find the start-ability in the traction diagram. There are two phenomena
that can limit start-ability: propulsion system or road friction. Also, in each case, we can theoretically
reach somewhat higher start-ability by allowing clutch or tyre to slip. However, in practice the start-
ability has to require “forward motion without significant slip in clutch or tyre”, because there will be a
lot of wear and heat in the slipping clutch or tyre. Hence, the lower curves in Figure 3-11 are used. The
reduction is however very small, since the resistance curves does not change very much in this speed
interval (the resistance curves in the figure have exaggerated slope; the air resistance can typically be
neglected for start-ability).
However, the energy loss (heat, wear) in clutch and tyre should be limited also during the starting
sequence. This can limit the start-ability more severely than the slope of the resistance curves, but it
cannot be shown in the traction diagram, since it is limited by energy losses in clutch or tyre, which is a
time integral of 𝑇 ∙ 𝑙 and 𝑇𝑤 𝑙 ∙ 𝑤 𝑙 . There can also be quite other limitations of start-ability,
such as deliberately limited engine torque at low gears, to save drive shafts.

191
 Longitudinal Dynamics

Start-ability limited by propulsion system Start-ability limited by road friction

Road slope of this curve
force

force
Longitudinal

Longitudinal
is the start-ability with
spinning tyres.
Friction limit Road slope of
for this curve is the
varying start-ability with
rolling tyres. 𝑅 for
road slope varying
Road slope of this curve is road slope
the start-ability (ending
with slipping clutch). Friction
limit
lowest gear lowest gear
Road slope of this curve is the start-
ability (ending with engaged clutch).
0 0
0 Speed, 𝒗 0 Speed, 𝒗

Figure 3-11: How Start-ability is read-out from Traction diagram.

3.2.7.2 Towing Capacity *

Function definition: Towing capacity is the maximum vehicle-external longitudinal force the vehicle can have
on its body and start and maintain a certain forward speed at a certain road friction and a certain up-hill gradient.
The driving situation for defining towing capacity is similar to the one for defining grade-ability. Tow-
ing capacity describes how much load the vehicle can tow, in Figure 3-12, on a certain up-hill gradi-
ent. Since towing a load is more relevant as part of a longer transport mission, it is normally also for a
particular constant speed, typically in range 80 to 100 km/h. Since the speed is that large, the air re-
sistance may not be neglected. It is also important consider air resistance of the trailer and that axle
loads can change, which changes both friction limitation and rolling resistance. A free-body diagram is
shown in Figure 3-12. It is noticeable, that there can also be an additional air resistance of the trailer
which will influence in a test of Towing capacity.
For pure off-road vehicles and agriculture tractors, the term “towing” can mean the maximum pulling
force at very low forward speeds at level ground. This is related but different to the above described
towing capacity for road vehicles.
mt*g

m*g

Figure 3-12: Towing Loads. The towing vehicle is front axle driven.

192
 Longitudinal Dynamics

3.3 Functions Over (Long) Cycles

Functions as fuel or energy consumption and emissions are relevant only over longer periods of driv-
ing, typically some minutes to hours of driving. A collective name for this kind of driving can be cycles.
There are different ways of defining such cycles.

3.3.1 Description Formats of Vehicle Operation

This section is about how to describe “all except the vehicle”, e.g. road, traffic, driver and payload. The
overall idea is to model the vehicle operation as independent of the vehicle, so that different vehicles, or
different designs or configurations of a certain vehicle, can be compared in a fair way.

3.3.1.1 (Traditional) Driving Cycles

One way to model vehicle operation is a so-called driving cycle; where the relevant variables are pre-
scribed as function of time. At least on defines speed as a function of time. Examples of commonly used
driving cycles are given in Figure 3-13 to Figure 3-14. In addition, it can also be relevant to give road
inclination as function of time. Engine temperature and selected gear as functions of time may also be
defined. For hybrid vehicles, the possibility to regenerate energy via electric machines is limited in
curves, so curvature radius can also be prescribed as function of time. For heavy vehicles, the weight of
transported goods can be another important measure to prescribe.
FTP and HFTP are examples of cycles derived from logging actual driving, mainly used in North Amer-
ica. NEDC is an example of a “synthetically compiled” cycle, mainly used in Europe. Worldwide harmo-
nized Light duty driving Test Cycle (WLTC) is a work with intention to be used world-wide, see Figure
3-16. WLTC exists in different variants for differently powered vehicle [power/weight].

80 km/h

Figure 3-13: New European Driving Cycle (NEDC). From (Boerboom, 2012)

Federal Test Procedure (FTP) Driving Trace

80 km/h
20 min

Figure 3-14: FTP cycle from http://www.epa.gov/oms/regs/ld-hwy/ftp-rev/ftp-tech.pdf

193
 Longitudinal Dynamics

3.3.1.2 Driving Pattern

A driving cycle can be condensed into a Driving pattern, i.e. a 2-dimensional function of speed and ac-
celeration, as shown in Figure 3-17. Note that the chronological order is no longer represented in such
representation; it is not a dynamic model. Figure 3-17 shows simply a scatter plot of time-sampled
combinations of speed and acceleration. Using the same diagram axis, such information can also be
shown as durations (in seconds or fractions of total time or frequency). A Driving pattern can only be
combined with a steady state model of the vehicle, such as “fuel consumption=function(speed, accelera-
tion)”, as opposed to a dynamic model of the vehicle. The Driving pattern itself includes the driver, so
no driver model is needed.

80 km/h

10 min

Figure 3-15: HFTP cycle from http://www.epa.gov/nvfel/methods/hwfetdds.gif

N/N(rated), %

Time, s
T/T(max), %

Time, s

194
 Longitudinal Dynamics

Figure 3-16: Top: WLTC cycle from http://www.unece.org, from 2015, for light vehicles.
Bottom: European Transient Cycle (ETC) for heavy commercial vehicles. The engine variants, 𝑇 ( )
and (𝑡), are used for certification from 2000. https://www.dieselnet.com/standards/cycles/etc.php

Figure 3-17: FTP cycle converted to a Driving pattern, i.e. a distribution of operating duration in
speed and acceleration domain. From http://www.epa.gov/oms/regs/ld-hwy/ftp-rev/ftp-tech.pdf.
Driving patterns can use more than 2 dimensions, such as [speed, acceleration, road gradient]. In prin-
ciple, they can also use less than 2 dimensions, maybe only [speed]. The (steady state) vehicle model
has to reflect the corresponding dimensions.

3.3.1.3 Operating Cycle

A vehicle independent description uses legal speed rather than actual speed, and it varies with posi-
tion along the route, 𝑠, rather than time. Also other parameters, like road grade, weight of transported
goods etc. are defined in position rather than time. For stops along the route, the stop duration or de-
parture time then has to be separately defined. This leads to a more realistic description (model) of the
vehicle usage, here called Operating cycle. Important is that an Operating cycle becomes vehicle inde-
pendent, so that different vehicles or different vehicle configurations can be fairly compared. See
(Pettersson, On Numerical Descriptions of Road Transport Missions, 2017).
Simulation with an Operating cycle requires some kind of driver model. A consequence is that different
driver models will give different results, e.g., different fuel consumption due to different driver pre-
ferred acceleration. Hence, the driver model itself can be seen as a part of the vehicle usage definition.

3.3.2 Rotating Inertia Influence on Acceleration

In Figure 3-5 it was shown that the acceleration cannot be found directly as a force difference (dis-
tance between curves) divided by vehicle mass. This is because the Traction Diagram does not contain
any dynamics, and dynamics are more complicated than accelerating the vehicle mass. The phenome-
non that occurs is referred to here as “Rotating Inertia Influence on Acceleration”. If no velocity losses,
the rotating part of the propulsion system, e.g. engine and wheels, must be synchronically accelerated
with the vehicle mass. This “steals” some of the power from the propulsion system. This affects the re-
quired propulsion force when following accelerations in a driving cycle.
Consider a wheel rolling which is ideally rolling (no slip), with a free-body diagram and notations as in
Figure 3-18. Setting up 2 equilibrium equations and 1 compatibility equation gives Eq [3.8].

195
 Longitudinal Dynamics

𝑇
𝑤 ∙
𝑤 ∙ 𝑇 ∙ 𝑅𝑤 [3.8]
𝑅𝑤 ∙ ⇒ 𝑅𝑤 ∙
Eliminate and gives:
𝑤 𝑤
( 𝑤 + 2
𝑤 ⁄𝑅𝑤 )∙ 𝑇⁄𝑅𝑤 or
2 [3.9]
𝑘 𝑤 ∙ 𝑇⁄𝑅𝑤 ℎ𝑒𝑟𝑒 𝑘 𝑤 𝑤 + ⁄𝑅𝑤
Figure 3-18: Rolling Note that we can keep and : ⁄𝑅𝑤 and 𝑇⁄(𝑘 𝑅𝑤 ) .
wheel
So, the rotating inertia makes the mass 𝑤 appear a factor 𝑘 larger and the reaction force (ex-
pressed in 𝑇) correspondingly lower. We call the factor 𝑘 the “rotational inertia coefficient”.
A vehicle with total mass will appear to have larger mass due to inertias in the propulsion system.
There are rotational inertias at two places: before transmission, i.e. rotating with same speed as en-
gine: and after transmission, i.e. rotating with same speed as the wheel: 𝑤 . The appearant mass, 𝑘 ∙
, will be dependent on the main transmission ratio 𝑟 as well:
𝑟 𝑤 ∙ 𝑟2
𝑘∙ ∙ 𝑇 ℎ𝑒𝑟𝑒 𝑘 ∙ + 2
+ 2
[3.10]
𝑅𝑤 𝑅𝑤 𝑅𝑤
Typically for a passenger car with traditional ICE propulsion, 𝑘 . . . .4 in the first gear and 𝑘 .
in the highest gear ( ≈ 0. [𝑘 2]
𝑟≈ 4 4 6 𝑅𝑤 0. [ ]). So, the phenomena is significant!
Typically for electrical propulsion of same vehicle, is smaller and there are fewer gears, so 𝑘 is lower
for low speed and higher for high speed.
We can now learn how to determine acceleration from the Traction Diagram, see Figure 3-19.
Propulsion
force,

• 𝑘 = rotating inertia coefficient on gear 1,

without engine intertia and without
wheel inertia, since slipping tyre, i.e.
Gear 1 Gear 2 and 𝑘 =1
Friction limit Slipping clutch
• 𝑘2 = rotating inertia coefficient on gear 1,
Resistance
• 𝑘3 = rotating inertia coefficient on gear 2,
without engine intertia, since slipping
Gear 2 clutch
𝑘2
𝑘

• 𝑘4 = rotating inertia coefficient on gear 2

Speed,
𝑘4
𝑘3

Figure 3-19: Acceleration in Traction Diagram. Rotating inertia effects are shown assuming that
the engine is operated on its maximum curve and gear selection and clutch operation are made for
highest acceleration.
When the clutch is slipping, there is no constraint between engine speed and vehicle, so the term with
disappears from Equation [3.10]. If the wheel spins, both terms and 𝑤 disappear. It should there-
fore be noted that, if tyre slip is modelled, the effect of rotating inertia is regarded without using the 𝑘
factor above. That is, if the mathematical model in [3.11] is used instead of [3.8]. Eq [3.11] gives an ex-
plicit form model with two states [ ] and it becomes increasingly computational demanding the
larger is. (Similar decoupling between inertias and 𝑤 happens if torque converter or elastic
driveshafts are modelled between engine inertia and wheel.)

196
 Longitudinal Dynamics

∙
∙ 𝑇 ∙ 𝑅𝑤
𝑅𝑤 [3.11]
𝑠 ℎ𝑒𝑟𝑒 𝑠
|𝑅𝑤 |

3.3.3 Four Quadrant Traction Diagram

When the driving cycle shows a deceleration, which is larger than can be achieved with resistance
force, we need to brake with a combination of engine braking and friction brakes. If only friction brak-
ing is used, it can be with engaged or disengaged clutch. That influences the rotating inertia coefficient
by using or not using the term in Equation [3.10], respectively. The traction diagram can be ex-
tended to also cover engine braking and friction braking. However, the friction brake system is seldom
limiting factor for how negative the longitudinal force can be. But the road friction is, see Figure 3-20.
Propulsion force, Fx
Braking friction limit, +
Gear 1
Propulsion friction limit,
+ 𝑣𝑛≈+ ⁄

Gear 2

Speed,

Marks upper and lower limits between

which operating points can be found
using propulsion and friction brake
Propulsion friction limit, systems, including slipping clutch and
𝑣𝑛≈ ⁄ spinning driven axle.

Reverse gear
Braking friction limit,

Figure 3-20: Traction Diagram in 4 quadrants. One of two axles is assumed driven, which limits
propulsion to ≈half of braking friction limit. Up-hill slope is assumed, which is seen as asymmetric
resistance.

3.3.4 Functions Over Cycles

3.3.4.1 Energy Consumption *
Function definition: Energy (or Fuel) Consumption is the amount of energy [ ] (or fuel [𝑘 ] or
[ 𝑡𝑟𝑒]) consumed by the vehicle per performed transportation amount. Transportation amount can e.g. be meas-
ured in km, 𝑘 𝑝𝑒𝑟𝑠𝑜𝑛 𝑘 𝑡𝑜𝑛 𝑘 or 3 𝑘 . The vehicle operation has to be defined, e.g. with a
certain driving cycle (speed as function of time or position), including road gradient, cargo load, road surface condi-
tions, etc.
The consumption arises in the prime mover, see Figure 2-71 and Figure 2-72, but as a cycle measure it
is dependent of the overall vehicle operation.
Driving cycles are used for legislation and rating for passenger cars. For commercial vehicles, the legis-
lation has been for the engine alone, and not for the whole vehicle. CO2 -rating for commercial vehicles
is newly introduced, see http://ec.europa.eu/clima/events/docs/0096/vecto_en.pdf.

197
 Longitudinal Dynamics

3.3.4.2 Assessing Energy Consumption and Other Cycle

Measures
The measures of all functions mentioned in 3.3.4 can be assessed in some sort of driving cycle/opera-
tion cycle computation. Such computations are described in the following, as exemplified by only one
measure, the Energy Consumption.
How to predict the consumption for a vehicle during a certain driving cycle is rather straight-forward
using what has been presented earlier in this chapter. Since a driving cycle is a prediction of how the
vehicle is moving, it actually stipulates the acceleration of a mass, which calls for an “inverse dynamic
analysis”. In such one assumes that the driving cycle is met exactly, which means that both required
wheel speed and required wheel torque can be calculated for each time instant. Then, via a propulsion
and brake system model, the corresponding fuel consumption in the engine can also be found for each
time instant. A summary of such an inverse dynamic algorithm for prediction of fuel or energy con-
sumption is given in Equation [3.12].
For each time step in the driving cycle:
• Calculate operating point for vehicle (speed and acceleration) from
driving cycle. Acceleration is found as slope of v(t) curve. Other quan-
tities, such as road slope, also needs to be identified;
• Select gear (and clutch state, tyre spin, friction brake state, etc) to
obtain this operating point. Select also friction brake, especially for
operating points which can be reached using only friction brake. If the
vehicle has an energy buffer, regenerative braking is also an option;
• Calculate required actuation from propulsion system on the driven wheel,
i.e. rotational speed and shaft torque; [3.12]
• Calculate backwards through propulsion system, from wheel to prime mo-
tor. It gives the operating point for prime mover (rotational speed and
torque);
• Read prime mover consumption [in kg/s or W=J/s, not specific consump-
tion, not efficiency] from prime mover consumption map;
• Sum up consumption [in kg or J] with earlier time steps, e.g. using the
Euler forward integration method: AccumulatedConsumption =
AccumulatedConsumption + Consumption*TimeStepLength;
end;

The final accumulated consumption [in kg or J] is often divided by the total covered distance in the
driving cycle, which gives a value in kg/km or J/km. If the fuel is liquid, it is also convenient to divide
by fuel density, to give a value in litre/(100*km). It can also be seen as a measure in 2, which is the
area of the “fuel pipe” which the vehicle “consumes” on the way.
For hybrid vehicles (with energy buffers) the same driving cycle can be performed (same (𝑡)) but
ending with different level of energy in the buffer. Also, the level when starting the driving cycle can be
different. This makes it unfair to compare energy consumption only as fuel consumption, one should
rather weight it to energy cost, €/𝑘 or €/(ton km).
3.3.4.2.1 Forward and Backward Simulation
We should note that the calculation scheme in Equation [3.12] does not always guarantee a solution.
An obvious example is if the driving cycle prescribes such high accelerations at such high speeds that
the propulsion system is not enough, i.e. we end up outside maximum torque curve in engine diagram.
This is often the case with “inverse dynamic analysis”, i.e. when acceleration of inertial bodies is pre-
scribed, and the required force is calculated. An alternative is to do a dynamic analysis, which means
that a driver model calculates the pedals in order to follow the driving cycle speed approximately, but
not exactly. Inverse dynamic analysis is often more computational efficient, but limits what can phe-
nomena that can be modelled in the propulsion and brake system. The computational benefit is espe-
cially large if state variables can be omitted, which is often the case but not always. Inverse dynamics
and dynamic simulations are sometimes referred to as backward and forward simulation, respectively;
see Reference (Wipke, Cuddy, & Burch, 1999).
§ Causality

198
 Longitudinal Dynamics

The causality (data flow directions) in an Explicit form model changes a lot if changing between For-
ward and Backward simulation. However, the Mathematical model experiences minimal changes, basi-
cally only the parameter 𝑡 𝑛 changes from 𝑡 𝑛 > 0 to 𝑡 𝑛 0, i.e. the driver can be seen as “ideal”,
i.e., in some “magic” way, making the vehicle follow what the Environment stipulates: . The Envi-
ronment is where the Driving Cycle (𝑡) or (𝑠) is modelled. A big difference is that the Backward
simulation can often lack solution, e.g. if propulsion system or brake system is weaker than driving cy-
cle demands.
Forward simulation:
”Non-ideal Driver” Vehicle
For instance:
Environ- 𝑒
ment 𝐷 𝑣 Driver
𝑡𝐷 𝑣 Actuators
Interpretator
𝑒 𝑓𝑢𝑛 𝐷 𝑣
with 𝑡𝐷 𝑣 > 0

Backward simulation:
”Ideal Driver” Vehicle
For instance:
Environ- 𝑒
ment 𝐷 𝑣 Driver
𝑡𝐷 𝑣 Actuators
Interpretator
𝑒 𝑓𝑢𝑛 𝐷 𝑣
with 𝑡𝐷 𝑣 0

Figure 3-21: § Data flow diagrams showing causality for Forward and Backward simulation.
Backward leads to changes marked with yellow and that integration ∫ 𝑡 changes to
differentiation ⁄ 𝑡 in Vehicle.
3.3.4.2.2 Stepped, CVT and Energy Buffering Main Transmissions
For a given driving cycle (𝑡) there is a certain [ ] for each time instant. For different types of
main transmissions, one can select operating point [𝑇 ] in prime mover diagram differently, see
Figure 3-22. CVTs (or tightly stepped transmissions) enables operation close to optimal curve of en-
gine. However, CVTs typically have more losses than stepped transmissions. So, for commercial vehi-
cles, a higher number of gears is more realistic than real CVTs.
As mentioned in Figure 3-22, one can operate in the optimal point by an energy buffer or by propelling
intermittent with a varying vehicle speed. The latter case can be seen as if the vehicle’s own kinetic en-
ergy ( 2⁄
) is used to buffer energy, see (Hedman, 1994) which shows fuel savings in the magni-
tude of 10%. Intermittent operation is on market as part of predictive cruise, see 3.5.2.2.1.

199
 Longitudinal Dynamics

optimal point (with

𝑇 energy buffer, one can
operate intermittent here)
gear 3
optimal ratio gear 2 (=optimal stepped
(reachable transmission gear)
with CVT)
gear 1

𝑙
optimal curve for varying
(reachable with CVT)

Figure 3-22: Conceptual difference between Stepped, Continuously Variable and Energy Buffering
transmission. Operating intermittent would only follow the driving cycle with same average speed.

3.3.4.3 Transport time *

Function definition: Transport Time is the time [𝑠] or [ℎ] for the vehicle to perform a certain transportation
mission. Transportation mission can e.g. be defined in terms of distance, payload, road topography, etc.
Transport Time is often in conflict with Energy Consumption. Either, one have independent require-
ments (constraints), but one can also formulate the total transport cost and minimize it, e.g.: 𝑜𝑠𝑡
𝑛𝑒𝑟 𝑦 𝑜𝑠𝑡[€⁄ ] 𝑛𝑒𝑟 𝑦 𝑜𝑛𝑠𝑢 𝑝𝑡 𝑜𝑛[ ] + 𝑇 𝑒 𝑜𝑠𝑡[€⁄ℎ ] 𝑇𝑟 𝑛𝑠𝑝𝑜𝑟𝑡𝑇 𝑒[ℎ] . (That cost model
is for exemplification and very simplified. Important other costs are investment and maintenance.)
For commercial freight traffic, 𝑇 𝑒 𝑜𝑠𝑡 is mainly the driver salary, so it can be quantified. For person
transport, especially private travels, it is much more difficult to motivate a number of the cost. One can
typically find studies which uses travel time cost≈ 0 𝑆 𝐾⁄ℎ ≈ €/ℎ for private travels in Sweden.
3.3.4.3.1 § Quantifying Personnel Transportation Time
The travel time cost (per hour) for transport of persons is very debatable how to quantify. When it is
needed, one should think of that it depends on transport mean. For instance, it is higher for air travels
than for train because the traveller is less free to do other things; travel with passenger car is in be-
tween. Increased driving automation, e.g. adding automatic steering to automatic longitudinal control,
should probably reduce the travel cost for passenger cars. References:
• Swedish Transport Administration (Trafikverket), Analysmetod och samhällsekonomiska
kalkylvärden för transportsektorn: ASEK 6.1, Kapitel 7 Värdering av kortare restid och transport-
tid, https://www.trafikverket.se/contentassets/4b1c1005597d47bda386d81dd3444b24/asek-
6.1/07_restid_o_transporttid_a61.pdf , in Swedish, 2018.
• Lyons and Urry, Travel time use in the information age, Transportation Research Part A 39
(2005) 257–276, doi:10.1016/j.tra.2004.09.004, 2004.

3.3.4.4 Emissions *
Function definition: As Energy consumption but amount of certain substance instead of amount of energy.
There are emission maps where different emission substances (NOx, HC, etc.) per time or per pro-
duced energy can be read out for a given speed and torque. This is conceptually the same as reading
out specific fuel consumption or efficiency from maps like in Figure 2-71 and Figure 2-72. A resulting
value can be found in mass of the emitted substance per driven distance.
Noise is also sometimes referred to as emissions. It is not relevant to integrate noise over the time for
the driving cycle, but maximum or mean values can have relevance. Noise emissions are very periph-
eral to vehicle dynamics.

200
 Longitudinal Dynamics

3.3.4.5 Tyre Wear *

Function definition: Tyre wear is the worn-out tyre tread depth on a vehicle per performed transportation
amount. Transportation amount can be measured as for Energy consumption. Tyre wear as a vehicle function has
to consider all tyres on the vehicle, e.g. as maximum over the wheels (assuming that all tyres are changed when one
is worn out) or average (assuming that single tyres are exchanged when worn out).
There are models for tyre wear (e.g. outputting “worn tread depth per time”), see Equation [2.49]. For
a certain driving cycle, we can integrate the 𝑊𝑒 𝑟𝑅 𝑡𝑒 [in mass/s or mm tread depth/s], over time
similar to energy consumption rate, which becomes worn material [in mass or mm tread depth]. The
wear rate per wheel is a function of the total slip, so it can include both longitudinal slip (propulsion
and brake) and lateral slip (from cornering and toe angles).
Generally, the worn material will be different for different axles, or wheels, so a tyre change strategy
might be necessary to assume to transform the worn material on several axles into one cost. The cost
will depend on whether one renews all tyres on the vehicle at once or if one change per axle. The tyre
wear is a cost which typically sums up with energy cost and cost of transport time (e.g. driver salary,
for commercial vehicles).

3.3.4.6 Range *
Function definition: Range [km] is the inverted value of Energy consumption [kg/km, litre/km or J/km], and
multiplied with fuel tank size [kg or litre] or energy storage size [J].
The range is how far the vehicle can be driven without refilling the energy storage, i.e. filling up fuel
tank or charging the batteries from the grid. This is in principle dependent on how the vehicle is used,
so the driving cycle influences the range. In principle, the same prediction method as for energy con-
sumption and substance emissions can be used. In the case of predicting range, you have to integrate
speed to distance, so that you in will know the travelled distance.

3.3.4.7 Acceleration Reserve *

Function definition: Acceleration reserve is the additional acceleration the vehicle will achieve within a certain
time (typically 0.1..1 s) without manual gear-shifting by pressing accelerator pedal fully, when driving in a certain
speed on level ground without headwind. For vehicles with automatic transmissions or CVTs the certain time set
can allow automatic gearshift (or ratio-change) or not. The reserve can also be measured in propulsion force.
In general terms, the lowest consumption is found in high gears. However, the vehicle will then tend to
have a very small reserve in acceleration. It will, in practice, make the vehicle less comfortable and less
safe to drive in real traffic, because one will have to change to a lower gear to achieve a certain higher
acceleration. The gear shift gives a time delay.
Figure 3-23 shows one way of defining a momentary acceleration reserve. The reserve becomes gener-
ally larger the lower gear one selects. A characteristic of electric propulsion systems is that an electric
motor can be run at higher torque for a short time than stationary, see Figure 2-70. On the other hand,
the stationary acceleration reserve is less gear dependent, since an electric motor can work at certain
power levels in large portions of its operating range.
One can calculate the acceleration reserve at each time instant over a driving cycle. However, integra-
tion of acceleration reserve, as we did with fuel, emissions and wear, makes less sense. Instead, a mean
value of acceleration reserve tells something about the vehicle’s driveability. Minimum or maximum
values can also be useful measures.
Acceleration reserve was above described as limited by gear shift strategy. Other factors can be limit-
ing, such as energy buffer state of charge for parallel hybrid vehicles or how much overload an electric
machine can take short term, see right part of Figure 3-23.

201
 Longitudinal Dynamics

Vehicle with conventional Vehicle with electric

Longitudinal

Longitudinal
force, 𝑭𝒙

force, 𝑭𝒙
propulsion system propulsion system
Mass∙Acceleration
Mass∙Acceleration gear 1 Reserve on gear 1
Reserve on gear 1 Mass∙Acceleration
gear 1 Mass∙Acceleration Reserve on gear 2
Reserve on gear 2

Short term
Acceleration
gear 2 Reserve

gear 2
0 0
0 Speed, 𝒗 0 Speed, 𝒗
Figure 3-23: Acceleration reserves for different gears. Large dots mark assumed operating points.

3.3.5 Load Transfer with Rigid Suspension

Longitudinal load transfer redistributes vertical force from one axle to the other. The off-loaded axle
can limit the traction and braking. This is because the propulsion and brake systems are normally de-
signed such that axle torques cannot always be ideally distributed.
For functions over longer events it is often reasonable to consider the suspension as rigid. We start
with the free-body diagram in Figure 3-24, which includes acceleration, .

x
y

Figure 3-24: Free Body Diagram for accelerating vehicle. Rolling resistance in 𝑓 and .
Note that the free-body diagram and the following derivation is very similar to the derivation of Equa-
tion [3.5], but we now include the fictive force ∙ .
Moment equilibrium, around rear contact with ground:
𝑓 ∙𝐿+ ∙ ∙ ( ∙ 𝑜𝑠( ) + ℎ ∙ 𝑠 𝑛( )) 𝑅 ∙ ℎ ∙ ∙ℎ 0 ⇒
∙ 𝑜𝑠( ) + ℎ ∙ 𝑠 𝑛( ) ℎ ℎ
⇒ 𝑓 ∙( ∙ ∙ ) 𝑅 ∙
𝐿 𝐿 𝐿
[3.13]
Moment equilibrium, around front contact with ground:
+ ∙𝐿 ∙ ∙ ( 𝑓 ∙ 𝑜𝑠( ) ℎ ∙ 𝑠 𝑛( )) 𝑅 ∙ ℎ ∙ ∙ℎ 0 ⇒
𝑓 ∙ 𝑜𝑠( ) ℎ ∙ 𝑠 𝑛( ) ℎ ℎ
⇒ ∙( ∙ + ∙ )+𝑅 ∙
𝐿 𝐿 𝐿
These equations confirm what we know from experience, the front axle is off-loaded under accelera-
tion with the load shifting to the rear axle. The opposite occurs under braking.

202
 Longitudinal Dynamics

The load shift has an effect on the tyre’s grip. If one considers the combined slip conditions of the tyre
(presented in Chapter 2), a locked braking wheel limits the amount of lateral tyre forces. The same is
true for a spinning wheel. This is an important problem for braking as the rear wheels become off-
loaded. This can cause locking of the rear wheels if the brake pressures are not adjusted appropriately.
See more in 3.4.4.

3.3.5.1 § Influence of Aerodynamic Resistance

Influence of Aerodynamic Resistance becomes different depending on the engineering problem. For
𝜕
instance, assuming 0 and 𝑓 , the influence on from 𝑅 can be expressed as 𝑟𝑧
𝜕 𝑖𝑟
𝜕 𝑚 𝑖𝑟 𝑖𝑟 𝑚 𝜕 𝑖𝑟 𝑖𝑟 𝑚
( + + ) ( + ).
𝜕 𝑖𝑟 2 𝐿 𝐿 𝜕 𝑖𝑟 𝐿 𝐿
𝜕
For an “inverse dynamic” problem, is given (independent). Then 𝜕 0 because a change in 𝑅
𝑖𝑟
𝜕 𝑟𝑧
is assumed to be compensated with 𝑓 + . Therefore 𝜕 𝐿
𝑖𝑟
.
𝑖𝑟

If the problem is instead a “natural causality” problem, e.g. simulation of acceleration. Then is given
𝜕
(independent) and is dependent of 𝑅 , via 𝑚
𝑖𝑟
hence 𝜕 𝑟𝑧 𝑖𝑟
𝐿
.
𝑖𝑟

So, depending on which of these engineering problem one has in mind, it can either be true that 𝑅
influences only if ℎ ≠ 0 or only if ℎ ≠ ℎ.

3.3.5.2 § Influence of Acceleration of Rotational Parts

The influence of 𝑓 𝑓 and on load transfer will now be studied. See the FBD (and equilibrium)
of the whole vehicle in Figure 1-22. If included, it will appear as terms 𝑓 𝑓 and in negative
pitch direction. If we simplify to assume flat ground and 𝑓 𝐿⁄ and 𝑅 0:
Moment equilibrium, around rear contact with ground:
𝐿
𝑓 ∙𝐿+ ∙ ∙ ∙ ∙ℎ 𝑓 𝑓 0 ⇒

𝑓 ≈ ≈ ⁄𝑅𝑤 ℎ
⇒ 𝑓 ≈{ }≈ ( ∙( ))
𝑓 ≈ ≈ 𝐿 ℎ 𝑅𝑤
With realistic values for a passenger vehicle, ≈ 500 𝑘 and 0.5 𝑘 2
, the influence from
wheel rotational inertia is ( )⁄( ℎ 𝑅𝑤 ) ( 0.5)⁄( 500 0.5 0. ) ≈ 0.005 0.5% on the
load transfer which in most cases can be considered as neglectable. In a strong take-off or brake apply,
when ≫ ⁄𝑅𝑤 , the equation would give a much higher influence but just for a short while. To cap-
ture the short transient in such cases, one should model also suspension as in 3.4.5.2.2.

3.3.5.3 Varying Road Pitch

The model in 3.4.4..3.4.5.2 assumes flat but not level road, i.e. is constant. An example where
varies is when passing a crest or a sag, see Figure 3-25. If negotiating a curve at the same time as a
crest, a vehicle can lose vertical force under tyres so that lateral grip is affected.
Moment equilibrium, around rear and front wheel contact with ground gives:

″ 2)
∙ cos( ) + h ∙ sin( ) ℎ ℎ
𝑓 ∙ (( + ∙ ∙ ) 𝑅 ∙
𝐿 L L
[3.14]
″ 2) 𝑓 ∙ cos( ) h ∙ sin( ) ℎ ℎ
∙ (( + ∙ + ∙ )+𝑅 ∙
𝐿 𝐿 𝐿
2
Assuming that we have the road as (𝑠), then arctan ( ) ≈ and ″
2 .
Note that this model is assuming that vertical variations of road are much larger than wheelbase and
track width and same on left and right side of the road/vehicle. Else the variation would be called road
unevenness, which will be more treated in Chapter 5.

203
 Longitudinal Dynamics

If models with body vertical and pitch motion and suspension springs, such as in 3.4.5 and 3.4.5.2 it is
often suitable to express the vertical fictive force, ̈ with instead of ″ 2
. The fictive
force downwards will then be 2
𝜅 instead. This can be understood from basic
geometry, ″ ≈ 𝜅 , where 𝜅 is the road pitch curvature [ ⁄ 𝑒𝑛 𝑡ℎ], see Figure 3-25.

crest sag
s
𝑦
′ ′
s s

• The variable s is the distance along the road.

• Road gradient versus inertial coordinate
system, ≈ ′, is a function of s

• where ′ is a function of s.
• Vertical acceleration in inertial coordinate
system, ̈ ≈ 2 ,
̈

2
• where 2 is a function of s.

Figure 3-25: Free Body Diagram for driving over non-flat vertical road profile.

3.3.6 Acceleration
Acceleration performance like, typically, 0-100 km/h over 5..10 s, will be addressed in this section.
These accelerations are relatively steady state (vehicle pitch and heave are relatively constant), so the
suspension compliance is not considered.
Accelerations will also be covered in 0, as being shorter events. The vehicle pitch and heave vary more
and, consequently, the suspension compliance becomes important to model. This modelling is also
more suited for braking, which typically involve suspension more than propulsion.

3.3.6.1 Acceleration Performance *

Function definition: Acceleration performance is the time needed to, with fully applied accelerator pedal,
increase speed from a certain speed to another certain higher speed, at certain road friction on level ground without
headwind and certain load.

3.3.6.2 Solution using Integration over Time

A front-wheel-drive passenger car with a stepped gearbox should accelerate from 0 to 100 km/h. A
Matlab code is given in Equation [3.15], which simulates the acceleration uphill from stand-still, using
simple numerical integration. The code calculates the possible acceleration in each of the gears, and
one mode with slipping clutch. In each time step it selects that which gives the highest acceleration.
The numerical data and results are not given in the code, but some diagrams are shown in Figure 3-26.
The code is not fully documented, only using equations so far presented in this compendium.
dt=0.1; t_vec=[0:dt:10]; vx_vec(1)=0;
for i=1:length(t_vec)
vx=vx_vec(i);
Rres=-m*g*sin(p)+froll*m*g*cos(p)+0.5*roh*A*cd*vx*vx;
%if gear 1 (clutch engaged)
ratio=ratios(1);
we=vx*ratio/radius; [3.15]
Te=interp1(Engine_w,Engine_T,we);
Fx=Te*ratio/radius;
ax=(Fx-Fres)/(m+(Jw+Je*(ratio^2))/(radius^2));
Ffz=m*(g*lr/L-ax*h/L);
if Fx>mu*Ffz
%if gear 2

204
 Longitudinal Dynamics

Fx=mu*Ffz;
ax=(Fx-Rres)/m;
end
ax1=ax; %(clutch engaged)
ratio=ratios(2);
… then similar as for gear 1
ax2=ax;
%if gear 3 (clutch engaged)
ratio=ratios(3);
… then similar as for gears 1 and 2
ax3=ax;
%if clutch slipping on gear 2
ratio=ratios(2);
wc=vx*ratio/radius; %speed of output side of clutch
Te=max(Engine_T);
we=Engine_w(find(Engine_T>=Te)); %engine runs on speed where max torque
Fx=Te*ratio/radius;
ax=(Fx-Fres)/(m+Jw/(radius^2));
Ffz=m*(g*lr/L-ax*h/L);
if Fx>mu*Ffz
Fx=mu*Ffz;
ax=(Fx-Rres)/m;
end
if wc>we %if vehicle side (wc) runs too fast, we cannot slip on clutch
ax=-inf;
end
ax0=ax;
[ax,gear_vec(i)]=max([ax0,ax1,ax2,ax3]); vx_vec(i+1)=vx+ax*dt;
end
Phenomena that are missing in this model example are:
• Gear shifts are assumed to take place instantly, without any duration
• The option to use slipping clutch on 1st and 3rd gear is not included in model
• The tyre slip is only considered as a limitation at a strict force level, but the partial slip is not
considered for simplification. The code line “we=vx*ratio/radius;” is hence not fully cor-
rect. Including the slip, the engine would run at somewhat higher speeds, leading to that it
would lose its torque earlier, leading to worse acceleration performance.
• Load transfer is assumed to take place instantly quick; delays due to Suspension compliance, as
described in 3.3, are not included.

205
 Longitudinal Dynamics

Engine 4
Traction diagram
x 10
2
Torque[Nm] gear 1
160
Power[kW] 1.8 gear 2
gear 3
140 1.6
resistance
slip limit
120 1.4

Longitudinal force [N]

max achievable
1.2
100

1
80
0.8
60
0.6
40
0.4
20
0.2

0 0
0 100 200 300 400 500 600 700 0 5 10 15 20 25 30 35 40
rotational speed [rad/s] speed [m/s]

Simulation result
14

12

10

vx [m/s]
6
gear (0=clutch slips on gear 2)

0
0 1 2 3 4 5 6 7 8 9 10
t [s]

Figure 3-26: Example of simulation of acceleration, using the code in Equation [3.15].

3.3.6.3 § Solution using Integration over Velocity

In some cases, with simple vehicle models and monotonously varying velocity, there is an alternative
way to find the relation between and 𝑡 is to separate the differential equation:
∙
∙ ∙ ( ) 𝑅 ( ) ⇒ 𝑡⇒
𝑡 ( ) 𝑅 ( )
𝑣 𝑣
∙ ∙
⇒ ∫ ∫ 𝑡⇒ 𝑡 𝑛 ∫
( ) 𝑅 ( ) ( ) 𝑅 ( )

If simple mathematic functions can be used to describe ( ) and ( ) the solution can even be on
closed form.

3.3.6.4 § Solution using Simulation over Position

In some cases, with simple vehicle models and monotonously varying position 𝑠, we rewrite to a, initial
value problem with position instead of time. The following table shows an example of driving over a
crest and set speed is reduced on the way down. The ′ denotes / 𝑠.

206
 Longitudinal Dynamics v_xReq v_x a_x
30

20 v_xReq v_x a_x

30
10
20
With time 𝑡 as With position 𝑠 as 0
independent variable independent variable 10
-10
0

′ -20
∙ 𝑅 𝑅 -10 0 10 20 30

( )⁄𝑡𝐷 𝑣 ( )⁄𝑡𝐷 𝑣 -20

t [s]

𝑅 𝑅 (𝑠 ) 𝑅 𝑅 (𝑠 ) 0
s
10
t [s]
20 30

𝑠 𝑡′ ⁄
200
′ s
0
200
0 10 20 30
0 t [s]

Figure 3-27: § Comparing simulation with time and position0 as independent

10 variable.
20 30
t [s]
It is often possible to study transport missions between two positions with 𝑠 as independent variable.
An advantage is in those cases one does not have to solve the state variable 𝑡. If one need to solve for 𝑡,
the model has the drawback of being singular for stand-still ( 0) due to 𝑡 ′ ⁄ . It can be over-
come by using 𝑡 as independent variable for velocities close to 0. Generally, the step selection can be
difficult also if only close to 0; it can then be an idea to judge which time step Δ𝑡 would be suitable
if time was independent variable, and then select a varying position step Δ𝑠 Δ𝑡.

3.4 Functions in (Short) Events

This section targets models and methods to define and verify functions in a certain and shorter time
frame, typically 0.5 to 5 seconds. It can be both acceleration and deceleration. (Friction) Brake system
and phenomena as load transfer then becomes important, why these are presented early. But first,
some typical driving manoeuvres are presented.

3.4.1 Typical Test Manoeuvres

When applying the longitudinal actuator systems (propulsion system and brake system) there are a
couple of different situations which are typical to consider:
• Straight line maximum braking from, typically 100 km/h to stand-still for passenger cars.
• Braking in curve with significant lateral acceleration, see References (ISO, 2006) and (ISO,
2011).
• Straight line acceleration, typically 0 to 100 km/h and 80-100 km/h.
• Accelerating in curve with significant lateral acceleration.
For these four main situations, one can also vary other, typically:
• At high road friction and at low friction, often called “hi-mu” and “lo-mu”
• At different road friction left and right, often called “split-mu”
• At sudden changes in road friction, called “step-mu” or “step-up” and “step-down” when going
to higher and lower friction, respectively.
• At high speed, typically 200 km/h, for verifying lateral stability
• At different up-hill/down-hill gradients
• At different road banking (slope left to right)
A propelled or braked wheel or axle develops a longitudinal force, , counter-acting the rotation. is
limited by the road friction: | |𝑚 ∙ . (not | |𝑚 ∙ 𝑓 ∙ , see Figure 2-15).
Braking Coefficient ⁄ , is a property defined for an axle or a single wheel. It can be seen as the
utilized friction coefficient, 𝑙 = brake force/normal load. The 𝑙 should not be mixed up with
(available) friction coefficient, ; the relation between them is 𝑙 ≤ .

207
 Longitudinal Dynamics

3.4.2 Deceleration Performance

There are some different functions that measures braking performance or deceleration performance.

3.4.2.1 Braking Efficiency *

Function definition: Braking Efficiency is the ratio of vehicle deceleration and the best brake-utilized axle (or
wheel), while a certain application level of the brake pedal at a certain speed straight ahead, at certain road friction
on level ground without head-wind and certain load at a certain position in the vehicle.
In equation form, the Braking Efficiency becomes . If Braking Efficiency =
max( 𝑖 𝑢 𝑖 ) max( 𝑖 / 𝑖𝑧 )
∀𝑖 ∀𝑖
1 = 100%, the distribution of braking is optimal.
An alternative definition which is only relevant for maximum brake pedal apply is Braking Efficiency
. This definition requires that all tyres have the same and known characteristics at the particular
𝑝 𝑘
ground. The value of 100% then means that the braking is maximum.

3.4.2.2 Braking Distance *

Function definition: Braking Distance is the distance travelled during braking with fully applied brake pedal
from a certain speed straight ahead to another certain lower speed, at certain road friction on level ground without
headwind and certain load at a certain position in the vehicle.
For passenger cars one typically brakes fully from 100 km/h and then the braking distance is typically
around 40 m (average 9.65 ⁄𝑠 2 ). For a truck it is typically longer, 51..55 m ( 7.5. . 7 ⁄𝑠 2 ).

3.4.2.3 Stopping Distance *

Function definition: Stopping Distance is the distance travelled from that an obstacle becomes visible to driver
have taken the vehicle to stand-still. Certain conditions, as for Braking Distance, have to be specified, but also a
certain traffic scenario and a certain driver to be well defined.
Stopping Distance is the braking distance + the “thinking/reaction distance”, which depends on the
speed and the reaction time. The reaction time of a driver is typically 0.5..2 seconds.

3.4.3 Pedal Functions

3.4.3.1 Pedal Response *
Function definition: Accelerator pedal response is how vehicle acceleration varies with accelerator pedal po-
sition, for a certain vehicle speed and possibly certain gear, on level ground without pressing the brake pedal.
Function definition: Brake pedal response is how vehicle deceleration varies with brake pedal force, for a cer-
tain vehicle speed, on level ground without pressing the accelerator pedal.
These functions, together with the functions in 3.4.3.2, enable the driver to operate the vehicle longitu-
dinally with precision and in an intuitive and consequent way. The requirements based on above func-
tion definitions, are typically that the translation of pedal position (or force) to vehicle acceleration (or
deceleration) should be consistent, progressive and oscillation-free.
For accelerator pedal steps, there should be enough acceleration, but also absence of “shunt and shuf-
fle” (driveline oscillations). When accelerator pedal is suddenly lifted off, there shall be certain decel-
eration levels, depending on vehicle speed and gear selected.

3.4.3.2 Pedal Feel *

Function definition: Accelerator pedal feel is the pedal’s force response to pedal position.
Function definition: Brake pedal feel is the pedal’s position response to pedal force.
These functions, together with the functions in 0, enable the driver to operate the vehicle longitudi-
nally with precision and in an intuitive and consequent way.

208
 Longitudinal Dynamics

3.4.4 Brake Proportioning

For brake performance it is important that both axles are used as much as possible during braking. But
one also should consider that, in most driving situations, it is preferred that the front wheels lock first,
because:
• A vehicle with locked front wheels ( 𝑓 0) tends to be yaw stable. However, steering ability
is lost, so vehicle continues straight, incapable of curving its path.
• A vehicle with locked rear wheels ( 0) tends to be yaw unstable. It turns around and
ends up sliding with the rear first.
Hence, there are trade-offs when designing the wheel torque distribution. Same reasoning works for
propulsion, if “locked“ is replaced by “spin”, meaning large positive longitudinal tyre slip. Spin at front
makes vehicle more yaw stable than spin rear. The yaw stability then has a trade-off with acceleration
performance.
Wheel torques is influenced simultaneously by both propulsion system and (friction) brake system,
especially if regenerative braking with electric propulsion system. So, coordination of brake and pro-
pulsion systems might be needed.
The basic function of a brake system is that brake pressure (hydraulic on passenger cars and pneu-
matic on trucks) is activated so that it applies brake pads towards brake discs or drums. In a first ap-
proximation, the pressure is distributed with a certain fraction on each axle. For passenger cars this is
typically 60..70% of axle torque front. In heavy trucks, the proportioning varies a lot, e.g. 90% for a
solo tractor and 30% for heavy off-road construction rigid truck. The intention is to utilize road fric-
tion in proportion to the normal load, but not brake too much rear to avoid yaw instability.
If neglecting air resistance and road grade in Eq [3.13], the vertical axle loads can be calculated as
function of deceleration ( ). An ideal brake distribution would be if each axle always utilize same
𝑓 𝑟 𝐴𝑠𝑠𝑢 𝑒 𝑓 𝑓𝑧 𝑓
fraction of available friction: ⇒ { } ⇒ ⇒
𝑓 ∙ 𝑓𝑧 𝑟 ∙ 𝑟𝑧 𝑓 𝑟 𝑟𝑧 𝑟
𝑚∙( ∙ 𝑟 ∙ ) ∙𝑙𝑟 ∙
𝐿
𝑓
𝐿
Combining with 𝑓 + ∙ gives the optimal 𝑓 and :
𝑚∙( ∙ + ∙ ) ∙𝑙𝑓 + ∙
L L

𝑙 𝑙𝑓
𝑓 ∙ ∙(𝑟 ) and ∙ ∙( + )
𝐿 𝐿 𝐿 𝐿

or, if eliminating : 𝑟
𝑓𝑥 √𝑟 4 ℎ 𝐿
𝑓𝑥
[3.16]
ℎ ℎ
𝑙𝑟 𝑙𝑓
or: 𝑓 ⁄
𝐿 𝑚 𝐿
and ⁄
𝐿
+
𝑚 𝐿

If including non-zero and non-zero (but small) : 𝑓 ⁄ ⁄𝐿 ∙ ℎ⁄( ∙ 𝐿) + 𝑅 ∙

(ℎ ℎ )⁄( ∙ 𝐿) and ⁄ 𝑓 ⁄𝐿 + ∙ ℎ⁄( ∙ 𝐿) 𝑅 ∙ (ℎ ℎ )⁄( ∙ 𝐿) . Air re-
sistance 𝑅 and road gradient , of course, influences so that we need to adjust to reach a certain
. However, road gradient does not influence the distribution of longitudinal tyre force be-
tween axles and air resistance only if ℎ ≠ ℎ .
Eq [3.16] is plotted for variation in centre of gravity height and longitudinal position in Figure 3-28.

209
 Longitudinal Dynamics

⁄𝑁

Conceptual purpose
with pressure limiting
valve or EBD

𝑓 ⁄𝑁
Figure 3-28: Brake Proportioning diagram. The curved curves mark optimal distribution for some
variation in position of centre of gravity.
The proportioning is done by selecting pressure areas for brake calipers, so the base proportioning
will be a straight line, marked as “Hydrostatic brake proportioning”. For passenger cars, one typically
designs this so that front axle locks first for friction below 0.8 for lightest vehicle load and worst vari-
ant. For heavier braking than 0.8 , or higher (or front-biased) centre of gravity, rear axle will lock
first, if only designing with hydrostatic proportioning.
To avoid rear axle lock-up, one restricts the brake pressure to the rear axle. This is done by pressure
limiting valve, brake pads with pressure dependent friction coefficient or Electronic Brake Distribu-
tion (EBD). In principle, it bends down the straight line as shown in Figure 3-28. With pressure de-
pendent values one gets a piece-wise linear curve, while pressure dependent friction coefficient gives
a continuously curved curve. EBD is an active control using same mechatronic actuation as ABS. EBD is
the design used in today’s passenger cars, since it comes with ABS, which is now a legal requirement
on most markets.
On heavy vehicles with EBS (Electronic Brake System) and vertical axle load sensing, the brake pres-
sure for each axle can be tailored. For modest braking (corresponding to deceleration ≤ ⁄𝑠 2) all
axles are braked with same brake torque, to equal the brake pad wear which is importance for vehicle
maintenance. When braking more, the brake pressure is distributed more in proportion to each axle’s
vertical load.

3.4.5 Heave and Pitch

So far, in 3.3.5 and 3.4.4, we modelled transfer of vertical forces between axles, but neither heave and
pitch motion nor displacement. This will be added in 3.4.5. In 3.4.5.1, the load transfer is steady state
and the linkage “trivial”. In 3.4.5.2, the load transfer is transient and the linkage “non-trivial”.

3.4.5.1 Steady State Load Transfer and Trivial Linkage

Additional to that the axle vertical loads change due to acceleration , there are also change in out-of-
road-plane motion (heave and pitch). In the following section, we study constant acceleration, e.g.
when mild braking for a long time. We propose the steady state model in Figure 3-30. The model dif-
fers between the “unsprung mass” (wheels and the part of the suspension that does not heave) and the
“sprung mass” or “body” (parts that heaves and pitches as one rigid body). The wheels are assumed to
be linked to the body through “trivial linkage” as in Figure 2-61.

210
 Longitudinal Dynamics

Ideal curve

ECE regulation
limits to this region

[data from BMW 320i E46]

Figure 3-29: Brake Proportioning. From (Boerboom, 2012). If looking carefully, the “HydroStatic”
curve is weakly degressive, thanks to brake pad material with pressure dependent friction
coefficient.
Suspension model with “trivial suspension”

𝑥 𝑓

ℎ
ℎ

0 𝑓 0 𝑓
𝑓

𝐿
𝑥 𝑓 , are displacements from a Quasi steady-state assumed, so that longitudinal
static stand-still position. acceleration ( ) may be non-zero, but vertical
𝑓 0 means that road is smooth. and pitch acceleration are zero.

Figure 3-30: Model for steady state heave and pitch due to longitudinal wheel forces.
There is no damping included in model, because their forces would be zero, since there is no displace-
ment velocity, due to the steady-state assumption. As constitutive equations for the compliances
(springs) we assume that displacements are measured from a static condition and that the compli-
ances are linear. The road is assumed to be smooth, i.e. 𝑓 0.

𝑓 𝑓 + 𝑓∙( 𝑓 𝑓) 𝑛 + ∙( )
[3.17]
ℎ𝑒𝑟𝑒 𝑓 + ∙ 𝑛 𝑓 ∙ 𝑓 ∙ 0

211
 Longitudinal Dynamics

We see already in free-body diagram that 𝑓 and always act together, so we rename 𝑓 +
𝑤 , where w refers to wheel. The assumption of “trivial linkage” explains how longitudinal forces are
transferred between wheels and body. Equilibrium then gives:
∙ + 𝑤 0
∙ 𝑓 0 [3.18]
∙ 𝑓 ∙ 𝑓 𝑤 ∙ℎ ∙ (ℎ ℎ) 0
Compatibility, to introduce body displacements, and 𝑝 , gives:

𝑓 𝑓 ∙ 𝑛 + ∙ [3.19]
Solving constitutive relations, equilibrium, compatibility using Matlab Symbolic toolbox gives:
>> clear, syms zf zr Ffz Frz Ffz0 Frz0 ax z py
>> sol=solve( ...
'Ffz=Ffz0-cf*zf', ...
'Frz=Frz0-cr*zr', ...
'Ffz0+Frz0=m*g', ...
'Ffz0*lf-Frz0*lr=0', ...
'-Fair-m*ax+Fxw=0', ...
'm*g-Ffz-Frz=0', ...
'Frz*lr-Ffz*lf-Fxw*h-Fair*(hair-h)=0', ...
'zf=z-lf*py', ...
'zr=z+lr*py', ...
zf, zr, Ffz, Frz, Ffz0, Frz0, ax, z, py);
The solution is given in Eq [3.20]:
𝑤

𝑓 ∙ 𝑓 ∙
∙( 𝑤 ∙ℎ+ ∙ (ℎ ℎ )) [3.20]
𝑓 ∙ ∙ 𝐿2
𝑓 +
𝑝 ∙( 𝑤 ∙ℎ+ ∙ (ℎ ℎ ))
𝑓 ∙ ∙ 𝐿2
In agreement with intuition and experience the body dives (positive pitch) when braking (negative
𝑤 ). Further, the body centre of gravity is lowered (negative z) when braking and weaker suspension
front than rear ( 𝑓 ∙ 𝑓 < ∙ ), which is normally the chosen design for cars.
The air resistance force is brought into the equation. It can be noted that for a certain deceleration,
there will be different heave and pitch depending on how much of the decelerating force that comes
from air resistance and from longitudinal wheel forces. But, as already noted, heave and pitch does not
depend on how wheel longitudinal force is distributed between the axles.

3.4.5.2 Transient Load Transfer and Non-Trivial Linkage

Compared to 3.4.5.1, we will now model also the transients of load transfer. If we study longer events
when the wheel force is applied and then kept constant for a longer time (1-5 s), it is often a good
enough model. But if the wheel forces vary more, we need to capture the transients better. Then it is
important to consider that the linkage can transfer some of the wheel longitudinal forces. When study-
ing the transients, it is also relevant to consider the damping.
There are basically two modelling ways to include the suspension linkage in the load transfer: through
a pitch centre or through a pivot point for each axle, see Figure 3-31.
3.4.5.2.1 Model with Pitch Centre
This model will not be deeply presented in this compendium. It has drawbacks in that it has only one
suspended degree of freedom. Also, it does not take the distribution of longitudinal wheel forces be-
tween the axles into account. These shortcomings is not very important for studying dive and squat,
but they are essential when studying rapid individual wheel torque changes in time frames of 0.1 s,
such as studying ABS or traction control. So, since the model with axle pivot points is more generally

212
 Longitudinal Dynamics

useful and not much more computational demanding (and probably easier intuitively), that model is
prioritized in this compendium.

Model with pitch center Model with Axle Pivot Points

vz
wx vx

h-hPC
h-hPC

PC

lPCf lr lf
lPCf
PC=Pitch Centre

Figure 3-31: Models for including suspension effects in longitudinal load transfer
3.4.5.2.2 Model with Axle Pivot Points
Behold the free-body diagram in Figure 3-32. The road is assumed flat, 𝑓 0. The force play in
the rear axle is shown in more detail. and are reaction forces in the pivot point. The is the
force in the elasticity, i.e. where potential spring energy is stored. The torque 𝑇 is the shaft torque, i.e.
from the propulsion system. Both torque from propulsion and brake system contribute to . But
torque from friction brake 𝑇 is not visible in free-body diagram, unless decomposed in suspension
and wheel, as in the right-most part of Figure 3-32. This is because the friction braking appears as in-
ternal torque (or, depending on the brake design, probably forces) between brake pad and brake calli-
per. Any part of the longitudinal wheel force that is applied via reaction torque to unsprung parts will
not add to shaft torque, such as an electric motor mounted on unsprung parts or propulsion via longi-
tudinal propeller shaft to a final gear (as usual for rigid propelled axles).
The term is easy to forget, but it does influence especially when is large. The term can be ex-
plained as the other centripetal accelerations (giving centrifugal (fictive) forces) in 4.4.2.3. The term
is generally much smaller. Both the terms appear due to that velocities and accelerations are
expressed as components in the vehicle fix (and hence rotating) coordinate system. We could intro-
duce also velocities and accelerations in ground fix coordinate system, with subscripts differing be-
tween [ 𝑣 𝑣 ] and [ ] and between [ 𝑣 𝑣 ] and [ ], in a similar way shown in
4.4.2.3.3 for yaw rotation where it is much more important to differ between vehicle and ground fix.
We assume that displacements are measured from the forces 𝑓 and , respectively, and that the
compliances are linear. The total constitutive equations become:

𝑓 𝑓 + 𝑓 ∙( 𝑓 𝑓) 𝑛 + ∙( ) [3.21]
Now, there are two ways of representing the dynamics in spring-mass systems: Either as second order
differential equations in position or first order differential equations in velocity and force. We select
the latter, because it is easier to select suitable initial values. Then we need the differentiated versions
of the compliance’s constitutive equations:
𝑓 0+ 𝑓 ∙( 𝑓 𝑓) 𝑓 ∙ 𝑓 𝑛 0+ ∙( ) ∙ [3.22]

213
 Longitudinal Dynamics

Decomposed in smaller
Rear axle: free body diagrams, to
+ explain 𝑇 and 𝑇 .
𝑓 𝑓
Suspension:
+

ℎ 𝑇

𝑒𝑓 𝑒
Wheel:
0 𝑓 0
0 𝑓 0 𝑓 𝑇
𝑓
𝑓
𝐿 𝑓

𝑥 𝑓 are displacements from a static stand-still position. 𝑓 𝑓 0 means flat road.

Figure 3-32: Free-body diagram for model with Axle Pivot Points. (The model assumes drive shafts
from propulsion system mounted on body. If rigid shaft with longitudinal propeller shaft, 𝑇 0.)
The damper forces are denoted 𝑓 and . They will appear in the equilibrium equations quite simi-
lar to 𝑓 and . Note that the damping coefficients, 𝑓 and , are the effective ones, i.e. the ones de-
fined at the wheel contact point with ground, as opposed to the physical ones defined for the actual
physical damper. C.f. effective stiffness in 3.4.5.

𝑓 𝑓 ∙( 𝑓 𝑓) 𝑓 ∙ 𝑓 𝑛 ∙( ) ∙ [3.23]
Now, 3 equilibria for whole vehicle and 1 moment equilibria for each axle around its pivot point gives:
∙ + 𝑓 + 0 ( 𝑡ℎ ≈ )
∙ ∙ + 𝑓 + 0 ( 𝑡ℎ )
∙ + ∙ 𝑓 ∙ 𝑓 ( 𝑓 + )∙ℎ 0 [3.24]
( )∙ ∙𝑒 +𝑇 0
( 𝑓 + 𝑓 𝑓 )∙ 𝑓 𝑓 ∙ 𝑒𝑓 + 𝑇 𝑓 0
It can be noted that the unsprung parts are considered massless. Also note, that a “trivial suspension
model” (as used in Eq [3.17] and see also Figure 2-61) is a special case which falls out from the equa-
tions by letting → , so that + . With such trivial suspension, there is no difference in
vehicle motion if actuation is done with shaft torque 𝑇 or via reaction to unsprung parts, 𝑇 .
Equilibria for wheels on each axle:
𝑇 +𝑇 𝑅𝑅 𝑅𝑤
𝑅𝑤 𝑤 0
{ ⇒
𝑇𝑓+𝑇 𝑓 𝑓 𝑓𝑤 𝑓 𝑅𝑅 𝑓 𝑅𝑤
𝑅𝑤 0
[3.25]
neglect rotational inertia 𝑇 +𝑇 𝑅𝑤 0
⇒{ }⇒ {
neglect rolling resistance 𝑇𝑓 +𝑇 𝑓 𝑓 𝑅𝑤 0
Compatibility, to connect to body displacements, and , gives:

𝑓 𝑓 ∙ 𝑛 + ∙
𝑓 𝑓 ∙ 𝑛 + ∙ [3.26]
𝑛
By combining constitutive relations, equilibrium and compatibility we can find explicit function so:
• 𝑆𝑡 𝑡𝑒𝐷𝑒𝑟 𝑡 𝑒𝑠 𝑥𝑝 𝑡 𝑜𝑟 𝑢𝑛 𝑡 𝑜𝑛(𝑆𝑡 𝑡𝑒𝑠 𝐼𝑛𝑝𝑢𝑡𝑠)
• States: [ 𝑓 ]
• State Derivatives: [ 𝑓 ] (if wheel rotational inertias neglected)
• Inputs: [𝑇 𝑓 𝑇 𝑇 𝑓 𝑇 ] or [ 𝑓 𝑇𝑓 𝑇 ]
The ExplicitFormFunction can be integrated with well-established methods for numerical ODE. Such
simulation of is shown in 3.4.8.1. Note that the model is linear.

214
 Longitudinal Dynamics

3.4.5.2.3 Additional Phenomena

It is relevant to point out the following, which are not modelled in this compendium:
• Stiffness and damping may be dependent of wheel (vertical) displacement and wheel steer
angle. One way of inserting this in the model is to make the coefficients varying with spring
force, which is a measure of how much compressed the suspension is. Here, non-linearity
within spring working range, as well as bump stops, can be modelled. Also, position of pivot
points (or pitch and roll centres) can be dependent of wheel displacement steer angle.
• Dampers are often deformation direction dependent, i.e. different damping coefficients are
suitable to use for compression and rebound. Typical is 2..4 times softer (smaller d [N/(m/s)])
in compression than in rebound.

3.4.6 Steady State Heave and Pitch, Non-Trivial

Linkage
If we study long term steady state for the model described in 3.4.5.2.2 we will find a steady state model
comparable with the model in 3.4.5. So, Equations [3.21] to [3.26] are combined. We also neglect air
resistance force for clarity of equations. Equation [3.21] becomes, in Matlab format:
>> sol=solve( ...
Fsf==Fsf0-cf*zf, Fsr==Fsr0-cr*zr, ...
Fsf0+Fsr0==m*g, Fsf0*lf-Fsr0*lr==0, ...
-m*ax+Ffx+Frx==0, -m*0-m*g+Ffz+Frz==0, ...
-J*0+Frz*lr-Ffz*lf-(Ffx+Frx )*h==0, ...
(Frz-Fsr-0 )*gr-Frx*er+Tsr==0, ...
(Fsf+0-Ffz )*gf-Ffx*ef+Tsf==0, ...
zf==z-lf*py, zr==z+lr*py, ...
Fsf0,Fsr0, Ffz, Frz, zf,zr, ax, Fsf,Fsr, z,py)

%Results:
% sol.ax = (Ffx+Frx)/m
% sol.z = -(Tsr*cf*gf*lf^2 -Tsf*cr*gr*lr^2 +Tsr*cf*gf*lf*lr
[3.27]
-Tsf*cr*gr*lf*lr -Frx*cf*er*gf*lf^2 +Ffx*cr*ef*gr*lr^2
+Ffx*cf*gf*gr*h*lf +Frx*cf*gf*gr*h*lf -Ffx*cr*gf*gr*h*lr
-Frx*cr*gf*gr*h*lr -Frx*cf*er*gf*lf*lr +Ffx*cr*ef*gr*lf*lr)
/(cf*cr*gf*gr*(lf + lr)^2)
% sol.py = -(Tsr*cf*gf*lf +Tsr*cf*gf*lr +Tsf*cr*gr*lf
+Tsf*cr*gr*lr +Ffx*cf*gf*gr*h +Frx*cf*gf*gr*h
+Ffx*cr*gf*gr*h +Frx*cr*gf*gr*h -Frx*cf*er*gf*lf
-Ffx*cr*ef*gr*lf -Frx*cf*er*gf*lr -Ffx*cr*ef*gr*lr)
/(cf*cr*gf*gr*(lf+lr)^2)
% sol.Ffz = -(Ffx*h +Frx*h -g*lr*m)/(lf+lr)
% sol.Frz = +(Ffx*h +Frx*h +g*lf*m)/(lf+lr)
The solution can be compared with corresponding solution in Equation [3.20]. The is exactly the
same. Then, a general reflection is that the displacement, z and py, in Equation [3.27] follows a com-
plex formula, but that they are dependent on how the 𝑤 𝑓 + is applied: both dependent on
distribution between axles and dependent on how much of the axle forces ( 𝑓 and ) that are actu-
ated with shaft torques (𝑇 𝑓 and 𝑇 , respectively). In Figure 3-33, dashed lines show the solutions
from Equation [3.20].

3.4.7 Pitch Functions at Transient Wheel Torques

3.4.7.1 Dive at Braking *
Function definition: Dive at braking is pitch angle of the vehicle body when applying a step in deceleration re-
quest to a certain level. Either peak pitch or quasi-steady state pitch angle can be addressed.

215
 Longitudinal Dynamics

Now, study the suspension at front axle in Figure 3-32. When the axle is braked, 𝑓 will be negative
and push the axle rearwards, i.e. in under the body. The front of the vehicle will then be lifted as in pole
jumping. This means that this design counter-acts the (transient) dive of the front. (Only the transient
dive will be reduced, while the dive after a longer time of kept braking is dependent only on the stiff-
nesses according to Equation [3.20].) The design concept for front axle suspension to place the pivot
point behind axle and above ground is therefore called “anti-dive”.
If the braking is applied without shaft torque 𝑇 𝑓 , a good measure of the Anti-dive mechanism is 𝑒𝑓 / 𝑓 .
This is the normal way for braking, since both the action and reaction torque acts on the axle. For in-
board brakes, or braking via propulsion shaft, the reaction torque is not taken within the axle, but the
reaction torque is taken by the vehicle body. The action torque 𝑇 𝑓 then appears in the equilibrium
equation for the axle, as shown in Equation [3.24]. If we neglect the wheel rotational dynamics for a
while, we can insert 𝑇 𝑓 𝑓 ∙ 𝑅𝑤 in the equation with 𝑇 𝑓 in Equation [3.24]:

( 𝑓+ 𝑓 𝑓 )∙ 𝑓 𝑓 ∙ 𝑒𝑓 + 𝑇 𝑓 0 𝑡ℎ 𝑇 𝑓 𝑓 ∙ 𝑅𝑤 ⇒
⇒( 𝑓+ 𝑓 𝑓 )∙ 𝑓 𝑓 ∙ 𝑒𝑓 + 𝑓 ∙ 𝑅𝑤 0 ⇒ [3.28]
⇒( 𝑓+ 𝑓 𝑓 )∙ 𝑓 𝑓 ∙ (𝑒𝑓 𝑅𝑤 ) 0
We can then see that a good measure of the Anti-dive mechanism is (𝑒𝑓 𝑅𝑤 )/ 𝑓 instead.

3.4.7.2 Squat at Propulsion *

Function definition: Squat at propulsion is pitch angle of the vehicle body when applying a step in acceleration
request to a certain level. Either peak pitch or quasi-steady state pitch angle can be addressed.
Now, study the suspension at rear axle in Figure 3-32. When the axle is propelled, will push the
axle in under the body. This means that this design reduces the rear from squatting (transiently). The
design concept for rear axle suspension to place the pivot point ahead of axle and above ground is
therefore called “anti-squat”.

3.4.7.3 Anti-dive and Anti-Squat Designs

With Anti-dive front and Anti-squat rear, we avoid front lowering at braking and rear lowering at ac-
celeration, respectively. But how will the designs influence the parallel tendencies: that rear tend to lift
at braking and front then to lift at propulsion? Well, they will luckily counteract also these: Braking at
rear axle will stretch the rear axle rearwards and upwards relative to the body. When propelling the
front axle, the propulsion force will stretch the front axle forwards and upwards relative to the body.
(If one brakes at one axle and propels at the other, the reasoning is not valid. This mode can be desired
for a hybrid vehicle with ICE on front axle and electric motor on rear axle, if one would like to charge
batteries “via the road”.)
In summary: Anti-dive and anti-squat refer to the front diving when braking and the rear squatting
when acceleration. Anti-dive and anti-squat can be measured in fractions: Anti-dive for =𝑒𝑓 / 𝑓 or
(𝑒𝑓 𝑅𝑤 )/ 𝑓 and Anti-squat=𝑒 / or (𝑒 𝑅𝑤 )/ . Normal values are typically 0.05..0.15.

3.4.8 Acceleration and Deceleration

Acceleration performance like, typically, 0-100 km/h over 5..10 s, was addressed in 3.3.6. In present
Section we address the similar functionality but include larger transients, such as when wheel longitu-
dinal wheel force is changed more rapidly, typically changing ± ⁄ during 0.2-0.5 s.

3.4.8.1 Deceleration Performance *

Function definition: See 3.4.2.2.
Deceleration performance can now be predicted, including the suspension mechanisms. It is a very im-
portant function, and every decimetre counts when measuring braking distance in standard tests like
braking from 100 to 0 km/h. The active control of the brake torques (ABS function) is then very

216
 Longitudinal Dynamics

important, and this is so fast dynamics that the suspension mechanisms of Anti-lift and Anti-dive influ-
ences. The position of the load in the vehicle will influence, since it influences the load transfer.
We will now set up a mathematical model, see Equation [3.29], which shows how the normal forces
change during a braking event. It is based on the physical model in Figure 3-32. Driving resistance con-
tributes normally with a large part of the deceleration, but we will neglect this for simplicity, just to
show how the suspension mechanism works. The equations in the model are presented in the dynamic
modelling standardized format “Modelica”, and are hence more or less identical to Equation [3.22] to
[3.26]. (The term is included but makes no visible difference.)
//Actuation:
Ffx = if 1 < time and time < 3 then -0.4*m*g else 0;
Frx = if 3 < time and time < 7 then -0.4*m*g else 0;
Tsf/Rw = 0;
Tsr/Rw = if 5 < time and time < 7 then -0.4*m*g else 0;

//Motion equations:
der(z) = vz; der(py) = wy;
//Constitutive equations for the springs:
der(Fsf) = -cf*vfz; der(Fsr) = -cr*vrz;
//Constitutive equations for the dampers: [3.29]
Fdf = -df*vfz; Fdr = -dr*vrz;
//(Dynamic) Equilibrium equations:
-m*(der(vx)-vz*wy) + Ffx + Frx = 0;
-m*(der(vz)-vx*wy) - m*g + Ffz + Frz = 0;
-Jy*der(wy) + Frz*lr - Ffz*lf - (Ffx + Frx)*h = 0;
(Frz - Fsr - Fdr)*gr - Frx*er + Tsr = 0;
(Fsf + Fdf - Ffz)*gf - Ffx*ef + Tsf = 0;
//Compatibility:
zf = z - lf*py; zr = z + lr*py;
vfz = vz - lf*wy; vrz = vz + lr*wy;
The simulation results are shown in Figure 3-33. It shows a constant deceleration, but it is changed
how the decelerating force is generated. At time=3 s, there is a shift from braking solely on front axle
to solely on rear axle. The braking is, so far, only done with friction brakes, i.e. generating torque by
taking reaction torque in the axle itself. At time=5 s, there is a shift from braking with friction brakes to
braking with shaft torque. It should be noted that if we shift axle or shift way to take reaction torque,
gives transients even if the deceleration remains constant.
One can also see, at time=1 s, that the normal load under the braked axle first changes in a step. This is
the effect of the Anti-dive geometry. Similar happens when braking at rear axle, due to the Anti-squat
geometry. Since brake performance is much about controlling the pressure rapidly, the transients are
relevant, and the plots should make it credible that it is a control challenge to reach a high braking effi-
ciency.

3.4.8.2 Acceleration Performance *

Function definition: See 3.3.6.1.
The model presented in Eq [3.29] can also be used to predict acceleration performance in a more accu-
rate way compared to 3.3.6. Especially, the more accurate model is needed when propelling or braking
on the limit of tyre to road adhesion, since the normal load of each tyre then is essential. It is a chal-
lenge to control the propulsion and brake wheel torques to utilize the varying normal loads under
each axle.

217
 Longitudinal Dynamics
w ithCentrifugalForce.vx
50

0
0 1 2 3 4 5 6 7 8 9
w ithCentrifugalForce.z [m] w ithCentrifugalForce.py [rad] w ithCentrifugalForce.z
0.03

Different displacements, both

0.02
transiently and steady state

0.01
py= /[𝑟 ]
0.00 /[ ]
/[ ]
-0.01
𝑓 /[ ]
-0.02

0 1 2 3 4 5 6 7 8 9
w ithCentrifugalForce.Fzf w ithCentrifugalForce.Fzr w ithoutCentrifugalForce.Fzf
1.4E4

1.2E4 𝑓
1.0E4

8.0E3
Same steady state vertical
6.0E3
forces, but transiently different
4.0E3
0 1 2 3 4 5 6 7 8 9
Braking on front axle Braking on rear axle Braking on rear axle
with friction brakes with friction brakes with shaft torque

Figure 3-33: Deceleration with constant vehicle deceleration but varying ways of actuation. (Eq
[3.29]. With the centripetal term (solid) and without (dashed). Dotted shows without anti-
dive/-squat geometry, i.e. 𝑓 . The term makes no visible difference.)

3.4.9 Other Functions

There are many more longitudinal functions, originating from the attribute Driving dynamics, which
could have fitted in 0. Examples of such:
• Off-road accessibility: Ability to pass obstacles of different kind, such as uneven ground, ex-
treme up- and down-hills, mud depth, snow depth, etc.
• Shift quality: Quick and smooth automatic/automated gear shifts
• Shunt & shuffle: Absence from oscillation for quick pedal apply, especially accelerator pedal.

218
 Longitudinal Dynamics

3.5 Control Functions

Some control functions will be described. First, some general aspects on control are given.

3.5.1 Longitudinal Control

Some of the most important sensors available in production vehicles and used mainly for longitudinal
control are listed below. (Sensors for instrumented vehicles for testing can be many more.)
• Wheel Speed Sensors, WSS. For vehicle control design, one can often assume that “sensor-
close” software also can supply information about longitudinal vehicle speed.
• Vehicle body inertial sensors. There is generally a yaw velocity gyro and a lateral accelerome-
ter available, but sometimes also a longitudinal accelerometer. The longitudinal accelerometer is
useful for longitudinal control and longitudinal velocity estimation.
• Pedal sensors. Accelerator pedal normally has a position sensor and brake pedal force can be
sensed via brake system main pressure sensor. Heavy vehicles often have both a brake pedal po-
sition and brake pressure sensors.
• Today’s vehicles have environment sensors (camera, radar, GPS with electronic map, etc.) that
can give information (relative distance and speed, etc.) about objects ahead of subject vehicle. It
can be both fixed objects (road edges, curves, hills, …) and moving objects (other road users, ani-
mals, …). See also 2.7 Environment Sensing System.
• Information about what actuation that is actually applied at each time instant is available, but
it should be underlined that the confidence in that information often is questionable. Infor-
mation about axle propulsion torque is generally present, but normally relies on imprecise mod-
els of the whole combustion process and torque transmission, based on injected amount of fuel
and gear stick position. (Electric motors can typically give better confidence in estimation, espe-
cially if motor is close to the wheel without too much transmission in between.) Wheel individ-
ual friction brake torque is available, but normally rely on imprecise models of the brake sys-
tems hydraulic/pneumatic valves and disc friction coefficient, based on brake main cylinder
pressure.
• Information about what actuation levels that are possible upon request (availability or capa-
bility) is generally not so common. It is difficult to agree of general definitions of such infor-
mation, because different functions have so different needs, e.g. variations in accepted time de-
lay for actuation.
• Sometime wheel/axle forces can be sensed. One case is when pneumatic suspension. More ex-
treme variants are under development, such as sensors in the wheel bearings which can sense
forces (3 forces and roll and yaw moment) and sensors in shafts.

3.5.2 Longitudinal Control Functions

3.5.2.1 Pedal Driving *
Function definition: See 3.4.3.
Pedal driving is often not seen as comparable with other control functions, but they become more and
more relevant to define as such, since both accelerator and brake pedals tend to be electronically con-
trolled, and hence they become increasingly tuneable. Also, more and more functions, such as those
below in 3.5.2, will have to be arbitrated with the pedals.
In modern passenger vehicles, Accelerator pedal is normally electronically controlled but the Brake
pedal is basically mechanical. In modern heavy commercial vehicles, both functions are electronically
controlled.
The functions in “3.4.3.2 Pedal Feel *” are normally not actively controlled, but in there are concept
studies with active pedals, where also the pedal feel can be actively changed to give feedback to driver.

219
 Longitudinal Dynamics

3.5.2.2 Cruise Control and Adaptive Cruise Control (CC, ACC) *

Function definition: Cruise Control, CC, controls the vehicle’s longitudinal speed. Driver can activate the func-
tion and decide desired speed.
Function definition: Adaptive Cruise Control, ACC, is an addition to CC. ACC controls the vehicle’s time
gap to a lead vehicle. Driver can activate the function and decide desired gap. When there is no lead vehicle, CC
controls the vehicle’s speed.
The purpose of CC is to keep the vehicle at a driver selected longitudinal speed, so that driver does not
need to adjust the accelerator pedal himself. The actuator used is the propulsion system. In heavy ve-
hicles, also the braking system (both retarders and service brakes) is used to maintain or regulate the
vehicle speed.
ACC is an addition to CC. The purpose of ACC is to keep a safe distance to the lead vehicle (vehicle
ahead of subject vehicle). ACC uses also friction brake system as actuator, but normally limited to a de-
celeration of . .4 /𝑠 2.
The safe distance which ACC aims for, is often expressed as a time gap, driver adjustable in the range
. . 𝑠. A model behind this is that the time gap is the driver reaction time and the subject vehicle can
decelerate as much as the object vehicle. More advanced models can allow smaller time gap in certain
situations. This is desired because it reduces the risk that other vehicles cut in between subject and
leading object vehicle. Such models can consider, e.g., acceleration of the object vehicle, pedal opera-
tion of subject vehicle, road gradient, road curvature, road friction and deceleration capabilities of sub-
ject and object vehicles.
Pedals and CC/ACC has to be arbitrated. Typically, accelerator pedal wins temporarily if pressed to
higher request than CC/ACC, while any use of brake pedal turns off (wins permanently over) the
CC/ACC .
CC is normally only working down to 0.. 40 𝑘 /ℎ. ACC can have same limitation, but with automatic
transmission, good forward-looking environment sensors, brake actuators and speed sensing, ACC can
be allowed down to stand-still.
3.5.2.2.1 Predictive CC
Contribution from Toheed Ghandriz

A development of CC has varying set speed trajectory ahead. The trajectory is typically optimized for a
predicted operation, about a minute ahead. The optimization typically considers road topography. In-
termittent propulsion and traffic are other phenomena that can be considered in optimization. Such
products are on the market, e.g. Volvo iSee and Scania Active Prediction for heavy vehicles.
§ Concepts of Predictive Optimization for Predictive CC
A predictive CC works in principle as figure below. It is an automated longitudinal driving function. It
selects an optimal trajectory for a certain horizon ahead, e.g. 1 km ahead and minimizing transport
cost including energy cost. The figure assumes it defines candidates to select by trying different (𝑠),
but there are also many other ways to define candidates, such as different driver model parameters.
The figure indicates an optimization by trial-and-error, but more sophisticated methods can be used,
such as Dynamic Programming, Quadratic Programming, Linear Programming, MPC, etc. Other quanti-
ties, such as gear shifting, including Neutral gear selection, 𝑒 𝑟(𝑠) and state of charge 𝑆𝑜 (𝑠) if the
vehicle is a hybrid vehicle, can be treated in primarily two different ways. They can be part of candi-
date definition, which increases the search space for the optimization. Alternatively, they can be calcu-
lated (e.g. to give momentaneous optimum) in the model used for optimization, which makes them
available as additional optimum trajectories beside 𝑂𝑝 (𝑠).
Macro-level information about traffic density and traffic mean velocity or the route ahead can be con-
sidered in optimization. But taking micro-level traffic information, such as nearest vehicle ahead, is
more challenging, especially predicting the velocity of the object vehicles.

220
 Longitudinal Dynamics

RouteManager (executes seldom, e.g. once per 𝑠, and predicts over a horizon, e.g. 𝑘 ) Vehicle
Model (simulated to calculate 𝑜𝑠𝑡) Vehicle & Actuators Model
𝑠
Environment “Ideal following
Model (map with model” Actuators
route and/or
environment sensors) 𝑛 𝑠

𝑛 𝑠 𝑜𝑠𝑡 ( 𝑛 𝑜𝑡ℎ𝑒𝑟 𝑡𝑟 𝑗𝑒 𝑡𝑜𝑟 𝑒𝑠)

Optimizer (tries different (𝑠) and selects the one which gives minimum 𝑜𝑠𝑡)
Sampler
𝑂𝑝 𝑠 (and/or other trajectories)
𝑠 𝑂𝑝 𝑠
𝑠
e.g.
(Human) Driver VMC, part1
Actuators
𝐴
Environ 𝑒 Driver
For instance: Arbitrator VMC,
-ment Interpretator
part2
𝐷 𝑣
𝑡𝐷 𝑣
𝑒 𝑓𝑢𝑛 (VMC=VehicleMotionControl)
𝐷 𝑣

Figure 3-34: § Environment, Driver and Vehicle, where the vehicle has a Predictive CC. See also
Figure 3-42.
§ Optimization by Varying Acceleration Trajectory
An example of how the cost (as sum of energy cost and fuel cost) can be reduced is given in the 3 fol-
lowing figures.

Figure 3-35: § Simulation with full model, including a driver model and actuator delays.

221
 Longitudinal Dynamics

Figure 3-36: § Simulation with the model used for optimization with varying 𝑛 (𝑠). The
acceleration trajectory from previous figure used, discretized every 25 m.

Figure 3-37: § Simulation with the model used for optimization with varying 𝑛 (𝑠). The
acceleration trajectory from optimization used, discretized every 25 m. Cost reductions of ≈40%
and ≈10%, respectively, is found with the parameters used in this example. The candidates
are marked to envision that a next step optimization could be done with a model including
actuator dynamics.
The optimizations are made with Matlab function fmincon, with (𝑠) (discretized to parameters
[ (𝑠 ) ⋯ (𝑠𝑁 )]) as parameters to vary. The cost function included a simulation of the system, includ-
ing calculation of cost and constraints. Note that the simulations in the cost functions can include “in-
verse dynamics” if the candidate prescribe the whole trajectory of a state, which can enable more sta-
ble derivative approximations.

222
 Longitudinal Dynamics

Note also that the above figure with optimisation result is not using the full model, i.e. delays in VMC2
and actuators are not considered. So, the end result in cost reduction will not be seen until running it
in a more accurate vehicle model or a real vehicle.
§ Optimization by Varying Force Request Trajectory
If we instead use the (𝑠) as theway to define the candidate, we can still pick an initial guess from
same simulation as before. But the

Figure 3-38: § Simulation with the model used for optimization with varying 𝑛 (𝑠),
discretized every 25 m.

3.5.2.3 Anti-Lock Braking System, ABS *

Function definition: Anti-lock Braking System, ABS, prohibits driver to lock the wheels while braking.
The wheel brake torques requested are limited by ABS in a way that each individual wheel’s longitudinal slip
stays above a certain (negative) value. An extended definition of ABS also includes vehicle deceleration requested
by other functions than pedal braking, such as AEB.
The purpose of ABS is to avoid losing vehicle brake force due to that the tyre force curve drops at high
slips AND to leave some friction for steering and cornering, see 2.2.1.6 and 2.2.4.7.5. ABS is a wheel slip
closed loop control, active when driver brakes via brake pedal. It keeps the slip above a certain value,
typically -15..-20 %. ABS uses the friction brakes as actuator.
Each wheel is controlled individually, but all wheels’ speed sensors contribute to calculation of vehicle
longitudinal speed, for calculation of actual slip. In the ABS function, it may be included how slip are
distributed between the wheels, such as normally the front axle is controlled to a slip closer to locking
than the rear axle. Also, a sub-function called “select-low” which makes the wheel closest to locking
decide the pressure also for the other wheel at the same axle. Select-low is typically used at rear axles.

223
 Longitudinal Dynamics

Figure 3-39: ABS control. Principle and control sequence, from Ref (Gillespie, 1992)

3.5.2.4 Electronic Brake Distribution, EBD *

Function definition: Electronic Brake Distribution prohibits driver to over-brake the rear axle while brak-
ing. An extended definition of EBD also includes vehicle deceleration requested by other functions than pedal
braking, such as AEB. EBD only uses friction brake as actuator.
With a fix proportioning between front and rear axle braking, there is a risk to over-brake rear axle
when friction is very high, since rear axle is unloaded so much then. Before electronic control was
available, it was solved by hydraulic valves, which limited the brake pressure to rear axle when pedal
force became too high. In today’s cars, where electronic brake control is present thanks to legislation
of ABS, the software base function EBD fulfils this need. EBD is primarily a feedforward control of
and 𝐴 𝑙 . Good estimates of mass, CoG location in x and z direction would be useful, but
these are seldom available. EBD can also have feedback of longitudinal slip difference between axles.
In heavy vehicles, pneumatic valves are used that limits the brake pressure in relation to the rear axle
load (deflection of mechanical spring suspension or air pressure in air suspension).
"DataLog_Braking.xls","ABS straight"
0.05

-0.05

-0.1
sxfl/[1]
-0.15 sxfr/[1]
sxrl/[1]
-0.2 sxrr/[1]

-0.25

1 1.5 2 2.5 3 3.5 4 4.5 5

"DataLog_Braking.xls","ABS straight"
140

120

100

80

60
pfl/[bar]
40 pfr/[bar]
prl/[bar]
20 prr/[bar]
vx/[km/h]
0
1 1.5 2 2.5 3 3.5 4 4.5
time [s] 5
Figure 3-40: ABS control, Data log from passenger car test.

224
 Longitudinal Dynamics

There are other side functions enabled by having ABS on-board. Such are “select low”, which means
that the brake pressure to both wheels on an axel is limited by the one with lowest pressure allowed
from ABS. So, if one wheel comes into ABS control, the other gets the same pressure. This is most rele-
vant on rear axle (to reduce risk of losing side grip) but one tries to eliminate the need of it totally, be-
cause it reduces the brake efficiency when braking in curve or on different road friction left/right.
It is often difficult to define strict border between functions that is a part of ABS and which is part of
EBD, which is why sometimes one say ABS/EBD as a combined function.

3.5.2.5 Traction Control, TC *

Function definition: Traction Control prohibits driver to spin the driven axle(s) in positive direction while ac-
celerating. An extended definition of TC also includes vehicle acceleration requested by other functions than pedal
braking, such as CC. TC uses both friction brakes and propulsion system as actuators.
The purpose of Traction Control is to maximise traction AND to leave some friction for lateral forces
for steering and cornering. Traction control is similar to ABS, but for keeping slip below a certain
value, typically +(15..20)%.
Traction control can use different ways to control slip, using different actuators. One way is to reduce
engine torque, which reduces slip on both wheels on an axle if driven via differential. Another way is to
apply friction brakes, which can be done on each wheel individually. Vehicles with propulsion on sev-
eral axles can also redistribute propulsion from one axle to other axles, when the first tends to slip. Ve-
hicles with transversal differential clutch or differential lock can redistribute between left and right
wheel on one axle.

3.5.2.6 Engine Drag Torque Control, EDC *

Function definition: Engine Drag Torque Control prohibits over-braking of the driven axle(s) while en-
gine-braking. EDC uses both friction brakes and propulsion system as actuators.
The purpose of Engine Drag Torque Control is as the purpose of ABS, but the targeted driving situation
is when engine braking at low road friction, when engine drag torque otherwise can force the wheels
to slip too much negative. Similarly to ABS, it keeps the slip above a certain negative value. However, it
does it by increasing the engine torque from negative (drag torque) to zero (or above zero for a short
period of time).

3.5.2.7 Automatic Emergency Brake, AEB *

Function definition: Automatic Emergency Brake decelerates vehicle without driver having to use brake
pedal when probability for forward collision is predicted as high.
The purpose of AEB is to eliminate or mitigate collisions where subject vehicle collides with a lead ve-
hicle. AEB uses friction brake system as actuator, up to full brake which would be typically 10 m/s2.
An AEB system is often limited by not trigger too early, because driver would be disturbed, or it could
actually cause accidents. Therefore, in many situations, AEB will rather mitigate than avoid collisions.
Conceptually, an AEB algorithm can be assumed to know physical quantities as marked in Figure 3-41.
The quantity time-to-collision, TTC, can then be defined as 𝑇𝑇 𝑥 ⁄( ), which means the
time within a collision will appear if no velocities changes. Subscript 𝑠 and 𝑜 means subject and object
vehicle, respectively.
Also, one can define enhanced time-to-collision measure, eTTC, which considers the present accelera-
tions of the lead vehicle. TTC shall not be mixed up with “time gap” (TG), which is the time when sub-
ject vehicle will reach the present position of the lead vehicle, 𝑇 𝑥 ⁄ .
AEB function shall, continuously, decide whether or not to trigger AEB braking. AEB shall intervene by
braking when driver can be assumed to collide without intervention. If no other information, this can
be predicted as when driver can NOT avoid by normal driving. Avoidance manoeuvres that have to be
considered are (normal) deceleration and (normal) lateral avoidance to the left and to the right. What
to assume as normal driving is a question of tuning; here the following limits are used | | < 𝑛
𝑒. . 4 ⁄𝑠 2 and | | < 𝑛 𝑒. . 6 ⁄𝑠 2 . The concept of a physical model-based algorithm is this:

225
 Longitudinal Dynamics

• Normal deceleration ( 𝑛 4 ⁄𝑠 2 ) leads to collision if:

2
min(𝑥 (𝑡)) < 0 ⇒ min(𝑥 + ∙𝑡 ( ∙𝑡+ 𝑛 ∙ 𝑡 ⁄ )) < 0 ⇒
> >
⇒ (𝑥 + ∙𝑡 ( ∙𝑡+ 𝑛 ∙ 𝑡 2 ⁄ ))| 𝑣
= 𝑜
𝑣 <0 ⇒
𝑣 𝑣𝑜 𝒗 𝒙 𝒗𝒐𝒙
⇒ 𝑥 ∙( )2 < 0 ⇒ 𝑜
< 2∙( ⇒ 𝑻𝑻𝑪 <
2∙( ) 𝑣 𝑣𝑜 ) 𝟖
• Normal avoidance to the left ( 𝑛 6 ⁄𝑠 2 ) leads to collision if:
𝑤 2 𝑤
(𝑦 𝑙 (𝑡) + )| < 0 ⇒ (𝑦 𝑙 𝑛 ∙ + 2 )| 𝑜
<0 ⇒
2 𝑜=
2 =
𝑣 𝑣𝑜

𝑜 𝑜 +𝑤/2 .6+ .8/2

⇒ <√ ∙ {𝑒. . } √ ≈ 0.4 𝑠 ⇒ 𝑻𝑻𝑪 < 𝟎. 𝟒
𝑣 𝑣𝑜 𝑦 3
• Normal avoidance to the right ( 𝑛 6 ⁄𝑠 2 ) leads to collision if:
𝑜 𝑜𝑟 𝑤 ⁄2
⋯ ⇒ 𝑣 𝑣𝑜
<√ ∙ ⋯ ⇒ 𝑻𝑻𝑪 < 𝟎. 𝟒
𝑦

• Assuming that the AEB intervention decelerates the vehicle with 𝐴𝐸 8 ⁄𝑠 2 , a

forward collision can be avoided if AEB intervenes AND if:
𝒗 𝒗 𝑣 𝑣𝑜 𝒗 𝒗
⋯ ⇒ 𝑣 𝑜𝑣 > 𝟐∙( 𝒙𝒂 𝒐𝒙 ) 6
⇒ 𝑻𝑻𝑪 > 𝒙𝟏𝟔 𝒐𝒙
𝑜 𝒙𝑨𝑬𝑩

Figure 3-41 shows a diagram where different condition areas are marked. The sectioned area shows
where AEB will be triggered, using above rules. The smaller of the sectioned areas shows where it also
will be possible to trigger AEB so timely that a collision is actually avoided; with the assumed num-
bers, this is for speeds up to 6.4 m/s≈23 km/h.

𝑥
(or lead vehicle)

𝑇𝑇
object vehicle

𝑥 0.4 𝑠
Collision if normal
lateral avoidance
𝑦 𝑚
( 6 2)
𝑦 𝑙

. /𝑠 6.4 /𝑠
AEB triggered AND avoids collision
(or own vehicle)

⁄ ⁄
subject vehicle

AEB triggered AND mitigates (not avoids) collision

“Mild AEB” or FCW might be triggered, since too late
for normal lateral avoidance
“Mild Automatic Lane Change” or FCW might be triggered,
since too late for normal longitudinal deceleration
Figure 3-41: Left: Quantities known for an AEB algorithm in the subject vehicle, assuming
“symmetrically behind” (𝑦 𝑙 𝑦 ). Right: Model based decision of triggering AEB and
effectiveness of AEB if triggered.
The reasoning above is very simplified. Additional information can improve effectiveness of AEB, such
as knowing if a lateral avoidance on one side of object vehicle is blocked, the acceleration of the object
vehicle, pedal operation of subject vehicle, road gradient, road curvature and road friction is. The vehi-
cle dynamics model used is simply a point mass with predicted constant velocity and certain assumed
acceleration capability, which of course can be extended a lot; both with taking actual accelerations
into account and more advanced vehicle dynamics models. The decision to intervene is dependent of

226
 Longitudinal Dynamics

many pieces of information and simple models; vehicle dynamics and driver behaviour (in both sub-
ject and object vehicles) as well as road characteristics. AEB function has to be designed together with
other similar functions, such as ACC and Forward Collision Warning (FCW).
AEB is legal requirement for both passenger vehicles and heavy vehicles (ISO19377, 2017).
Related functions are, e.g., an automatic extra force assistance in brake pedal when driver steps
quickly onto brake. Another related function is automatic braking triggered by a first impact and in-
tended to mitigate or avoid secondary accident events, starts to appear at market, see Reference (Yang,
2013). In semantic meaning, this could be seen as AEB, but they are normally not referred to as AEB;
AEB normally refers to functions that use environment sensors (forward directed radar, camera, etc.).
When designing and evaluating AEB, it is important to also know about the function Forward Collision
Warning, FCW. FCW is a function that warns the driver via visual and/or audio signals when a forward
collision is predicted. FCW is typically triggered earlier than AEB.
AEB as described above could be called “Rear-end AEB”. Another similar functionality, not included in
today’s AEB, could be called “Intersection AEB” and include braking for intersecting traffic, see
(Sander, 2018).

3.5.3 Longitudinal Motion Function Architecture

All control functions controls have to cooperate, and they have to be transferable between platforms
and vehicle variants. It is very complex to take all functions into consideration, but with a scope lim-
ited to the longitudinal Motion functionality
Figure 3-42 can be drawn as a solution within the reference architecture.
By using a reference architecture, it can be illustrated that Adaptive Cruise control and cruise control
can be seen as part of Traffic Situation Layer (ACC=CC if no vehicle ahead). The Traffic Situation Layer
has the purpose and scope to understand the ego vehicle’s surrounding traffic by looking at e.g. For-
ward Sensors. The forward-looking sensor is in this case part of Vehicle Environment sensors.
Vehicle Motion and Coordination Layer would include the arbitration of Driver’s Acceleration and
Brake pedal input and Traffic functionality, see Figure 3-42. In addition, Vehicle Motion and Coordina-
tion Layer would perform the powertrain coordination and brake distribution. The coordinated re-
quests are then sent to Motion Support Device Layer.
The Human Machine Interface would include the services available for the driver to activate or re-
quest, E.g. ACC activation to Traffic Situation or Deceleration by pressing the brake pedal.

227
 Longitudinal Dynamics

Route Managment (or Layer) Long horizon planning SpeedTrajecotry

functions, e.g. Predictive CC
Position
Vehicle Environment Interface

Maps
… Traffic Situation Layer Forrward Direction

Human Machine Interface

Dist. Ctrl (ACC) ACCbtn
FwdSens Dist
AEB Speed Ctrl (CC) Dist btn

Arb(Min)

Arb(sum) APedInterp APed

Vehicle Motion and Coordination Layer
Speed Arb (prio BPed, BPedInterp
else Max) BPed
Energy Managament
Powertrain Coord ABS &
TC ESC RSC
EBD
Arb(min)
Arb
Arb(min)
Coord
Coord
Arb (switching) Arb (switching)
Legend
Motion Support Device Layer
Border of Vehicle
Vehicle EngAct Level Functionality
GboxAct FrntBrkAct RearBrkAct
Motion
Sens SW connection (signals,
here 2 scalar signals)
HW connection
Sensors & Actuators
End Customer Function
Arbitration
Coordination

Figure 3-42: Functional architecture for conventional front axle driven passenger car. Mainly
longitudinal functions (plus ESC, RSC) are shown, e.g. no steering. Cf. Figure 1-56.
If a reference architecture is used, it can assist function developers from OEM’s Electrical, Powertrain,
and Chassis departments and suppliers to have a common view of how vehicle’s embedded motion
functionality is intended to be partitioned and to understand how different functions relate and inter-
act with each other and what responsibilities they have.

228
 Lateral Dynamics

4 LATERAL DYNAMICS
4.1 Introduction
The lateral motion of a vehicle is needed to follow the road curves, select route in intersections and
laterally avoid obstacles, which all involve steering. Vehicle steering is studied mainly through the ve-
hicle degrees of freedom: yaw rotation and lateral translation .
A vehicle can be steered in different ways:
• Applying steer angles on road wheels. Normally both of front wheels are steered with approxi-
mately same angle. Steering system described in 0.
• Applying longitudinal forces on road wheels; directly by unsymmetrical between left and right
side of vehicle, e.g. one-sided braking, or indirectly by deliberately use up much friction longi-
tudinally on one axle in a curve, so that that axle loses lateral force.
• Articulated steering, where the axles are fixed mounted on the vehicle body, but the vehicle
itself can bend.
The turning manoeuvres of vehicles encompass two sub-attributes. Handling is the driver’s perception
of the vehicle’s response to the steering input. Cornering is the physical response (open loop) of the
vehicle independent of how it influences the driver.
The lateral dynamics of vehicles is often experienced as the most challenging for the new automotive
engineer. Longitudinal dynamics is essentially motion in one plane and rectilinear. Vertical dynamics
may be 3 dimensional, but normally the displacements are small and in this compendium the vertical
dynamics is mainly studied in one plane as rectilinear. However, lateral dynamics involves motion in
the vehicle coordinate system which introduces curvilinear motion since the coordinate system is ro-
tating as the vehicle yaws.
The chapter introduces the models, more or less, in order of increasing number of states. For some
readers, it might be comprehensive to start with a look-ahead on the “linear one-track model” in 4.4.2.
This is the simplest model which yet captures the essential lateral motion quantities and as
states. The models earlier in the chapter can then be seen as simplifications.

4.1.1 Lateral Model Categorization

The chapter is organised as shown in Table 4.1. The table also shows a high-level categorization of
functions/models in the sections. Generally speaking, the input is steer angle and the output is yaw
rate and lateral velocity . The longitudinal speed is often treated as a parameter ( 0).
Table 4.1: Approximate high-level categorization of lateral functions/models
Section Longitudinal Typical input variables Typical output variables
velocity 𝒗𝒙 or parameters
Parameter, Path limitations, like Path 𝑥 (𝑠) 𝑦(𝑠) (𝑠), i.e.
4.2 Low Speed
0+ 𝑜𝑟 0 [ ⁄𝑠] road width and radius integrated over the manoeuvre
4.3 Steady State Cor- Parameter, Steady states (𝑅𝑝 )
and Path radius 𝑅𝑝
nering at High Speed 0. . 0 [ ⁄𝑠] ( ) ( ) and ( )
4.4 Stationary Oscil- Parameter, , Steering amplitude , Stationary oscillating responses
̂
lating Steering 0. . 0 [ ⁄𝑠] Frequency 𝑓 ̂ ( ̂ 𝑓) and ̂ ( )
Initial , Steering (𝑡),
Variable, ≠ 0 <Numerous and various>
4.5 Transient Driving Wheel torques (𝑻(𝑡))
Parameter, Pure and simplified handling manoeuvres, such as 4.5.4
0. . 0 [ ⁄ 𝑠 ]
There are other categorizations in 1.6.2, such as irp/oorp and one-track/two-track. Categorization of
tyre models is found in 2.2.6.

229
 Lateral Dynamics

4.1.2 References for this Chapter

• 0 and “Chapter 25 Steering System” in Ref (Ploechl, 2013).
• “Chapter 27 Basics of Longitudinal and Lateral Vehicle Dynamics” in Ref (Ploechl, 2013).
• “Chapter 8: Electronic Stability Control” in Ref (Rajamani, 2012)

4.2 Low Speed Manoeuvres

This section is about operating vehicles in low speeds, including stand-still and reverse. Specific for
low speed models is that inertia effects can be neglected, i.e. one neglect ∙ , ∙ and in
equilibrium. The function measures often requires that one find the paths [𝑥(𝑠) 𝑦 𝑠 of the vehicles
( )]
body edges or wheels. The paths can be obtained from CoG path with orientation, see 1.6.1.6.
Single unit vehicle are treated in 4.2.1..4.2.4, while 4.2.5 treats articulated vehicles.

4.2.1 Low Speed Model, Ackermann, without

Forces
Low speed manoeuvres are characterised by that the inertial forces are neglected, i.e. ∙ 0. If the
geometry is according to Ackermann and the forces on the vehicle are small, it is reasonable to assume
that the tyre forces also are small, and we can use ideally tracking tyre model for wheels or axles, see
2.2.6. We can see this as Eq [2.32] with infinite cornering stiffness: Eq [2.32] ⇒ 𝑠 →
𝑠 ⇒𝑠 0 ⇒ 𝑤 0 . The tyre force can be any (finite) value, determined by other part of the
system than the tyre. This tyre model can be seen as a compatibility relation (or “kinematic model”),
since it relates velocities to each other without involving any force. However, in this compendium we
still consider it as a constitutive model, keeping in mind that it is definitely invalid if > .
The effects of ideal tracking are that the intersection point of the wheels rotational axes coincides with
the instantaneous centre of vehicle rotation in ground plane. This directly relates steer angles and path
radius to each other. For the one-track model in Figure 1-53 this relation becomes:
𝐿
tan( 𝑓 ) 𝐿 𝐿 𝐿
𝑅 }⇒ 𝑓 arctan ∙ sign(𝑅) ≈ ∙ sign(𝑅) ≈
𝑅 [4.1]
𝑅2 𝑅 + 2
2
√ 2 2
√𝑅2 2
( 𝑅 )
ℎ𝑒𝑟𝑒 𝑅 > 0 𝑒 𝑛𝑠 𝑡ℎ 𝑡 𝑛𝑠𝑡 𝑛𝑡 𝑛𝑒𝑜𝑢𝑠 𝑒𝑛𝑡𝑟𝑒 𝑜𝑓 𝑟𝑜𝑡 𝑡 𝑜𝑛 𝑠 𝑒𝑓𝑡 𝑜𝑓 𝑡ℎ𝑒 𝑒ℎ 𝑒

4.2.2 Low Speed Functions

4.2.2.1 Turning Diameter *
Function definition: Turning diameter is the diameter of the smallest possible circular path obtained steady
state at low speed, measured to a certain point at the vehicle. The certain point can be either outer-most point on
wheel (Kerb Turning diameter) or outer-most point on body (Wall Turning diameter).
The end of the simulation in Figure 4-5 is made with constant steer angle. If we assume that it is the
maximum steer angle, the circle actually shows the turning circle for centre of gravity. If we add the
path for the outermost wheels, we get the kerb turning diameter, see Figure 4-1. If we add the path for
the outermost point at the vehicle body we get the wall turning diameter, also shown in Figure 4-1.
The outermost point at the vehicle body is normally the front outer corners of the vehicle body.

4.2.2.2 Swept Path Width, SPW *

Function definition: Swept path width is the distance between the outermost and innermost paths of wheels (or
body points). The paths are then from a certain turning or lane change manoeuvre at a certain speed.
For manoeuvrability, there is a function which is complementary to turning diameter. It is “Swept Path
Width” (SPW), see Figure 4-1. It can be defined for circle driving with maximum steering angle, as

230
 Lateral Dynamics

Turning diameter, but also for any other certain manoeuvre. It can be defined for kerb and wall; dis-
tance between wheels or body points. The SPW should be small for good manoeuvrability.

kerb turning diameter

wall turning diameter

y [m]

• CoG path diameter=8.54 m

• front right wheel path diameter=11.19 m
• real left wheel path diameter=6.29 m

x[m]
Figure 4-1: Paths for wheels and body points (added to result in Figure 4-5).

4.2.2.3 Low Speed Off-Tracking, LSOT *

Function definition: Off-tracking is the distance between the outermost and innermost paths of mid-points of
axles. The paths are then from a certain turning or lane change manoeuvre at a certain speed.
Another measure of manoeuvrability
is “Off-tracking”, see Figure 4-2. It is
like Swept Path Width, but for the L
mid points of the axles. It is also
used for higher speeds, and then the
df
rear axle often tracks on a larger ra-
dius than front axle. A variant of def-
inition of Off-Tracking uses (lateral)
midpoints of body and it can be
much larger than the first definition
for vehicles with long front or rear
overhangs.

Figure 4-2: Off-tracking (added to result in Figure 4-5).

4.2.2.4 Circle and Manoeuvre Measures

SPW and Off-tracking is most relevant for vehicles with several units, such as truck with trailer. They
can be defined for driving several rounds in a circle with constant steering angle (“Circle SPW/Off-
tracking”). It is a well-defined measure since it is a steady state with respect to articulation angle.
However, circle driving is seldom the most relevant manoeuvre. So, one often set requirements on
“Manoeuvre SPW/Off-tracking”, e.g. when driving from straight, via curve with certain outer radius,
to a new straight in a certain angle from the first straight, e.g. 12.5 m and 90 deg is common. A way to
predict SPW/Off-tracking for such a manoeuvre is to simulate with time integration, see 4.2.5.2. One
can integrate in travelled distance 𝑠 instead of time, but time is often easier and allows stand-still parts

231
 Lateral Dynamics

in the manoeuvre. The states in the simulation are the path coordinates with orientation (𝑥 𝑦 ) and
articulation angles ( 2 …).
The steady state is typically approached asymptotically, so the corresponding circle values requires
either long simulations or inserting state derivatives zero and algebraic solution. From geometry in
Figure 4-2 on can find an expression for (Circle) Off-tracking Δ:

(𝑅𝑓 𝐿) 𝑅𝑓 𝑅 𝑅𝑓 √𝑅𝑓2 𝐿2
[4.2]
( 𝑓 𝐿) 𝑅𝑓 𝑅 𝐿⁄sin( 𝑓 ) 𝐿⁄tan( 𝑓 )

4.2.2.5 Steering Effort at Low Speed *

Function definition: Steering effort at low speed is the steering wheel torque needed to turn the steering
wheel a certain angle at a certain angular speed at vehicle stand-still on high road friction.
At low or zero vehicle speed, it is often difficult to reach a low steering wheel torque, due to:
• Castor offset in Figure 2-95 gives the wheel a side slip when steering and hence a tyre lateral
force is developed. Tyre lateral forces times castor offset increases the steering wheel torque.
• Additionally, there is a spin moment in the contact patch, 𝑀𝑍𝑇 in Figure 2-6. It does not influ-
ence very much, except for at very low vehicle speed. Quantitative models for 𝑀𝑍𝑇 are not pre-
sented in this compendium.
A critical test for steering effort at low speed is to steer a parked vehicle with a certain high steering
wheel rotational speed, typically some hundred 𝑒 /𝑠. The steering wheel torque is then required to
stay under a certain design target value, normally a couple of Nm. The torque needed will be depend-
ent on lateral force, spin moment and steering geometry and dependent on the capability of the power
steering system (which is dependent on steering, due to delays in the steering assistance actuator). A
failure in this test is called “catch-up”, referring to that driver catches up with the power steering sys-
tem. It can be felt as a soft stop and measured as a step in steering wheel torque.

4.2.3 Low Speed Model, Ackermann, with Forces

4.2.3.1 Implications of Forces on Turning Circle
If we have a vehicle with Ackermann geometry, it is tempting to model without involving forces, using
Eq [4.1]. But forces can influence low speed paths. When introducing rolling resistance, the force equi-
librium is obtained by counter-directed tyre-longitudinal forces on the two axles. Due to the steer an-
gle, a lateral force on the front wheel is required, which gives a lateral tyre slip, 𝑓 ≠ 0, see igure 4- .
This changes the motion compared to Figure 1-53. Road grade resistance influences in same way and
superimposes on rolling resistance.
For heavy combination vehicles, these effects can be significant, adding also rolling resistance from
towed units and non-Ackermann effects of several non-steered axles on some units. Then, the function
to turn can be quantified by required road friction on steered and driven axle, as opposed to radius at
maximum steer angle.

Rear axle propulsion: Front axle propulsion:

𝑓 𝑤 𝑝 𝑝
𝑓 𝑓 𝑓
𝑓
𝑓
𝑓 𝑓

𝑓 𝑓

𝑝 𝑝 𝑓 𝑙𝑙 𝑓 𝑤 𝑓 𝑓
𝑓 𝑤 𝑓 𝑙𝑙

Figure 4-3: Smaller turning circle diameter for front axle propulsion, as compared to rear axle
propulsion due to rolling resistance on the un-driven axle.

232
 Lateral Dynamics

4.2.3.2 Model
The model in Eq [4.1] predicts a motion without involving forces, or actually assuming forces are zero.
To get a more complete model, where more variables can be extracted, we can set up the model in Fig-
ure 4-4.
Velocities: 𝑓 arctan 𝑠 𝑤
𝑓 arctan
𝑓 𝑣 Forces:
𝑓 𝑣 𝑓 𝑣
𝑓 𝑓

𝑓 𝑣
0

𝑓 𝑣
𝑟𝑦
𝑓 𝑓
𝑓 𝑣
𝑓 𝑓
𝑓 𝑓
𝐿 𝐿

Figure 4-4: One-track model with ideally tracking axles. Lower view of front wheel shows
conversion between wheel and vehicle coordinate systems.
The “physical model” in Figure 4-4 gives the following “mathematical model”:
Equilibrium (longitudinal, lateral and yaw around CoG):
0 𝑓 𝑣 +
0 𝑓 𝑣 +
0 𝑓 𝑣 ∙ 𝑓 ∙
Transformation between vehicle and wheel coordinate systems:
𝑓 𝑣 𝑓 𝑤 ∙ cos( 𝑓 ) 𝑓 𝑤 ∙ sin( 𝑓 )
𝑓 𝑣 𝑓 𝑤 ∙ sin( 𝑓 ) + 𝑓 𝑤 ∙ cos( 𝑓 )
𝑓 𝑣 𝑓 𝑤 ∙ cos( 𝑓 ) 𝑓 𝑤 ∙ sin( 𝑓 )
𝑓 𝑣 𝑓 𝑤 ∙ sin( 𝑓 ) + 𝑓 𝑤 ∙ cos( 𝑓 )
Compatibility between CoG and axles:
𝑓 𝑣 𝑛 𝑓 𝑣 + 𝑓∙
𝑛 ∙ [4.3]
Ideal tracking (Constitutive relation, but with infinite lateral slip stiffness):
𝑓 𝑤 0 𝑛 0
Path with orientation (compatibility), from Eq [1.5]:
𝑥 ∙ cos( ) ∙ sin( )
𝑦 ∙ cos ( ) + ∙ sin( )

Controls (driver or actuation):

( 5 ∙ 𝜋⁄ 80) ∙ sin(0.5 ∙ ∙ 𝜋 ∙ 𝑡) 𝑓 𝑡 < 4.5
𝑓 {
5 ∙ 𝜋/ 80 𝑒 𝑠𝑒
Constitution (Rear axle undriven, which gives drag from roll resistance):
00 sign( )
Note that, since we have Ackermann geometry, we can (mathematically) use the ideal tracking tyre
model ( 𝑤 0 ) even if forces are now introduced. For some engineering problems, it might be moti-
vated to use 𝑤 𝑠 𝑤 ⁄|𝑅 |≈ 𝑤 ⁄| 𝑤 | , but we leave this to 4.2.4.
The compatibility in Eq [4.3] neglects the influence of steering axis offsets at ground, see 0. The terms
neglected are of the type LateralOffset ∙ in the equation for 𝑓 𝑣 and LongitudinalOffset ∙ in the
equation for 𝑓 𝑣 . This is generally well motivated for normal road vehicles, except for very quick
steering when vehicle is close to stand-still.
Equation [4.3] is written in Modelica format in Equation [4.4]. Comments are marked with //. The sub-
scripts v and w refer to vehicle coordinate system and wheel coordinate system, respectively. The

233
 Lateral Dynamics

actual assumption about ideal tracking lies in that 𝑓 𝑤 0. Global coordinates from Figure
1-49 is also used. A driving resistance of 100 N is assumed on the rear axle ( 00 ) to exemplify
that forces do not need to be zero.
//Equilibrium:
0 = Ffxv + Frx;
0 = Ffyv + Fry;
0 = Ffyv*lf - Fry*lr;
//Ideal tracking (Constitutive relation, but without connection to forces):
vfyw = 0; vry = 0;
//Compatibility:
vfxv = vx; vfyv = vy + lf*wz;
vrx = vx; vry = vy - lr*wz;
//Transformation between vehicle and wheel coordinate systems:
Ffxv = Ffxw*cos(df) - Ffyw*sin(df);
Ffyv = Ffxw*sin(df) + Ffyw*cos(df); [4.4]
vfxv = vfxw*cos(df) - vfyw*sin(df);
vfyv = vfxw*sin(df) + vfyw*cos(df); //or atan(vfyv/abs(vfxv))=df+atan(sfy); sfy=0;
//Path with orientation:
der(x) = vx*cos(pz) - vy*sin(pz);
der(y) = vy*cos(pz) + vx*sin(pz);
der(pz) = wz;
// Prescription of actuation:
df = if time < 4.5 then (35*pi/180)*sin(0.5*2*pi*time) else 35*pi/180;
//Rear axle undriven, which gives drag from roll resistance:
Frx = -100;

4.2.3.3 Simulation
The longitudinal speed is a parameter, 0 𝑘 /ℎ. A simulation result from the model is shown in
Figure 4-5. It shows the assumed steer angle function of time, which is an input. It also shows the re-
sulting path, 𝑦(𝑥).
df [rad] y [m]

35 deg

x[m]
time [s]
Figure 4-5: Simulation results of one-track model with ideal tracking tyres.
The variables 𝑥 𝑦 𝑝 are the only state variables of this simulation. If not including the path
model (Eq [1.5]), the model would be only an algebraic system of equations. That system of equations
could be solved isolated for any value of steer angle without knowledge of time history.

4.2.4 Low Speed Model, Non-Ackermann

If Non-Ackermann geometry, we have to include the forces. Non-Ackermann geometry can be seen as
lateral wheel or tyre forces fight each other; either between left and right wheels or between axles.

234
 Lateral Dynamics

Examples are a two-axle vehicle which has parallel steering and truck with 3 axles, whereof the two
rear are non-steered, respectively.

4.2.4.1 Non-Ackermann between Axles

We will go through model changes needed for the latter example. In order to compare the models as
closely as possible, we simply split the rear axle into two rear axles, in the example in 4.2.1. The physi-
cal model becomes as in Figure 4-6. The measures appear in Figure 4-8, and you see that it is not a typ-
ical truck, but a very unconventional vehicle of passenger car size but with two rear axles.

velocities: vy vfxv forces: Ffxv

vr2y 𝑓
vr1y wz Fr2y Fr1y 𝑓
vfyv

Ffyv
vx df Fr2x Fr1x df
vr2x vr1x
𝑓
lr lf 𝑓
L
Δ𝐿𝑟 Δ𝐿𝑟
2 2 DeltaL

Figure 4-6: Non-Ackermann geometry, due to un-steered rear axles. Top: Rigid Truck with 3 axles,
whereof only the first is steered. Bottom: One-track model.
The changes we have to do in the model appear as underlined in Equation [4.5]. There has to be double
variables for , denoted 1 and 2 respectively. Also, we cannot (mathematically) use
𝑓 𝑤 0 anymore, but instead we introduce a lateral tyre force model, as described in 2.2.4.

//Equilibrium:
0 = Ffxv + Fr1x + Fr2x; //grade resistance could be added here, e.g. “+500”
0 = Ffyv + Fr1y + Fr2y;
0 = Ffyv*lf - Fr1y*(lr - DLr/2) - Fr2y*(lr + DLr/2);
//Constitutive relation (with slip, as opposed to Ideal tracking):
Ffyw = -Cf*sfy; sfy = vfyw/abs(vfxw);
Fr1y = -Cr1*sr1y; sr1y = vr1y/abs(vr1x);
Fr2y = -Cr2*sr2y; sr2y = vr2y/abs(vr2x);
//Compatibility:
vfxv = vx;
vfyv = vy + lf*wz;
vr1x = vx;
vr2x = vx;
vr1y = vy - (lr – DLr/2)*wz;
vr2y = vy - (lr + DLr/2)*wz; [4.5]
//Transformation between vehicle and wheel coordinate systems:
Ffxv = Ffxw*cos(df) - Ffyw*sin(df);
Ffyv = Ffxw*sin(df) + Ffyw*cos(df);
vfxv = vfxw*cos(df) - vfyw*sin(df);
vfyv = vfxw*sin(df) + vfyw*cos(df);
//Path with orientation:
der(x) = vx*cos(pz) - vy*sin(pz);
der(y) = vy*cos(pz) + vx*sin(pz);
der(pz) = wz;
// Prescription of steer angle:
df = if time < 4.5 then (35*pi/180)*sin(0.5*2*pi*time) else 35*pi/180;
//Rear axles undriven, which gives drag from rolling resistance:
Fr1x = -100/2*sign(vx);
Fr2x = -100/2*sign(vx);

235
 Lateral Dynamics

The new result is shown in Figure 4-7, which should be compared to Figure 4-5. The radius in the cir-
cle increases a little, which is intuitive, since the double rear axle makes turning less easy.
df [deg] y [m]
35 deg

vx=2.778m/s

vy=1.036m/s

wz=0.652rad/s

time [s] x[m]

Figure 4-7: Simulation results of one-track model. Non-Ackermann geometry due to two non-
steered rear axles. Tyres models are linear, as opposed to ideally tracking.
4.2.4.1.1 Equivalent Wheelbase
We can also draw the different locations of the in-
stantaneous centre for both cases. This is shown, in • instantaneous centre for
scale, in Figure 4-8. 3-axle vehicle
We could tune the steer angle required to reach ex- • instantaneous centre for
actly the same path radius as for the 2-axle reference 2-axle reference vehicle
vehicle. Then, we would have to steer a little more
than the 35 degrees used, and we could find a new
instantaneous centre, and we could identify a so-
called Equivalent wheelbase. This leads us to a defi-
nition: The equivalent wheelbase of a multi-axle ve-
𝐿𝑟

𝐿𝑟
2

hicle is the wheel base of a conventional two-axle ve-

hicle which would exhibit the same turning behav- df
iour as exhibited by the multi-axle vehicle, given
same steer angle and similar axle cornering stiff-
nesses. In 4.2.4 it is shown that ideally tracking tyre lr=1.5 lf=1.3
model is not enough for non-Ackermann geometry; Figure 4-8: Instantaneous centre with a 3-
at least a linear tyre model is needed with ≠ . axle vehicle, with corresponding 2-axle
vehicle as reference.
Figure 4-9: Two-track model with arbitrary angles , so that non-Ackermann steering can be
modelled.
Figure 4-10: Simulation result showing Low Speed Manoeuvrability.

236
 Lateral Dynamics

4.2.5 Articulated Vehicles

4.2.5.1 Ackermann Geometry for Articulated vehicles
For vehicle without articulation, there has to be one common instantaneous centre for all points on the
vehicle body. For articulated vehicles, each unit can have its own instantaneous centre. But these are
dependent of each other through the coupling points. So, adding units with one axle does not disqual-
ify the Ackermann property of the resulting combination vehicle. See “transient state” in Figure 4-11.

4.2.5.2 Transients in Articulation Angle

For articulated vehicles, also the low speed case has transients in the sense that the articulation angles
change transiently. The models are steady state with respect to velocities, but transient with respect to
articulation angle, see “transient state” in Figure 4-11. Consider the case of instantaneous step steer.
For a vehicle without articulation, a steady state is reached directly, since inertia is not considered. But
for an articulated vehicle it takes some time (or travelled distance, since it can be studied independent
of speed and time) before a steady state articulation angle is achieved, see “steady state” in Figure
4-11.
So, for articulated vehicles, a scalar requirement on turning radius is not so relevant as for two axle
vehicles. The functions “4.2.2.2 Swept Path Width, SPW *” and “4.2.2.3 Low Speed Off-Tracking, LSOT
*” are better, assuming one defines a certain road geometry, e.g. through outer radius and total angle
of turning.
Vehicle (non-Ackermann geometry)
Model as Ackermann
geometry by collapsing each
axle group as one axle AND
each axle as one-track
semi-trailer

Model in transient state Model in steady state

2: Instantaneous centre of (driving in circle)
rotation for dolly (1 axle,
2 coupling points)
Instantaneous centre of
1: Instantaneous centre of
rotation, common for all units
rotation for truck (2 axles,
(4 axles, 2 coupling points)
1 coupling point)

3: Instantaneous
centre of rotation for
semi-trailer (1 axle and
1 coupling points)

Figure 4-11: Instantaneous centre of rotation for truck with trailer.

A general model, covering arbitrarily many units and Non-Ackermann or forces, is presented in
4.5.2.2.2.2.

237
 Lateral Dynamics

4.2.5.3 § Model for Ackermann, without forces

Articulated vehicles with Ackermann and without forces moves as indicated in left lower part of Figure
4-11; they have one instantaneous centre of motion per unit and these centres are related through the
coupling points. Figure 4-11 is therefore very close to a simulation model, it is just to set up the geo-
metrical relations and they will be just many enough:
• If formulated as dynamic system with time as independent variable: input variables will be one
steering angles for each steered axle (group) and the longitudinal velocity of one unit, e.g. the
first towing unit. At least each articulation angle will be a state variable. If adding Eq [1.5], also
positions of any selected point and orientation of each unit can be solved for and then they also
becomes state variables.
• The model can be reformulated to a dynamic system with position (travelled distance) 𝑠 of a
point at the vehicle as independent variable instead of time, see Eq [1.6]. The time then disap-
pears. The input longitudinal velocity of first unit, (𝑡), is replace by the discrete variable
sign( (𝑡)).

4.2.6 Reversing
Low speed manoeuvring is often about both driving forward and reversing. The derived models work
formally also for < 0. Assume that, after some driving forward, is changed to a negative value.
Assume that steering is same for same position along the path, (𝑠). The vehicle will then reverse
in approximately same path as it first drove forward. If ideal tracking tyres, such as Eq [4.1], the paths
will be identical. If adding forces to the model, the reverse path can deviate from forward path. A small
such deviation can be seen in figure due to a large rolling resistance coefficient (𝑅𝑅 0. 0).
In reality, the reverse path for an articulated vehicle often differs more from the forward path. This is
mainly due to back-lash in couplings; even if only some centimetre backlash it can influence a lot. The
backlash can be modelled by letting coupling point position be dependent of sign(coupling force).

4.2.6.1 Stability when Reversing

Single unit vehicles do not become unstable in low speed, neither for forward nor reverse driving. But
articulated vehicles do, when reversing. To study this, we derive a linear model, by assuming small an-
gles. Note that the stability analysis for low speed differs from stability analysis at higher speeds, see
4.4.3.5, in that the low speed models do not include inertial terms 𝑠𝑠 𝑒 𝑒𝑟 𝑡 𝑜𝑛.

4.2.6.2 Linear Low Speed Model for Tractor with Semi-trailer

Figure 4-12 shows a physical model for tractor with trailer, assuming small angles , and .
2

𝑓
𝑓
𝐿 𝑓
𝐿

Figure 4-12: A low speed model of tractor with semi-trailer. Small angles 𝑓 are assumed.
The mathematical model for lateral position variables (𝑦 ) becomes:
𝑙
𝑦 𝐿 𝑓+ ∙ (from Eq [1.5], with small )
1
𝑣1 𝛿
𝐿1
(rotational velocity = tangential speed / radius to instantaneous centre)

238
 Lateral Dynamics

𝑣1 𝛿 𝑣2𝑦
𝐿1 𝐿2
(articulation angle is difference between units)
𝑙
2 +𝐿 (units have same velocity in coupling point, small , small 𝑓)
1

Eliminating 2 gives:
𝑦 0 0 𝑦 ⁄𝐿
[ ] [0 0 0 ] [ ]+ [ ⁄𝐿 ] 𝑓
⏟0 0 ⁄𝐿2 ⁄𝐿 ⁄(𝐿 𝐿2 )
𝑨
The eigenvalues to 𝑨 becomes 𝜆 2 ±0 𝜆3 ⁄𝐿2 . So, the system is unstable when < 0, be-
cause it makes e(𝜆3 ) > 0. It is probably intuitive for may readers that the vehicle is unstable, as the
semi-trailer pushed rearwards via a moment-free joint. However, the analysis was included in the
compendium to show how stability appears for low speed models, i.e. without inertial terms 𝑠𝑠
𝑒 𝑒𝑟 𝑡 𝑜𝑛. The model can be extended with other multiple-unit vehicles, multiple axles in axle
groups and driver models. The state variables will remain as 𝑦 and one articulation angle for
each coupling.

4.3 Steady State Cornering at High Speed

Steady state cornering refers to that all time derivatives of vehicle speeds ( ) are zero. The
physical understanding is then that the vehicle drives on a circle with constant yaw velocity, see Figure
4-13. Alternatively, this can be described as driving with constant tangential speed ( ), on a constant
path radius (𝑅) and with a constant side slip angle ( ).

4.3.1 Steady State Driving Manoeuvres

When testing steady state function, one usually runs on a so called “skid-pad” which appears on most
test tracks, see Figure 1-53. It is a flat circular surface with typically 100 m diameter and some concen-
tric circles marked. A general note is that tests in real vehicles are often needed to be performed in
simulation also, and normally earlier in the product development process.
Typical steady state tests are:
• Constant path radius. Driven for different longitudinal speeds.
• Constant longitudinal speed. Driven for different path radii.
• Constant steering wheel angle. Then increase accelerator pedal (or apply brake pedal) gently.
(If too quick, the test would fall under transient handling instead.)

Steady state cornering can be defined by either 3

or 2 quantities, depending on assumptions:
• For a general vehicle it is fully defined only by
all 3 quantities, e.g. or 𝑅𝑝 .
• For a certain vehicle it is often considered as
fully defined by 2 inputs from driver:
accelerator pedal and steering wheel. (Then,
one have neglected other inputs from driver,
such as brake pedal and parking brake. One
have also neglected other possible automatic
actuation, such as all wheel drive distribution
and ESC-braking one wheel.)
Figure 4-13: Steady state cornering. ( will be negative for larger , i.e. vehicle will point
inwards.)
Relevant standards for these test manoeuvres are: References (ISO 4138) and (ISO 14792).

239
 Lateral Dynamics

Vehicle Handling
Dynamics track
Area

Skid pad

High speed
track
Figure 4-14: An example of test track and some parts with special relevance to Vehicle Dynamics.
The example is Hällered Proving Ground, Volvo Car Corporation.

=Vehicle Dynamics Area

Figure 4-15: An example of test track. The example is AstaZero (Active Safety Test Arena), owned
by Research Institute of Sweden (RISE) and Chalmers University of Technology.

240
 Lateral Dynamics

High
speed
circle
Skid pad

Low
friction
Vehicle Dynamics Area
strips
Hill
strips

Figure 4-16: An example of test track. The example is CASTER’s (virtual) test track. Used for
CASTER’s driving simulator at Chalmers University of Technology.

4.3.2 Steady State One-Track Model

In steady state we have neither inertial effect from changing the total vehicle speed ( √ 2+ 2
𝑜𝑛𝑠𝑡 𝑛𝑡) nor from changing the yaw velocity ( 𝑜𝑛𝑠𝑡 𝑛𝑡). However, the inertial “centrifugal” ef-
fect of the vehicle must be modelled. The related acceleration is the centripetal acceleration, 𝑅𝑝 ∙
2 2
/𝑅𝑝 ∙ .
A vehicle model for this is sketched in Figure 4-17. The model is a development of the model for low-
speed in Figure 4-4 and Equation [4.4], with the following changes:
• Longitudinal and lateral accelerations are changed from zero to components of centripetal ac-
celeration, ac, as follows (see Figure 4-17):
o sin( ) sin( )
o + cos( ) + cos( ) +
• The constitutive relations for the axles are changed from ideal tracking to a (linear) relation
between lateral force and lateral slip:
o 𝑓 𝑤 𝑓 ∙ 𝑠𝑓 ; where 𝑠𝑓 𝑓 𝑤 ⁄|𝑅𝑤 𝑓 | ≈ 𝑓 𝑤 ⁄| 𝑓 𝑤 |
o ∙ 𝑠 ; where 𝑠 ⁄ | 𝑅𝑤 | ≈ ⁄| |;
The constitutive relations above capture the slip characteristics for the tyres, see 2.2.4, but they can
also capture steering system compliance, side force steering, and roll steering, see 4.3.7.3.
Note that 𝑓 𝑤 and in Figure 4-17 are drawn as positive for the sign convention of (ISO 8855). It
can help understanding to note that these velocities get negative values if vehicle steady state turns to
the left, as indicated by the drawn steering angle 𝑓 .

241
 Lateral Dynamics

accelerations: velocities: forces:

vehicle
𝑓 𝑓
sin( ) path

𝑓 𝑣

𝑓 𝑣
+ cos( ) centre
𝑓
vrx 𝑓 𝑣 𝑓 𝑣
𝐿
𝑓 𝑓 𝑓
cos
sin
b
Figure 4-17: One-track model for Steady State Cornering. Dashed forces are “fictive forces”.
The model in Figure 4-17 has the mathematical form in Eq [4.6] (in Modelica format). Longitudinal
speed is assumed to be positive. The subscripts v and w refer to vehicle coordinate system and
wheel coordinate system, respectively. A driving resistance of 100 N is assumed on the rear axle
(Frx=100;). The longitudinal speed is a parameter, 00 𝑘 /ℎ.
//Equilibrium:
m*ax = Ffxv + Frx; //Air and grade resistance neglected
m*ay = Ffyv + Fry; Jz*0 = Ffyv*lf - Fry*lr; // der(wz)=0
-ax = wz*vy; +ay = wz*vx;
//Constitutive relation, i.e. Lateral tyre force model:
Ffyw = -Cf*sfy; sfy = vfyw/vfxw;
Fry = -Cr*sry; sry = vry/vrx;
//Compatibility:
vfxv = vx; vfyv = vy + lf*wz;
vrx = vx; vry = vy - lr*wz;
//Transformation between vehicle and wheel coordinate systems:
Ffxv = Ffxw*cos(df) - Ffyw*sin(df); [4.6]
Ffyv = Ffxw*sin(df) + Ffyw*cos(df);
vfxv = vfxw*cos(df) - vfyw*sin(df);
vfyv = vfxw*sin(df) + vfyw*cos(df);
//Path with orientation (from Eq [1.5]):
der(x) = vx*cos(pz) - vy*sin(pz);
der(y) = vy*cos(pz) + vx*sin(pz);
der(pz) = wz;
// Prescription of steer angle:
df = if time < 2.5 then (5*pi/180)*sin(0.5*2*pi*time) else 5*pi/180;
// Rear axle undriven, which gives drag from roll resistance:
Frx = -100;
A simulation result from the model is shown in Figure 4-18. Note that steering start to the left, but ve-
hicle path starts bending to the right. This comes from that it is a steady state model but used in a tran-
sient manoeuvre. The steady state cornering condition is found directly and turning left has the steady
state directed outwards, to the right, due to centrufugal force.

242
 Lateral Dynamics

df [rad] y [m]
5 deg 2 +
2
=
8. 7 /𝑠

Ffyw
Ffxw

Frx-100 Nm

vx = 100 km/h=27.78 m/s

wz = 0.3995 rad/s
vy = -5.761 m/s
x[m]
time [s]
Figure 4-18: Simulation results of steady state one-track model. The vehicle sketched in the path
plot is not in scale, but correctly oriented.
Now, the validity of a model always has to be questioned. There are many modelling assumptions
which could be checked, but in the following we only check the assumption instead of the
more correct + , which we will learn in “4.4.2 One-Track Model”. Comparison of the
terms gives | |
𝑚
≈ | | 𝑚 ≈ 0 ⁄𝑠 2 , so | | is large and this jeopardizes the validity.
Large | | happens during 0 < 𝑡 <≈ 𝑠, so the model is not very valid there. But, at 𝑡 >≈ 𝑠, the
model is valid, at least in this aspect, since there | | ≈ 0 ≪ | |. So, the model is not so valid dur-
ing the initial sinusoidal steering. This shows that a steady state models should not be trusted outside
steady state conditions.
Eq [4.6] is a complete model suitable for simulation, but it does not facilitate understanding very well.
We will reformulate it assuming small 𝑓 (i.e. cos( 𝑓 ) , sin( 𝑓 ) 0, and 𝑓2 0). Eliminate slip, all
forces that are not wheel longitudinal, and all velocities that are not CoG velocities:
( +( + 𝑓 ) 𝑓)
𝑓 𝑤 ( +( + 𝑓 ) 𝑓) + 𝑓 ( + 𝑓 ) 𝑓 + ( +( + 𝑓 ) 𝑓)
( +( + 𝑓 ) 𝑓)
𝑣𝑦 𝑙𝑟 𝜔𝑧
𝑓 ( 𝑓 +( + 𝑓 )) + 𝑓 𝑤 𝑓 ( +( + 𝑓 ) 𝑓) [4.7]
𝑣

𝑓 ( 𝑓 +( + 𝑓 )) 𝑓 𝑓 𝑤 𝑓 𝑓

( +( + 𝑓 ) 𝑓)

Eq [4.7] is a complete model, which we can see as a dynamic system without state variables.
• Actuation: Steering and wheel torque on each axle: 𝑓 𝑓 𝑤 .
• Motion quantities:
For propulsion on both axles, it can be a reasonable case that we know 𝑓 𝑓 𝑤 and want to calcu-
late . Eq [4.8] is a rearrangement of Eq [4.7] for this purpose. It can be used for rear axle

243
 Lateral Dynamics

drive using 𝑓 𝑤 0 and calculating how large needs to be. Front axle drive requires more rear-
rangements.
𝑓 𝐿+ 𝐿 𝑓 𝑤
2 2 𝑓
𝑓 𝐿 + ( 𝑓 𝑓)
( 2 ) + ( 2)
𝑓 𝐿 𝑓 𝐿 𝑓 𝑓 𝑤
𝐿2 + ( 2 𝑓 [4.8]
𝑓 𝑓 𝑓)

𝑓 ∙( + 𝑓 ∙ )∙ 𝑓
∙ ∙ 𝑓 𝑤
(( + 𝑓 ∙ )∙ 𝑓 + )

§ Eq [4.8] is an approximation of Eq wz wz_Approx

[4.6]. For instance, the figure to the right 0.6

shows the difference in yaw velocity in 0.4

same simulation as in Figure 4-18. The
approximation is good for small front 0.2
axle steering angle 𝑓 .
0.0

-0.2

-0.4
time [s]
-0.6
0 1 2 3 4

Figure 4-19: § Difference between Eq [4.6] and Eq [4.8]

4.3.2.1 Relation 𝜹𝒇 𝒗𝒙 and 𝑹𝒑

Solving the first equation in Equation [4.8] yields:

𝑓 ∙ ∙ 𝐿2 + ( ∙ 𝑓 ∙ 𝑓) ∙ ∙ 2
𝑠𝑠𝑢 𝑒 𝑠 ⇒
𝑓 ∙ ≈{ }≈
𝑓 ∙ ∙𝐿+( ∙ 𝑓 + ∙ )∙ 𝑓 𝑤
⇒ ⁄ ≈ 𝑅𝑝
∙ ∙ 2 𝑠𝑠𝑢 𝑒:
𝐿 𝑓 𝑓 ∙
≈ ∙ + ∙ ≈{ }≈
+ 𝑓 𝑤⁄ 𝑓 𝑅𝑝 ( 𝑅𝑝 𝑤⁄ 𝑓 ≈ 0
+ 𝑓 𝑤) ∙ ∙𝐿 𝑓
𝑓
2
[4.9]
𝐿 ∙ 𝑓 ∙ 𝑓 ∙ ∙ 𝑓 ∙ 𝑓 𝑓
≈ + ∙ {𝑢𝑠𝑒: 𝐾 }
𝑅𝑝 𝑓 ∙ ∙𝐿 𝑅𝑝 𝑓 ∙ ∙𝐿 𝑓 ∙𝐿 ∙𝐿
2
𝐿 ∙
+𝐾 ∙
𝑅𝑝 𝑅𝑝
The coefficient 𝐾 is the understeer gradient and it will be explained more in 4.3.3.

4.3.2.2 Relation 𝜹𝒇 𝒗𝒙 and 𝜷

Solving the second equation in Equation [4.8] yields:

𝑓 ∙ ∙ 𝐿2 + ( ∙ 𝑓 ∙ 𝑓) ∙ ∙ 2
𝑓 2)
∙ ≈
𝑓 ∙( ∙𝐿 𝑓 ∙ ∙ 2) + ( 𝐿 𝑓 ∙ 𝑓 𝑤
𝑓 ∙ ∙ 𝐿2 ( 𝑓 ∙ 𝑓 ∙ )∙ ∙ 2
≈ {𝑢𝑠𝑒: 𝑓 𝑤 0} ≈ 2
∙ ≈
∙ ∙ ∙𝐿 ∙ 𝑓∙ ∙
𝑓 𝑓 [4.10]
𝐿
𝑓 → ∙
𝑣 →
⇒ ∙ ∙ 𝐿
𝑓 𝑓
𝑓 → ∙ 𝐾 ∙ ∙ ∙
{ 𝑣 →∞ 𝑓 ∙ 𝑓 𝑓

244
 Lateral Dynamics

𝑣𝑦
We can see that there is a speed dependent relation between steer angle and side slip, . The side slip
𝑣
𝑦 𝑣
can also be expressed as a side slip angle, arctan ( 𝑣 ). Since normally 𝐾 > 0, the side slip changes
sign, when increasing speed from zero to sufficient high enough. This should feel intuitively correct, if
agreeing on the conceptually different side slip angles at low and high speed, as shown in Figure 4-20.
We will come back to this equation in context of Figure 4-31.

4.3.2.3 Relation 𝒗𝒙 𝑹𝒑 and 𝜷

If we approximate 𝑓 𝑤 0 and use both equations in Equation [4.8] to eliminate 𝑓 we get:
2 2
𝐿 𝑓 𝑓
≈ 𝑅𝑝 tan( ) ⇒
𝐿 𝐿
2 [4.11]
𝑓
⇒ tan( ) ( )
𝐿 𝑅𝑝
path of
low speed front axle
high speed
𝜷>𝟎 path of 𝜷 𝜷
rear axle 𝜷<𝟎

v v
vehicle
vehicle path
path centre
centre

path of
front axle
path of
rear axle
Figure 4-20: Body Slip Angle for Low and High Speed Steady State Curves

4.3.2.4 Less General Derivation of Model

A less general and more intuitive way to reach almost same expression as first expression in Eq [4.8] is
given in Figure 4-21. Here, the simplifications are introduced earlier, already in physical model, which
means e.g. the influence of 𝑓 𝑤 is not captured. The simpler compatibility from 1.6.1.5 is used and
> 0 is assumed.
Physical model: Mathematical model:
• Path radius >> the vehicle. Then, all forces (and centripetal 𝑣 2
Equilibrium: ≈ 𝑓 + 0≈ 𝑓 𝑓
acceleration) are approximately co-directed.
• Small tyre and vehicle side slip. Constitution: 𝑓 𝑓 𝑠𝑓 𝑠
Then, angle=sin(angle)=tan(angle). Compatibility:
𝑣 2
(Angles are not drawn small, 𝑠𝑓 ≈ + 𝑓 ⁄ 𝑓
which is the reason why the 𝑝
𝑓 𝑠 ≈ ⁄
forces not appear 𝑓
co-linear in figure.) ≈ ⁄𝑅𝑝
𝑓
𝑓 Eliminate 𝑓 , 𝑠𝑓 𝑠 yields:
𝐿 2
𝑓 𝑓
𝑓 ≈ +𝐾 ℎ𝑒𝑟𝑒 𝐾
𝑓 𝑅𝑝 𝑅𝑝 𝑓 𝐿
𝐿
Figure 4-21: Simpler derivation final step in Equation [4.9].
Note that 𝑓 and in Figure 4-21 are drawn as positive for the sign convention of (ISO 8855). It can
help understanding to note that these angles get negative values if vehicle steady state turns to the left,
as indicated by the drawn steering angle 𝑓 .

245
 Lateral Dynamics

4.3.2.5 Steady State Cornering for Non-Ackermann Geometry

Ackermann geometry has been assumed above. We will now try a vehicle like in 4.2.4 and Figure 4-6.
Further on, we assuming 𝑓 𝑤 0 ( + 2 )⁄ Δ𝐿 2 𝐿 𝑓+ 𝑛 2
⁄ . This gives an equation, comparable with Eq [4.9], as follows:
2 2
𝐿 Δ𝐿𝑟 𝐿 𝑓 𝑓
+( ) ( + ) +𝐾 ℎ𝑒𝑟𝑒 𝑛 𝐾
𝑓
𝑅
⏟𝑝 𝐿 𝑓 𝑅𝑝 𝑅𝑝 𝑓 𝐿 [4.12]
𝛿

Note: We can still identify an 𝐾 , but the reference angle 𝐴 , see also 4.3.3, is not as simple as 𝐿⁄𝑅𝑝 .

4.3.2.6 Model Validity

The validity of the steady state models described in 4.3.2, is of course limited by if the manoeuvre is
transient which would mean that steady state is not reached, e.g. if driving above critical speed. But it
is also limited by if the assumption of linear tyre characteristics, ∙ 𝑠 , is violated. Therefore,
one should check if some axle uses too much of available friction, √ 2
+ 2
> 𝑓𝑟 𝑡 𝑜𝑛 ∙ ∙ .
Referring to the tyre brush model with uniform pressure distribution one can argue for using 0.5 as
this limiting fraction, since the tyre force is linear with slip up to this value.

4.3.3 Under-, Neutral- and Over-steering *

Function definition: Understeering (gradient) is the additional steer angle needed per increase of lateral force
(or lateral acceleration) when driving in high speed steady state cornering on level ground and high road friction.
Additional refers to low speed. The gradient is defined at certain high-speed steady state cornering conditions, in-
cluding straight-line driving. Steer angle can be either road wheel angle or steering wheel angle.
The first term in Eq [4.9], 𝐿/𝑅, can be seen as a reference steer angle 𝐴, which is the Ackermann steer
angle. The (ISO 8855) defines 𝐴 as the steer angle which would be needed to give same instantaneous
centre of rotation if the vehicle would have had two axles, perfect Ackermann steering and no tyre side
slip. So, 𝐴 corresponds to L/R for a two-axle vehicle. An extended definition of 𝐴 is the steering angle
needed for a certain curvature /𝑅𝑝 at low speed. The understeer gradient is then understood as a
measure of how this is changed with increasing speed .
The understeering gradient, 𝐾 , is normally positive, which means that most vehicles require more
steer angle for a given curve, the higher the speed is. Depending on the sign of 𝐾 a vehicle is said to be
oversteered (if 𝐾 < 0), understeered (if 𝐾 > 0) and neutral steered ( 𝑓 𝐾 0). In practice, all vehi-
cles are designed as understeered, because over steered vehicle would become unstable and difficult
to control.
The 𝐾 in Eq [4.9] is called “understeer gradient” and has hence the unit rad/N or 1/N. Sometimes one
can see slightly other definitions of what to include in definition of understeer gradient, which have
different units, see 𝐾 2 and 𝐾 3 in Eq [4.13].
𝐿 𝑟 ∙𝑙𝑟 𝑓 ∙𝑙𝑓 𝑚∙ ∙𝑣 2 𝑟 ∙𝑙𝑟 𝑓 ∙𝑙𝑓 𝐿 𝑣 2
𝑓 + ∙ {𝐾 2 ∙ ∙ [ 𝑜𝑟 𝑟 ]} +𝐾 2 ∙
𝑓 ∙ 𝑟 ∙𝐿 ∙ 𝑓 ∙ 𝑟 ∙𝐿 ∙
[4.13]
𝐿 𝑟 ∙𝑙𝑟 𝑓 ∙𝑙𝑓 𝑚∙𝑣
2 𝑟 ∙𝑙𝑟 𝑓 ∙𝑙𝑓 𝐿 𝑣 2
𝑓 + ∙ {𝐾 3 ∙ [𝑚⁄ 2 𝑜𝑟 𝑚⁄ 2]} +𝐾 3∙
𝑓 ∙ 𝑟 ∙𝐿 𝑓 ∙ 𝑟 ∙𝐿

𝐾 3 is the definition used in (ISO 8855). For 𝐾 3 , one can sometimes see the unit “rad/g” used, which
present compendium recommended to not use.
If vertical loads on axles are only due to gravity ( (𝐿 )⁄𝐿) and tyres linear with vertical
load ( ) we can express 𝐾 2 ⁄ 𝑓 ⁄ .

246
 Lateral Dynamics

4.3.3.1 Understeering as a Fix Built-In Measure

The understeering gradient 𝐾 can be understood as how much additionally to the reference steer an-
gle one has to steer, to reach a certain centrifugal force, ∙ 2 ⁄𝑅 (or, if using 𝐾 3 , a certain accel-
eration 2⁄
𝑅):
∙ 2 𝑓 𝐴 𝑓 𝑓
𝑓 𝐴 + 𝐾 ∙ 𝐴 + 𝐾 ∙ ⇒ 𝐾 𝑜𝑟 𝐾 3
𝑅
Understeering is a steady state property and does depend on which axle is steered, see Figure 4-22.

1. If small (Low speed) 𝑓 𝑤

2. If increases, 𝑓
increases…

2a: …and if 𝑓 ≪ ⇒ side 2b: …and if ≪ 𝑓 ⇒ side

slip changes mainly on front ⇒
𝑤
slip changes mainly on rear ⇒
curve radius increases, which curve radius decreases, which
we call under-steering we call over-steering

Figure 4-22: Under- and over-steering for a two-axle vehicle with 𝑓 . It does not depend on
which axle is steered, but which axle is first in the direction of motion. Figure drawn for vehicle
driving forward.

4.3.3.2 Understeer Gradient as Varying with Steady State Lat-

eral Acceleration
So far, the understeering gradient is presented as a fix vehicle parameter. There is nothing that says
that a real vehicle behaves linear, so in order to get a well-defined value of 𝐾 , the 𝑓 and the
should be small. However, if we accept that 𝐾 can change with , 𝐾 can be defined as a differential
quantity. 𝐾 can also be understood as how much the additional steer angle, 𝑓 , has to increase per
increased centrifugal force, , or per centrifugal acceleration, :
𝜕( 𝑓) 𝜕 𝜕 𝑓 𝜕( 𝑓) 𝜕 𝑓
𝐾 ( 𝑓 𝐴) 𝑜𝑟 𝐾 3 [4.14]
𝜕 𝜕 𝜕 𝜕 𝜕
Equation [4.14] shows the understeering gradient as a function of , rather than a scalar parameter.
But it is still fix and built-in in the vehicle. If assessing understeering for a lateral forces up to near road
friction limit, Equation [4.14] is more relevant than Equation [4.9], because it reflects that understeer-
ing gradient changes.

4.3.3.3 Understeering as a Varying Quantity during a Transient

Manoeuvre
A third understanding of the word understeering is quite different and less strictly defined. It is to see
the understeering as a variable during a transient manoeuvre. For instance, a vehicle can be said to un-
dersteer if tyre side slip is larger on front axle than on rear axle, | 𝑓 | > | |, and over-steer if opposite,
| | > | 𝑓 |. This way of defining understeering and oversteering is not built-in in vehicle but varies
over time through a (transient) manoeuvre. E.g., when braking in a curve a vehicle loses grip on rear
axle due to temporary load transfer from rear to front. Then the rear axles can slide outwards signifi-
cantly, and the vehicle can be referred to as over-steering at this time instant, although the built-in un-
dersteering gradient is >0. This “instantaneous” under-/over-steering (binary, not an understeer gra-
dient) can be approximately found from log data with this simple approximation:

247
 Lateral Dynamics

𝐿 𝐿∙ 𝐿∙
𝑛 𝑙 ≈{ ≈𝑅∙ }≈ ≈{ ≈ ∙ }≈ 2 [4.15]
𝑅
If the actual vehicle has | 𝑓 | < | 𝑛 𝑙 | the vehicle oversteers, and vice versa. This is often very prac-
tical since it only requires simply logged data, 𝑓 and . Note that when 𝑓 and 𝑛 𝑙 have differ-
ent signs, neither understeer nor oversteer is suitable as classification, but it can sometimes be called
“counter-steer”. An example of applying Eq [4.15] is shown in Figure 4-23, where one also see that the
ESC system does not follow the Eq [4.15] when deciding ESC interventions; ESC has more advanced
“reference models”, see 4.6.2.1.
A second look at Equation [4.9] tells us that we have to assume absence of propulsion and braking on
front axle, 𝑓 𝑤 0, to get the relatively simple final expression. When propulsion on front axle
( 𝑓 𝑤 > 0), the required steer angle, 𝑓 , will be smaller; the front propulsion pulls in the front end of
the vehicle. When braking on front axle ( 𝑓 𝑤 < 0), the required steer angle, 𝑓 , will be larger; the
front braking hinders the front end to turn in. To keep constant, which is required within definition
of steady state, one have to propel the vehicle because there will always be some driving resistance to
overcome. Driving fast on a small radius is a situation where the driving resistance from tyre lateral
forces becomes significant, which is a part of driving resistance which was only briefly mentioned in
3.2.

Figure 4-23: Log data from passenger car with ESC in a double lane change. Upper: Vehicle
motion. Middle: 𝑛 𝑙 from Eq [4.15] used to define “instantaneous under-/over-steering”
(US/OS). Lower: Pressure to each wheel brake.

4.3.3.4 Neutral Steering Point

An alternative measure to understeering coefficient is the longitudinal position of the neutral steering
point. The point is defined for lateral force disturbance during steady state straight-ahead driving, as
opposed to steady state cornering without lateral force disturbance. The point is where a vehicle-ex-
ternal lateral force, such as wind or impact, can be applied on the vehicle without causing a yaw veloc-
ity ( 0), i.e. only causing lateral velocity ( ≠ 0). From this definition, we can derive a formula for
calculating the position of the neutral steering point, see Figure 4-24.
The result is con- ∙ 𝑓 ∙ 𝑓 𝑓 ∙ ∙ 𝑓 ∙ 𝑓
densed in Eq [4.16]. 𝐾 ∙ 𝐿 ℎ𝑒𝑟𝑒 𝐾 [4.16]
𝑓 + 𝑓 + 𝑓 ∙ ∙𝐿

248
 Lateral Dynamics

We can see that the understeer gradient from steady state cornering model appears also in the for-
mula for neutral steering point position, . Since 𝑓 𝑛 𝐿 are positive, the neutral steering point is
behind of CoG for understeered (two-axle) vehicles, and in front of CoG for oversteered (two-axle) ve-
hicles. This is why and 𝐾 can be said to be alternative measures for the same vehicle function/char-
acter, the yaw balance.
Physical model: Mathematical model:
• Steady state ( 0) Equilibrium:
• Straight ahead driving (
𝑓 +
0) 0 +
• No steering
• Small tyre and vehicle side slip. Then,
0 𝑓 𝑓
angle=sin(angle)=tan(angle). Constitution: 𝑓 𝑓 𝑠𝑓 and 𝑠
𝑣𝑦
Fey Compatibility: 𝑠𝑓 𝑠
Fry Ffy 𝑣

𝑟 𝑙𝑟 𝑓 𝑙𝑓
Eliminate 𝑓 , 𝑠𝑓 𝑠 yields:
b 𝑟 𝑙𝑟
𝑓+ 𝑟
𝑓 𝑙𝑓
Identify understeering gradient, 𝐾
𝑓 𝑟 𝐿
0 Then:
𝑓 𝑓
L 𝐾 𝐿
𝑓 +
Figure 4-24: Model for definition and calculation of neutral steering point.

4.3.4 Required Steer Angle

A fundamental property of the vehicle is what steer angle that is required to negotiate a certain curva-
ture (=1/path radius /𝑅𝑝 ). This value can vary with longitudinal speed and it can be normalized
with wheelbase, 𝐿. From Equation [4.9], we can conclude:
∙ 𝑅𝑝 2
𝑓 ∙
𝑵𝒐𝒓 𝒂𝒍𝒊𝒛𝒆𝒅 𝒓𝒆𝒒 𝒊𝒓𝒆𝒅 𝒕𝒆𝒆𝒓𝒊𝒏𝒈 𝒂𝒏𝒈𝒍𝒆 +𝐾 ∙ [4.17]
𝐿 𝐿
The normalized required steer angle is plotted for different understeering gradients Figure 4-25. It is
the same as the inverted and normalized curvature gain, see 4.3.6.2.

2.5

Ku = 2.525e-6 [1/N]
Understeered
2
required steering, df*R/L [rad]

1.5

Neutral steered
Ku = 0e-6 [1/N]
1

Oversteered
0.5 Ku = -1.794e-6 [1/N]

Characteristic speed (for Critical speed (for

understeered vehicle) oversteered vehicle)
0
0 5 10 15 20 25 30 35 40 45 50
vx [m/s]

Figure 4-25: Normalized steer angle ( 𝑓 ∙ 𝑅 ⁄𝐿) for Steady State Cornering

249
 Lateral Dynamics

4.3.5 Critical and Characteristic Speed *

Function definition: Critical speed is the speed above which the vehicle becomes unstable in the sense that the
yaw velocity grows largely for a small disturbance in, e.g., steer angle.
Function definition: Characteristic speed is the speed at which the vehicle requires twice as high steer angle for
a certain path radius as required at low speed (Figure 4-25). (Alternative definitions: The speed at which the yaw
velocity gain reaches maximum (Figure 4-28). The speed at which the lateral acceleration gain per longitudinal
speed reaches its highest value. (Figure 4-30).)
We can identify that zero steer angle is required for the over-steered vehicle at 28 m/s. This is the so-
called Critical Speed, which is the speed where the vehicle becomes unstable. It can be noted that there
are stable conditions also above critical speed, but one has to steer in the opposite direction, often re-
ferred to as “drifting”. Normal vehicles are built understeered, which is why a Critical speed is more of
a theoretical definition. However, if studying (quasi-steady state) situations where the rear axle is
heavily braked, the cornering stiffness rear is reduced, and a critical speed can be relevant.
For understeered vehicles, we can instead read out another measure, the Characteristic Speed. The un-
derstanding of Characteristic Speed is, so far just that required steering increases to over twice what is
needed for low speed at the same path radius. A better feeling for Characteristic Speed is suggested in
4.3.6.3.
The vehicle is unstable for speeds above the critical speed. Instability is further discussed in 4.4.3.5.
From Equation [4.9], we can find a formula for critical and characteristic speeds:
2 ∙ ∙ 𝐿2
𝐿 ∙ 𝐿 𝑓
𝑓 +𝐾 ∙ 0 ⇒ √ √
𝑅𝑝 𝑅𝑝 𝐾 ∙ ( 𝑓 ∙ 𝑓 ∙ )∙
[4.18]
2 ∙ ∙ 𝐿2
𝐿 ∙ 𝐿 𝐿 𝑓
𝑓 +𝐾 ∙ ∙ ⇒ √ √
𝑅𝑝 𝑅𝑝 𝑅𝑝 𝐾 ∙ ( ∙ 𝑓 ∙ 𝑓) ∙

4.3.5.1 Quasi-Steady State Longitudinal Acceleration

We will now leave the strict steady state, by claiming that longitudinal load transfer due to ≠
0, although is constant. This is refered to as “Quasi-Steady State”.
Longitudinal load transfer show that braking increases over-steering tendency. It is actually so, that
the critical speed √𝐿⁄( 𝐾 ∙ ) (see Equation [4.18]) can come down to quite reachable lev-
els when braking hard; i.e. hard braking at high speed may cause instability. This is especially so for
front biased CoG location. See Figure 4-26, inspired by Reference (Drenth, 1993).
Using definition of understeer gradient:
𝑜𝑛 𝑡𝑢 𝑛
𝑓 𝑓
𝐾 { 𝑜 𝑡𝑟 𝑛𝑠𝑓𝑒𝑟}
𝑓 ∙𝐿 ∙𝐿 𝑓 ∙ 𝑓 ∙𝐿 ∙ ∙𝐿
∙ ∙ ℎ⁄𝐿
𝑓
ℎ 𝑓 ℎ
𝑓 ∙( ∙ ∙ ∙ ∙ )∙𝐿 ∙( ∙ ∙𝐿+ ∙ ∙ 𝐿) ∙ 𝐿
𝐿 𝐿
[4.19]
𝑓 𝑓 𝑓
( ) { }
𝑓
𝑓 ∙( ∙ ℎ) ∙(𝑓+ ∙ ℎ)

𝑓
( ) { } 0
ℎ ℎ 0
( ∙ ) ( + ∙ )
𝑓

250
 Lateral Dynamics

However, the cornering stiffness varies degressively, e.g. 𝑘𝑝∙ 𝑘 ∙ 2 . This is further stud-
ied in Reference (Drenth, 1993).
If taking the degressiveness of tyre cornering stiffness into account, the weight distribution plays a
role also without longitudinal load transfer; front biased weight distribution gives under-steered vehi-
cles and vice versa. Also, the number of wheels per axle influence stronger; single wheel front (or dou-
ble-mounted rear) gives under-steered vehicles and vice versa.
It should be noted that if the longitudinal acceleration is due to wheel torques, as opposed to road
grade or aerodynamic forces, the tyre combined slip effects will influence the curves which is not con-
sidered in Figure 4-26; the cornering stiffness of an axle will decrease with increased longitudinal
force.
𝑓⁄ 𝑓 𝑓 ⁄ =

Figure 4-26: Left: Under-steering gradient as function of longitudinal acceleration, , and static
load distribution, 𝑓 ⁄𝐿. Right: Critical and characteristic velocity as function of acceleration and
load distribution.
4.3.5.1.1 § Influence of Longitudinal Tyre Forces
With known combined slip effect, we could distribute longitudinal force so that 𝐾 is unchanged
and/or we could avoid braking so that becomes lower than the actual speed . To examplify
this, we use the simple combined slip model in Eq [2.46] and require 𝐾 0. We also assume flat road
and no air resistance. Resulting distribution is compared with two other distribution rules in next fig-
ure. For low | |, we have to apply wheel torque in different directions to keep 𝐾 0.

251
 Lateral Dynamics

⁄
⁄

⁄
⁄

⁄
𝑓
⁄
𝑓

⁄𝑓
⁄
𝑓

Figure 4-27: Three wheel torque distribution strategies: Distribution as static distribution of
vertical forces, Equal utilization of (as “ideal curve” in 3.4.4) and Neutral steered.

4.3.6 Steady State Cornering Gains *

Function definition: Steady state cornering gains are the amplification from steer angle to certain vehicle
response measures for steady state cornering at a certain longitudinal speed.
From Equation [4.9], we can derive some interesting ratios. We put steer angle in the denominator, so
that we get a gain, in the sense that the ratio describes how much of something we get “per steer an-
gle”. If we assume 𝑓 𝑤 0, we get Equation [4.20].
𝒀𝒂𝒘 𝒗𝒆𝒍𝒐𝒄𝒊𝒕𝒚 𝒈𝒂𝒊𝒏
⁄𝑅𝑝
{𝑢𝑠𝑒: ≈ }≈ 2 2
𝑓 𝑅𝑝 𝐿 ∙ 𝐿+𝐾 ∙ ∙
𝑅𝑝 + 𝐾 ∙ 𝑅𝑝
𝑪 𝒓𝒗𝒂𝒕 𝒓𝒆 𝒈𝒂𝒊𝒏
κ ⁄𝑅𝑝 ⁄𝑅𝑝
𝑓 𝑓 𝐿 ∙ 2
𝐿+𝐾 ∙ ∙ 2 [4.20]
𝑅+𝐾 ∙ 𝑅𝑝
𝑳𝒂𝒕𝒆𝒓𝒂𝒍 𝒂𝒄𝒄𝒆𝒍𝒆𝒓𝒂𝒕𝒊𝒐𝒏 𝒈𝒂𝒊𝒏
𝑢𝑠𝑒: 2
∙ 𝑅𝑝 2
≈ 2 2
𝑓 𝑛 ≈ 𝐿 ∙ 𝐿+𝐾 ∙ ∙
{ 𝑅𝑝 } 𝑅𝑝 + 𝐾 ∙ 𝑅𝑝
Yaw velocity gain is also derived for 𝑓 𝑤 ≠ 0, and then we get Equation [4.21].
𝒀𝒂𝒘 𝒗𝒆𝒍𝒐𝒄𝒊𝒕𝒚 𝒈𝒂𝒊𝒏 ( 𝑡ℎ 𝑓 𝑤 𝑡 𝑘𝑒𝑛 𝑛𝑡𝑜 𝑜𝑢𝑛𝑡)
𝑓 𝑤 [4.21]
( + ) 2
𝑓 𝑓 𝐿+𝐾 ∙ ∙

252
 Lateral Dynamics

4.3.6.1 Yaw Velocity Gain

The yaw velocity gain gives us a way to understand Characteristic Speed. Normally one would expect
the yaw velocity to increase if one increases the speed along a circular path. However, the vehicle will
also increase its path radius when speed is increased. At the Characteristic Speed, the increase in ra-
dius cancel out the effect of increased speed, so that yaw velocity in total decrease with increased
speed. One can find the characteristic speed as the speed where one senses or measures the highest
value of yaw velocity for a fix steer angle. The curves for 𝑓 𝑤 +0. 5 ∙ 𝑓 and 𝑓 𝑤 0. 5 ∙ 𝑓 in
Figure 4-28 are generated using Eq [4.21]. Note that critical and characteristic speed is independent of
𝑓 𝑤.

30

Oversteered
25
/

Ku = -1.794e-6 [1/N]
wz/df [(rad/s)/rad]

20
⁄ 𝑓

Neutral steered
Ku = 0e-6 [1/N]
Characteristic
gain,

15
speed (for
yaw rate gain,

understeered
yaw velocity

vehicle)
10

Understeered
Ku = 2.525e-6 [1/N]
5
Critical speed (for
oversteered vehicle)
0
0 5 10 15 20 25 30 35 40 45 50
longitudinal vx
velocity,
[m/s] ⁄𝑠
Figure 4-28: Yaw velocity gain ( / 𝑓 ) for Steady State Cornering. Each “cluster of 3 curves”:
Mid curve 𝑓 𝑤 0. Upper 𝑓 𝑤 +0.5 ∙ 𝑓 . Lower 𝑓 𝑤 0.5 ∙ 𝑓 .

4.3.6.2 Curvature Gain

If driving on a constant path radius, and slowly increase speed from zero, an understeered vehicle will
require more and more steer angle (“steer-in”), to stay at the same path radius. For an over-steered
vehicle, one has to steer less (“open up steering”) when increasing the speed.

Oversteered
Ku = -1.794e-6 [1/N]
2.5
⁄ [1/(m*rad)]
𝑚

2
curvature gain, (1/R)/df 𝑓

1.5
curvature gain,

Neutral steered
0.5 Ku = 0e-6 [1/N]

Ku = 2.525e-6 [1/N] Understeered

0
0 5 10 15 20 25 30 35 40 45 50
longitudinalvxvelocity,
[m/s] ⁄𝑠
⁄
Figure 4-29: Curvature gain ( ) for Steady State Cornering
𝛿𝑓

253
 Lateral Dynamics

4.3.6.3 Lateral Acceleration Gain

Figure 4-31 shows the lateral acceleration gain as function of vehicle speed. The characteristics speed
is once again identified in this diagram, and now as the speed when lateral acceleration per longitudi-
nal speed (( ⁄ 𝑓 )⁄ ) reaches its highest value. This is an alternative definition of characteristic
speed, c.f. 4.3.5.

500
Oversteered
𝑚⁄ 2

450 Ku = -1.794e-6 [1/N] Neutral steered

lateral acceleration gain, ay/df [(m/(s*s))/rad]

Ku = 0e-6 [1/N]
400
⁄ 𝑓

350

300
lateral acceleration gain,

250
Understeered
200
Ku = 2.525e-6 [1/N]
150

100 Characteristic speed is

where gain curve has a
50 tangent through origin
0
0 5 10 15 20 25 30 35 40 45 50
longitudinal velocity,
vx [m/s] ⁄𝑠
𝑦
Figure 4-30: Lateral acceleration gain ( ) for Steady State Cornering
𝛿𝑓

From the previous figures the responsiveness of the vehicle can be identified for different understeer
gradients. In all cases the vehicle which is understeered is the least responsive of the conditions. Both
the yaw velocity and lateral acceleration cannot achieve the levels of the neutral steered or over-
steered vehicle. The over-steered vehicle is seen to exhibit instability when the critical speed is
reached since small changes in the input result in excessive output conditions. In addition, the over-
steered vehicle will have a counter-intuitive response for the driver. To maintain a constant radius
curve, an increase in speed requires that the driver turns the steering wheel opposite to the direction
of desired path. The result of these characteristics leads car manufacturers to produce understeered
vehicles that are close to neutral steering to achieve the best stability and driver feedback.

4.3.6.4 Side Slip Gain as Function of Speed

All gains above can be found from solving from Eq [4.9]. If instead solving the other unknown, ,
we can draw “side slip gain” instead. Eq [4.22] shows the formula for this.
2
𝑓 ∙ ∙ ∙𝐿 𝑓 ∙ 𝑓 ∙ ∙
[4.22]
∙ 𝑓 𝑓 ∙ ∙ 𝐿2 ( 𝑓 ∙ 𝑓 ∙ )∙ ∙ 2

It is not solely the understeering gradient that sets the curve shape, but we can still plot for some real-
istic numerical data, which are under-, neutral and over-steered, see Figure 4-31.
All cases in Figure 4-31 goes from positive side slip to negative when speed increases. This is the same
as we expected already in Figure 4-20.
We can also calculate and plot the longitudinal location of the motion centre, i.e. 𝑥𝑀 ⁄ , by
combining Eqs [4.22] and [4.20]. Note that 𝑅 is independent of 𝑓 , while the longitudinal location of
the motion centre, 𝑦𝑀 ⁄ , is not.

4.3.6.5 Limitations due to Road Friction

The formulas with the gains in Eq [4.20] does not include the limitation due to maximum road friction,
i.e. the peak of the tyres’ (𝑠 ) curves. This limitation is a lateral acceleration 𝑚 where on of the
axles reach friction limit. Yaw and lateral equilibrium requires 𝑓 ∙ ⁄𝐿 and ∙

254
 Lateral Dynamics

𝑓 ⁄𝐿 , so the limit due to front axle is ∙ 𝑚 𝑓 ⁄𝐿 𝑓 𝑓 𝑓 ⁄𝐿 ⇒ 𝑚 𝑓

𝑓 and corresponding for rear axle. The total limit is 𝑚 min( 𝑓 ) Inserting this in Eq
[4.20] and eliminate 𝑓 , one finds:

𝑚 √ 𝑚 𝑅𝑝 √min( 𝑓 ) 𝑅𝑝 [4.23]

For example, if negotiating a curve with 𝑅 00 and 𝑓 gives 𝑚 ≈ ⁄𝑠 ≈

𝑘 ⁄ℎ and if either (or both) of 𝑓 and are just 0.5, 𝑚 ≈ ⁄𝑠 ≈ 80 𝑘 ⁄ℎ. So, we should
see the curves in Figure 4-25 and in Figure 4-28 to Figure 4-31 as invalid over certain speed 𝑚 ,
which depends on max road friction and curvature.
Reaching 𝑚 means loss of steerability if 𝑓 < , and loss of yaw stability if < 𝑓 . So, also an
under-steered vehicle becomes unstable, if < 𝑓 and driving faster than 𝑚 . And, an oversteered
vehicle might be limited from reaching its critical speed if 𝑓 < and driving on a certain radius 𝑅.
m = 2000 [kg]; lf = 1.3 [m]; lr = 1.5 [m] m = 2000 [kg]; lf = 1.3 [m]; lr = 1.5 [m]
1 20
understeered
lr/L

𝑓
0.5 neutral steered
oversteered

𝑦𝑀
15
⁄𝑟
⁄ 𝑓 [1/rad]

0
𝑥𝑀 ⁄
[1/rad]
⁄𝑟
⁄ (vy/vx)/df

-0.5
10
Neutral steered
xMC = -vy/wz

-1
slip gain",

-1.5 5
gain,
side slip"side

-2
𝑥𝑀

0
understeered
-2.5
neutral steered
oversteered
-3 -5
0 5 10 15 20 25 0 5 10 15 20 25
longitudinalvx velocity,
[m/s] ⁄𝑠 longitudinal
vx velocity,
[m/s] ⁄𝑠
𝑣𝑦 𝑣𝑦
Figure 4-31: Left Side slip gain (𝑣 ∙𝛿𝑓
). Right: Motion centre longitudinal location (𝑥𝑀 𝜔𝑧
). For
Steady State Cornering.

4.3.7 How Design Influences Steady State Gains

The cornering stiffness for each axle are only abstract design parameters which influence steady state
gains. The cornering stiffness is a combined effect from various, more concrete, design parameters.
Such more concrete design parameters are presented in 4.3.7. Also, some vehicle operation which af-
fects understeering gradient, such as hard braking in 4.3.7.1, is mentioned.

4.3.7.1 Tyre Design, Inflation Pressure and Number of Tyres

The cornering stiffness of each tyre is an obvious parameter which influences the axle cornering stiff-
ness. The cornering stiffness of an axle is influenced by the sum of cornering stiffness for all tyres.
There are normally two tyres per axle, but there are also vehicles with one tyre (e.g. bicycles) or 4 (2
double mounted tyres on each side on heavy trucks).
Tyre design influences, which is geometrical dimensions and material selection. Inflation pressure is in
here to be seen as a design parameter; a vehicle is sold with a recommended tyre inflation pressure.
In a first approximation, tyre cornering stiffness is approximately proportional to vertical load:
∙ . For a vehicle with same tyres front and rear ( 𝑓 ), this means that it will be neutral
steered, neglecting body forces (air and grade). This is because, in steady state cornering, vertical
loads are distributed over the axles in the same relation as lateral loads.

255
 Lateral Dynamics

4.3.7.2 Roll Stiffness Distribution between Axles

During cornering, the vertical load is shifted towards the outer wheels. Depending on the roll stiffness
of each axle, the axles take differently much of this lateral load transfer. This also influences the yaw
balance. The more roll stiff an axle is, the more of the lateral load shift it takes. Tyre cornering stiffness
varies degressively with vertical load. Together, this means that increasing the roll stiffness on the
front axle, leads to less front cornering stiffness, see Figure 2-39, and consequently more understeered
vehicle. Increasing roll stiffness on rear axle makes the vehicle more oversteered. The total roll stiff-
ness of the vehicle does not influence the understeering gradient. Normally one makes the front axle
more roll stiff than the rear axle. This means that vehicle becomes more and more understeered for
increased lateral acceleration, e.g. more steer angle is needed to maintain a certain path radius if speed
increases. One can change the roll stiffness of an axle by changing roll centre height, wheel stiffness
rate and anti-roll bar stiffness.

.7 05 𝑁/𝑟
. 05 𝑁/𝑟

Both wheels, if no load transfer.

. 05 Axle cornering stiffness ≈
𝑁/𝑟 ≈ . 6. 05 𝑁/𝑟
Cornering
stiffness

Left and right wheel, if load transfer ± 5 𝑘𝑁.

Axle cornering stiffness ≈ . + .7 5.8 05 𝑁/𝑟

± 5 𝑘𝑁 ± 5 𝑘𝑁

Figure 4-32: The wheels cornering stiffness (𝜕 ⁄𝜕 ( )| ) changes degressively with vertical
𝑦=
load. The axle cornering stiffness therefore decreases with increased load transfer.

4.3.7.3 Suspension and Steering Compliances

4.3.7.3.1 Side-Force Steer Gradient
Side-force steer gradient, , is defined for an axle and it is how much the wheels on an axle steers
[deg] negatively per lateral force [N]. Negative is chosen since the normal case is that pivot point is
ahead of wheel, so that wheel steers negative for a positive force. Also, a non-steered axle steers due to
side force steering, which depends on the compliance of the suspension bushings.
It can be modelled as an extra compliance, with the constitutional equations: 𝑓 𝑓 ∙ Δ 𝑓 and
∙ Δ , where the Δ marks additional steer angle due to the lateral force. These extra com-
pliances come into play as series connected with the tyre cornering compliances. If we update Equa-
tion [4.9] with side force steering it becomes:
2
𝐿 ∙
𝑓 +𝐾 ∙
𝑅 𝑅
𝑓 ∙ 𝑓 ∙ 𝑓
ℎ𝑒𝑟𝑒 𝐾
𝑓 ∙𝐿 ∙𝐿 𝑓 ∙ ∙𝐿 [4.24]

ℎ𝑒𝑟𝑒 𝑓 𝑛
+ +
𝑓 𝑓

For vehicles with largely varying vertical axle load (such as heavy trucks), one has to consider that the
contribution from tyre to axle cornering stiffness is rather proportional to vertical axle load, while the
contribution from side-force steering comes from suspension elasticities and is rather constant. So,
utilizing side-force steering makes the vehicles lateral manoeuvrability inconsistent with vertical load.
§ Influence of Translatory Compliance

256
 Lateral Dynamics

One way to deliberately design for side-force steering is to let the axle or wheels have yaw pivot points
ahead or behind. This also leads to a lateral translatory compliance of the axle or wheels which,
strictly, also gives one more state variable. For lower frequencies, such as for handling, this compliance
can often be neglected.
4.3.7.3.2 Roll Steer Gradient
Roll steer gradient, 𝑘 , is defined for an axle and it is how much the wheels on an axle steers [deg]
per vehicle roll angle [deg]. Also, a non-steered axle can steer due to roll-steering. Roll-steering de-
pends on the suspension linkage geometry. The added steer angle can be expressed: Δ 𝑘 ∙ .
We will now derive the influence on steady state cornering. Add steering on rear axle to Eq [4.9]:
𝐿 𝑚∙𝑣 2 𝑟 ∙𝑙𝑟 𝑓 ∙𝑙𝑓
𝑓 ≈ +𝐾 𝑛 ∙ 𝐾 𝑛 (subscript “noRS” means “no Roll-Steer”)
𝑝 𝑝 𝑓 ∙ 𝑟 ∙𝐿

If we see as built up by one angle from the steering system and one part coming from the suspen-
sion, via roll-steering Δ :
𝐿 𝑚∙𝑣 2
𝑓 +Δ 𝑓 ( +Δ )≈ +𝐾 𝑛 ∙
𝑝 𝑝

Then, we can express Δ in :

Δ {Δ 𝑘 } 𝑘 ∙ 𝑓𝑜𝑟 𝑓 𝑟.
We can also express the relation between and :
ℎ 𝑚 𝑦 𝑣2 𝑘𝑖 𝑆 𝑚 𝑣2
𝑘 ∙ { } 𝑘 ∙ { } 𝑓𝑜𝑟 𝑓 𝑟.
𝑝

Insertion identifying an understeering gradient with roll-steering gives:

2 ℎ (𝑘 𝑘𝑓 )
𝐿
𝑓 ≈ +𝐾 𝐾 𝐾 𝑛 + [4.25]
𝑅𝑝 𝑅𝑝

4.3.7.3.3 Steering System Compliance

The steering system is also compliant: 𝑓 𝑆𝑡𝑒𝑅 𝑡 𝑜 𝑤 𝑇 𝑤⁄ where [𝑁 / 𝑒 ]
is the compliance between steering wheel and road wheel steering. The steering system compliance
and the side-force steering compliance are very closely related. They are two series coupled compli-
ances: compliance in steering system (e.g. steering column torsion bar) and in suspension (e.g. bush-
ings), respectively.
4.3.7.3.4 Quantified Combined Effect
Side force steering and roll-steering are similar but have different time scales. Roll-steering requires
that sprung body changes roll angle, which takes significantly longer time; typically roll eigen-fre-
quency is 1..2 Hz. Side force steering does not require a roll angle change, so side-force steering has
much less time delay. Roll-steering also comes into play for one-sided road unevenness, i.e. also with-
out cornering and without body roll.
The combined effect from side force steering (4.3.7.3.1), roll-steering (4.3.7.3.2) and steering system
compliance (4.3.7.3.3), often represent a significant part of the front axle cornering compliance, e.g.
reduces compliance with 20..40% compared to tyre cornering compliance only, Reference (Wedlin,
Tillback, & Bane, 1992). Most of this is due to steering system compliance.
On rear axles on passenger cars, the influence is typically less and in opposite direction, e.g. increase
1..5% compared to tyre cornering compliance only.
Rear axles on heavy vehicles are typically designed without significant side-force steering. However,
the frame compliance can cause a relevant amount of side-force oversteering due to that the whole
frame steers curve-outwards rear and curve-inwards front. The frame compliance is especially influ-
ential on a tractor with fifth wheel ahead of rear axle(s); clearly larger than sideforce compliance on a
single axle and on a rigid truck about the same, depending on the body-build on the frame.
Patents exist for making the rear axle suspension on heavy vehicle’s trailers sideforce steering so that
axle becomes less cornering compliant, reducing compliance with typically 1/3 compared to tyre cor-
nering compliance only. This increases yaw stability, which is very much same concept as using side-

257
 Lateral Dynamics

force understeering rear at a two-axle vehicle. Using this concept can lead to very yaw stable vehicles.
The drawback is reduced yaw agility. If really exaggerated, it can take the rear axle to effectively nega-
tive cornering stiffness, which makes vehicle unstable.

4.3.7.4 Camber Steer

Negative camber (wheel top leaning inwards) increases the cornering stiffness. One explanation to this
is that curve outer wheel gets more vertical load than the curve inner wheel. Hence, the inwards di-
rected camber force from outer wheel dominates over outwards directed camber force from the inner
wheel. Negative camber is often used at rear axle at passenger cars. Drawback with non-zero camber is
tyre wear.

4.3.7.5 Toe Angle

Toe has some, but limited effect on an axle’s cornering stiffness. Non-zero toe increases tyre wear. Toe-
angle: When rolling ahead, tyre side forces pre-tension bushes.
If toe (=toe-in) is positive there are tyre-lateral forces on each tyre already when driving straight
ahead, even if left and right cancel out each other: ( 𝑙𝑓 )∙
2
0 . Then, if the axle
takes a side force, the vertical loads of the wheels are shifted between left and right wheel, which also
changes the tyre cornering stiffnesses. The outer wheel will get more cornering stiffness. Due to posi-
tive toe, it will also have the largest steer angle. So, the axle will generate larger lateral force than with
zero toe. For steady-state cornering vehicle models, this effect comes in as an increased axle cornering
stiffness, i.e. a linear effect.

4.3.7.6 Wheel Torque Effects

Wheel torque give tyre longitudinal force, directed as the wheel is directed. If the wheel is steered, the
wheel longitudinal forces can influence the yaw balance, see also 𝑓 𝑤 in Equation [4.9].
Unsymmetrical wheel torques (left/right) will give a direct yaw moment in the yaw equilibrium in
Equation [4.6]. The actuated yaw moment around CoG is then of the magnitude of wheel longitudinal
wheel force times half the track width. ESC and Torque vectoring interventions have such effects.
High longitudinal utilization of friction on an axle leads to that lateral grip is reduced on that axle. The
changed yaw moment, compared to what one would have without using friction longitudinally, can be
called an indirect yaw moment. The actuated change in yaw moment around CoG is then of the mag-
nitude of change in wheel lateral wheel force times half the wheelbase. It influences the yaw balance.
That is the reason why a front axle driven vehicle may be more understeered than a rear axle driven
one. On the other hand, the wheel-longitudinal propulsion force on the front axle does also help the
turn-in, which acts towards less understeering.

4.3.7.7 Transient Vehicle Motion Effects on Yaw Balance

The effects presented here are not so relevant for steady state understeering coefficient. However,
they affect the yaw balance in a more general sense, why it is relevant to list them in this section.
• Longitudinal load transfer changes normal forces. E.g. strong deceleration by wheel forces
helps against under-steering, since front axle gets more normal load. This effect has some de-
lay. Also, it vanishes after the transient.
(This effect can be compared with the effect described in 4.3.7.2, which is caused by tyre cor-
nering stiffness varying degressively with vertical load, while the longitudinal load transfer ef-
fect can be explained solely with the proportional variation.)
• Change of longitudinal speed helps later in manoeuvre. E.g. deceleration early in a manoeuvre
makes the vehicle easier to manoeuvre later in the manoeuvre. It is the effect of the term
that is reduced.

4.3.7.8 Some Other Design Aspects

High cornering stiffness is generally desired for controllability.

258
 Lateral Dynamics

Longer wheelbase (with unchanged yaw inertia and unchanged steering ratio) improves the transient
manoeuvrability, because the lateral forces have larger levers to generate yaw moment with.

4.3.8 Manoeuvrability and Stability

The overall conclusion of previous section is that all gains become higher the more over-steered (or
less understeered) the vehicle is. Higher gains are generally experienced as a sportier vehicle and they
also improves safety because they improve the manoeuvrability. A higher manoeuvrability makes it
easier for the driver to do avoidance manoeuvres. This motivates a design for low understeering gradi-
ent.
However, there is also the effect that a vehicle with too small understeer gradient becomes very sensi-
tive to the steering wheel angle input. In extreme, the driver would not be able to control the vehicle.
This limits how small the understeering gradient one can design for. Generally, vehicles are built un-
dersteered.
It is not impossible for a driver to keep an unstable vehicle (𝐾 < 0 and > ) on an intended
path, but it requires an active compensation with steering wheel. If adding support systems, such as
yaw damping by steering support or differentiated propulsion torques, it can be even easier. If one
could rely on a very high up time for such support systems, one could move today’s trade-off between
manoeuvrability and stability. This conceptual design step has been taken for some airplanes, which
actually are designed so that they would be unstable without active control. See also 4.4.3.5.

4.3.9 Handling Diagram

There are many frequently used graphical tools or diagrams to represent vehicle characteristics. One
is the “handling diagram”. A handling diagram tells maximum cornering 𝑀 and understeering gra-
dient at different . A handling diagram is essential a plot of same data as in Lateral Acceleration gain
in Figure 4-30, but the curve ( 𝑓 ) for one instead of ⁄ 𝑓 ( ). For the linear model it would be a
straight line, with ⁄ 𝑓 𝑜𝑛𝑠𝑡 𝑛𝑡 2 ⁄(
𝐿 + 𝐾 ∙ ∙ 2 ). But for a real vehicle test, or a more
advanced model, it becomes more different.
A handling diagram is constructed as follows. Same simplifying assumptions are done as in Figure
4-21, with the exception that we don’t assume linear tyre models.
Equilibrium:
2
𝑓
{ ∙ ∙ 𝑓 + 0 𝑓 ∙ 𝑓 ∙ }⇒ 𝑓 ∙ ∙ ∙ ∙
𝑅 𝐿 𝐿
Constitution:
𝑓 𝑓 ( 𝑓) ⇒ 𝑓 𝑓 ( 𝑓 ) ( )⇒ ( )
Solving for 𝑓 yields:
𝑓
𝑓 ( ) 𝑓 ( 𝑓 ) ( ∙ ∙ ) 𝑓 ( ∙ ∙ )
𝐿 𝐿
So, we can plot 𝑓 as function of . This relation is interesting because compatibility ( 𝑓 + 𝑓
𝐿⁄𝑅 ) yields 𝑓 𝑓 𝐿⁄𝑅 𝑓 𝐴 . And 𝑓 𝐴 is connected to one of the understand-
𝜕
ings of 𝐾 in Equation [4.14], (𝐾 𝜕 𝑦
( 𝑓 𝐴) ). If we plot 𝑓 𝑓 𝐴 on abscissa axis and
on ordinate axis, we get the most common way of drawing the handling diagram, see Figure 4-33.
The axle’s constitutive relations can be used as graphical support to construct the diagram, but then
𝐿
the constitutive relations should be plotted as: ( ) ∙ ∙ ( ) . The quantity can be seen
𝐿 𝑙𝑖 𝑚
as the lateral force scaled with a certain fraction of vehicle mass, where the certain fraction is such that
both axles’ values correspond to the same vehicle lateral acceleration.
Figure 4-33 shows the construction of a handling diagram from axle slip characteristics. Figure 4-34
show examples of handling diagrams constructed via tests with simulation tools. Handling diagrams

259
 Lateral Dynamics

can be designed from real vehicle tests as well. The slope in the handling diagram corresponds to un-
dersteering gradient 𝐾 3 in Equation [4.13].

𝐿
front axle, 𝑓 𝑓 𝑓
𝑙𝑟 𝑚
𝐿
rear axle,
𝑙 𝑚

Using the axle

models above,
Handling diagram the vehicle
becomes
Linear axle models, oversteered
understeered vehicle

Normal vehicle stays

understeered

𝑓 𝐴 ≈ 𝑓 Try to get some

measured as well.
From
Figure 4-33: Construction of the “Handling diagram”. The axle’s slip characteristics VCC, Saab?
(upper
diagram) are chosen so that vehicle transits from understeer to over-steer with increased
longitudinal speed, . The dashed shows two other slip characteristics.

From: Daniel A. Fittanto, et al. “Passenger Vehicle Steady-State Directional Stability Analysis
Utilizing EDVSM and SIMON”, Copyright 2004 by Engineering Dynamics Corporation

Figure 4-34: Example of handling diagram.

4.3.10 Lateral Load Transfer in Steady State Cornering

In the chapter about longitudinal dynamics we studied (vertical tyre) load transfer between front and
rear axle. The corresponding issue for lateral(EDVSM
dynamics is load
and SIMON aretransfer between
two different left
simulation and right side of
tools.)
the vehicle. Within the steady state lateral dynamics, we will cover some of the simpler effects, but
save the more complex suspension linkage dependent effects to 4.4.3.5.

260
 Lateral Dynamics

The relevance to study the load transfer during steady state cornering is to limit the roll during corner-
ing (for comfort) and yaw balance (understeering gradient, see 4.3.7.2). Additionally, the load transfer
influence the transient handling; see 4.4 and 4.4.3.5.

4.3.10.1 Load Transfer between Vehicle Sides

Without resolving into front and rear, we do not need to involve suspension in model:
View from rear:

curve-inner side curve-outer side

h
(if > 0, i.e. (if > 0, i.e.
turning left) 𝑙
turning left)

𝑙
w/2 w/2
Figure 4-35: A cornering vehicle. The is a fictive force. Subscript l and r mean left and right.

Moment equilibrium, around left contact with ground: ⇒

ℎ
⇒ ∙ ∙ + ∙ ∙ℎ ∙ 0⇒ ∙( + ∙ ) [4.26]
Moment equilibrium, around right contact with ground: ⇒ 𝑙 ∙( ∙ )
2 𝑤

These equations confirm what we know from experience, the curve-inner side if off-loaded.

4.3.10.2 Body Heave and Roll Due to Lateral Wheel Forces

Now, we shall find out how much the vehicle rolls and heaves during steady state cornering. First, we
decide to formulate the model in “effective stiffnesses”, in the same manner as for longitudinal load
transfer in previous chapter.
There is no damping included in model, because their forces would be zero, since there is no displace-
ment velocity, due to the “quasi-steady-state” assumption. As constitutive equations for the compli-
ances (springs) we assume that displacements are measured from a static condition and that the com-
pliances are linear. The road is assumed to be smooth, i.e. 𝑙 0 (2nd subscript means road).

𝑙 𝑙 + ∙( 𝑙 𝑙) + ∙ (( 𝑙 𝑙) ( ))
{ 𝑙 0} 𝑙 ( + )∙ 𝑙 + ∙
[4.27]
⋯ ( + )∙ + ∙ 𝑙
ℎ𝑒𝑟𝑒 𝑙 + ∙ 𝑛 𝑙 ∙ ⁄ ∙ ⁄ 0
The stiffnesses and ( 𝑟 means anti-roll bar) are effective stiffnesses as measurable under
the wheels. The physical springs are mounted inside in some kind of linkage and have different stiff-
ness values, but their effect is captured in the effective stiffnesses. Some examples of different physical
spring and linkage design are given in 2.3.3.1 and 2.3.4.1.
We see already in free-body diagram in Figure 4-36 that 𝑙 and always act together, so we rename
𝑙 + . We see in Figure 2-61 that we have to assume something about how the lateral forces
are transferred from road to body. The “trivial linkage” from Figure 2-61 is assumed. Equilibrium
then gives:
∙ 0
∙ 𝑙 0 [4.28]
𝑙 ∙( ⁄ ) ∙ ( ⁄ ) + ∙ ℎ + ∙ ∙ ( 𝑦) 0
The term ∙ ∙ ( 𝑦) is taken as ∙ ∙ (ℎ ℎ ) ∙ sin( ) ≈ ∙ ∙ (ℎ ℎ ) ∙ . It assumes a height
for the point where the roll takes place, ℎ . We don’t know the value of it, until below where we study

261
 Lateral Dynamics

the suspension design, but it can be mentioned already here that most vehicles have an ℎ ≪ ℎ. This
causes a “(roll) pendulum effect”, especially significant for heavy commercial vehicles due to their
large ℎ..
Compatibility, to introduce body displacements, and , gives:
𝑙 +( ⁄ )∙ 𝑛 ( ⁄ )∙ [4.29]
-y
(Anti-roll is drawn, only in left
picture and beside the vehicle,
ay z for better clarity in drawing.)
y m*ay
zl zr
, anti-roll bar
torsional spring
h m*g

Fyl Fyr hRC

zlr=0 Fzr
zrr=0 Fzl
w/2 w/2 w/2 w/2
Steady-state assumed, so that
z,y,px= ,zl,zr, are displacements lateral acceleration (ay) may be
from a static stand-still position. non-zero, but vertical and roll
zlr=zrr=0 means that road is smooth. acceleration are zero.
Figure 4-36: Model for steady state heave and roll due to lateral acceleration. Suspension model is
no linkage (or “trivial linkage”) and without difference front and rear.
Combining constitutive relations, equilibrium and compatibility, gives, as Matlab script:
syms …; sol=solve( ...
Flz==Flz0-(cside+carb)*zl+carb*zr, ...
Frz==Frz0-(cside+carb)*zr+carb*zl, ...
Flz0+Frz0==m*g, Flz0*w/2-Frz0*w/2==0, ...
Fy-m*ay==0, ...
m*g-Flz-Frz==0', ... [4.30]
Flz*(w/2)-Frz*(w/2)+Fy*h+m*g*(h-hRC)*px==0, ...
zl==z+(w/2)*px, zr==z-(w/2)*px, ...
Fzl1==-1/((Fzl/m-g/2)*w/(ay*h)), ...
zl, zr, Flz, Frz, Flz0, Frz0, Fy, z, px);
The results from the Matlab script in Equation [4.30]:
∙ 0
2 𝑚∙ 𝑦∙
𝑝 𝑥 ( 𝑖 )∙𝑤 2 2 𝑚∙ ∙(
+2∙ 𝑟 𝐶)
𝑦∙ 2 𝑚∙ 𝐶 [4.31]
𝑙 ∙ (2 ⁄( ∙ ))
𝑤 𝑖 +2 𝑟 𝑤2
𝑦∙ 2 𝑚∙ 𝐶
∙( + ⁄( ∙ ))
2 𝑤 𝑖 +2 𝑟 𝑤2

In agreement with intuition and experience the body rolls with positive roll when steering to the left
(positive 𝑤 ). Further, the body centre of gravity is unchanged in heave (vertical motion, ). The for-
mula uses ℎ which we cannot estimate without modelling the suspension. Since front and rear axle
normally are different, we could expect that ℎ is expressed in some similar quantities for each of
front and rear axle, which also is the case, see Equation [4.38].
4.3.10.2.1 Steady-State Roll-Gradient *
Function definition: Steady state roll-gradient is the body roll angle per lateral acceleration for the vehicle
during steady state cornering with a certain lateral acceleration and certain path radius on level ground.

262
 Lateral Dynamics

4.3.10.3 Lateral Load Transfer Models of Suspension Linkage

For longitudinal load transfer, during purely longitudinal dynamic manoeuvres, the symmetry of the
vehicle makes it reasonable to split vertical load on each axle equally between the left and right wheel
of the axle. However, for lateral dynamics it is not very realistic to assume symmetry between front
and rear axle. Hence, the suspension has to be considered separately for front and rear axle. The prop-
erties that are important to model for each axle is not only left and right elasticity (as we modelled the
whole vehicle in Figure 4-36). It is also how the lateral tyre forces are transmitted from road contact
patches to the vehicle body. We end up with conceptually the same two possible linkage modelling
concepts as we found for longitudinal load transfer, see Figure 3-31. Either we can introduce 1 roll
centre heights for each axle (c.f. pitch centre in 0) or we can introduce two pivot points for each axle (1
per wheel if individually suspended wheels or 2 per axle if rigid axle) (c.f. axle pivot points in 0). A dif-
ference for lateral load transfer, compared to longitudinal load transfer, is that it is significant also at
steady state (due to centrifugal force). The two modelling ways to include the suspension in the lateral
load transfer are shown in Figure 4-37. Generally speaking, they can be combined, so that one is used
on front axle and the other on rear axle. In this compendium, we will select the roll-centre model when
modelling.
One should differ between roll-centre heights and roll-centres. One can say that roll-centre heights ex-
ists while roll-centres is only a model concept. If a vehicle was actually designed with a roll-centre as
being a real pivot point between axle and body, that vehicle would be totally rigid in vertical direction.
4.3.10.3.1 Load Transfer Model with Two Pivot Points per Axle
This model will not be deeply presented in this compendium. However, it should be mentioned as hav-
ing quite a few advantages:
• The model has both heave and roll degree of freedom. (Roll centre model is restricted to roll
around roll centre.)
• If wheel independent suspension, the distribution of lateral wheel forces between left and right
side is considered. (Roll centre model only uses the sum of lateral forces per axle and needs
involvement of tyre model to resolve into individual left and right side forces.)
Generally spoken, this model is more accurate and not much more computational demanding and
probably easier to intuitively understand, since it does not constrain heave motion.
Note that non-individual, rigid axles or beam axles, the pivot-point model does not have one pivot
point for each wheel, but instead two pivot points for the whole axle: heave and roll. The roll-centre
height affects as the height where lateral force is transferred between axle and body.
Cases when this model is recommended as opposed to the model with roll centres are:
• Steady state and transient manoeuvres where the heave displacement is important.
• When large differences between lateral load on left and right wheels are present, such as:
o Large load transfer, i.e. high CoG and large lateral accelerations. One example is when
studying wheel lift and roll-over tendencies.
o Large differences between longitudinal slip, while axle skids sideways. Then one wheel
might have zero lateral force, due to that friction is used up longitudinally, while the other
can have a large lateral force.
o If individual steering within an axle would be studied. One could think of an extreme case
if actuating a sudden toe-in or toe-out, which would cause large but counter-directed lat-
eral forces on left and right wheel.

263
 Lateral Dynamics

2 Pivot Points per Axle 1 Roll Centre per Axle

Transversal sections from rear over front axle:
approximation
wheel individual rigid axle with vertically
suspension suspension rigid suspension

RC h h h

ef ef hRCf

gf
w/2 w/2 w/2 w/2
Transversal sections from rear over rear axle:
Similar, but with subscript “r” instead of “f”. Typically, ℎ >ℎ 𝑓

Figure 4-37: Two alternative models for including suspension linkage effects (kinematics) in
lateral load transfer. Anti-roll bars not drawn.

4.3.10.3.2 Load Transfer Model with Roll Centre (One Pivot Point) per Axle
The model with 1 roll centres has some drawback as listen before. To mention some advantages, it is
somewhat less computational demanding. However, the main reason why the compendium uses this
model is to cover two different concepts with longitudinal and lateral load transfer.
Study the free-body diagrams in Figure 4-38.

displacements, speeds, accelerations: forces, on body:

ay m*g
z
zfl=zrl=0 y zfr=zrr Only one anti-roll bar m*ay
drawn, but there is one
per axle: 𝑓 and .
Pry
Pfy

Fsfl+Fsrl Fsfr+Fsrr
zflr=zrlr=0
+

zfrr=zrrr=0
𝑓

forces, forces, on front axle: forces, on rear axle:

on vehicle: m*g Fsfl Fsfr Fsrl Fsrr
m*ay 𝑟
𝑓
Pry
h Faf Pfy Faf Far
Far

Ffly+Frly Ffly Ffry Frly Frry

Ffry+Frry
Fflz Ffrz Frlz Frrz
Fflz+Frlz Ffrz+Frrz

Figure 4-38: Model for steady state heave and roll due to lateral acceleration, using roll centres,
which can be different front and rear.

264
 Lateral Dynamics

The road is assumed to be flat, 𝑓𝑙 𝑓 𝑙 0. In free-body diagram for the front axle,
𝑓 and 𝑓 are the reaction force in the rear roll-centre. Corresponding reaction forces are found for
rear axle. Note that roll centres are free of roll moment, which is the key assumption about roll centres.
The 𝑓𝑙 𝑓 𝑙 and are the forces in the compliances, i.e. where potential spring energy is
stored. One can understand the roll-centres as also unable to take vertical force, as opposed to con-
straining vertical motion (as drawn). Which of vertically force-free or vertically motion-free depends
on how one understands the concept or roll-centre, and it does not influence the equations.
Note carefully that the “pendulum effect” is NOT included here, in 4.3.10.3, as it was in 4.3.10.2. The
motivation is to get simpler equations for educational reasons.
There is no damping included in model, because their forces would be zero, since there is no displace-
ment velocity, due to the steady-state assumption. As constitutive equations for the compliances
(springs) we assume that displacements are measured from a static condition and that the compli-
ances are linear. Note that there are two elasticity types modelled: springs per wheel ( 𝑓𝑤 per front
wheel and 𝑤 per rear wheel) and anti-roll bars per axle ( 𝑓 front and rear). The road is assumed
to be smooth, i.e. 𝑓𝑙 𝑓 𝑙 0. The stiffnesses 𝑓𝑤 𝑤 𝑓 and are effective stiff-
nesses per wheel. We see already in free-body diagram that 𝑓𝑙 and 𝑓 always act together, so we
rename 𝑓𝑙 + 𝑓 𝑓 and 𝑙 + .

𝑓𝑙 𝑓𝑙 + 𝑓𝑤 ∙ ( 𝑓𝑙 𝑓𝑙 )
𝑓 𝑓 + 𝑓𝑤 ∙ ( 𝑓 𝑓 )
𝑙 𝑙 + 𝑤∙( 𝑙 𝑙)
+ 𝑤 ∙( )
[4.32]
𝑓 0+ 𝑓 ∙ (( 𝑓𝑙 𝑓𝑙 ) ( 𝑓 𝑓 ))
0+ ∙ (( 𝑙 𝑙) ( ))
∙ ∙ ∙ ∙ 𝑓
ℎ𝑒𝑟𝑒 𝑓𝑙 𝑓 𝑛 𝑙
∙𝐿 ∙𝐿
Equilibrium for whole vehicle (vertical, lateral, yaw, pitch, roll) neglecting body forces (air resistance
and gravity components in road plane) gives:
𝑓𝑙 + 𝑓 + 𝑙 + ∙
∙ 𝑓 +
0 ∙ ∙
𝑓 𝑓 [4.33]
( 𝑓𝑙 + 𝑓 )∙ 𝑓 +( 𝑙 + )∙ 0
( 𝑓𝑙 + 𝑙 )∙ ( 𝑓 + )∙ +( 𝑓 + )∙ℎ 0

Equilibrium for each axle (roll, around roll centre). Note that the axles are considered massless:
( 𝑓𝑙 𝑓𝑙 + 𝑓) ∙ ( 𝑓 𝑓 𝑓) ∙ + 𝑓 ∙ℎ 𝑓 0
[4.34]
( 𝑙 𝑙 + )∙ ( )∙ + ∙ℎ 0

Compatibility, to introduce body displacements, z, and , gives:

𝑓𝑙 + ∙ 𝑓 ∙ 𝑛 𝑓 ∙ 𝑓 ∙

+ ∙ + ∙ 𝑛 ∙ + ∙ [4.35]
𝑙

𝑓𝑙 + 𝑓 0 𝑛 𝑙 + 0
The measure h is redundant and can be connected to the other geometry measures as follows. The
geometrical interpretation is given in Figure 4-39.

265
 Lateral Dynamics

∙ℎ 𝑓 + 𝑓 ∙ℎ
ℎ ℎ [4.36]
𝐿
Combining Equations [4.32] to [4.36] gives, as Matlab script and solution:
syms …; sol=solve( ...
Fsfl==Fsfl0-cfw*zfl, Fsfr==Fsfr0-cfw*zfr, ...
Fsrl==Fsrl0-crw*zrl, Fsrr==Fsrr0-crw*zrr, ...
Faf==0-caf*(-zfl+zfr), Far==0-car*(-zrl+zrr), ...
Fsfl0==(1/2)*m*g*lr/L, Fsfr0==(1/2)*m*g*lr/L, ...
Fsrl0==(1/2)*m*g*lf/L, Fsrr0==(1/2)*m*g*lf/L, ...
Fflz+Ffrz+Frlz+Frrz==m*g, ...
m*ay==Ffy+Fry, ...
0==Ffy*lf-Fry*lr, ...
-(Fflz+Ffrz)*lf+(Frlz+Frrz)*lr==0, ...
(Fflz+Frlz)*w/2-(Ffrz+Frrz)*w/2+(Ffy+Fry)*h==0, ... [4.37]
(Fflz-Fsfl+Faf)*w/2-(Ffrz-Fsfr-Faf)*w/2+Ffy*hRCf==0, ...
(Frlz-Fsrl+Far)*w/2-(Frrz-Fsrr-Far)*w/2+Fry*hRCr==0, ...
zfl==z+(w/2)*px-lf*py, zfr==z-(w/2)*px-lf*py, ...
zrl==z+(w/2)*px+lr*py, zrr==z-(w/2)*px+lr*py, ...
zfl+zfr==0, zrl+zrr==0, ...
dh==h-(lr*hRCf+lf*hRCr)/(lf+lr), ...
zfl, zfr, zrl, zrr, Fsfl, Fsfr, Fsrl, Fsrr, ...
Faf, Far, Fsfl0, Fsfr0, Fsrl0, Fsrr0, ...
Fflz, Ffrz, Frlz, Frrz, Ffy, Fry, z, px, py, h);
The result from the Matlab script in Equation [4.37], but in a prettier writing format:
𝑙𝑟 𝑙𝑓
𝑓 ∙ ∙ 𝐿
𝑛 ∙ 𝐿
∙
𝑚∙ 𝑦 ∙ ( 𝑓𝑦 + 𝑟𝑦 )∙
0 𝑛 𝑝 𝑛 𝑝 0
𝑣 𝑖 𝑟𝑜 𝑣 𝑖 𝑟𝑜

∙ ℎ 𝑓 ∙ ℎ 𝑓 𝑙𝑙
𝑓𝑙 ∙( ∙( + ∙ ))
∙𝐿 𝐿∙ 𝑣 𝑙 𝑙𝑙

∙ ℎ 𝑓 ∙ ℎ 𝑓 𝑙𝑙
𝑓 ∙( + ∙( + ∙ ))
∙𝐿 𝐿∙ 𝑣 𝑙 𝑙𝑙

∙ 𝑓 ℎ ∙ 𝑓 ℎ 𝑙𝑙
𝑙 ∙( ∙( + ∙ )) [4.38]
∙𝐿 𝐿∙ 𝑣 𝑙 𝑙𝑙
∙ 𝑓 ℎ ∙ 𝑓 ℎ 𝑙𝑙
∙( + ∙( + ∙ ))
∙𝐿 𝐿∙ 𝑣 𝑙 𝑙𝑙

where, roll stiffnesses are:

𝑤 2 𝑁𝑚
𝑓 𝑙𝑙 ∙( 𝑓𝑤 + ∙ 𝑓) ∙ ( ) 2
[ ]
𝑤 2 𝑁𝑚
𝑙𝑙 ∙( 𝑤 + ∙ )∙( )
2
[ ]
𝑁𝑚
𝑣 𝑙 𝑙𝑙 𝑓 𝑙𝑙 + 𝑙𝑙 [ ]

The axle roll stiffnesses, 𝑓 𝑙𝑙 and 𝑙𝑙 are identified beside vehicle roll stiffness 𝑣 𝑙 𝑙𝑙 . We
should compare Equation [4.38] with Equation [4.31]. Eq [4.31] considers the “(roll) pendulum effect”,
but not the differentiation between front and rear suspension. Eq [4.38] does the opposite.
Assume ℎ ℎ and look at the sum of vertical force on one side, 𝑙 in Eq [4.31]. Compare 𝑙 in Eq
[4.31] and 𝑓𝑙 + 𝑙 in Eq [4.38]; the equations agree if:
ℎ 𝑓 ∙ +ℎ ∙ 𝑓 ℎ ℎ
𝑓𝑙 + 𝑙 𝑙 ⇒ ∙( ∙( + )) ∙( ∙ ) 𝑙 ⇒
𝐿∙
ℎ 𝑓 ∙ +ℎ ∙ 𝑓 ℎ ℎ
⇒ + ⇒ℎ 𝑓 ∙ +ℎ ∙ 𝑓 (ℎ ℎ) ∙ 𝐿
𝐿∙

266
 Lateral Dynamics

This is exactly in agreement with the definition of the redundant geometric parameter h, see Eq
[4.36]. This means that a consistent geometric model of the whole model is as drawn in Figure 4-39.
Here the artefact roll axis is also defined.
The terms of type ℎ ∙ ⁄(𝐿 ) in Eq [4.38] can be seen as the part of the lateral tyre forces that
goes via the stiff linkage. The terms of type (Δℎ ⁄ ) ( 𝑤 ⁄( 𝑤 + 𝑤 )) in Eq [4.38] can be seen as the
part of the lateral tyre forces that goes via the compliance. The latter part is distributed in proportion
to roll stiffness of the studied axle, as a fraction of the vehicle roll stiffness. This should agree with in-
tuition and experience from other preloaded mechanical systems (load distributes as stiffness).
Body rolls with positive roll when steering to the left, as long as CoG is above roll axle. Further, the
body centre of gravity is unchanged in heave (vertical z) because the model does not allow any vertical
displacements, which is a drawback already mentioned.
front axle
CoG roll centre
front axle

h
hRCf

rear axle
roll centre

h
rear axle

ℎ 𝑓 + 𝑓 ℎ
hRCr

ℎ ℎ
𝐿

Figure 4-39: Roll axis for a two-axle vehicle. (Note that the picture may indicate that the roll
centres and roll axis are above wheel centre, but this is normally not the case.)
Eq [4.44] was derived as a steady state out-of-road-plane model, but only the ratio between roll stiff-
nesses influence the lateral load transfer. So, if the roll stiffnesses are large, they can be neglected (con-
sidered infinite), if the ratios are given. Then, Eq [4.44] works also for transient manoeuvres.
4.3.10.3.3 Steady State Longitudinal and Lateral Distribution
If the vehicle has a steady state acceleration with combined and , we can combine Eqs [4.38]and
[3.27] (with 𝑓𝑥 + 𝑟𝑥
( ) ) to Eq [4.39]. Note:
• Body forces (air resistance and gravity components in road plane) are neglected.
• We have also assumed symmetric longitudinal load transfer. This is reasonable if no roll pre-
tension. Pre-tension could appear if uneven ground or unusual suspension, e.g. pre-tensioned
(or “active”) anti-roll bar.
∙𝑙 𝐶𝑓 ∙𝑙𝑟 𝑓 𝑟𝑜
𝑓𝑙 ∙ ( 2∙𝐿𝑟 2𝐿
∙( 𝐿∙𝑤
+ 𝑤
∙ ))
𝑣 𝑖 𝑟𝑜

∙𝑙 𝐶𝑓 ∙𝑙𝑟 𝑓 𝑟𝑜
𝑓 ∙ ( 2∙𝐿𝑟 2𝐿
+ ∙( 𝐿∙𝑤
+ 𝑤
∙ ))
𝑣 𝑖 𝑟𝑜
[4.39]
∙𝑙𝑓 𝐶𝑟∙𝑙𝑓 𝑟 𝑟𝑜
𝑙 ∙ ( 2∙𝐿 + 2𝐿
∙( 𝐿∙𝑤
+ 𝑤
∙ ))
𝑣 𝑖 𝑟𝑜

∙𝑙𝑓 𝐶𝑟∙𝑙𝑓 𝑟 𝑟𝑜
∙ ( 2∙𝐿 + 2𝐿
+ ∙( 𝐿∙𝑤
+ 𝑤
∙ ))
𝑣 𝑖 𝑟𝑜

267
 Lateral Dynamics

4.3.11 High Speed Steady State Vehicle Functions

4.3.11.1 Steering Feel *
Function definition: Steering feel is the steering wheel torque response to steering wheel angle. The function is
used in a very wide sense; on a high level, it is a measure of steering wheel torque, or its variation, for certain driv-
ing situations. Often, it can only be subjectively assessed.
At steady state driving at high speed, there are basically three aspects of steering feel:
• Lateral steering feel feedback at cornering: The steering wheel torque is normally desired to
increase monotonously with lateral forces on the front axle. This is basically the way the me-
chanics work due to castor trail. Some specifications on steering assistance system is however
needed to keep the steering wheel torque low enough for comfort.
• Steering torque drop when cornering at low friction: It is built into the mechanics of the castor
trail and the pneumatic trail that steering wheel torque drops slightly when one approaches
the friction limit on front axle. This is normally a desired behaviour because it gives driver
feedback that the vehicle is approach the limits.
• On-centre feel in straight line driving: When the vehicle is driven in straight line, the steering
wheel is normally desired to return to centre position after small perturbations. This is a com-
fort function, which OEMs works a lot with, and it is often rather subjectively assessed.

4.3.11.2 High Speed Steady-State Off-tracking *

Function definition: High speed steady-state off-tracking is the lateral offset between the paths of the
centre of the front axle and the centre of the most severely off-tracking axle of any unit in a steady turn at a certain
friction level and a certain constant longitudinal speed. From Reference [ (Kati, 2013)].
The function is mainly of interest for long combination vehicles, as illustrated in Figure 4-40. Off-track-
ing was also mentioned in 4.2. It measures the lateral road space required. High speed Off-tracking,
which is an outboard off-tracking, can be either determined in a steady state turn or in a transient ma-
noeuvre such as lane change; the latter is termed as high speed transient off-tracking, see 4.5.6.2.

Figure 4-40: Illustration of high speed off-tracking. From (Kharrazi , 2012).

4.3.11.3 Tracking-Ability on Straight Path, TASP *

Function definition: Tracking-ability on straight path, TASP is the swept width between outer-most axle
centres when driving at a road with certain cross-fall and certain road friction at a certain speed.
The axles on any vehicle driving at a road with cross-fall will not track exactly in each other axle’s tra-
jectories. This is especially pronounced if long vehicles with many articulation points. The driving situ-
ation is straight steady state low or high speed.

268
 Lateral Dynamics

Tracking ability
road and 𝑟 𝑡𝑦
is this swept width
translation
direction

cross-fall
direction

2 2

2
𝑤
2 2
2 2

+ 3
Figure 4-41: Tracking ability on straight path, TASP, for an “A-double”. Longitudinal forces
neglected. Axles within an axle group lumped together, ⋯ 5. (Longitudinal dimensions as
in Figure 1-50.)
An example is seen in Figure 4-41. If we neglect combined tyre slip and assume same cornering coeffi-
cient, , on all axles we can derive these equations, one for each axle group:
• 𝑤 ( )
• 2 2
• + + 𝑓𝑜𝑟 𝑢𝑛 𝑡 . .4
Since the levers for moment equilibria in road 𝑥𝑦-plane and in road 𝑥 -plane are equal, the distribu-
tion of the axles’ vertical forces and lateral forces becomes identical. So, relation between lateral force
𝑔𝑗𝑦 𝑔1𝑦𝑣 𝑔1𝑦𝑤
and vertical force becomes tan( )≈ for all axle groups 𝑗 . .5 and ≈ ≈ :
𝑔𝑗𝑧 𝑔1𝑧 𝑔1𝑧

•
𝑔1𝑦𝑤
≈ ≈ ( )
𝑔1𝑧

•
𝑔2𝑦
≈ ≈
𝑔2𝑧

•
𝑔 𝑖+1 𝑦
≈ ≈ 𝑓𝑜𝑟 𝑢𝑛 𝑡 . .4
𝑔 𝑖+1 𝑧

Solving for side slip angles and steer angle:

𝜑𝑟
• ≈ 𝑓𝑜𝑟 𝑢𝑛 𝑡 . .4
𝜑𝑟 𝜑𝑟 𝜑𝑟
• ≈ + ≈ + ≈ 0 (the model is too approximative to give a good values of )
The swept width becomes: 𝑆 𝑒𝑝𝑡𝑊 𝑡ℎ 𝑇𝐴𝑆 ≈
≈ (0 ) +( 2 3) 2+( 3 4) 3 +( 4 5) 4 ≈
𝜑𝑟
≈ ((0 )+( 2 3) + ( 3 4) +( 4 5 ))

This can be summarized as:

𝑇𝐴𝑆 ≈ [ 𝑠𝑡 𝑛 𝑒 𝑒𝑡 𝑒𝑒𝑛 𝑓 𝑟𝑠𝑡 𝑛 𝑠𝑡 𝑥 𝑒 𝑟𝑜𝑢𝑝] [4.40]

The model is very approximative, since it does not take the following aspects into account:
• Axles within an axle group can take different vertical force.
• Axles can have different cornering coefficient .
• Large road grades. This can influence via longitudinal forces and both articulation angles and
combined tyre slip. The effect would be larger if combined with low road-friction.
More accurate value of TASP can be found by simulation to a steady-state pose of the vehicle.

269
 Lateral Dynamics

4.3.12 Roll-Over in Steady State Cornering

When going in curves, the vehicle will have roll angles of typically some degrees. At that level, the roll
is a comfort issue. However, there are manoeuvres which can cause the vehicle to roll-over, i.e. roll ≥
90 𝑒 so that vehicle body crashes into ground. Roll-over can be seen as a special event, but if sorting
into the chapters of this compendium it probably fits best in present chapter, about lateral dynamics.
One can categorize roll-overs in e.g. 3 different types:
• Tripped roll-over. This is when the car skids sideways and hits an edge, which causes the roll-
over. It can be an uprising edge, e.g. pavement or refuge. It can be the opposite, a ditch or loose
gravel outside road. In both these cases, it is strong lateral forces on the wheels on one side of
the vehicle that causes the roll-over. Tripped roll-over can also be when the vehicle is exposed to
large one-sided vertical wheel forces, e.g. by running over a one-sided bump. A third variant of
tripped roll-over is when the vehicle is hit by another vehicle so hard that it rolls over.
• Un-tripped roll-over or on-road roll-overs. These happen on the road and triggered by high
tyre lateral forces. This is why they require high road friction. For sedan passenger cars, these
events are almost impossible, since road friction seldom is higher than approximately 1. For
SUVs, un-tripped roll-overs can however occur but require dry asphalt roads, where friction is
around 1. For trucks, un-tripped roll-over, can happen already at very moderate friction, like 0.4,
due to their high CoG in relation to track width. Within un-tripped roll-overs, one can differ be-
tween:
o Steady state roll-over. If lateral acceleration is slowly increased, e.g. as running with into a
hairpin curve or a highway exit, the vehicle can slowly lift off the inner wheels and roll-over.
This is the only case of roll-over for which a model is given in this compendium.
o Transient roll-over. This is when complex manoeuvres, like double lane changes or sinus-
oidal steering, are made at high lateral accelerations. This can trigger roll eigen-modes,
which can be amplified due to unlucky timing between the turns. Models from 4.4.3.5 can be
used as a start, but it is required that load transfer is modelled carefully and includes wheel
lifts, suspension end-stops and bump stops.

4.3.12.1 Roll-Over Threshold Definitions

An overall requirement on a vehicle is that the vehicle should not roll-over for certain manoeuvres.
Heavy trucks will be possible to roll-over on high-mu conditions. The requirement for those is based
on some manoeuvres which not utilize the full road friction. For passenger cars, it is often the intended
design that they should be impossible to roll-over, even at high mu. Any requirement needs a defini-
tion of what exactly roll-over is, i.e. a Roll over threshold definition. Candidates for Roll over threshold
definition are:
• One wheel lifts from ground
• All wheels on one side lift from ground
• Vehicle CoG reaches its highest point (point of no-return towards roll ≥ 90 deg)
Note that:
• It is the 3rd threshold which really is the limit, but other can still be useful in requirement set-
ting. To use the 3rd for requirement setting makes the verification much more complex, of
course in real vehicles but also in simulation.
• The 1st is not a very serious situation for a conventional vehicle with 4 wheels. However, for a
3-wheeled vehicle, such as small “Tuctucs” or a 3-wheel moped, it is still a relevant threshold.
• The 2nd threshold is probably the most useful threshold for two-tracked vehicles, because it
defines a condition from which real roll-over is an obvious risk, and still it is relatively easy to
test and simulate. For 3-wheeled vehicle, 2nd and 3rd threshold generally coincide. The 2nd
threshold will be used in this compendium.
Figure 4-42 shows how the inner wheels lift off subsequently during a slowly increasing lateral force
(or lateral acceleration) build-up. Before any wheel is lifted, the load transfer is proportional to roll-
centre heights and roll stiffnesses, as shown in Equation [4.38]. But every time a wheel lifts, the

270
 Lateral Dynamics

distribution changes, so that a “knee” on the curves appears, see Figure 4-42. So, the relation of type as
Equation [4.38] is no longer valid. For instance, it is not physically motivated to keep the roll-centre
model for an axle which has lifted one side. So, the prediction of critical lateral acceleration for roll-
over is not trivial, especially for heavy vehicles which has many axles, and often also a fifth wheel
which can transfer roll-moment to a certain extent. There are approximate standards for how to calcu-
late steady state roll-over thresholds for such vehicle, e.g. UN ECE 111 (http://www.unece.org/filead-
min/DAM/trans/main/wp29/wp29regs/r111e.pdf).
model Vehicle_1Unit_NAxles Vertical force on
…
inner wheels [N]veh.F_alz[2]

first inner

second inner
wheel lift
equation veh.F_alz[1] veh.F_alz[3]

wheel lift
//Test case (Steady state cornering, incr. a_y, until left lifts):
der(p_x)=1; // p_x=0.01; // 5E4
when sum(Lift)>=na then
terminate("Both left wheels lifted");
end when; 4E4

last inner
wheel lift
//Whole unit (=vehicle in this case):
//Equilibria irp: //Lateral: 3E4
0= sum(F_ay) - m*a_y;
//Yaw:
0 = sum(F_ay.*(x_a-x_CoG*ones(na))); 2E4
//Equilibria oorp: //Vertical:
0 = sum(F_alz+F_arz) - m*g;
//Pitch:
0 = sum((F_alz+F_arz).*(x_a-x_CoG*ones(na))); 1E4 Lateral
//Roll, around mid ground:
0=m*a_y*h + sum(F_alz)*w/2 - sum(F_arz)*w/2; acceleration
0E0
//For each axle:
for i in 1:na loop
//Equilibria: //Roll, around pivot point: -1E4
0 = F_ay[i]*h_aRC[i] + (F_alz[i] - F_arz[i])*w/2 - M_ax[i]; 0.0 2.5 5.0
//Constitution: //Tyre:
F_ay[i]=-CC_y*(F_alz[i]+F_arz[i])*s_ay[i]; veh.a_y [m/s²]
s_ay[i]=v_ay[i]/v_x;
v_ay[i]=v_y+(x_a[i]-x_CoG)*w_z;
//Suspension: Lateral force
der(M_ax[i]) = c_a[i]*(der(p_ax[i]) - der(p_x));
if Lift[i] > 0.5 then on axles [N]
veh.F_ay[1] veh.F_ay[2] veh.F_ay[3]
der(F_alz[i]) = 0;
else der(p_ax[i]) = 0; end if;
when (pre(Lift[i]) < 0.5) and F_alz[i] < 0 then
Lift[i] = 1; reinit(p_ax[i], 0); end when; 4E4
end for; Lateral
//Suspension Ctrl. Generates eqs only if number of axles=na>2:
for i in 2:(na-1) loop
acceleration
der(F_alz[i])+der(F_arz[i])=der(F_alz[i+1])+der(F_arz[i+1]); 0E0
end for;
end Vehicle_1Unit_NAxles; 0.0 2.5 5.0
veh.a_y [m/s²]

Figure 4-42: Example of 3-axle vehicle steady state roll-over wheel lift diagram.
4.3.12.1.1 § Roll-Over Model
One way to assess the steady state roll-over lateral acceleration is to model a dynamic system with roll
angle as independent variable (instead of time) and then simulate (i.e. study over a lapse with increas-
ing roll angle). When all wheels on one side lifts from ground, we define it as steady state roll-over lat-
eral acceleration. The model is then much more intuitively physical than, e.g. UN ECE 111 formulas.
Figure 4-42 shows a Modelica model of a 1-unit N-axles vehicle. Note that, since 3 axles means non-
Ackermann steering geometry, it necessarily involves the lateral tyre slip stiffnesses also. The model
can, with maintained physical interpretation, be extended to more units with roll-free or roll-rigid con-
nections. Phenomena as (roll) pendulum effect and lateral deformation of tyres can also be added.
An alternative way of computing the critical lateral acceleration is to use the same model, but not sim-
ulate it. Instead one could solve the states where assume that all axles except one are lifted and then
see how much lateral acceleration is needed to lift the last axle. Trying each axle to be the last to lift
gives several lateral acceleration values, whereof the smallest is the critical lateral acceleration.

4.3.12.2 Static Stability Factor, SSF

One very simple measure of the vehicles tendency to roll-over is the Static Stability Factor, SSF. It is
proposed by NHTSA, http://www.nhtsa.gov/cars/rules/rulings/roll_resistance/, and it is simply de-
fined as:
𝐻 𝑓 𝑇𝑟 𝑘𝑊 𝑡ℎ ⁄
SS [4.41]
𝐻𝑒 𝑡ℎ 𝑓 𝑜 ℎ
A requirement which requires SSF>number cannot be directly interpreted in terms of certain manoeu-
vre and certain roll-over threshold. It is not a performance-based requirement, but a design-based (or
prescriptive) requirement. However, one of many possible performance-based interpretations is that

271
 Lateral Dynamics

the vehicle shall not roll-over for steady-state cornering on level ground with an enough friction coeffi-
cient. Another is that it should not roll-over in a tilt-table. Since the requirement is not truly perfor-
mance based, each interpretation will also stipulate a certain verification method; here it would be
theoretical verification using a rigid suspension model. Such model and threshold are shown in Figure
4-43.
The derivation of the SSF based requirement looks as follows:

∙ + ∙ ∙ℎ ∙ ∙
ℎ∙
𝑀𝑜 𝑒 : { + ∙ }⇒ ∙ ∙( )
⇒
∙ ∙( + )
𝑅𝑒𝑞𝑢 𝑟𝑒 𝑒𝑛𝑡: ≥0 } [4.42]

ℎ∙
⇒ 𝑅𝑒𝑞𝑢 𝑟𝑒 𝑒𝑛𝑡: > ⇒ 𝑆𝑆 >
∙ℎ
Maximum road friction, , is typically 1, which is why SSF>=1 would be a reasonable. However, typi-
cal values of SSF for passenger vehicles are between 0.95 and 1.5. For heavy trucks, it can be much
lower, maybe 0.3..0.5, much depending on how the load is placed. There are objections to use SSF as a
measure, because SSF ignores suspension compliance, handling characteristics, electronic stability
control, vehicle shape and structure.
view from rear, when turning left

ay
m*g

m*ay
h
Fy
Fiz≥0
Foz
(=roll-over w/2
threshold)
w
Figure 4-43: Model for verification of requirement based on Static Stability Factor, SSF.

4.3.12.3 Steady-State Cornering Roll-Over *

A function defined for requirement setting can be:
Function definition: Steady state cornering roll-over acceleration is the maximum lateral acceleration
the vehicle can take in steady state cornering without lifting all inner wheels. On level ground with enough road fric-
tion and certain weight and position of payload.
For a long combination-vehicle with several articulation points, one often need to drive a long distance
after a steer angle change before steady state values on articulation angles are reached. Hence, it can
be more relevant to formulate a corresponding roll-over function in terms of curvature to follow, total
yaw angle for turn and longitudinal speed. A common way is also the, somewhat artificial, tilt-table
test, which means that one measure steady-state roll-over with a (real or virtual) tilt-table where the
maximum road pitch angle before wheel lift on one side is the measure to set requirement on. An even
simpler way to handle steady-state roll-over is to set requirement on the SSF.
Consider a roll-stiff vehicle in steady state cornering. Assume lateral acceleration is subsequently in-
creased. If the vehicle is a two-axle vehicle, Eq [4.38] is valid until first axle lifts its inner wheel, since
for larger lateral accelerations, the constitutive equation Eq [4.32] is invalid for the inner wheel, since
the lifted inner wheel has zero force from ground. We can identify the terms of one of the inner wheels
(if > 0) equations in Eq Eq [4.38] as follows:

272
 Lateral Dynamics

∙ ℎ 𝑓 ∙ ℎ 𝑓 𝑙𝑙
𝑓𝑙 ∙( ∙( + ∙ ))
∙𝐿 𝐿∙ 𝑙𝑙 𝑣 𝑙
∙ ∙ ℎ 𝑓 ℎ 𝑓 𝑙𝑙
∙ ∙ ∙ ⏟∙ ∙ ∙
⏟ ∙𝐿 ⏟ 𝐿 ⏟ 𝑙𝑙 𝑣 𝑙
𝐿 𝑡𝑒𝑟 𝑜𝑟 𝑒 𝑛𝑉𝑒ℎ 𝑒
𝑉𝑒𝑟𝑡 𝑜𝑟 𝑒 𝑛𝐴𝑥 𝑒 𝐿 𝑡𝑒𝑟 𝑜𝑟 𝑒 𝑛𝐴𝑥 𝑒 𝑅𝑜 𝑆𝑡 𝑓𝑓𝑛𝑒𝑠𝑠 𝑟 𝑡 𝑜𝑛 𝑛𝐴𝑥 𝑒
ℎ ℎ 𝑙𝑙
𝑦( )∙ 𝑦( )∙ ∙
𝑙𝑙 𝑣 𝑙
If the vehicle has more axles, Eq [4.38] is generalized to Eq [4.43], which also is valid until first inner
wheel lifts from ground.
For axle of a roll-stiff vehicle:
ℎ ℎ 𝑙𝑙 [4.43]
𝑙 ( )∙ ( )∙ ∙
𝑙𝑙 𝑣 𝑙

For a vehicle with >2 axles, the parameter Δℎ can not be calculated from Eq [4.36], but can still be un-
derstood as the vertical distance between roll axis and the axles roll centre. It should be noted that the
pendulum effect is not included in Eq [4.43], and this is often a significant approximation if applied on
high CoG vehicles, like heavy trucks.
4.3.12.3.1 Model Assuming All Inner Wheels Lift at the Same Lateral Accelerat-
ion
An approximation of Steady state cornering roll-over acceleration (lateral acceleration when all
inner wheels lifted) can be found for vehicles where Eq [4.43] gives 𝑙 0 for all axles at same .
Then, summing the Eq [4.43] for all axles leads to the Eq [4.44] which is the same as the simple
SSF model in Figure 4-43 and 4.3.12.2 gives.

[4.44]
ℎ
In the following, we will elaborate with 4 additional effects, which marked in Figure 4-44.
• The tyre will take the vertical load on its outer edge in a roll-over situation. This suggests a
𝑦 𝑤+𝒘𝒕𝒚𝒓𝒆 𝑤+𝒘𝒕𝒚𝒓𝒆
change of performance and requirement to: < 2∙ and 2∙ > . This effect is accen-
tuated when low tyre profile and/or high inflation pressure. This effect decreases the risk for
roll-over.
• Due to suspension and tyre lateral deformation, the body will translate laterally outwards,
𝑦 𝑤 𝑫𝒆𝒇𝒚 𝑤 𝑫𝒆𝒇𝒚
relative to the tyre. This could motivate < 2∙
and 2∙
> . This effect increases the
risk for roll-over.
• Due to suspension linkage and compliances, the body will roll. Since the CoG height above roll
𝑦 𝑤 𝒉∙ 𝑤 𝒉∙
axis, ℎ, normally is positive, this could motivate < 2∙ 𝒙 and 2∙ 𝒙 > . This effect in-
creases the risk for roll-over. At heavy vehicle this “(roll) pendulum effect” is large.
• Due to suspension linkage and compliances, the body will also heave. This requires a suspen-
sion model with pivot points per wheel, as opposed to roll-centre per axle. The heave is nor-
𝑤 𝑦 𝑤
mally positive. This could motivate 2∙( + ) > and < 2∙( + ). The effect is sometimes called
“jacking” and it increases the risk for roll-over.
• Road leaning left/right (road banking) or driving with one side on a different level (e.g. out-
side road or on pavement) also influence the roll-over performance.

273
 Lateral Dynamics

view from rear, when turning left

𝑥
ay m*g

ℎ m*ay

h
Fy
𝐷𝑒𝑓𝑦
Fiz=0
Foz
(=roll-over w/2
threshold) wtyre/2
w
Figure 4-44: Steady-state roll-over model, with fore/aft symmetry. The measures 𝐷𝑒𝑓 ℎ∙
𝑛 mark effects additional to what is covered with a simple SSF approach.
4.3.12.3.2 Model with Sequential Lifts of Inner Wheels
A model which does not assume wheel lift at same lateral acceleration will be sketched. For each axle
that has lifted, the equations have to be changed. Instead of simply the constitutive equation
( 𝑙 ) ⁄ 𝑙𝑙 one need to assure 𝑙 0 . The axle will then position itself so that it
keeps 𝑙 + . That means both roll and vertical translation of the axle centre, why also
the vertical suspension compliance needs to be modelled. A new position variable has to be declared,
e.g. the lift distance of the inner wheel, 𝑓𝑡 𝑙 . This variable is constrained to 𝑓𝑡 𝑙 0 before lift, but
after lift the constraint is 𝑙 0 . So, the model is suitably implemented as a state event model with
the event “ ℎ𝑒𝑛 𝑙 𝑒 𝑜 𝑒𝑠 < 0”. If is swept from zero and upwards, the result will be something
like shown in Figure 4-42.
4.3.12.3.3 Using a Transient Model for Steady-State Roll-Over
Another work-around to avoid complex algebra is to run a fully transient model, including suspension,
and run it until a steady state cornering conditions occur. If then, the lateral acceleration is slowly in-
creased, one can identify when or if the roll-over threshold is reached. Lateral acceleration increase
can be through either increase of longitudinal speed or steer angle. It should be noted that the model
should reasonably be able to manage at least lift of one wheel from the ground. This way of verifying
steady state cornering roll-over requirements has the advantage that, if using tyre models with friction
saturation, the limitation discussed in 4.3.12.3.2 does not have to be checked separately.

4.3.12.4 Roll-Over and Understeering/Propulsion

With the above formulas for roll-over there will always be a certain lateral acceleration that leads to
roll-over, because neither limitation due to road friction nor propulsion power modelled yet. Since ve-
hicles generally are understeered, they are limited to develop lateral acceleration, see Figure 4-30. For
propulsion-weak vehicles, there is also the limitation of lateral acceleration due to limited longitudinal
speed, which in turn is due to driving resistance from the steered wheels (=wheel lateral force *
sin(steer angle)) and loosing propulsion power due to longitudinal wheel slip. However, one should
consider that the propulsion limitation is less in down-hill driving, which increases the roll-over risk
again. Also, if the vehicle goes relatively quickly into steady state cornering, the longitudinal speed will
not have time to decelerate to its real (longitudinal) steady state value.
For heavy trucks, the critical lateral acceleration is typically (0. . .0.4) , which is quite possible to
reach during normal road conditions, because road friction is around 1. For passenger cars, the critical
lateral acceleration is typically in the region of 1, so it is often not possible to reach the roll-over-criti-
cal lateral acceleration. This is also the case for heavy trucks on low road friction.

274
 Lateral Dynamics

4.4 Stationary Oscillating Steering

In between steady state and transient manoeuvres, one can identify stationary oscillations as an inter-
mediate step. Generally, a mechanical system can be excited with a stationary oscillating excitation.
The response of the system is, after possible transients are damped out, a stationary oscillation. How
this response vary with excitation frequency will be called Frequency Response. If staying within the
linear region for the system and the excitation is harmonic (sinus and cosine), the ratio between the
response amplitude and the excitation amplitude is only dependent of the frequency. The ratio is
called transfer function.
For lateral vehicle dynamics, the excitation is typically steering wheel angle and the response is ampli-
tudes of yaw velocity, curvature or lateral acceleration. The corresponding transfer functions are a fre-
quency version of the gains defined in Equation [4.20]. Also, there will be a delay between excitation
and response. This is another important measure, beside the amplitude ratio.

4.4.1 Stationary Oscillating Steering Tests

When testing stationary oscillating steering functions, one usually drives on a longer part of the test
track. It might be a high-speed track, see Figure 1-53, because the track rather needs to be long than
wide, since one is often not too close to lateral grip limits. If the available Vehicle Dynamics Area, see
Figure 1-53, is long enough this is of course a safer option. A Vehicle Dynamics Area is a flat surface
with typically 100..300 m diameter. It normally has entrance roads for accelerating up to a certain
speed.
Typical tests for Stationary oscillating steering functions:
• Sweeping frequency (chirp) and/or amplitude
• Random frequency and amplitude
There are ISO standards for sweeping and random tests. The Frequency Response will be very depend-
ent of the vehicle longitudinal speed, why the same tests are typically done at different such speeds.

4.4.2 One-Track Model for Two-Axle Vehicle for Tran-

sient Dynamics
Stationary oscillating steering can often be of interest also with so small amplitudes that a linear
model is accurate enough. Such linear model will be derived in 4.4.2. A transient one-track model will
first be derived and then approximated to become linear. The resulting linear model is found Eq [4.50]
and a less general derivation of it is found in Figure 4-48. Note that road is assumed to be level, i.e.
down-grade (or pitch grade) of road 0. Therefore, neither or yaw angle appears in
model.
The vehicle model is sketched in Figure 4-46. The model is similar to the model for steady state cor-
nering in Figure 4-17, with the following changes:
• Longitudinal and lateral accelerations have both components of centripetal acceleration (
and ) and the derivatives and :
∙
+ ∙ [4.45]
(Often, ≪ , why ∙ is more neglectable than ∙ .)
• The yaw acceleration, , is no longer zero.
• The speed is no longer defined as a parameter, but a variable. Then, one more prescription
is needed to be a consistent model. For this purpose, an equation that sets front axle propul-
sion torque to 1000 Nm is added.

275
 Lateral Dynamics

velocities: forces:
𝑓 𝑣 𝑓 𝑓 𝑣 𝑓

𝑓 𝑣

𝑓 𝑣
𝑓 𝑓

𝑓 𝑓
𝑓
𝐿

𝒗
sin

Figure 4-46: One-track model for transient dynamics. Dashed show fictive forces. Compare to
Figure 4-17. Note that road is assumed to be level in 4.4, i.e. 0.
It is assumed that the longitudinal tyre forces and slips are small, so 𝑤 (𝑠 ) ≈ 𝑤 (𝑠 𝑤 ). It is also
assumed that lateral tyre forces are far below road friction saturation, 𝑤 (𝑠 𝑤 ) ≈ 𝑠 𝑤.
It can be difficult to understand the difference between acceleration [ ] and derivatives [ ] of
velocities [v v ]. Some explanations are proposed in 4.4.2.3.
The Mathematical model shown in Eq [4.46] in Modelica format. The subscript and refers to vehi-
cle coordinate system and wheel coordinate system, respectively. The model is developed from the
model for steady state in Eq [4.6], with the changes marked with yellow and underlined text.
//(Dynamic) Equilibrium:
m*ax = Ffxv + Frx; m*ay = Ffyv + Fry; J*der(wz) = Ffyv*lf - Fry*lr;
ax=der(vx)-wz*vy; ay=der(vy)+wz*vx;

//Constitutive relation (Lateral tyre force model):

Ffyw=-Cf*sfyw; Fry=-Cr*sryw;
sfyw=vfyw/vfxw; sryw=vry/vrx;

//Compatibility:
vfxv = vx; vfyv = vy + lf*wz;
vrx = vx; vry = vy - lr*wz;

//Transformation between vehicle and wheel coordinate systems:

Ffxv = Ffxw*cos(df) - Ffyw*sin(df);
Ffyv = Ffxw*sin(df) + Ffyw*cos(df); [4.46]
vfxv = vfxw*cos(df) - vfyw*sin(df);
vfyv = vfxw*sin(df) + vfyw*cos(df); //can be exactly replaced with
// atan2(vfyv,vfxv)=df+atan(sfyw); or approximately with vfyv/vfxv=df+sfyw;

//Path with orientation:

der(x) = vx*cos(pz) - vy*sin(pz);
der(y) = vy*cos(pz) + vx*sin(pz);
der(pz) = wz;

// Prescription of steer angle:

df = if time < 2.5 then (5*pi/180)*sin(0.5*2*pi*time) else 5*pi/180;
//Shaft torques:
Ffxw = +1000; // Front axle driven.
Frx = -100; // Rolling resistance on rear axle.
Position variables [𝑥 𝑦 p ] and speed variables [ w ] are selected state variables. The
input variables are 𝑓 𝑓 𝑤 . The only non-zero initial value is 00 𝑘 /ℎ. Simulation result is

276
 Lateral Dynamics

shown in Figure 4-47. The manoeuvre selected is same steering wheel function of time as in Figure
4-18, for better comparison of the different characteristics of the models.
exact.df_deg
10
0
𝑓 [deg] exact.y
-10 100
0 10 20 30

y [m]
exact.vx exact.w z_dps 90
smalAngleApprox.vx small_vy_w z_Approx.vx smalAngleApprox.w z_dps small_vy_w z_Approx.w z_dps small_vy_
40
80

30 70

60
20
50

10 40

30
0
20

-10 10

0
-20
x[m]
-10
time [s] 0 50 100

-30
exact.x
0 10 20 30

Figure 4-47: Simulation results of transient one-track models. Solid: Eq [4.46] or [4.47]. Dashed:
Eq [4.48], which employs the small angle approximation. Dotted: Eq[4.46]=[4.47] but without the
term .
Eq [4.46] is a complete model suitable for simulation. Removing [𝑥 𝑦 p ] and the 3 equations for
“path with orientation”, eliminating some variable and rewrite (no added approximations) gives:
Equilibrium: ∙( ∙ ) 𝑓 𝑤 ∙ cos( 𝑓 ) 𝑓 𝑤 ∙ sin( 𝑓 ) +
∙( + ∙ ) 𝑓 𝑤 ∙ sin( 𝑓 ) + 𝑓 𝑤 ∙ cos( 𝑓) +
∙ ( 𝑓 𝑤 ∙ sin( 𝑓 ) + 𝑓 𝑤 ∙ cos( 𝑓 )) ∙ 𝑓 ∙
Constitution: 𝑓 𝑤 𝑓 ∙ 𝑠𝑓 𝑤 ∙𝑠 𝑤
[4.47]
Compatibility: 𝑠𝑓 𝑤 𝑓 𝑤 | 𝑓 𝑤 | ∎ and 𝑠 𝑤
⁄ ( ∙ )⁄| |
Transformation from vehicle to wheel coordinate system on front axle:
𝑓 𝑤 ( + 𝑓 ∙ ) ∙ sin( 𝑓 ) + ∙ cos( 𝑓 ) ∎
𝑓 𝑤 ( + 𝑓 ∙ ) ∙ cos( 𝑓 ) ∙ sin( 𝑓 ) ∎
Typically, this model is used for simulation, where 𝑓 , 𝑓 𝑤 and are input variables. Suitable state
variables are , and . It is a non-linear model suitable for arbitrary transient manoeuvres and we
will come back to this in 4.4.3.5.

4.4.2.1 Known Longitudinal Velocity and Small Angles

Model in Eq [4.47] is not linear, but we will derive a linear model from it now, through some approxi-
mations. The angle sum approach from 1.6.1.5 is used to replace the 3 equations (∎) with 2 new and
replace the 2 unknowns 𝑓 𝑤 𝑓 𝑤 with one new 𝑓 .
Equilibrium: ∙( ∙ ) 𝑓 𝑤 ∙ cos( 𝑓 ) 𝑓 𝑤 ∙ sin( 𝑓 ) +
∙( + ∙ ) 𝑓 𝑤 ∙ sin( 𝑓 ) + 𝑓 𝑤 ∙ cos( 𝑓 ) +
∙ ( 𝑓 𝑤 ∙ sin( 𝑓 ) + 𝑓 𝑤 ∙ cos( 𝑓 )) ∙ 𝑓 ∙
Constitution: 𝑓 𝑤 𝑓 ∙ 𝑠𝑓 𝑤 ∙ 𝑠 𝑤

277
 Lateral Dynamics

𝑣𝑦 +𝑙𝑓 𝜔𝑧 𝑣𝑦 𝑙𝑟 ∙𝜔𝑧
Compatibility: 𝑓 𝑓 + arctan(𝑠𝑓 𝑤) 𝑓 |𝑣 |
𝑠 𝑤 |𝑣 |
Now, we assume small angles: steering angle 𝑓 , wheel side slip angles ≈ arctan( 𝑤 ⁄ 𝑤 )
arctan(𝑠 𝑤 ), and body side slip angle over steered axle 𝑓 ≈ arctan( 𝑓 ⁄ ). These are “trigonometric
approximations” during the Mathematical modelling stage, motivated if not too sharp turning.
Equilibrium: ∙( ∙ ) 𝑓 𝑤 + 𝑓 𝑤 ∙ 𝑓
∙( + ∙ ) 𝑓 𝑤 ∙ 𝑓 + 𝑓 𝑤 +
∙ ( 𝑓 𝑤 ∙ 𝑓 + 𝑓 𝑤) ∙ 𝑓 ∙ [4.48]
Constitution: 𝑓 𝑤 𝑓 ∙ 𝑠𝑓 𝑤 ∙ 𝑠 𝑤
𝑣𝑦 +𝑙𝑓 𝜔𝑧 𝑣𝑦 𝑙𝑟 ∙𝜔𝑧
Compatibility: 𝑓 ≈ 𝑓 + 𝑠𝑓 𝑤 𝑓 ≈ |𝑣 |
𝑠 𝑤 |𝑣 |

(Note that 𝑤 ∙ can be employed as approximation already during tyre modelling. This
would not change the resulting Eq [4.48].)
Figure 4-47 shows a simulation with model in Eq [4.48], for comparison with the model without small
angle approximations, i.e. from Eq [4.47].
Eliminating 𝑓 𝑤 𝑠𝑓 𝑤 𝑠 𝑓 gives:
𝑣𝑦+𝑙𝑓 𝜔𝑧
∙( ∙ ) 𝑓 𝑤 + 𝑓 ∙( |𝑣 | 𝑓) ∙ 𝑓 +
𝑣𝑦 +𝑙𝑓 𝜔𝑧 𝑣𝑦 𝑙𝑟∙𝜔𝑧
∙( + ∙ ) 𝑓 𝑤 ∙ 𝑓 𝑓 ∙ ( |𝑣 | 𝑓) ∙ |𝑣 |
𝑣 +𝑙 𝜔 𝑣𝑦 𝑙𝑟 ∙𝜔𝑧
∙ ( 𝑓 𝑤 ∙ 𝑓 𝑓 ∙ ( 𝑦 |𝑣𝑓| 𝑧 𝑓 )) ∙ 𝑓 + ∙ |𝑣 | ∙
Yet another approximation which we can do is to assume that centripetal acceleration is directed
purely lateral in vehicle and hence remove the term ∙ . Figure 4-47 shows a simulation of Eq
[4.47] without the term ∙ for judging the importance of it; it has considerable influence on
over time. Also, the influence of the force 𝑓 𝑤 ∙ 𝑓 in longitudinal direction can be neglected. Eq [4.49]
shows the resulting model.
∙ 𝑓 𝑤 +
+ 𝑓 ∙
∙( + ∙ ) 𝑓 𝑤 ∙ 𝑓 𝑓 ∙( 𝑓) ∙
| | | | [4.49]
+ 𝑓 ∙
∙ ( 𝑓 𝑤∙ 𝑓 𝑓 ∙( 𝑓 )) ∙ 𝑓 + ∙ ∙
| | | |
For stationary oscillation steering, 0, so can be seen as a manoeuvre-parameter. This is often a
good approximation also when turning during mild acceleration and deceleration. But, instead of set-
ting 0, we keep it more generic: is considered as a known function of time (𝑡), i.e. and
are known functions of time, or is an input and a state. So, we can continue to select as
states, but inputs become 𝑓 𝑓 𝑤 :
The first scalar equation (the longitudinal equilibrium) in Eq [4.49] can be used to calculate what is
needed to keep the prescribed (𝑡): ∙ (𝑡) 𝑓 𝑤 . Note that neither aerodynamic, grade nor
rolling resistance are considered.
The two latter scalar equations can be written as a linear model on matrix form, with 2 states
and 2 inputs 𝑓 𝑓 𝑤 𝑓 , see Eq [4.50], where the output is chosen as 𝒚𝑻 [ ]
[ + ]. The (𝑡) is often a scalar constant .

278
 Lateral Dynamics

𝑓
[ ] 𝑨∙[ ]+𝑩∙[ ]
𝑓𝑥 𝑓 𝒙 𝑨∙𝒙+𝑩∙
or {
𝑓 𝒚 𝑪∙𝒙+𝑫∙
[ ] 𝑪∙[ ]+𝑫∙[ ]
{ 𝑓𝑥 𝑓
𝑓+ 𝑟 𝑓 ∙ 𝑙𝑓 𝑟 ∙ 𝑙𝑟
+ ∙ (𝑡)
0 |𝑣 ( )| |𝑣 ( )|
[4.50]
Where 𝑨 [ ] ∙[ 2
+ 𝑟 ∙ 𝑙𝑟 2
]
0 𝑓 ∙𝑙𝑓 𝑟 ∙𝑙𝑟 𝑓 ∙ 𝑙𝑓
|𝑣 ( )| |𝑣 ( )|

0 𝑓
and 𝑩 [ ] ∙[ ]
0 𝑓 ∙ 𝑓 𝑓
0 0 0 0 0
and 𝑪 [ ]∙𝑨+[ (𝑡)]; and 𝑫 [ ] 𝑩;
0 0 0
The model above is approximated only for driving situation where tyre forces are far from saturation
and angles are small. If studying driving conditions which do not fulfil this, we can still approximate
with linearized models if variation from the driving condition is small. The method to linearize is to
use only the first terms in a Taylor series expansion of the model 𝒙 𝒇(𝒙 ) around [𝒙 ] [𝒙𝟎 𝟎 ]:
𝒅𝒇 𝒅𝒇
𝒙 𝒇(𝒙 ) ≈ 𝒇(𝒙𝟎 𝟎 ) + | (𝒙 𝒙𝟎 ) + | ( 𝟎)
𝒅𝒙
⏟ 𝒙𝟎 𝟎 𝒅
⏟ 𝒙𝟎 𝟎
𝑨𝒍𝒊𝒏 𝑩𝒍𝒊𝒏
𝒅𝒇 𝒅𝒇 𝒅𝒇 𝒅𝒇
(𝒇(𝒙𝟎 𝟎) | 𝒙𝟎 | 𝟎) + | 𝒙+ |
⏟ 𝒅𝒙 𝒙𝟎 𝟎
𝒅 𝒙𝟎 𝟎
𝒅𝒙
⏟ 𝒙𝟎 𝟎
𝒅
⏟ 𝒙𝟎 𝟎
𝒇𝟎 𝑨𝒍𝒊𝒏 𝑩𝒍𝒊𝒏
This can be used when studying small input offsets at a steady state cornering ( 𝑓 𝑤 𝑓 𝑓 𝑤 𝑓
and 𝑓 𝑓 ). It can also be used for finding eigen solutions.

4.4.2.1.1 § Convert to 2nd Order Single State Model

We can eliminate one of the 2 states ( ) in Eq [4.50], but then we have to accept that the ODE be-
comes 2 order. We can start from the last 2 scalar equations in Eq [4.49]. We here simplify by limit-
nd

ing to 0 and 𝑓 𝑤 0. To eliminate (𝑡), we first eliminate in the eq “ ∙ ⋯” using the eq

“ ∙ ( + ∙ ) ⋯”:
( 𝑓 𝑓) | | +( 𝑓 + ) | | +
2
+( 𝑓 (𝑓+ ) ( 𝑓 𝑓 ) | |) 𝑓 (𝑓+ ) | | 𝑓 ∎
To eliminate (𝑡) completely, we also need to eliminate . For this reason, we need to differentiate
the eq “ ∙ ⋯”, which gives:
+ 𝑓
∙ ̈ 𝑓 ∙( 𝑓) ∙ 𝑓 + ∙ ∙
| | | |
With equation “ ∙ ̈ ⋯” and ∎, we can now eliminate :
2
∙ ∙ ∙ ̈ +
2 2)
+ (( 𝑓 + ) ∙ + ( 𝑓 ∙ 𝑓 + ∙ ∙ )∙| |∙ +
2
+( 𝑓 ∙ ∙(𝑓+ ) +( ∙ 𝑓 ∙ 𝑓) ∙ ∙ ∙ | |) ∙
2
𝑓 ∙ 𝑓 ∙ ∙ ∙ 𝑓 (𝑓+ )∙ 𝑓 ∙ ∙| |∙ 𝑓

The coefficients in front of ̈ 𝑓 and 𝑓 can be compared with coefficients in simpler damped
mass and spring systems, see 1.5.2.2.1. Note, for instance, that the coefficients are dependent of and
that 𝑓 and cannot be simply identified as either springs or dampers.
Fourier transform gives:
∙ ∙ 2 2 ℱ( ) +

279
 Lateral Dynamics

2 2)
+ (( 𝑓 + )∙ +( 𝑓 ∙ 𝑓 + ∙ ∙ )∙| |∙𝑗 ℱ( )+
2
+( 𝑓 ∙ ∙(𝑓+ ) +( ∙ 𝑓 ∙ 𝑓) ∙ ∙ ∙ | |) ∙ ℱ( )
2
𝑓 ∙ 𝑓 ∙ ∙ ∙𝑗 ℱ( 𝑓 ) (𝑓+ )∙ 𝑓 ∙ ∙ | | ∙ ℱ( 𝑓 ) ⇒
ℱ( )
⇒ 𝐻𝛿𝑓 →𝜔𝑧 ( )
ℱ( 𝑓 )
2∙
𝑓 ∙𝑙𝑓 ∙𝑚∙𝑣 𝜔+(𝑙𝑓 +𝑙𝑟 )∙ 𝑓 ∙ 𝑟 ∙|𝑣 |
2 2 2 2
𝐽∙𝑚∙𝑣 𝜔2 (( 𝑓 + 𝑟 )∙𝐽+( 𝑓 ∙𝑙𝑓 + 𝑟 ∙𝑙𝑟 )∙𝑚)∙|𝑣 |∙ 𝜔 ( 𝑓 ∙ 𝑟 ∙(𝑙𝑓 +𝑙𝑟 ) +( 𝑟 ∙𝑙𝑟 𝑓 ∙𝑙𝑓 )∙𝑚∙𝑣 ∙|𝑣 |)

This gives the same transfer function as in 4.4.3.3.1, which is derived with matrix formula for the “1 st
order, 2 state model”. A single state, 2nd order model for can be derived similarly.

4.4.2.2 Less General Derivation of Linear One-Track Model

A less general and more intuitive derivation of Eq [4.50] is made in Figure 4-48. The approximations
are introduced earlier, already in the physical model. Therefore, the influence of 𝑓 𝑤 is not reflected
and the longitudinal equilibrium is neglected. It is also assumed that > 0.
Physical model: Mathematical model:
• Path radius >> the vehicle. Then, all forces (and centripetal Equilibrium:
acceleration) are approximately co-directed. + ≈ 𝑓 +
• Small tyre and vehicle side slip ⇒ angle=sin(angle)=tan(angle). ≈ 𝑓 𝑓
(Angles are not drawn small, which is why the forces not appear
Constitution:
parallel in figure.)
𝑓 𝑓 𝑠𝑓 𝑠

Compatibility:
bf 𝑓 + 𝑓
𝑓 𝑓 + 𝑠𝑓 ≈ 𝑓 + 𝑓 𝑓 ≈
af
b df 𝑠 ≈ ≈
br
Eliminate 𝑓 𝑓 𝑓 yields:
𝑓+ 𝑓 𝑓
𝑓 + + + ≈ 𝑓 𝑓

𝐿 𝑓 𝑓 𝑓
2
𝑓 +
2
+ + ≈ 𝑓 𝑓 𝑓

Figure 4-48: Less general derivation of the Linear One-Track Model, i.e. Eq [4.50].
One can add more axles to the derivation. E.g., a truck with double rear axle will get an added term
proportional to in the moment equilibrium.

4.4.2.3 Accelerations and Velocity Derivatives

Contribution from Toheed Ghandriz

See Equation [4.45]. This section provides some explanations to the difference between acceleration
[ ] and derivatives [ ] of velocities [ ].
4.4.2.3.1 Theoretical explanation
First, think of and as geometric vectors (acceleration and velocity are vectors in physics). Accelera-
tion is acceleration in inertial frame, i.e. accelerations over ground. Velocity is velocity in inertial
frame, i.e. velocity over ground. A driver or accelerometers, attached to the vehicle body, will experi-
ence [ ]. The ground, observed from the vehicle through a hole in the floor, would be experienced
as moving with [ ].
It is suitable to decompose and in vehicle directions, since most equations (tyres, steering, propul-
sion, braking, etc) are expressed in those directions. With 𝑢
⃗ and 𝑢
⃗ as unit vectors in vehicle

280
 Lateral Dynamics

directions: 𝑢
⃗ + ⃗ and
𝑢 𝑢
⃗ + 𝑢⃗ , or shorter, [ ] and [ ], which are
mathematical vectors. Now, it is important to remember that the vehicle is rotating with .
The accelerations are used in equilibrium equations, where the fictive force is . Since
we express in components in vehicle coordinate system directions, the differentiation is not as sim-
ple as differentiation component by component, [ ]. Instead: [ ] [ ]+
[ ∙ ∙ ] . The terms proportional to are centrifugal accelerations.
This can be mathematically shown as differentiation of the geometrical vector :
( 𝑢
⃗ + ⃗ ) (
𝑢 𝑢
⃗ + ⃗ )+(
𝑢 𝑢
⃗ + ⃗ )
𝑢
( 𝑢
⃗ + ⃗ )+(
𝑢 𝑢
⃗ ⃗ )
𝑢 (⏟ ) 𝑢
⃗ + (⏟ + ) 𝑢
⃗
𝑦

The constitutive and compatibility equations are typically expressed in velocities, not accelerations.
For high index models, some of the compatibility equations (a.k.a. constraint equations) also needs to
be differentiated, see example in 4.5.2.2.
4.4.2.3.2 Practical Explanation
Another explanation is given in left part of Figure 4-49 and the following reasoning. We can express
the velocity in direction of the x axis at time t, at two time instants:
• Velocity at time 𝑡:
• Velocity at time 𝑡 + 𝑡: ( + ) ∙ cos( ) ( + ) ∙ sin( ) ( ∙ cos( ) +
∙ cos( )) ( ∙ sin( ) + ∙ sin( )) ≈ { 𝑠 }≈( + )
( ∙ + ∙ )≈{ ∙ 𝑠 }≈( + ) ∙
Using these two expressions, we can express as the change of that speed per time unit:
{(𝑣 + 𝑣 ) 𝑣𝑦 ∙ 𝜓} 𝑣 𝑣 𝑣𝑦 ∙ 𝜓 𝑣 𝜓
• Acceleration=Velocity change per time ≈
∙ ∙
Corresponding for the lateral direction gives the Equation [4.45]. In Equation [4.45], the term ∙ is
generally more important than the term ∙ . This is because is generally much larger than .
Comparison with mass on
𝑦 𝑙 𝑙
𝑡 𝑒 𝑡 + Δ𝑡: +Δ circular path, with centrifugal
𝑥 acceleration
𝑦 Δ
Δ Δ
𝑡 𝑒 𝑡: 𝒂𝒄𝒆𝒏𝒕𝒓𝒊𝒑𝒆𝒕𝒂𝒍

𝑦
Δ
𝑥

2
𝑅 2
𝑛 𝑝 𝑙
𝑥 𝑙 𝑙 𝑅
𝑅
Figure 4-49: How to understand the acceleration term for vehicle motion in ground plane.
Left: Two consecutive time instants. Right: Comparison with circular motion, to identify the
centripetal acceleration.
4.4.2.3.3 Model with Velocity Components in Ground Fixed Directions
An alternative mathematical model is to express the velocity in components in ground fix (inertial) di-
rections ( 𝑢
⃗ + ⃗ , subscript for ground) instead of vehicle fix directions (
𝑢 𝑣
𝑢
⃗ 𝑣 + 𝑣 ⃗ 𝑣 , subscript for vehicle). The fictive forces are till
𝑢 , but we should express in
[ ] instead of [ 𝑣 𝑣 ], which becomes: ( 𝑢
⃗ + 𝑢
⃗ ) 𝑢
⃗ + 𝑢
⃗ .

281
 Lateral Dynamics

This is a simpler differentiation than using the rotating unit vectors 𝑢

⃗ 𝑣 and 𝑢
⃗ 𝑣 . We get
and instead of 𝑣 𝑣 + 𝑣 and 𝑣 𝑣 𝑣 .

velocities: accelerations:

𝑣
𝑣 𝑣 𝑣 ≠ 𝑣
𝑣 𝑣 + 𝑣 ≠ 𝑣

Figure 4-50: Velocity and Acceleration Components in Vehicle fix and Ground Fixed Directions.
Then, we replace the 2 translation equilibrium equations in Eq [4.47]. We also add 2 transformation
equations and a compatibility in yaw. The remaining equations can be kept, but it is proposed to add
subscript to and . The two models, side-by-side becomes as in Figure 4-51.
Model in… …vehicle fix directions, Eq [4.47] … ground fixed directions

Translation ∙( 𝑣 ∙ 𝑣) 𝑓 𝑣 + 𝑣 ( 𝑓 𝑣 + 𝑣) cos( ) ( 𝑓 𝑣 + 𝑣) sin( )

Equilibrium ∙( 𝑣 + ∙ 𝑣) 𝑓 𝑣 + 𝑣 ( 𝑓 𝑣 + 𝑣) sin( )+( 𝑓 𝑣 + 𝑣) cos( )
Velocity + cos( )+ sin( )
𝑣
Transformation, <not needed>
ground to vehicle 𝑣 sin( )+ cos( )

Compatibility, yaw <not needed> (adds as state)

Yaw Equilibrium ∙ 𝑓 𝑣 ∙ 𝑓 𝑣 ∙

Constitution 𝑓 𝑤 𝑓 ∙ 𝑠𝑓 𝑤 𝑣 ∙𝑠

Compatibility 𝑓 + 𝑠𝑓 𝑤 ≈( 𝑣 + 𝑓 )⁄| 𝑣| 𝑠 ( 𝑣 ∙ )⁄| 𝑣|

Force transfor-
mation, steered 𝑓 𝑣 𝑓 𝑤 𝑓 𝑤 ∙ 𝑓 𝑓 𝑣 𝑓 𝑤 𝑓 + 𝑓 𝑤
wheels to vehicle
From Eq [4.46]: 𝑥
Path with orienta- 𝑥 𝑣 𝑜𝑠( ) 𝑣 𝑠 𝑛( )
𝑦
tion: 𝑦 𝑣 cos( )+ 𝑣 𝑠 𝑛( )

Figure 4-51: Comparison between model with velocities in vehicle fixed and ground fixed
directions.
Using the velocity components in inertial directions brings yaw angle into the differential equations
for velocities. So, we have to add as state: we get 4 states ( ) instead of 3 ( 𝑣 𝑣 ).
The need of transformation manifests that vehicle directions is more natural directions for the consti-
tutive equations (tyre models). However, the equations for path becomes simpler with the ground
fixed directions. The two models gives, of course, the same result, e.g. (𝑡) 𝑣 (𝑡) cos( (𝑡))
𝑣 (𝑡) sin( (𝑡)), but the model in vehicle fixed directions is more frequently used.
A similar comparison between model in vehicle and ground fix and components can be done for
pitch rotation in 3.4.5 and 5.7.2 but the difference there is less important since pitch angle is much
more limited than yaw angle.

282
 Lateral Dynamics

4.4.2.4 Validity of Model

When only studying the Frequency Response as yaw velocity and curvature response, it is easy to for-
get that one very easily comes into manoeuvres where road friction is limited, i.e. where the linear tyre
model is not valid. Hence it is good to look at lateral acceleration Frequency Response, because we can
roughly say that for | | in the same magnitude as ∙ ∙ , it is doubtful if the model is valid. If the
wheel torques are significant, the validity limit is even lower. For high CoG vehicles, another invalidat-
ing circumstance is wheel lift, which can be approximately checked by checking that | | ≪ 𝑆𝑆
∙ ⁄( ℎ ).
If one needs to include nonlinear tyre models in stationary oscillation response, one can simulate using
time integration (same method as usually used for transient handling) over several excitation cycles,
until the response shows a clear stationary oscillation. This consumes more computational efforts and
the solutions become approximate and numerical.

4.4.3 Using the One-Track Model

Use the linear matrix model in Eq [4.50] to study “stationary oscillating steering”:
[ ] 𝐴∙[ ] + 𝐵 ∙ (𝑡) 𝐴∙[ ] + 𝐵 ∙ ̂ ∙ cos( ∙ 𝑡)

(Note similar notation for vehicle yaw velocity, , and steering angular frequency, .) Knowing ̂ and
, it is possible to calculate the responses ̂ , ̂ , and :
̂ 𝑣𝑦 ̂ 𝜔𝑧
[ ] [ ] ∙ cos ( ∙ 𝑡 [ ]) 𝑛 [ ] [ ̂ ] ∙ cos ( ∙ 𝑡 [ ])
̂ 𝜔 𝑧 𝑦

Different methods are available for calculation of the Frequency Responses:

• Fourier transform
• Complex mathematics, using 𝑒 ∙𝜔∙
• Real trigonometry, using cos( ∙ 𝑡 + 𝑝ℎ 𝑠𝑒) sin( ∙ 𝑡 + 𝑝ℎ 𝑠𝑒) or cos( ∙ 𝑡) + sin( ∙ 𝑡).
Typically, the problem has at least dimension 2, which makes matrix algebra efficient. Matrix formula-
tion and Fourier transform is very convenient with tools as Matlab if numerical solutions are accepted.

4.4.3.1 Single Frequency Response

4.4.3.1.1 Solution with Fourier Transform
Eq [4.50] can be transformed to the frequency domain (ℱ denotes Fourier transform, i.e. ℱ(𝜉(𝑡))
∫ 𝑒 𝑗∙ ∙𝑡 ∙ 𝜉(𝑡) ∙ 𝑡):

𝑗∙ ∙ 𝓕 ([ ]) 𝑨 ∙ 𝓕 ([ ]) + 𝑩 ∙ ℱ( 𝑓 )
{
𝓕 ([ ]) 𝑪 ∙ 𝓕 ([ ]) + 𝑫 ∙ ℱ( 𝑓 )
Solving for states and outputs, using ℱ ( ) 𝑗∙ ℱ ( );, gives:

[ ] ℱ (𝓕 ([ ])) ℱ ((𝑗 ∙ ∙𝑰 𝑨) ∙ 𝑩 ∙ ℱ( 𝑓 ))

[ ] ℱ (𝓕 ([ ])) ℱ (𝑪 ∙ 𝓕 ([ ]) + 𝑫 ∙ ℱ( 𝑓 ))
{
Expressed as transfer functions:
(𝑗 ∙ ∙𝑰 𝑨) ∙ 𝑩 ∙ ℱ( 𝑓 )
𝑯 𝑣𝑦 𝓕 ([ ]) (𝑗 ∙ ∙𝑰 𝑨) ∙𝑩
𝛿𝑓 →[𝜔 ] ℱ( 𝑓 ) ℱ( 𝑓 )
𝑧
[4.51]
𝑯 𝜔𝑧 𝓕 ([ ]) 𝑪∙𝑯 𝑣𝑦 +𝑫 𝑪 ∙ (𝑗 ∙ ∙𝑰 𝑨) ∙𝑩+𝑫
𝛿𝑓 →[
𝑦
] ℱ( 𝑓 ) 𝛿𝑓→[𝜔 ]
𝑧

283
 Lateral Dynamics

We have derived the transfer functions. The subscript tells that the transfer function is for the vehicle
operation with excitation=input 𝑓 and response=output [ ]𝑇 and output [ ]𝑇 . The
transfer function has dimension 2x1 and is complex. It operates as follows:
̂
[
] |𝐻 𝑣𝑦 | ∙ ̂𝑓
̂ 𝛿𝑓 →[𝜔 ]
𝑧
Amplitudes:
̂
[ ] |𝐻 𝜔𝑧 | ∙ ̂𝑓
{ ̂ 𝛿𝑓→[
𝑦
]
[4.52]
arg( ) )
{arg( 𝑓 }
𝑣𝑦
[ ] [ ] [ ] ∙ arg( 𝑓 ) arg (𝐻 𝑣𝑦 )
𝜔𝑧 arg( ) 0 𝛿𝑓 →[𝜔 ]
𝑧
Phase delays:
arg( )
{arg( 𝑓 )}
𝜔𝑧
[ ] [ ] [ ] ∙ arg( 𝑓 ) arg (𝐻 𝜔𝑧 )
{ 𝑦 arg( ) 0 𝛿𝑓→[
𝑦
]

4.4.3.1.2 Solution with Complex Mathematics

This section avoids requiring skills in Fourier transform. This makes the derivation quite long to reach
the final results Eqs [4.57] and [4.56]. With Fourier Transform, the expression for the Transfer Func-
tion, H, can be reached with less algebra. Knowing H, it can be used in Eq [4.56].
The fundamental situation for steering frequency response is that the excitation is: 𝑓 ̂𝑓 ∙
cos( ∙ 𝜋 ∙ 𝑓 ∙ 𝑡) {𝑒 ∙
cos( ) + 𝑗 ∙ sin( )} e( ̂𝑓 ∙ 𝑒 ∙2∙𝜋∙𝑓∙ ), where f is the (time) frequency in Hz
and 𝑗 is the imaginary unit. We rewrite ∙ 𝜋 ∙ 𝑓 as (angular frequency), which has to be carefully dis-
tinguished from (yaw velocity). Insert this in Eq [4.50] and neglecting the longitudinal force 𝑓 𝑤 .
The full (complex) equation is used:
𝑓 + 𝑓 ∙ 𝑓 ∙
+ ∙
0 𝑓
e [ ]∙[ ]+ 2 2 ∙[ ] [ ] ∙ ̂𝑓 ∙ 𝑒 𝜔∙
[4.53]
0 𝑓 ∙ 𝑓 ∙ 𝑓 ∙ 𝑓 + ∙ 𝑓 ∙ 𝑓

[ [ ] ]
We intend to solve the complex equation, and then find the solutions as real parts: Re[ ] and
Re[ ] . (Subscript c means complex.)
If only interested in the stationary solution, which is valid after possible initial value dependent transi-
ents are damped out, we can assume a general form for the solution.
̂ ∙𝜔∙ ̂ ∙𝜔∙
[ ] [ ]∙𝑒 ⇒ [ ] 𝑗∙ ∙[ ]∙𝑒 [4.54]
̂ ̂

284
 Lateral Dynamics

Inserting the assumption in the differential equation gives:

0 ̂ ∙𝜔∙
[ ]∙𝑗∙ ∙[ ]∙𝑒 +
0 ̂
𝑓 + 𝑓 ∙ 𝑓 ∙
+ ∙
̂
+ 2 2 ∙[ ]∙𝑒 ∙𝜔∙ [ ]∙ 𝑓 ∙ ̂𝑓 ∙ 𝑒 ∙𝜔∙
⇒
𝑓 ∙ 𝑓 ∙ 𝑓 ∙ 𝑓 + ∙ ̂ 𝑓

[ ]
̂
⇒[ ]
̂ [4.55]

𝑓 + 𝑓 ∙ 𝑓 ∙
+ ∙
0
[ ]∙𝑗∙ + 2 2 ∙[ ]∙ 𝑓 ∙ ̂𝑓
0 𝑓 ∙ 𝑓 ∙ 𝑓 ∙ 𝑓 + ∙ 𝑓

( [ ])
𝐻 𝑣𝑦 ∙ ̂𝑓
𝛿𝑓→[𝜔 ]
𝑧

Then, we can assume we know ̂ and ̂ from Equation [4.55], and consequently we know and
from Eq [4.54]. We have derived the transfer function, 𝐻 𝑣𝑦 . The subscript tells that the trans-
𝛿𝑓 →[𝜔 ]
𝑧

fer function is for the vehicle operation with excitation=input 𝑓 and response=output [ ] case.
This transfer function has dimension 2x1 and it is complex. It operates as follows:
̂ | |
Amplitudes: [ ] [ ] |𝐻 𝑣𝑦 | ∙ | 𝑓 |
̂ | | 𝛿𝑓 →[𝜔 ]
𝑧
[4.56]
arg( )
Phase delays: [ ] [ ] ∙ arg( 𝑓 ) {arg( 𝑓 ) 0} arg (𝐻 𝑣𝑦 )
arg( ) 𝛿𝑓→[𝜔 ]
𝑧

However, we can derive expressions for and on a real (non-complex) form, 𝑝 𝑡𝑢 𝑒 ∙

cos( 𝑛 𝑢 𝑟 𝑓𝑟𝑒𝑞𝑢𝑒𝑛 𝑦 ∙ 𝑡 𝑝ℎ 𝑠𝑒 𝑒 𝑦), without involving transfer function. That is done in the
following:
e( ) e( ̂ ∙ 𝑒 ∙𝜔∙ ) ⋯ | ̂ | ∙ cos( ∙ 𝑡 + arg( ̂ ))
The same rewriting can be done with , so that in total:

̂ ∙ cos( ∙ 𝑡 𝑣 ) | ̂ | ∙ cos ( ∙ 𝑡 ( arg( ̂ )))

[ ] [ ] [ ] [4.57]
̂ ∙ cos( ∙ 𝑡 𝜔 ) | ̂ | ∙ cos( ∙ 𝑡 ( arg( ̂ )))
Equation [4.55] and Equation [4.57] now gives us the possibility to find vehicle response amplitude
and phase delay. The ratios between amplitude of responses and amplitude of excitation, ̂ ⁄ ̂𝑓 and
̂ ⁄ ̂𝑓, are called gains. The difference in argument is the phase delay.

4.4.3.2 Random Frequency Response

Solutions to harmonic excitation of linear dynamic systems are superimposable. This is also why the
response from a mixed frequency excitation can be spliced into separate frequencies, e.g. using Fourier
transformation. Hence, a common way to measure the frequency response diagrams is to log data from
a random steering test. The frequency response diagram can then be extracted from this test.

4.4.3.3 Frequency Responses on Steering *

Function definition: Frequency Responses on Steering are amplification and delay of steer angle to vehicle
responses’ measures (yaw velocity, lateral acceleration etc.), for stationary oscillating harmonic steering at certain lon-
gitudinal speed and varying steering frequency.

285
 Lateral Dynamics

4.4.3.3.1 Lateral Velocity and Yaw Velocity Frequency Response *

The frequency response for the two states, Lateral Velocity and Yaw velocity, can be plotted using Eq
[4.51] and only one input ( 𝑓 instead of [ 𝑓 𝑓 𝑤 𝑓 ]𝑇 ), see Figure 4-52. The curves show
Frequency Response for same vehicle, but different speed. The yaw velocity gain curve has a knee at
0.5..1 Hz. The decrease after that is a measure of yaw damping. The curve for high speed actually has a
weak peak just before the knee. This is not desired, because the vehicle might feel a bit nervous. Yaw
damping can also be how fast yaw velocity decays after a step response, see 4.4.3.5.
From Equation [4.18] we can calculate that characteristic speed for the vehicle is 120 km/h. With an-
other understeering coefficient, we could have calculated a critical speed. Studying Figure 4-52 to Fig-
ure 4-55, one can find these special speeds in another appearance:
• For an understeered vehicle, speeds above the characteristic speed gives a negative yaw veloc-
ity delay for low steering frequencies will be negative.
• For over-steered vehicles, speeds above the critical speed gives a yaw velocity delay larger
than 80 𝑒 and yaw velocity amplitudes which are very large for low steering frequencies.
From Equation [4.18] we can calculate that characteristic speed for the vehicle is 120 km/h.
In Figure 4-53 the curves are for same speed and constant understeering gradient, but they show the
Frequency Response for different sums of cornering stiffness 𝑓 + . Increasing the stiffness in-
creases the yaw velocity gain (agility) at high frequencies.
In Figure 4-54 the curves are for same speed and constant sum of cornering stiffness ( 𝑓 + ), but
they show the Frequency Response for different values of understeering gradient (Ku). Increasing un-
dersteer gradient decreases the yaw velocity gain (agility) at low frequencies. But, also note that in-
creased understeering increases the peak in YVFR at around 0.5 Hz, which means low yaw damping.
LateralLateral
Velocity Frequency
Velocity Response Response Yaw Velocity
Yaw Frequency Response
Rate Response
120 10

100 8
wz gain [(rad/s)/rad]
vy gain [(m/s)/rad]

80
6
60
4
40

20 2

0 0
-2 -1 0 1 2 -2 -1 0 1 2
10 10 10 10 10 10 10 10 10 10

vx=150 km/h
vx=100 km/h
2 2
vx=50 km/h
1 vx=1 km/h
1.5
wz delay [rad]
vy delay [rad]

0
1
-1
0.5
-2

-3 0

-4 -0.5
-2 -1 0 1 2 -2 -1 0 1 2
10 10 10 10 10 10 10 10 10 10
frequency [Hz] frequency [Hz]
Figure 4-52: Frequency response to steer angle. Vehicle data: 000 𝑘 000 𝑘
2
𝑓 . .5 𝑓 8 400 𝑁/𝑟 78000 𝑁/𝑟 (𝐾 . 6 𝑟 /𝑀𝑁).

286
 Lateral Dynamics

LateralLateral
Velocity Frequency
Velocity Response Response Yaw Velocity Frequency
Yaw Rate ResponseResponse
80 8
Cf+Cr=119 kN/rad
Cf+Cr=159 kN/rad

wz gain [(rad/s)/rad]
vy gain [(m/s)/rad]
60 Cf+Cr=199 kN/rad 6

40 4

20 2

0 0
-2 -1 0 1 2 -2 -1 0 1 2
10 10 10 10 10 10 10 10 10 10

2 2

1
1.5

wz delay [rad]
vy delay [rad]

-1 1

-2
0.5
-3

-4 0
-2 -1 0 1 2 -2 -1 0 1 2
10 10 10 10 10 10 10 10 10 10
frequency [Hz] frequency [Hz]

Figure 4-53: Frequency response to steer angle. Same vehicle data as in Figure 4-52, except
varying 𝑓 and but keeping understeering gradient 𝐾 constant. Vehicle speed = 100 km/h.

LateralLateral
Velocity Frequency
Velocity ResponseResponse Yaw Velocity
Yaw Frequency Response
Rate Response
400 40
Ku= -1.26 rad/kN
wz gain [(rad/s)/rad]

Ku= -0.01 rad/kN

vy gain [(m/s)/rad]

300 30
Ku= 0.01 rad/kN
Ku= 1.26 rad/kN
200 20

100 10

0 -2 -1 0 1 2
0 -2 -1 0 1 2
10 10 10 10 10 10 10 10 10 10

2 2

1
1.5
wz delay [rad]
vy delay [rad]

-1 1

-2
0.5
-3

-4 -2 -1 0 1 2
0 -2 -1 0 1 2
10 10 10 10 10 10 10 10 10 10
frequency [Hz] frequency [Hz]

Figure 4-54: Frequency response to steer angle. Same vehicle data as in Figure 4-52, except
varying understeering gradient Ku but keeping 𝑓 + constant. Vehicle speed = 100 km/h.

287
 Lateral Dynamics

4.4.3.3.2 Lateral Acceleration Frequency Response *

The lateral acceleration Frequency Response is another useful measure to study and set requirements
on. Actually, yaw velocity and lateral acceleration are the most frequently used measures, since they
are easily measured, e.g. from ESC sensors in most vehicles.
The transfer function is found (here using Fourier transform and previous results):
+ ∙ 𝑗∙ ∙ + ∙
Amplitude: ̂ | | {𝑢𝑠𝑒: } | |
𝑓
| ∙ [𝑗 ∙ ]∙𝐻 𝑣𝑦 ∙ 𝑓| |[𝑗 ∙ ]∙𝐻 𝑣𝑦 | ∙ [4.58]
𝛿𝑓 →[𝜔 ] 𝛿𝑓→[𝜔 ]
𝑧 𝑧

Phase delay: arg( ) arg( 𝑓 ) {𝑢𝑠𝑒: arg( 𝑓 ) 0} arg ([𝑗 ∙ ]∙𝐻 𝑣𝑦 )

𝛿𝑓→[𝜔 ]
𝑧

Lateral acceleration Frequency Response is plotted for different vehicle speeds in Figure 4-55.
Lateral AccelerationLateral Acceleration FrequencyLateral
Response Response
Acceleration Response
250 2

vx=150 km/h
vx=100 km/h 1.5
200 vx=50 km/h
vx=1 km/h 1
ay gain [(m/(s*s))/rad]

0.5
150
ay delay [rad]

100
-0.5

-1
50
-1.5

0 -2
-2 -1 0 1 2 -2 -1 0 1 2
10 10 10 10 10 10 10 10 10 10
frequency [Hz] frequency [Hz]

Figure 4-55: Lateral acceleration frequency response to steer angle. Vehicle data:
2
000 𝑘 000 𝑘 𝑓 . .5 𝑓 8 400 𝑁/𝑟
78000 𝑁/𝑟 (𝐾 . 6 𝑟 /𝑀𝑁).
§ Influence of Tyre Relaxation
If relaxation is taken into account using 0 and 𝐿 50% 𝑜𝑓 ℎ𝑒𝑒 𝑟 𝑢 𝑓𝑒𝑟𝑒𝑛 𝑒 0.5 𝜋 𝑅 𝑅
0. , we get the following versions of the diagrams in Figure 4-52 .. Figure 4-55:

288
 Lateral Dynamics

We see that we get similar frequency responses, but the gain peaks at 𝑓 ≈ 0. . .08 𝐻 becomes a bit
higher. This is expected, since we add a delay in the system. The delay influence generally more if the
vehicle is articulated, such as long heavy combination vehicles.
It should be noted that the Lateral Acceleration Frequency Response gain approaches zero asymptoti-
cally when frequency increases, which was not the case without relaxation. The approach to zero is
truer and more intuitive since, with relaxation, the steered axle will not have time to develop forces
when its steering angle oscillates with high frequency.
4.4.3.4 Other Frequency Responses to Oscillating Steering *
In principle, it is possible to study a lot of other responses, such as Path Curvature Frequency Re-
sponse, Side slip Frequency Response and Lateral Path Width Frequency Response etc. For combina-
tion-vehicles it is common to plot Rearward amplification (RA) as function of frequency. The compen-
dium identifies some alternative definitions: 𝑅𝐴𝜔 max( ̂ ⁄ ̂ ) where numbers the vehicle
units. If the manoeuvre is not stationary oscillations, but e.g. a single lane change, we instead propose
𝑅𝐴𝜔 max (max| |⁄max| |) . RA can also be defined for the worst frequency, e.g. 𝑅𝐴𝜔
max (max( ̂ ⁄ ̂ )) . (Some argue for alternative definitions with instead of . However, these
𝑓
are more difficult to agree on; which point, vertical and longitudinal, one should measure in.)

289
 Lateral Dynamics

4.4.3.5 Lateral Stability

Eq [4.50] can explain instability. If one of the eigen solutions grows unlimited, and/or grows un-
limited. We call the property of such growth lateral instability. Simplified, instability can have two
modes: Front axle maintains grip but rear axle slides sideways (oversteering in 4.3.3.3) and vice versa
which makes vehicle unsteer-able (understeering in 4.3.3.3),
4.4.3.5.1 Locking One Axle
Locking one axle, , can be modelled by replacing with . This can be proven by linearization of
Eq [4.47] with the locked axle’s constitution replaced by “ ∙ ⁄| | ”. Since ≈
and cornering compliance, , is typically 5..10 (trucks) and 10..15 (passenger cars), we can con-
clude that is normally ≪ . Realistic variation of road friction coefficient is 0.2..1. So, a locked
axle can be seen as having much (typically 5..75 times) lower cornering stiffness on the locked axle. So,
locking front gives extreme under-steering and locking rear gives extreme over-steering.
4.4.3.5.2 Frequency Responses for Different Understeering Gradients
Eq [4.50] can explain instability. Instability means here that vehicle gets infinitely growing solutions
although 𝑓 0. Eq [4.68] shows the explicit solution. We set 𝑓 0 and a small yaw disturbance as
initial conditions: [ ] [0 𝜀 ] and get the following expression for the solution:
̂ ̂ 𝜆1 ∙
0 ] ∙ [[ ̂ ] [ ̂ 2 ]]
[ ] [[ ] [ 2 ]] ∙ [𝑒 ∙[ ]
̂ ̂ 2 0 𝑒 𝜆2 ∙ ̂ ̂2
̂ ̂ 2 𝜆 0
ℎ𝑒𝑟𝑒 [[[ ] [ ]] [ ]] eig(𝑨)
̂ ̂ 2 0 𝜆2
In Figure 4-56 we sweep for normal passenger car data and try the following cornering stiffnesses:
• Locked front axle: 𝑓 𝑓 ≈ 0. (extremely under-steered)
• Under-steered: 𝑓 𝑓 ≈ 0.9 (moderatly under-steered
• Neutral-steered: 𝑓 𝑓 ≈
• Over-steered: 𝑓 𝑓 ≈ . (moderatly over-steered
• Locked rear axle: 𝑓 𝑓 ≈ 0 (extremely over-steered)
In Figure 4-56 we see that only over-steered (incl. lock rear) gives an unstable vehicle. And that only
under-steered vehicles (incl. lock front) gives oscillating solutions. Also, we see that all vehicles are
stable for low enough speed.
Limit to instability (subscript 𝑟 𝑡) is when eal(𝜆 ) 0. Since 𝜆 𝜆( ), we can find
√ 𝑓 𝐿2 ⁄( 𝑓 𝑓 ) . This is the same expression as we found already with the steady
state model in Eq [4.18]. An advantage with using the transient model in Eq [4.50] is that we can also
express how fast the solutions grows towards infinity, if they are oscillating or exponential and how
the eigenmodes look.
We can express the condition for instability in other quantities. An example is when we want to limit
regenerative braking with electric motor on rear axle. Then we solve for :
𝑓 ⁄ (𝐿2⁄ 2
𝑓 ⁄ 𝑓 ) . Together with a combined slip tyre model (e.g. Eq [2.46]), we can express in-
2
stability in: √( )2 ( 𝑓 ⁄((𝐿2 ⁄ 2
𝑓⁄ 𝑓) | )) .
=

290
 Lateral Dynamics

lock rear
neutral-steered
under-steered

lock front
under-steered
neutral-steered
over-steered
lock rear

lock front

neutral-steered

Figure 4-56: Stability analysis of passenger car with varying cornering stiffnesses.

4.4.3.6 Driver Influence on Stability

A driver can improve or disturb the vehicle. The gains for steady state cornering and the frequency re-
sponses for stationary oscillating steering does not measure the real stability since they do not include
the dynamics of the driver, neither human nor automated driver. However, studying critical speed
through a transient model enables to add a dynamic driver model and quantify the influence of driver.
𝑠𝑠𝑢 𝑒 𝑦
A driver model can be, e.g.: 𝑓 𝑘 ( ) { 0} 𝑘 [0 𝑘] [ ]

Adding this equation to Eq [4.50] gives: [ ] 𝑨∙[ ]+𝑩∙ 𝑓 (𝑨 + 𝑩 ∙ [ 0 𝑘]) [ ]

Eigenvalue analysis on matrix (𝑨 + 𝑩 ∙ [0 𝑘]) for varying speed and driver gain gives:

𝑓 . stable for driver with k=+0.05

.5
000
000

𝑓 5 stable for
𝐿
driver with
k=-0.05
𝑓
0
𝐿

Figure 4-57: Influence of driver on stability. Driver model used: 𝑓 𝑘 ( ) 𝑘 .

291
 Lateral Dynamics

4.5 Transient Driving

This section addresses both transient handling and arbitrary transient driving, where the first is
mainly lateral, while longitudinal dynamics is added in the latter. Generally, this can be turning and
braking/accelerating at the same time through a manoeuvre.

4.5.1 Transient Driving Manoeuvres *

From the manoeuvres defined in this section, there are a large number of relevant function possible to
define. Here, only a few is mentioned:
Function definition: Double Lane Change Capability is the highest entry speed that a certain driver (or
driver model) can manage without hitting any cones. A certain cone pattern (e.g. as Figure 4-58 or Figure 4-59)
and a certain road friction has to be defined. A certain operation principle of pedals needs to be specified, e.g. re-
lease both pedals at entry or full brake pedal apply.
Function definition: Over-speed into Curve Capability is the highest entry speed that a certain driver (or
driver model) can manage without hitting any cones. A certain cone pattern e.g. as Figure 4-60) and a certain road
friction has to be defined. A certain operation principle of pedals needs to be specified, e.g. release both pedals at
entry or full brake pedal apply.
Function definition: Steer-In and Release Accelerator Pedal stability is a measure (e.g. side slip rear
axle peak) after a certain simultaneous steer-in and release of accelerator pedal, starting from a steady state corner-
ing at a certain path radius and speed. A certain road friction has to be defined.
When testing Transient driving manoeuvres, the typical part of the test track is the Vehicle Dynamics
Area or a Handling Track, see Figure 1-53. A Handling Track is a normal width road, intentionally
curved and with safety areas beside the curves for safety in case of run-off road during tests.
Typical transient tests are:
• Step steer, where one can measure transient versions of
o Yaw velocity response
o Lateral acceleration response
o Curvature response
o Yaw damping
• Lateral avoidance manoeuvres:
o Single Lane Change SLC cone track
o Double Lane Change DLC cone track, see Figure 4-23, Figure 4-58, and Figure 4-59
o Lane change while full braking
o Sine with dwell
o Steering effort in evasive manoeuvres
• Tests from steady state cornering
o Brake or accelerate in curve
o Lift accelerator pedal and steer-in while cornering
o Over-speeding into curve, see Figure 4-60
• Handling type tests
o Slalom between cones
o Handling track, general driving experience
• Roll-over tests
Standards which are relevant to these test manoeuvres are, e.g. References (ISO 3888), (ISO 7401),
(ISO 7975, 2006), (ISO 11026), (ISO 14791), (ISO 14793), (ISO 14794, 2011) and (NHTSA).
There are many standards for DLC beside ISO, such as Consumer Union. Generally speaking, the DLCs
with short distance in 2nd lane is often used for ESC-like tests.

292
 Lateral Dynamics

[m]

𝑊 + 0. 5

𝑊 + 0. 5

𝑊 + 0. 5
1
.

.
.5
driving direction

.
15 30 25 25 15+15

Figure 4-58: Passenger cars - Test track for a severe lane-change manoeuvre - Part 1: Double lane-
change. Reference (ISO 3888).

. 𝑊 + 0. 5
[m]

(but not <3)

𝑊 + 0. 5
𝑊 + 0. 5

1
driving direction
.

12 13.5 11 12.5 12
Figure 4-59: Passenger cars - Test track for a severe lane-change
manoeuvre - Part 2: Obstacle avoidance. Reference (ISO 3888).

Figure 4-60: Cone track for one standardized test of Over-speeding into curve

4.5.2 One-Track Models, without Lateral Load

Transfer
4.5.2.1 Two-Axle Vehicle
This section starts from the model derived in 4.4.2 and Figure 4-46 and Equations [4.46] and [4.47].
However, in the context of transient dynamics it is more relevant to use the model for more violent

293
 Lateral Dynamics

manoeuvres, and also active control such as ESC interventions. Hence, we extend the model in three
ways:
• The constitutive relation is saturated, to reflect that each axle may reach friction limit, friction
coefficient times normal load on the axle. See max functions in Equation [4.59].
• To be able to do mentioned limitation, the longitudinal load transfer is modelled, but only in
the simplest possible way using stiff suspension models. Basically, it is the same model as given
in Equation [3.13].
• A yaw moment representing (left/right) unsymmetrical braking/propulsion. See the term
𝑀 in yaw equilibrium in Equation [4.59]. This is much better modelled in a two-track
model, where 𝑀 does not appear, but we instead can have different longitudinal tyre forces
on left and right side.
It should be noted that the model lacks lateral load transfer and the transients in longitudinal load
transfer and the reduced cornering stiffness and reduced max friction due to load transfer and utiliz-
ing friction for wheel longitudinal forces.
Equilibrium in road plane (longitudinal, lateral, yaw):
∙( ∙ ) 𝑓 𝑤 ∙ cos( 𝑓 ) 𝑓 𝑤 ∙ sin( 𝑓 ) +
∙( + ∙ ) 𝑓 𝑤 ∙ sin( 𝑓 ) + 𝑓 𝑤 ∙ cos( 𝑓 ) +
∙ ( 𝑓 𝑤 ∙ sin( 𝑓 ) + 𝑓 𝑤 ∙ cos( 𝑓 )) ∙ 𝑓 ∙ +𝑀
Equilibrium out of road plane (vertical, pitch):
𝑓 + ∙ 0
𝑓 ∙ 𝑓 + ∙ ( 𝑓 𝑤 ∙ cos( 𝑓 ) 𝑓 𝑤 ∙ sin( 𝑓 ) + )∙ℎ 0
Constitution:
[4.59]
𝑓 𝑤 sign(𝑠𝑓 ) ∙ min( 𝑓 ∙ |𝑠𝑓 | ∙ 𝑓 )
sign(𝑠 ) ∙ min( ∙ |𝑠 | ∙ )
Compatibility, slip definitions:
𝑓 𝑤 ∙
𝑠𝑓 𝑛 𝑠
𝑓 𝑤
Compatibility, transformation from vehicle to wheel coordinate system:
𝑓 𝑤 ( + 𝑓 ∙ ) ∙ sin( 𝑓 ) + ∙ cos( 𝑓 )
𝑓 𝑤 ( + 𝑓 ∙ ) ∙ cos( 𝑓 ) ∙ sin( 𝑓 )
A Modelica model is given in Eq [4.60]. Changes compared to Eq [4.46] are marked as underlined code.
//Equilibrium, in road plane:
m*(der(vx)-wz*vy) = Ffxv + Frx;
m*(der(vy)+wz*vx) = Ffyv + Fry;
J*der(wz) = Ffyv*lf - Fry*lr + Mactz;
//Equilibrium, out of road plane:
Ffz + Frz - m*g = 0;
-Ffz*lf + Frz*lr -(Ffxv + Frx)*h = 0;
//Compatibility:
vfxv = vx; vfyv = vy + lf*wz;
vrx = vx; vry = vy - lr*wz;
//Lateral tyre force model:
Ffyw = -sign(sfy)*min(Cf*abs(sfy), mu*Ffz); sfy = vfyw/vfxw; [4.60]
Fry = -sign(sry)*min(Cr*abs(sry), mu*Frz); sry = vry/vrx;
//Transformation between vehicle and wheel coordinate systems:
Ffxv = Ffxw*cos(df) - Ffyw*sin(df);
Ffyv = Ffxw*sin(df) + Ffyw*cos(df);
vfxv = vfxw*cos(df) - vfyw*sin(df);
vfyv = vfxw*sin(df) + vfyw*cos(df);
//Shaft torques
Ffxw = +1000; // Front axle driven.
Frx = -100; // Rolling resistance on rear axle.
Mactz=0;
A simulation of this model is seen in Figure 4-61. Same steering input as used in Figure 4-47. 𝑀 is
zero. Cornering stiffnesses are chosen so that the vehicle is understeered in steady state. Road friction

294
 Lateral Dynamics

coefficient is 1. We can see that the vehicle now gets unstable and spins out with rear to the right. This
is mainly because longitudinal load transfer unloads the rear axle, since the kept steer angle deceler-
ates the vehicle. In this manoeuvre, it would have been reasonable to model also that the rear corner-
ing stiffness decreases with the decreased rear normal load, and opposite on front. Such addition to
the model would make the vehicle spin out even more. On the other hand, the longitudinal load shift is
modelled to take place immediately. With a suspension model, this load shift would require some
more time, which would calm down the spin-out. In conclusion, the manoeuvre is violent enough to
trigger a spin-out, so a further elaboration with how to control 𝑀 could be of interest. However, it
is beyond the scope of this compendium.
df=df [deg]

y [m]

wz [deg/s]

x[m]
yz=pz [deg]

time [s]
x[m]
Figure 4-61: Simulation results of one-track model for transient dynamics. The vehicle drawn in
the path plot is not in proper scale, but the orientation is approximately correct.
The vehicle reaches zero speed already after 7 seconds, because the wide side slip decelerates the ve-
hicle a lot. The simulation is then stopped, because the model cannot handle zero speed. That vehicle
models become singular at zero speed is very usual, since the slip definition becomes singular due to
zero speed in the denominator. The large difference compared to Figure 4-47 is due to the new consti-
tutive equation used, which shows the importance of checking validity region for any model one uses.
A simplified version of the mathematical model in Eq [4.59] follows in Eq [4.61]; assuming constant
and small angles and small propulsion force and no 𝑀 .

295
 Lateral Dynamics

Equilibrium in road plane (lateral, yaw):

∙( + ∙ ) 𝑓 +
∙ 𝑓 ∙ 𝑓 ∙
Equilibrium out of road plane (vertical, pitch):
𝑓 + ∙ 0
𝑓 ∙ 𝑓 + ∙ 0
[4.61]
Constitution:
𝑓 sign(𝑠𝑓 ) ∙ min( 𝑓 ∙ |𝑠𝑓 | ∙ 𝑓 )
sign(𝑠 ) ∙ min( ∙ |𝑠 | ∙ )
Compatibility:
+ 𝑓∙ ∙
𝑓 + 𝑠𝑓 𝑛 𝑠

4.5.2.2 Articulated Vehicles

A model for an articulated vehicle will now be derived in the same spirit as in 4.4.2. The model can rep-
resent a car with trailer or rigid truck with centre-axle trailer or tractor with semitrailer (if < 𝐿 ).

𝑓𝑦
𝑟𝑦

𝑟𝑥 df

𝑓 𝑣
𝐿

𝑓 𝑣
2 arctan 2 ⁄ 2 part of free body free body diagram part of free body
arctan ⁄ diagram of trailer: of coupling: diagram of trailer:

Figure 4-62: Model of two-unit articulated vehicle.

Model equations for the 1st unit:
Equilibrium of 1st unit (longitudinal, lateral, yaw around CoG):
( ) cos( 𝑓 ) 𝑓 𝑤 sin( 𝑓 ) 𝑓 𝑤+ +
( + ) sin( 𝑓 ) 𝑓 𝑤 + cos( 𝑓 ) 𝑓 𝑤+ +
(sin( 𝑓 ) 𝑓 𝑤 + cos( 𝑓 ) 𝑓 𝑤) (𝐿 )+
+( ) ( )
Constitution for axles on 1st unit:
[4.62]
𝑓 𝑤 𝑓 𝑠 𝑓 𝑠
Compatibility, within 1st unit:
𝑠𝑓 𝑓 𝑤 ⁄| 𝑓 𝑤 | 𝑠 ⁄| |
𝑓 𝑤 + cos( 𝑓 ) + sin( 𝑓 ) 𝑓 𝑣
𝑓 𝑤 sin( 𝑓 ) + cos( 𝑓 ) 𝑓 𝑣
𝑓 𝑣 +

296
 Lateral Dynamics

(𝐿 )
( ) ∎
Model equations for the 2nd unit:
Equilibrium of 2nd unit (longitudinal, lateral, yaw around CoG):
2 ( 2 2 2 ) 2 + 2
2 ( 2 + 2 2 ) 2 + 2
2 2 2 2 + 2 (2 2 )
Constitution for axles on 2nd unit:
[4.63]
2 2 𝑠2
Compatibility, within 2nd unit:
𝑠2 2 ⁄| 2 |
2 2 2 2
2 2 + 2 ( 2 2 ) ∎
Model equations for the coupling (see also1.6.1.1):
Equilibrium of coupling (x, y in 1st unit’s coordinates):
+ cos(θ) 2 + sin(θ) 2 0
sin(θ) 2 + cos(θ) 2 0
Compatibility of coupling: [4.64]
+ cos(θ) 2 + sin(θ) 2 ∎
sin(θ) 2 + cos(θ) 2 ∎
θ 2
𝑇
Let us assume [ 𝑓 𝑓 𝑤 ] are known inputs. Eqs [4.62].. [4.64] is a DAE system in with
2
24 equations and 24 unknowns: 2 2 2 θ 𝑓 𝑤 2 𝑠 𝑓 𝑠 𝑠2
2 2 𝑓 𝑤 𝑓 𝑤 𝑓 𝑣 2 2 . The equations can be written as they are into
a Modelica tool, which can find a possible state selection (such as 2 ), an explicit form
and perform simulations. Note that not only velocities but also an (relative angular) position becomes
states, as opposed to a one-unit vehicle. The use of a Modelica tool is very motivated for articulated ve-
hicles, since the articulation points makes the DAE system of equation to be of “high index”, meaning
that the Modelica tool identifies the constraints equations, i.e. the equations which have to be differen-
tiated, cf. 3.3.2. In this case, it is the equations marked with ∎ in Eqs [4.62]..[4.64].
4.5.2.2.1 § Explicit Form Model without DAE Tool
Contribution by Toheed Ghandriz, Vehicle Dynamics at Chalmers

It will now be explained how the explicit form model can be found using manual equation manipula-
tions or simpler symbolic tools, unable to handle DAEs (e.g. Matlab Symbolic Toolbox). We have to dif-
ferentiate the equations marked with ∎ in Eqs [4.62]..[4.64]. It leads to:
( )
2 2 + 2 (2 2 )
( sin(θ θ 2 + cos(θ) 2 ) + (cos(θ) θ 2 + sin(θ) 2 )
)
( cos(θ) θ 2 sin(θ) 2 ) + ( sin(θ) θ 2 + cos(θ) 2 )
So, we regard it as an ODE with 24+4=28 equations: . States and inputs as above DAE. The states are
regarded as known and we can count to 28 unknown variables, with notations as in 1.5.1.1.4.2:
{𝒙 𝒚𝟏 𝒚𝟐 𝒚𝟐 }
{ 2 }
{{ 𝑓 𝑤 2 𝑠 𝑓 𝑠 𝑠2 2 2 𝑓 𝑤 𝑓 𝑤 𝑓 𝑣 2 }}
{ 2 2 2 }{ 2 2 2 }
The expressions for state derivatives 2 θ and output variables (remaining of the 28 varia-
bles) can be found through algebraic manipulations, but in this case the symbolic tools will generate

297
 Lateral Dynamics

huge expressions (hundreds of thousands of tokens) or even fail. Therefore, we will reformulate model
and introduce approximations in 4.5.2.2.1.1.
A way to keep symbolic expressions smaller (still large, but not huge) without approximations is to
stay at the implicit form, and solve it with special time integration methods, see 1.5.1.1.4.2:
𝒓𝒆 𝒊𝒅 𝒂𝒍 ← 𝒇𝒊 𝒑𝒍𝒊𝒄𝒊𝒕 (𝒙 𝒙 𝒚𝟏 𝒚𝟐 𝒚𝟐 (𝑡) 𝑡) ℎ𝑒𝑟𝑒 𝒚𝟐 { 𝑦 𝑦 𝑦 𝑥}

Reformulated Model with Approximations

We will derive a model without introducing the constraint forces 2 2 . We select intuitively
the 5 states 2 and find the equations to express their derivatives: 4 equilibria and 1
compatibility relations which involves . We also introduce approximations as for the two axle one-
track model in 4.4.2: small angles 𝑓 arctan( 𝑤 ⁄| 𝑤 |), centrifugal acceleration perpen-
dicular to each unit ( 0) and steered axle lateral tyre force does not influence whole vehicle
longitudinally ( 𝑓 𝑤 ∙ 𝑓 0). Some new, but conceptually same, approximations can be done espe-
cially for articulated vehicles: small articulation angle and lateral tyre force on axles on 2nd unit does
not influence whole vehicle longitudinally ( 2 ∙ θ 0).
Equilibrium of whole vehicle (all units together: longitudinal to 1st unit, lateral to 1st unit):
0 + 𝑓 𝑤 cos( 𝑓 ) 𝑓 𝑤 sin( 𝑓 ) + +( 2 2 2 ) cos( )
+( 2 2 2 ) sin( )
≈ + 𝑓 𝑤 0+ +( 2 2 2 ) + (0 2 2 )
0 + 𝑓 𝑤 sin( 𝑓 ) + 𝑓 𝑤 cos( 𝑓 ) + ( 2 2 2 ) sin( )
+( 2 2 2 ) cos( )
≈ + 𝑓 𝑤 𝑓+ 𝑓 𝑤+ ( 2 2 2 )
+( 2 2 2 )
ℎ𝑒𝑟𝑒 ≈ 2 2 2 2 ≈ 2
+ 2 2 + 2 2
Equilibrium of 1st unit (yaw around coupling):
0 +( 𝑓 𝑤 sin( 𝑓 ) + 𝑓 𝑤 cos( 𝑓 )) + ( 𝐿 ) ( ) ≈
≈ +( 𝑓 𝑤 𝑓+ 𝑓 𝑤) + ( 𝐿 ) ( )
Equilibrium of 2nd unit (yaw around coupling):
0 2 2 + 2 2 (2 2 ) 2 2
Compatibility in coupling:
2
(Note: The same concept works for a vehicle with 𝑁 units: 2 translational equilibria for the whole vehi-
cle and 2 rotational equilibria (fore and aft part) per coupling point: + 𝑁 equilibria.)
Now, we realize that we have to find 2 compatibility relations that eliminate 2 2 . These may not
involve any new derivatives, only 2 2 and the state derivatives:
Differentiated Compatibility (to eliminate 2 2 ):

𝑥 ( cos( ) sin( )) ( cos( ) ( ( )) sin( ))

𝑡 𝑡
+ cos( ) sin( ) ( ( )) sin( )
( ( )) cos( ) ≈
≈+ ( ( )) ( ( ))

298
 Lateral Dynamics

2 ( 2 2 (2 2 )) ( cos( ) + sin( ) 2 ( 2 2 ))
𝑡 𝑡
(( ( )) cos( ) + sin( ) 2 (2 2 ))
𝑡
( ( )) cos( ) ( ( )) sin( ) +
sin( ) + cos( ) 2 (2 2 )≈
≈( ( )) ( ( )) +
+ + 2 ( 2 2 )
Now the constitution can be involved:
Constitution for axles (to eliminate lateral forces 𝑓 𝑤 2 ):
𝑓 𝑤 𝑓 𝑠 𝑓 𝑤
𝑠 𝑤
2 2 𝑠2
We now eliminate the slips 𝑠 𝑓 𝑤 𝑠 𝑤 𝑠2 and express them in state variables:
Compatibility (to eliminate 𝑠𝑓 𝑤 ):
arctan ( 𝑓 𝑣 | |) 𝑓 + arctan(𝑠 𝑓 𝑤) ⇒𝑠 𝑓 𝑤 ≈ 𝑓 𝑣 ⁄| | 𝑓
where 𝑓 𝑣 +
Compatibility (to eliminate 𝑠 ):
𝑠 ⁄| |
where (𝐿 )
Compatibility (to eliminate 𝑠2 ):
𝑠2 2 ⁄| 2 |
where 2 2 2 2
where 2 cos( ) + sin( ) ≈ +
where ( )
and 2 + cos( ) sin( ) ≈ +
If we perform the eliminations we get 5 equations:
𝒇𝑫𝑨𝑬 (𝒙 𝒙 ) 𝟎 ℎ𝑒𝑟𝑒 𝒙 [ 𝑥 𝑦 ] [ 𝑓 𝑓𝑥 𝑟𝑥 𝑥]
Using the exact (grey text) expressions, this is exactly same model as in Eqs [4.62]..[4.64].
Symbolically Linearized Explicit Form Model
We can reformulate to explicit form: 𝒙 𝒇𝑶𝑫𝑬 (𝒙 ) but this is the same huge expression, mentioned
above. Using the approximate (black text) expressions, the solution will be somewhat smaller, but
still very large. However, we can linearize around an operation point (𝒙𝟎 𝟎 ):
𝒙 ≈ 𝒇𝑶𝑫𝑬 (𝒙𝟎 𝟎 ) + 𝑨(𝒙𝟎 𝟎 ) (𝒙 𝒙𝟎 ) + 𝑩(𝒙𝟎 𝟎 ) ( 𝟎)
𝒇𝟎 + 𝑨𝟎 (𝒙 𝒙𝟎 ) + 𝑩𝟎 ( 𝟎)
(𝒇𝟎 𝑨𝟎 𝒙𝟎 𝑩𝟎 𝟎 ) + 𝑨𝟎 𝒙 + 𝑩𝟎 𝒌𝒙 + 𝑨𝟎 𝒙 + 𝑩𝟎
𝜕𝒇𝑶𝑫𝑬 𝒊 𝜕𝒇𝑶𝑫𝑬 𝒊
The elements of the matrices 𝑨𝟎 𝑩𝟎 is: 𝐴 𝜕𝒙𝒋
| 𝑛 𝐵
𝜕 𝒋
|
𝒙𝟎 𝟎 𝒙𝟎 𝟎
As example of relevant operating condition to linearize around, we take straight-line without signifi-
cant longitudinal tyre forces: 𝒙𝟎 [ 𝑥0 0 0 0 0]𝑇 𝟎 [0 0 0 0]𝑇 . Note that we do not assume
constant speed, i.e. we do not set 0, but is still a state variable.
Then we get symbolic expressions of only 𝑨𝟎 𝑩𝟎 of around 15 thousand tokens, which is manageable,
but not worth writing out in this compendium, except on this overviewing way:
0 0 0 0 0 0 ≠0 ≠0 ≠0
0 0 ≠0 ≠0 ≠0 ≠0 0 0 0
𝑨𝟎 0 0 ≠0 ≠0 ≠0 𝑩𝟎 ≠0 0 0 0
0 0 ≠0 ≠0 ≠0 ≠0 0 0 0
[0 0 0 ] [ 0 0 0 0 ]

299
 Lateral Dynamics

Now, we can simulate to compare the different models:

• “Exact” (“Modular” Eqs [4.62]..[4.64], which is identical to the equations with black text above)
• “Small Angles” (the equations above, with a mixture of black and grey text so that only using the
small angle approximations, but not using the “zeros”)
• “With All Physical Approximations, but not linearized” (the equations with black text above)
• “(Symbolically) Linearized” (𝒙 𝒌𝒙 + 𝑨𝟎 𝒙 + 𝑩𝟎 as described above in 4.5.2.2.1.2)
A simulation with these 4 models is shown in Figure 4-63.
d_f
20

0 𝑓 [deg]
-20
0 1 2 3 4 5

veh_Exact.v_1x veh_SmallAngleApprox.v_1x veh_BothApprox.v_1x veh_Linearized.v_1x veh_Linearized.v_1x0

22.0

21.5
[ ⁄𝑠]
21.0

20.5

20.0
0 1 2 3 4 5

veh_Exact.w _1z veh_Exact.w _2z veh_SmallAngleApprox.w _1z veh_SmallAngleApprox.w _2z veh_BothApprox.w _1z veh_BothApprox.w _2z veh_L
12

10

-2

-4

-6

-8
𝑡 [𝑠]
-10
0 1 2 3 4 5
Figure 4-63: Tractor and semitrailer. Comparison of simulations with the 4 models (thick solid:
”Exact”, thick dotted: “Small Angles”, thin dashed: “With All Physical Approximations, but not
linearized”, thin solid: “(Symbolically) Linearized”).
A similar linearization is done for a longer combination vehicle (A-double, i.e. Tractor+Semi-
trailer+Fulltrailer) in https://research.chalmers.se/publication/192958.
4.5.2.2.2 § Model Library for Articulated Vehicles
The first model (Eqs [4.62].. [4.64]) for articulated vehicle is written so that each equation belongs to
either a unit or a coupling. This opens up for systematic treatment of combination vehicles with more
than one articulation point. There are basically two conceptual ways:
• Modular library from which parts can be graphically dragged, dropped and connected.
• Vectorised formulation, see Reference (Sundström, Jacobson, & Laine, 2014) and 4.5.2.2.2.2.
§ Modular Library
A modular library is shown in figure below. Eq [4.62] is declared in “Unit” and Eq [4.64] in “Coupling”.

300
 Lateral Dynamics

drag and drop

connect

Figure 4-64: § Drag, drop and connect library for heavy combination vehicles. The model
example shows a so-called A-double, Tractor+SemiTrailer+Dolly+SemiTrailer.
An example of lane change manoeuvre, defined as lateral acceleration on 1 st axle follows a single sine-
wave, is shown in Figure 4-65. The natural input is prescribed steer angle ( 𝑓 sin(𝑡 𝑒) ), but since
modelled in Modelica, it is as easy to prescribe something else, e.g. lateral acceleration on first axle
( 𝑓 sin(𝑡 𝑒) ).

Figure 4-65: § One period sinus test of a Tractor+SemiTrailer.

§ Vectorised Model Format
Contribution from Toheed Ghandriz, Vehicle Dynamics at Chalmers and Mikko Karisaari, Oulo University, Finland
The model presented in this section can be run in 3 different modes (set by parameter DynOpCond):
Transient, SteadyState and LowSpeed.

301
 Lateral Dynamics

When modelling combination vehicles, there is a value of being able to use same model for vehicles
with different number of units. The following figure shows the physical model for a unit . . 𝑛𝑢,
where 𝑛𝑢 is the number of units. Only one axle 𝑗 . . 𝑛 , where 𝑛 is the maximum number of axles
per unit. There are 𝑛𝑢 couplings. The inputs to a model can be, e.g., 𝑛𝑢 𝑛 steering angles and
𝑛𝑢 𝑛 wheel-longitudinal forces 𝑤 . The latter can be actuated as 𝑤
(𝑇 𝑝 + 𝑇 𝑘 )⁄𝑅𝑤 𝑙 .
unit unit
𝑥 𝑀 𝑀
axle j 𝑀

rear coupling ℎ front coupling rear coupling

on unit on unit on unit

ℎ
ℎ

Coupling Eqs:
+ 0
𝑀 +𝑀 0

𝑦
𝑥

unit
rear coupling front coupling rear coupling
on unit on unit on unit

Subscripts:
unit number, . . 𝑛𝑢
𝑗 axle number, . . 𝑛
𝑢 unit
axle
Coupling Eqs:
wheel
R rear coupling
F front coupling + 0

Figure 4-66: § Physical model of a combination vehicle. Only one unit and one coupling drawn.
Notation “⋯
⃗⃗⃗ ” means 2D geometric vector in 𝑥𝑦-plane.
The Modelica model becomes rather compact, using Modelica’s vector, for and if notation:
model vehModel_LateralOneTrack
… //Vectorised parameters, e.g. nu=number of units, na=max number of axles per unit, m[nu], …
… //Vectorised variables, e.g. v_x[nu], F_az[nu,na], …
equation
//For each unit:
for i in 1:nu loop
//In-road-plane Equilibrium of units (longitudinal, lateral, yaw around CoG):
if DynOpCond==3 or DynOpCond==2 then
KinEnForce_x[i] =m[i]*((if DynOpCond==3 then der(v_x[i]) else 0) - w_z[i]*v_y[i]);
KinEnForce_y[i] =m[i]*((if DynOpCond==3 then der(v_y[i]) else 0) + w_z[i]*v_x[i]);
KinEnMoment_z[i]=J_z[i]*der(w_z[i]);
Else // if DynOpCond==1 then
KinEnForce_x[i] =0;
KinEnForce_y[i] =0;
KinEnMoment_z[i]=0;
end if;
KinEnForce_x[i] = sum(F_axu[i,:]) +F_cRx[i] +F_cFx[i] +m[i]*g*sin(phi_ry)*sin(phi_z[i])
-(if i==1 then HalfRhoACd*v_x[i]^2 else 0);
KinEnForce_y[i] = sum(F_ayu[i,:]) +F_cRy[i] +F_cFy[i] -m[i]*g*sin(phi_ry)*cos(phi_z[i]);
KinEnMoment_z[i]= sum(F_ayu[i,:].*(l_a[i,:]-l_CoG[i]*ones(na)))
+F_cRy[i]*(l_cR[i]-l_CoG[i]) +F_cFy[i]*(l_cF[i]-l_CoG[i]);
a_x[i]=der(v_x[i]) - w_z[i]*v_y[i];
a_y[i]=der(v_y[i]) + w_z[i]*v_x[i];
//Out-of-road-plane Equilibrium of units (vertical, pitch around point on ground under CoG):
0 = +sum(F_az[i,:]) +F_cRz[i] +F_cFz[i] -m[i]*g*cos(phi_ry);
0 = -sum(F_az[i,:].*(l_a[i,:]-l_CoG[i]*ones(na))) -F_cRz[i]*(l_cR[i]-l_CoG[i])
-F_cFz[i]*(l_cF[i]-l_CoG[i]) +F_cRx[i]*h_cR[i] +F_cFx[i]*h_cF[i]

302
 Lateral Dynamics

+M_cRy[i] +M_cFy[i] -m[i]*(der(v_x[i])-g*sin(phi_ry)*sin(phi_z[i])) *h[i];

//Compatibility for rear and front couplings:
v_cRx[i] = v_x[i]; v_cRy[i] = v_y[i] + w_z[i]*(l_cR[i] - l_CoG[i]);
v_cFx[i] = v_x[i]; v_cFy[i] = v_y[i] + w_z[i]*(l_cF[i] - l_CoG[i]);
//Constitution (suspension)
for j in 1:na loop
if AxleGroupNumber[i,j] == 0 then
F_az[i,j] = 0;
elseif j >= 2 then
if AxleGroupNumber[i,j] == AxleGroupNumber[i,j-1] then
F_az[i,j] = F_az[i,j-1];
end if;
end if;
end for;
//Front coupling on 1st unit and Rear coupling on last unit not coupled:
if i == 1 then
F_cFx[1] = 0; F_cFy[1] = 0; F_cFz[1] = 0; M_cFy[1] = 0;
end if;
if i == nu then
F_cRx[i] = 0; F_cRy[i] = 0;
end if;
//Rear coupling degrees of freedoms:
if VertFree_RearCoupl[i] or i == nu then
F_cRz[i] = 0;
end if;
if PitchFree_RearCoupl[i] then
M_cRy[i] = 0;
end if;
end for;

//For each coupling:

for i in 2:nu loop
//Compatibility of coupled couplings:
theta[i - 1] = phi_z[i - 1] - phi_z[i]; // der(theta[i - 1]) = w_z[i - 1] - w_z[i];
v_cFx[i] = +cos(theta[i - 1])*v_cRx[i - 1] - sin(theta[i - 1])*v_cRy[i - 1];
v_cFy[i] = +sin(theta[i - 1])*v_cRx[i - 1] + cos(theta[i - 1])*v_cRy[i - 1];
//Equilibrium of coupled couplings (longitudinal, lateral):
F_cFx[i] +(+cos(theta[i-1])*F_cRx[i-1] -sin(theta[i-1])*F_cRy[i-1]) = 0;
F_cFy[i] +(+sin(theta[i-1])*F_cRx[i-1] +cos(theta[i-1])*F_cRy[i-1]) = 0;
F_cFz[i] +F_cRz[i-1] = 0;
M_cFy[i] +M_cRy[i-1] = 0;
end for;

//For each axle:

for i in 1:nu loop
for j in 1:na loop
Rw[i, j] = R_w*w[i, j];
//Wheel Torque Actuation:
F_aa[i, j] = sign(F_aaReq[i, j])*min(abs(F_aaReq[i, j]),
min(F_aaMax,P_aaMax/max(abs(Rw[i, j]), v_eps)));

// if WhlTyreMod == 2 then
//Wheel, Rotational Equilibrium:
J_w*der(w[i, j]) = (F_aa[i, j] - RRC*F_az[i, j] - F_axw[i, j])*R_w;
//Tyre Constitution:
(F_axw[i, j],F_ayw[i, j]) = F_xy_RotatingWhl(F_z=max(F_az[i, j], F_eps), w=w[i, j],
v_xw=v_axw[i, j],v_yw=v_ayw[i, j],R_w=R_w,mu_peak=mu_peak);
s_x[i, j] = (Rw[i, j] - v_axw[i, j])/max(abs(Rw[i, j]), v_eps);
s_yt[i, j] = v_ayw[i, j]/max(abs(Rw[i, j]), v_eps);
alpha[i, j] = atan2(v_ayw[i, j], v_axw[i, j]);

//Compatibility for each axle:

v_axu[i, j] = +cos(delta[i, j])*v_axw[i, j] - sin(delta[i, j])*v_ayw[i, j];
v_ayu[i, j] = +sin(delta[i, j])*v_axw[i, j] + cos(delta[i, j])*v_ayw[i, j];
v_axu[i, j] = v_x[i];
v_ayu[i, j] = v_y[i] + w_z[i]*(l_a[i, j] - l_CoG[i]);
//Force equivalence or transformation:
F_axu[i, j] = +cos(delta[i, j])*F_axw[i, j] - sin(delta[i, j])*F_ayw[i, j];
F_ayu[i, j] = +sin(delta[i, j])*F_axw[i, j] + cos(delta[i, j])*F_ayw[i, j];
end for;
end for;
end vehModel_LateralOneTrack;

Note that the tyre model used takes the influence of vertical tyre load into account. This is important
for heavy vehicles which typically have large variation of load and position of load. It requires that also
the out-of-road-plane equilibria, vertical and pitch, are included. It also requires that different coupling

303
 Lateral Dynamics

degrees of freedom and axle lifting are modelled. Therefore, the coupling also have the parameters
“Boolean VertFree_RearCoupl[nu]”, “Boolean PitchFree_RearCoupl[nu]” and “In-
teger AxleGroupNumber[nu,na]”. Based on these parameters, and an assumption about the
suspension30 control, e.g. same vertical forces within an axle group, the vertical forces can be solved for.
Adouble_Dolly.veh.v_x[1]
Adouble_FullTrailer.veh.v_x[1]

It cannot be
20 guaranteed to work for all combination vehicles, but it works e.g. for the “A-double with
Dolly” and10 “A-double
0 1 with2 FullTrailer” 3 in4 Figure51-60. The
6 change
7 is only
8 to change
9 VertFree_Rear-
10
Adouble_Dolly.delta_f
Adouble_FullTrailer.delta_f
Coupl[2]0 from false to true and PitchFree_RearCoupl[3] from true to false. This
gives a change
0 due1 to that2 vertical3 force is 4 taken5 slightly
6 different,
7 see 8figure below.
9 10
Adouble_Dolly.veh.w_z[1]
Adouble_Dolly.veh.w_z[2]
Adouble_Dolly.veh.w_z[3]
Adouble_Dolly.veh.w_z[4]
Adouble_FullTrailer.veh.w_z[1]
Adouble_FullTrailer.veh.w_z[2]
Adouble_FullTrailer.veh.w_z[3]
Adouble_FullTrailer.veh.w_z[4]
0.6
3

0.4 2
4 solid=A-double with Dolly
dashed=A-double with FullTrailer
0.2

0.0

-0.2

-0.4

-0.6 𝑡 /[𝑠]
0 1 2 3 4 5 6 7 8 9 10

Figure 4-67: § Difference between “A-double with Dolly” and “A-double with FullTrailer”, see
Figure 1-60 . Single sine steering during < 𝑡 < at 0 /𝑠. The drawbar coupling carries no
vertical force in a FullTrailer, but some (here ≈ 𝑘𝑁) in the other case. Lateral acceleration,
peaking around 7. .8 /𝑠 2, is unrealistic for roll-over, but reducing to realistic . .4 /𝑠 2 gives
hardly any visible difference.
The difference does not become so large because the pitch moment of the dolly is taken by either the
pitch-rigid drawbar or the turntable of the 2nd SemiTrailer. The dynamic longitudinal load transfer is
considered by adding the term m[i]*der(v_x[i])*h[i] in pitch equilibrium of the units. This
generates more simultaneous equations (algebraic loops). So, neglecting the dynamic longitudinal load
transfer is an option for a simpler explicit form model.

4.5.2.3 Cambering Vehicle Model

The model below shows how a cambering vehicle can be modelled. The model is mainly made for lat-
eral dynamics, but it allows also longitudinal acceleration. The drawing shows a two-wheeler, but any
or both of the axles could have two wheels, as long as the suspension linkage is such that the axle does
not take any roll moment. The model is shown in Eq [4.65] in Modelica format. It is not modelled that
driver moves within the vehicle, which is why the inertial data is constant. Also, the chassis (frame) is
modelled as stiff and steering system as massless. The model lacks two equations, which is logical
since a driver model can add prescribed steer angle and prescribed .

304
 Lateral Dynamics

view from side view from rear

ℎ
𝑓 𝑣

𝑓 𝑣
𝑓
𝑓 𝑓

𝑓 𝑣

𝑓 𝑣

view from above

Figure 4-68: Model of cambering vehicle. (The stars marks point of moment equilibria in the
“mathematical model” derived from this “physical model”.)
//Equilibrium in road plane (x,y,rotz):
m*ax = Ffxv + Frx;
m*ay = Ffyv + Fry;
Jz*der(wz) = Ffyv*lf - Fry*lr - m*ax*h*px;
ax = der(vx) - wz*vy;
ay = der(vy) + wz*vx;
//Equilibrium out of road plane (z, rotx, roty):
m*g = Ffz + Frz;
Jx*der(wx) = m*g*h*px + m*ay*h;
0 = -Ffz*lf + Frz*lr - m*ax*h; [4.65]
wx = der(px);
//Constitution:
Ffyw = -CC*Ffz*sfy; Fry = -CC*Frz*sry;
//Compatibility, slip definition:
atan(sfy) = atan2(vy + lf*wz + h*wx, vx - h*px*wz) - d;
atan(sry) = atan2(vy - lr*wz + h*wx, vx - h*px*wz) - 0;
//Force coordinate transformation:
Ffxv = -Ffyw*sin(d);
Ffyv = +Ffyw*cos(d);
When entering a constant radius curve from straight driving one has to first steer out of the curve to
tilt the vehicle a suitable amount for the coming path curvature, 𝑅𝑝 . The suitable amount is hence
𝑙 ⁄ ⁄ 2⁄
(𝑅𝑝 ) . Then one steers with the turn and balances
(closed loop controls) to the desired roll angle. Systems like this, which has to be operated in opposite
direction initially is called “Non-minimum phase systems”. It is generally difficult to design a controller
for such systems. The two simulations shown in Figure 4-85 are done with the above model. Initial
speed is 0 /𝑠 and 0. Path radius, 𝑅𝑝 (𝑡) 𝑠𝑡𝑒𝑝 𝑓𝑢𝑛 𝑡 𝑜𝑛 𝑡 𝑡 0. , representing a sud-
denly curving road or path.
• One simulation (veh_driver) uses a closed loop driver model which first steers outwards
( < 0) and then continuously calculates the steer angle as a closed loop controller: 𝑛
( ) . It is not claimed that the driver model is representative for real drivers.
• In the other simulation (veh_inverse), the roll angle is prescribed as 2 ⁄(𝑅
𝑝 ) .
To prescribe the roll angle, instead of steer angle, is a way to avoid the “Non-minimum phase”
difficulties. The system becomes a normal Minimum phase system if actuated with roll angle
instead of steer angle. A price one has to pay for this is that the model equations has to go

305
 Lateral Dynamics

through a more advanced symbolic manipulation, e.g. differentiation, to solve for all variables,
including the steer angle . However, with a Modelica tool the symbolic manipulation is done
automatically. One can see this as a way to avoid a controller design and instead use an optimal
driver; optimal in the sense that it follows the path curvature with optimal yaw agility. The
road path curvature is a step function but has to be filtered twice (time constant 0.1 s is used in
Figure 4-85) to become differentiable. veh_driver.y
veh_inverse.y
10
The latter, optimal driver, negotiates the turn without overshot in yaw velocity, so it follows a sud-
denly curving path better. 8
veh_driver.d veh_inverse.d veh_driver.px veh_inverse.px

10
d= =steering angle [deg] 0
px= =roll angle [deg] 6

4
0 -10

2
-10 -20
0.0 0.4 0.8 1.2 0.0 1.6 0.4 2.0 0.8 1.2 1.6 2.0

0
veh_driver.Ffyv veh_inverse.Ffyv veh_driver.w z veh_inverse.w z
1000

1 -2

0
Ffyv= 𝑓 𝑣 = -4
0
vehicle-lateral force
-1000 on front tyre [N] wz= =yaw velocity [rad/s]
-6
-1
0.0 0.4 0.8 1.2 0.0 1.6 0.4 2.0 0.8 1.2 1.6 2.0 0

Figure 4-69: Simulations of entering a curve with a cambering vehicle. Blue solid curves without
dot-marker show a closed-loop driver model which actuates . Red curves with dot-marker shows
an optimal/inverse driver model which actuates .
Both driver models above only exemplify the low-level, roll-balancing, part of a driver. To run the
model in an environment with obstacles, one would also need a high-level, path selecting, part which
outputs desired, e.g., or . Additional driver model for longitudinal actuation is also needed.
It can be noted that the roll influences in two ways, compared to the non-cambering (roll-stiff) vehicles
previously modelled in the compendium:
• The roll motion itself is a dynamic motion, where the roll velocity becomes a state variable car-
rying kinetic energy.
• The roll influences the tyre slip, e.g. rear: 𝑠 ( +ℎ )⁄ ( ℎ ) . The
term ℎ can generally be neglected for non-cambering vehicles, but for a cambering vehicle,
such as a bike, it is essential. The term ℎ is only important at large roll angles, it is for
instance used as lever for longitudinal wheel forces in ESC-like control systems for motorbikes.

4.5.3 Two-Track Models, with Lateral Load Transfer

The models in this section model both transients longitudinal load transfer (as in 3.4.5.2) and transient
lateral load transfer (extension from 4.3.10.3, with transients and wheel-individual suspension).

4.5.3.1 § Physical and Mathematical Model

Compared to the model presented in Figure 4-38 and Equations [4.32]..[4.38], we add the following:
• Inertial term for roll rotation, i.e. . Note that sprung body roll inertia ( ), not whole
vehicle roll inertia ( ), is used in roll equilibrium because the unsprung parts does not partici-
pate in roll motion.
• Damping forces in parallel to each spring force, i.e. adding 𝑓𝑙 𝑓 𝑙 and .
Due to the axle roll centre model, as opposed to wheel pivot point model, the heave acceleration is
zero. Hence, no inertial force needs to be introduced.

306
 Lateral Dynamics

The free-body diagrams are given in Fig below, which should be compared to Figure 4-38.

displacements, speeds, accelerations: forces, on body:

ay m*g
z
zfl=zrl y
zfr=zrr m*ay
px
Pry
Pfy

Fsfl+Fsrl+ Fsfr+Fsrr+
zflr=zrlr=0
zfrr=zrrr=0 +Fdfl+Fdrl Pfz+Prz +Fdfr+Fdrr

forces, forces, on front axle: forces, on rear axle:

on vehicle: m*g Fsfl+Fdfl Fsfr+Fdfr Fsrl+Fdrl Fsrr+Fdrr
m*ay Pfz Prz
Pry
h

Ffly+Frly Ffly Ffry Frly Frry

Ffry+Frry
Fflz Ffrz Frlz Frrz
Fflz+Frlz Ffrz+Frrz

Figure 4-70: § Model for transient load transfer due to lateral acceleration, using axle roll centres.
The constitutive equations for the compliances (or springs) are as follows, see Eq [4.31]. Note that the
anti-roll bars are not modelled. Note also that we differentiate, since we will later use the spring
forces as state variables in a simulation.
𝑓𝑙 𝑓𝑙 + 𝑓𝑤 ∙ ( 𝑓𝑙 𝑓𝑙 ) ⇒ 𝑓𝑙 𝑓𝑤 ∙ 𝑓𝑙 𝑓𝑤 ∙ 𝑓𝑙
𝑓 𝑓 + 𝑓𝑤 ∙ ( 𝑓 𝑓 ) ⇒ 𝑓 𝑓𝑤 ∙ 𝑓 𝑓𝑤 ∙ 𝑓
𝑙 𝑙 + 𝑤 ∙ ( 𝑙 𝑙 ) ⇒ 𝑙 𝑤 ∙ 𝑙 𝑤∙ 𝑙
+ 𝑤 ∙( ) ⇒ 𝑤 ∙ 𝑤 ∙
∙ ∙ ∙ ∙ 𝑓
ℎ𝑒𝑟𝑒 𝑓𝑙 𝑓 𝑛 𝑙
∙𝐿 ∙𝐿
The constitutive equations for the dampers have to be added:
𝑓𝑙 𝑓𝑤 ∙ 𝑓𝑙
𝑓 𝑓𝑤 ∙ 𝑓
𝑙 𝑤∙ 𝑙
𝑤 ∙
As comparable with Equation [4.32], we get the next equation to fulfil the equilibrium. The change
compared to Equation [4.32] is that we also have a roll and lateral inertia terms and 4 damper forces,
acting in parallel to each of the 4 spring forces. Actually, when setting up equations, we also under-
stand that a model for longitudinal load transfer is needed, which is why the simplest possible such,
which is the stiff suspension on in Equation [3.13].
In-road-plane: Equilibrium for vehicle (longitudinal, lateral, yaw):
∙ ∙( ∙ ) 𝑓 +
∙ ∙( + ∙ ) 𝑓 + [4.66]
∙ 𝑓 ∙ 𝑓 ∙
Out-of-road-plane: Equilibrium for vehicle (vertical, pitch, roll):

307
 Lateral Dynamics

𝑓𝑙 + 𝑓 + + 𝑙 ∙ 0
( 𝑓𝑙 + 𝑓 ) ∙ 𝑓 + ( 𝑙 + )∙ ( 𝑓 + )∙ℎ 0
𝑤 𝑤
∙ ( 𝑓𝑙 + 𝑙 ) ∙ ( 𝑓 + )∙ +( 𝑓 + )∙ℎ
2 2
Equilibrium for each axle (pitch, around roll centre):
𝑤 𝑤
( 𝑓𝑙 ( 𝑓𝑙 + 𝑓𝑙 )) ∙ ( 𝑓 ( 𝑓 + 𝑓 )) ∙ + 𝑓 ∙ ℎ 𝑓 0
2 2
𝑤 𝑤
( 𝑙 ( 𝑙 + 𝑙 )) ∙2 ( ( + )) ∙ + ∙ℎ 0
2

Compatibility gives, keeping in mind that is the only non-zero out-of road plane velocity (i.e.
0):

𝑓𝑙 𝑓𝑙 + ∙ 𝑛 𝑓 𝑓 ∙

𝑙 𝑙 + ∙ 𝑛 ∙
Constitution for tyre forces versus slip in ground plane becomes per wheel, as opposed to per axle in
previous model:
𝑓 𝑓
𝑓 𝑤 sign(𝑠𝑓 ) ∙ (min ( ∙ |𝑠𝑓 | ∙ 𝑓𝑙 ) + min ( ∙ |𝑠𝑓 | ∙ 𝑓 ))

sign(𝑠 ) ∙ (min ( ∙ |𝑠 | ∙ 𝑙 ) + min ( ∙ |𝑠 | ∙ ))

Eqs above give a model very similar to the one in Eq [4.60]. Only the additional equations from this
new model are shown in Eq below. Note especially the new lateral tyre force model equations, which
now have one term per wheel, because one wheel on an axle can saturate independent of the other on
the same axle.
//Equilibrium, roll
Jsx*der(wx) = (Fflz+Frlz)*w/2 - (Ffrz+Frrz)*w/2 + (Ffyv+Fry)*h;
//Equilibrium for each axle (pitch, around roll centre):
(Fflz-(Fsfl+Fdfl))*w/2 - (Ffrz-(Fsfr+Fdfr))*w/2 + Ffy*hRCf = 0;
(Frlz-(Fsrl+Fdrl))*w/2 - (Frrz-(Fsrr+Fdrr))*w/2 + Fry*hRCr = 0;
//Constitutive relation for tyres (Lateral tyre force model):
Ffyw = -sign(sfy)*

(min((Cf/2)*abs(sfy), mu*Fflz) + min((Cf/2)*abs(sfy), mu*Ffrz));

Fry = -sign(sry)*
[4.67]
(min((Cr/2)*abs(sry), mu*Frlz) + min((Cr/2)*abs(sry), mu*Frrz));
sfy = vfyw/vfxw; sry = vry/vrx;
//Constitution for springs:
der(Fsfl) = -cfw*vflz; der(Fsfr) = -cfw*vfrz;
der(Fsrl) = -crw*vrlz; der(Fsrr) = -crw*vrrz;
//Constitution for dampers:
Fdfl = -dfw*vflz; Fdfr = -dfw*vfrz;
Fdrl = -drw*vrlz; Fdrr = -drw*vrrz;
//Compatibility, out of road plane:
vflz = +w/2*wx; vfrz = -w/2*wx; vrlz = +w/2*wx; vrrz = -w/2*wx;
A simulation of this model is shown in Figure 4-71 and should be compared with the simulation in Fig-
ure 4-47. When comparing these, we see a slight difference, which is that the axles saturate gradually
during 3<time<3.5, instead of both at once at time=3.25.

308
 Lateral Dynamics

df=df [deg]

y [m]
wz [deg/s]

x[m]
yz=pz [deg]

x[m]
time [s]
Figure 4-71: § Simulation results of one-track model for transient dynamics with lateral load
transfer. The vehicle drawn in the path plot is not in proper scale, but the orientation is
approximately correct.
Even if the load transfer model does not influence the vehicle path a lot in this case, it may be im-
portant to include it to validity check the model through checking wheel lift. Wheel lift can be identi-
fied as negative vertical wheel forces, which are why we plot some vertical wheel forces, see Figure
4-72. In this case we see that we have no wheel lift (which would disqualify the simulation). In the
right part of the figure we can also see the separate contribution from spring ( ), damper ( ) and
linkage ( 𝑙 𝑛𝑘 𝑓𝑙 𝑓𝑙 𝑓𝑙 𝑓𝑙 ).

Frrz [N]
Fsrr [N]
Ffrz [N]
Frrz [N]

Flink,rr,z [N]
Fflz [N]

Frlz [N] Fdrr [N]

time [s] time [s]

Figure 4-72: § Suspension vertical force plots from simulation with one-track model for transient
dynamics with lateral load transfer dynamics (same simulation as Figure 4-71). Left: Road contact
forces for all wheels. Right: Forces for one wheel, rear left.

309
 Lateral Dynamics

4.5.3.2 Example of Explicit Form Model; Two-Axle Vehicle,

Driver and Environment
The implementation of the model in 4.5.2.3 was done in Modelica. A Modelica tool automatically trans-
forms the model to explicit form which can be simulated, which is very efficient. But, as mentioned in
1.5.4.5, explicit form models can sometimes facilitate the understanding of the vehicle’s dynamics. This
is why the following model is presented. It is implemented in the data flow diagram tool Simulink. This
section explains how the states (or state variables) together with inputs (or input variables), influ-
ence the derivatives (or state derivatives).
The example model in this section is similar, but not identical, to the model in 4.5.2.3. The aim is to
model in-road-plane motion, due to transient actuation (wheel torques and wheel steer angles). Limi-
tations in this example model are:
• Influence from vertically uneven road is NOT modelled.
• Neither wheel lift nor suspension bump stop are modelled
• Control functions (such as ABS, TC and ESC) are NOT modelled
• The (roll) pendulum effect is NOT modelled, see 4.3.10.2
• Wheel camber and steer angle change with suspension travel is NOT modelled.
The model is a typical passenger vehicle, with driver and environment, see Figure 4-73. The driver in-
terface is the normal, 2 pedals, 1 steering wheel and a boolean request for direction of propulsion,
LongDir (=-1 or +1). The interface to environment is motion (variable position) in surrounding
world. To try out the vehicle model, driver and environment is also modelled. This includes also the
interface between them, which is motion of obstacles in environment, relative to subject vehicle. The
suspension is exemplified with wheel-individual suspension on both axles.

Environment Driver Vehicle

Figure 4-73: Top level of model with model tree structure. The Environment is a track test with
cone walls to go left and right around. Notation “irp” and “oorp” refers to in-road-plane and out-
of-road-plane, respectively.
As an initial overview, the states are presented. There are 21 states in total, and distributed:
• Driver: 0 states
• Vehicle:
o Vehicle Control & Actuation: 0 states
o Wheels, Tyres & Suspension: 12 states (4 wheels’ rotational speed, 4 Elastic parts of
vertical wheel forces, 4 Longitudinal tyre forces, 4 Lateral tyre forces)
o Vehicle Motion: 9 states (6 velocities and 3 positions)
• Environment: 0 states
The 4+4 tyre force states arise from modelled tyre relaxation, see 2.2.5.3.2.1.

310
 Lateral Dynamics

4.5.3.2.1 Submodel Environment

Generally, the environment model is where the surrounding to the driver and the vehicle should be
defined: road/road network, obstacles, other road users and the “driving task”/”driving instructions”.
In this example, it is very small and simple; it only captures stand-still point obstacles, each with in-
structions whether to be passed as obstacle left or right of the vehicle. Inputs to environment model is
the (subject) vehicles position, including orientation in global coordinates. Outputs are the relative po-
sition (𝑥 𝑦) to each obstacle, in (subject) vehicle coordinate system.
4.5.3.2.2 Submodel Driver
In this example, the driver model is very small and simple. Briefly described, it treats the longitudinal
dynamics very simple, as closed loop control towards a constant desired speed forward. The lateral
dynamics is divided into two parts:
• Driver planning: Based on how driver perceives the obstacles relative to the vehicle, one of the
objects is selected to mind for, which leads to where to be aim. Basically, the nearest obstacle
ahead of vehicle is selected as mind for and aim is, in principle, either half a vehicle width left
(or right) of this obstacle.
• Driver operation: Based on the driver’s motorics, the steering wheel angle is calculated. In the
example, it is simply an inverse model of an ideally tracking two-axle vehicle which calculates
which constant steering wheel angle that would be needed to make front axle run over the aim
obstacle.
It can be noted that a Driver model would also be a logical model part where to include calculation of
driver’s perception, such as steering effort and perceived safety during manoeuvre, etc. It can also be
noted that the border between Environment and Driver is sometimes not obvious, especially when it
comes to modelling the “driving task” in the Environment model, which can also be seen as a “driver
high-level decision” and then be a logical part of the Driver model. If Environment model includes sur-
rounding vehicles (object vehicles), it also includes driver models for those. For automatic functional-
ity, anything from cruise control to automated driving, there should also be a “button HMI output”
from Driver model, not only pedals and steering wheel. Such interface would turn on/off such func-
tionality in the Vehicle model.
4.5.3.2.3 Submodel Vehicle

9 states
12 states

Figure 4-74: Vehicle submodel.

The figure shows the decomposition of the vehicle into 3 parts:
• Submodel “Vehicle Control & Actuation” models the actuators (Propulsion system, Brake
system and Steering system, including “control functions”) that respond on requests from the
driver with wheel torques (T_s and T_us) and wheel steer angles (delta_w). The notation
ending “4” refers to that the quantities are vectors with 4 components, one per wheel. T_s is
shaft torque and T_us is torque applied from unsprung parts, e.g. friction brake torque from
brake calliper. One can think of very advanced models of these actuator systems, including e.g.

311
 Lateral Dynamics

propulsion system dynamics and control functions (ABS, ESC, TC, …). However, in this example
model it is only modelled very simple:
o Propulsion system outputs a fraction (determined by APed) of a certain maximum
power, distributed equally on front left and front right wheel. If brake pedal is applied,
the propulsion system outputs zero torque.
o Brake system outputs a fraction (determined by BPed) of a maximum brake torque (
/𝑅𝑤 ), distributed in a certain fix fraction between front and rear axle (70/30). The
distribution within each axle is equal on left and right wheel.
o There are no states modelled in the vehicle Control & Actuator submodel.
• Submodel “Wheels, Tyre & Suspension” models the part which pushes the tyres towards the
ground and consequently transforms the wheel torques and wheel steer angles, via the tyre, to
forces on the whole vehicle. F_xyv8 is the x and y forces in each of the 4 wheels, 2x4=8. F_z4
is the 4 vertical forces under each wheel. Submodel “Wheels, Tyre & Suspension” is further
explained in 4.5.3.2.5.
• Submodel “Vehicle Motion” models the motion of the whole vehicle in-road-plane and motion
of sprung body out-of-road-plane. The inertial effects (mass∙acceleration) of the unsprung
parts are considered for in-road-plane but not for out-of-road-plane. This submodel includes
integrators for the 3+3+3=9 states:
o Velocities in-road-plane, : v_irp3 (which is transformed to x- and y-veloci-
ties of each wheel and then fed back as v_irpv8)
o Position in-road-plane, 𝑥 𝑦 : pos_irp3 (which is only fed forward to “Environ-
ment”)
o Velocities out-of-road-plane, : v_oorp3 (which is transformed to z-velocities
of sprung body over each wheel and then fed back as v_z4)
4.5.3.2.4 Submodel Vehicle Control and Actuators

Figure 4-75: Vehicle Control & Actuators sub-model.

The example content of this sub-model is very minimalistic but can still be explained as two parts:
• Interpret pedals (including arbitrate between accelerator and brake pedal) to a sum over
wheels longitudinal force request ( ) and steering wheel angle to a front road wheel angle
request ( 𝑓 ).
• Coordinate and actuate propulsion and brake, i.e. allocate to 4 wheels’ propulsion tor-
ques and brake torques. Also allocate 𝑓 to each of left and right front wheels.
Vehicle variables used for the control are wheel rotational speeds. Since front axle propulsion is as-
sumed, the front rotational speeds are also input to the propulsion actuator modelling. No state

312
 Lateral Dynamics

variables are present in this minimalistic example, but in a more advanced actuation model there
could typically be states such as: engine speed, gear (discrete state), delay states for brake system and
elastic forces in steering system.
4.5.3.2.5 Submodel Wheels, Tyres and Suspension
This submodel is shown in Figure 4-76.
• The 2 sub-models “Coord Transf to Vehicle” and “Coord Transf to Wheels” are straight-for-
ward coordinate transformations, see Eq [1.3].
• The sub-model “Wheels” is also relatively straight-forward. For each wheel, the rotational equi-
librium is used as model: ∙ 𝑇 +𝑇 ∙ 𝑅𝑤 sign( ) ∙ 𝑅𝑅 ∙ ∙ 𝑅𝑤 where 𝑅𝑤 is wheel
radius and 𝑅𝑅 is rolling resistance coeeficient. The submodel will hence contain the 4 states:
Rotational speeds of each wheel: w_w4.
• The sub-model “Springs, Dampers & Linkage” models the springs (incl. anti-roll-bars) and
dampers and the linkage. For each wheel:
o Four states: Elastic part of vertical tyre force under each wheel: F_s4
o The derivatives are governed by the differentiated constitution of the springs: Conceptually
but involving both wheel spring and anti-roll-bar.
o The force in damper is governed by the damper’s constitutive relation:
o The contact forces are calculated in submodel “Suspension Equilibrium” in Figure 4-77.
They are calculated from moment equilibrium of unsprung parts around a 3-dimensional
pivot axis. The pivot axis is defined by two points, the pivot point in longitudinal load trans-
fer (see Figure 3-31) and the pivot point in longitudinal load transfer (see Figure 4-37). The
scalar equilibrium equation for one wheel can be expressed, with vector (cross) and scalar
(dot) products, in 𝑣 𝑣 𝑇 and point coordinates. From this, can be solved. It
should be noted that the general relation should use a screw joint along the pivot axis, see
https://en.wikipedia.org/wiki/Screw_theory, which is why the 𝑒 of the screw appears in
equations.

marks one scalar state

Figure 4-76: Wheels, Tyres & Suspension sub-model.

• The sub-model “Tyres” models the tyre mechanics, very much like the combined slip model in
Equation [2.43] and the relaxation model in Eq [2.48]. For each wheel:
o Four states: Magnitudes of tyre forces in ground plane ( ) for each wheel: F_xyw4

313
 Lateral Dynamics

o Unfortunately, the tyre forces and depend on . This could easily create algebraic
loops. However, since we also model relaxation, the tyre forces become state variables
which breaks such algebraic loops. Another way of getting rid of the algebraic loops could
have been to use “memory blocks”. “Memories” are such that value from last time instant is
used to calculate derivatives in present time instant. This is generally NOT a recommended
way of modelling.

There is one rotational equilibrium around pivot axis for the un- +
sprung part for each wheel (× means cross product, • means
scalar product): 𝑢 𝑓
𝑒 sin cos 0
𝑀𝑝 𝑣 𝑣 𝑣 • 𝑢𝑝 𝑣 shaft 𝑇
𝜋 axis
𝑟𝑙 𝑣 × 𝑣 𝑣 +
where 𝑀𝑝 𝑣 • 𝑢𝑝 𝑣
+𝑇 sin cos 0 𝑣 𝑣

where 𝑟𝑙 𝑣 𝑟𝑙 𝑟𝑙 𝑟𝑙
and 𝑢𝑝 𝑣 ⁄ 𝑟𝑝 𝑣 𝑟𝑝 𝑣 𝑟𝑝 𝑣 𝑟𝑝 𝑟𝑝 𝑟𝑝
If a linkage without screw effect on pivot axis, the 𝑒 is zero.

Figure 4-77: Sub-model “Springs, Dampers & Linkage”.

Figure 4-78: Sub-model “Tyres”.

314
 Lateral Dynamics

4.5.3.2.6 Sub-model Vehicle Motion

The sub-model is shown in Figure 4-79. It is divided in upper part in-road-plane (irp) and lower part
out-of-road-plane (oorp). The velocities in road plane (v_irp3 [ ]) is needed also in Sprung
body OORP because of the centripetal term, identified in 3.3.5.3 and used in 3.4.8.1.

• state

30

20 Figure 4-79: Vehicle Motion.

4.5.3.2.7 Simulation Example Double Lane Change
A double lane change between cones is used as simulation example, see Figure 4-80. The cones are run
over
10 and even passed on the wrong side because the driver model is very simple.
cone
cone

0 7

6.5

5.5
tyre force in ground plane
(vertical tyre force is radius
-10 5

of circle)
4.5

4
wheel hub translational
3.5
velocity
3
cone
-20
2.5

186 187 188 189 190 191

-30
150 160 170 180 190 200 210 220

Figure 4-80: Simulation results of a double lane change between cones.

315
 Lateral Dynamics

4.5.3.3 Additional Phenomena

It is relevant to point out the following, which are not modelled in this compendium:
• Same as pointed out as missing for longitudinal load transfer, see 3.4.5.2.3.
• Additionally, anti-roll arrangements (elastic connections between left and right wheel on one
axle, often built as torsion bar) are not modelled in this compendium. With same modelling
concept as used above, each such would be treated as a separate spring with one state variable,
e.g. 𝑓 (Force-antiroll-front). This force will act in parallel with 𝑓 and 𝑓 on each side. Note
that it will be added on one side and subtracted on the other.

4.5.3.4 Transient Roll-Over *

Function definition: Transient roll-over resistance is the most severe measure of a certain transient ma-
noeuvre that the vehicle can manage without lifting all wheels on inner side. The manoeuvre is typically on level
ground with high road friction and certain payload, loaded high.
The manoeuvre can typically be a double lane change, since the double triggers roll oscillations. The
severity measure of the manoeuvre can be the peak lateral acceleration, or a lane change width or lon-
gitudinal speed for given lane width.
If applying the function on articulated vehicles, it is often relevant to define the threshold as lifting all
inner wheels on a roll-stiff unit. Two units connected by a fifth wheel constitutes one roll-stiff unit,
since the fifth wheel is conceptually roll rigid.
The function is not relevant for cambering vehicles. These could rather roll-over inward in curve, trig-
gered by tyre loses lateral grip, which would be a completely different situation.

4.5.4 Step Steering Response *

Function definition: Step steering response is the response to a step in steering wheel angle measured in cer-
tain vehicle measures. The step is made from a certain steady state cornering condition to a certain steering wheel
angle. The response can be the time history or certain measures on the time history, such as delay time and over-
shoot.
§ In a simple model as the linear one-track model, a step from straight-line driving results in an initial
rotation around a point 𝑛 𝑙 ⁄( 𝑓 ) behind the vehicle CoG. For most vehilces, ≈
2
𝑓, 𝑛 𝑙 ≈ (𝐿⁄ ) ⁄ 𝑓 ≈ 𝐿⁄ , so the initial rotation appears somewhere close to rear
(𝐿⁄ ) ≈2 2

axle. In more detail: if < 𝑓 , we get 𝑛 𝑙 < , which means that will initially be directed
in same direction as 𝑓 , which means that rear axle rather helps than rotation. If smaller , the
the rear axle will help rather than .

4.5.4.1 Mild Step Steering Response

This section is to be compared with 4.5.4.2, which uses a more advanced model. In present section a less
advanced model will be used, which is enough for small steering steps.
The model used for single frequency stationary oscillating steering can also be used for other pur-
poses, as long as limited lateral accelerations. Most common interpretation is to make the steering step
from an initial straight-line driving. In reality, the step will be a quick ramp. Equation [4.50] allows an
explicit solution prediction of stationary oscillating steering, but also for step response.
Results for step steer to +3 deg are shown in Figure 4-81. Left diagram shows step steer from straight
line driving, while right diagram shows a step from steady state cornering with -3 deg steer angle. It
can be noted that, if the steering step is sized so that same steady state path radius, i.e. same steady
state yaw velocity, the understeered vehicle will require larger steering step, but it will respond
quicker in yaw velocity and lateral acceleration.

316
 Lateral Dynamics

Figure 4-81: Steering step response. Simulation with model from Equation [4.68].
4.5.4.1.1 Solution with Ansatz
Start from Equation [4.50] (assuming 𝑓 𝑤 0): [ ] 𝐀∙[ ]+𝐁∙ 𝑓

(0)
With steady state initial conditions: [ ] [ ] or [ ] 𝐀 ∙𝐁∙ 𝑓
(0)
̂ ̂ 𝜆1 ∙
] [ 2 ]] ∙ [𝑒 0 ]∙[ ] ⇒
∞
Make an ansatz: [ ] [ ] + [[
∞ ̂ ̂2 0 𝑒 𝜆2 ∙ 2
̂ ̂ 2 𝜆 ∙ 𝑒 𝜆1 ∙ 0
⇒ [ ] [[ ] [ ]] ∙ [ ]∙[ ]
̂ ̂2 0 𝜆2 ∙ 𝑒 𝜆2 ∙ 2

̂ ̂ 𝜆 ∙ 𝑒 𝜆1 ∙ 0
Insert: [[ ] [ 2 ]] ∙ [ ]∙[ ]
̂ ̂2 0 𝜆2 ∙ 𝑒 𝜆2 ∙ 2
̂ ̂ 𝜆1 ∙
] [ 2 ]] ∙ [𝑒 0 ] ∙ [ ]) + 𝐁 ∙
∞
𝐀 ∙ ([ ] + [[ 𝑓
∞ ̂ ̂2 0 𝑒 𝜆2 ∙ 2

Solve for each time function term (constant, 𝑒 𝜆1 ∙ and 𝑒 𝜆2 ∙ terms):

∞ ̂ ̂ 2 𝜆 0
[ ] 𝐀 ∙ 𝐁 ∙ 𝑓 𝑛 [[[ ] [ ]] [ ]] eig(𝐀)
∞ ̂ ̂2 0 𝜆2
The function ”eig” is identical to function ”eig” in Matlab. It is defined as eigenvalues and eigenvectors
for the matrix input argument. The eigenvalues are typically complex, which gives harmonic parts in
solution. The eigenvectors are also typically complex, so that the solution of physical variables be-
comes real.
∞ ̂ ̂
Initial conditions: [ ] [ ] + [[ ] [ 2 ]] ∙ [ ] ⇒
∞ ̂ ̂2 2
̂ ̂ ∞
⇒ [ ] [[ ] [ 2 ]] ∙ ([ ] [ ])
2 ̂ ̂2 ∞
The solution in summary:

317
 Lateral Dynamics

̂ ̂ 2 𝜆1 ∙
]] ∙ [𝑒 0 ]∙[ ]
∞
[ ] [ ] + [[ ] [
{ ∞ ̂ ̂2 0 𝑒 2∙
𝜆 2
𝜆1 ∙
+ ∙ 𝜆 ∙ ̂ ∙𝑒 ∙ + 𝜆2 ∙ ̂ 2 ∙ 𝑒 𝜆2 ∙ ∙ 2 + ∙
∞ ̂ ̂ 𝜆 0 [4.68]
where: [ ] 𝑨 ∙𝑩∙ 𝑓 and [[[ ] [ 2 ]] [ ]] eig(𝑨)
∞ ̂ ̂2 0 𝜆2
̂ ̂ ∞
and [ ] [[ ] [ 2 ]] ∙ ([ ] [ ])
2 ̂ ̂2 ∞

Another way to express or compute this is the “matrix exponential”, mentioned in 1.5.1.1.5.
4.5.4.1.2 Solution with Laplace transform
By Mats Jonasson, Volvo Cars and Vehicle Dynamics at Chalmers

Solution of same mathematical model as 4.5.4.1.1. But we use special case [ ] 𝟎.

Laplace transforms, 1.5.1.9, gives: 𝑠 ℒ ([ ]) [ ] ℒ (𝐀 ∙ [ ]+𝐁∙ 𝑓 (𝑡)) ⇒

⇒ 𝑠 ℒ ([ ]) 𝟎 𝐀 ∙ ℒ ([ ]) + 𝐁 ∙ ℒ ( 𝑓 (𝑡)) ⇒

⇒ (𝐀 𝑠 𝑰) ℒ ([ ]) 𝐁 ∙ ℒ ( 𝑓 (𝑡)) ⇒ ℒ ([ ]) ⏟(𝐀 𝑠 𝑰) 𝐁 ∙ ℒ ( 𝑓 (𝑡)) ⇒

𝑯( )

𝑓 𝑯(𝑠)
⇒ {ℒ(𝑠𝑡𝑒𝑝) 𝑠𝑡𝑒𝑝⁄𝑠} ⇒ ℒ ([ ]) ⏟(𝐀 𝑠 𝑰) 𝐁∙ ⇒ ℒ ([ ]) ∙ 𝑓 ⇒
𝑠 𝑠
𝑯( )
𝑯(𝑠)
⇒ [ ] ℒ ( ∙ 𝑓) ⇒
𝑠
inverse Laplace transform
𝑯( ) cos( 6 𝑡 + 7 )
⇒ [ ] ℒ ( )∙ 𝑓 ⋯{ e.g. using Matlab command } [ ] [ 3 ] 𝑒 5 [ ]
2 4 cos( 6 𝑡 + 8 )
ilaplace see .5.4.
where 2 ⋯ 8 are real parameters, expressed in model parameters 𝑓 𝑓 , initial condit-
ions and size of step in input variable 𝑓 . So, (𝑡 and
) (𝑡 can be computed and plotted, overlapping
)
in Figure 4-81. Note:
• The 𝑯(𝑠) is the Laplace transfer function. Each component of 𝑯(𝑠) is a surface over the com-
plex space [𝑅𝑒(𝑠) 𝐼 (𝑠)]. Compare with Fourier transfer function 𝑯( ) of which each compo-
nent is a one-dimensional curve over .
• Laplace transform can also handle other transient vehicle problems, such as engine torque step
or vertical road step. Laplace can also handle other tranisent disturbances, such as ramps.
• For stationary oscillations, Fourier transform is often enough and easier to use.

4.5.4.2 Violent Step Steering Response

This section is to be compared with 4.5.4.1 Mild Step Steering Response, which uses a model with lin-
ear tyre models without saturation. In present section, a more advanced model will be used, which
might be needed when the step steering is more violent.
Most common interpretation is to make the steering step from an initial straight-line driving. In real-
ity, the step will be a quick ramp. In simulations, an ideal step can be used.
The transients can easily be that violent that a model as Eq in 4.5.3 is needed. If ESC is to be simulated,
even more detailed models are needed (full two-track models, which are not presented in this com-
pendium). Anyway, if we apply a step steer to the model in 4.5.3, we can simulate as in Figure 4-82.

318
 df sfy sry

0.0

Lateral Dynamics
-0.2
0 1 2 3 4 5

Ffyw Fry
8000
vx
28
4000
26 vx[m/s]
0 1 2 3 4 5 0

0 1 2 3 4 5
df sfy sry
df[rad]
0.0 beta_deg wz_dps ay
15
sry[1] wz[deg/s]
-0.2 sfy[1] 10
0 1 2 3 4 5
ay[m/(s*s)]
5
Ffyw Fry
8000 Ffyw[N]
0
4000 Fry[N]
-5 beta[deg]
0
-10
0 1 2 3
time[s] 4 5 0 1 2 time[s]
3 4 5

Figure 4-82: Step

15
steer.
beta_deg 𝑒 onwz_dps
road wheels
ay at 00 km/h. Simulation with model in 4.5.3.

4.5.4.3 Steering Effort at High Speed *

10

Function definition:
5 Steering effort at high speed is the steering wheel torque (or subjectively assessed effort)
needed to perform
0
a certain avoidance manoeuvre at high road friction.
At higher vehicle speeds, the steering effort is normally less of a problem since unless really high steer-
ing wheel rate. Hence,
-5
steering wheel torque in avoidance manoeuvres in e.g. 70 km/h can be a rele-
vant requirement.-10
In these situations, the subjective assessment of steering effort can also be the
0 1 2 3 4 5
measure. Then, steering effort is probably assessed based on both steering wheel rate and steering
wheel torque.

4.5.5 Phase Portrait

The transient from one steady state to another is seen after the steps in Figure 4-81 and Figure 4-82.
Plotting several such transients as trajectories in a -diagram gives a phase portrait, which is a
graphical representation of how a vehicle stabilize itself or gets unstable. For transients that stays
within unsaturated tyre slip, the linear one-track model can be used, and the trajectories can then be
explicit time expressions using Eq[4.68]. This is exemplified in left part of Figure 4-83. Some states can
be confirmed stable already from this simple model. Simulations with higher fidelity model is exempli-
fied in right part of Figure 4-83. With that one can confirm some more stable areas.

319
 Lateral Dynamics

model valid w.r.t. <

oversteering
(transient)

small negative 𝑠𝑓 𝑠𝑓
large negative 𝑠

⁄𝑠
/𝑟
understeering

𝑠
(transient)

large negative 𝑠𝑓

small negative 𝑠 model valid w.r.t. <

𝑓 𝑓

Figure 4-83: Phase portrait for constant and constant steer angle 𝑓 . Bottom: Using simple
model in Eq [4.50]. Only black solid trajectories credible, since they are completely within model
validity. Upper: Using a model with larger validity. From Mats Jonasson, VCC.

4.5.6 Long Heavy Combination Vehicles High

Speed Functions
It is sometimes irrelevant to apply functions/measures from two axle vehicles on combinations of
units. This can be the case for passenger cars with a trailer, but it is even more obvious for long combi-
nations of heavy vehicles. Some typical measures for multi-unit combination vehicle are given in this
section.

4.5.6.1 Rearward Amplification, RA *

Function definition: Rearward Amplification for long heavy combination vehicles is the ratio of
the maximum value of the motion variable of interest (e.g. yaw velocity or lateral acceleration of the centre of grav-
ity) of the worst excited following vehicle unit to that of the first vehicle unit during a specified manoeuvre at a cer-
tain friction level and constant speed. From Reference [ (Kati, 2013)].
Figure 4-84 illustrates Rearward Amplification, RWA. RWA is defined for a special manoeuvre, e.g. a
certain lane change or step steer. RWA is the ratio of the peak value of yaw velocity or lateral accelera-
tion for the rearmost unit to that of the lead unit. This performance measure indicates the increased
risk for a swing out or rollover of the last unit compared to what the driver is experiencing in the lead
unit.

320
 Lateral Dynamics

Figure 4-84: Rearward amplification, P is peak value of motion variable of interest. From
(Kharrazi , 2012).
§ An alternative definition of RA is via the frequency response: 𝑅𝐴 |𝐻𝛿→𝜔1 ⁄𝐻𝛿→𝜔 |, where 𝐻 de-
nots transfer functions, so 𝐻 (𝑓), where 𝑓 is the frequency of an harmonic variation of . So, either we
see 𝑅𝐴 𝑅𝐴(𝑓) or 𝑅𝐴 max(𝑅𝐴(𝑓)). The 𝑓 where max appears is typically 0.4..0.6 Hz. The frequency
f
response definition is better (independent of amplitude) for measuring how difficult it is for driver
to keep the lateral stability of the last unit when driving straight ahead with small lateral disturbances.
It is a relevant measure both for human driver and virtual driver, such as LKA, 4.6.2.3.

4.5.6.2 High Speed Transient Off-tracking, HSTO *

Function definition: High speed transient off-tracking for long heavy combination vehicles is
the overshoot in the lateral distance between the paths of the centre of the front axle and the centre of the most se-
verely off-tracking axle of any unit in a specified manoeuvre at a certain friction level and a certain constant longi-
tudinal speed. From Reference [ (Kati, 2013)].
Figure 4-40 illustrates a manoeuvre where this Off-tracking can be defined. Off-tracking can be either
determined in a steady state turn or in a transient manoeuvre such as lane change. The steady state
version is described in 4.3.11.2.

4.5.6.3 Yaw Damping, YD *

Function definition: Yaw Damping for long heavy combination vehicles is the ratio of decay of the least damped
articulation joint’s angle of the combination vehicle during free oscillations excited by actuating the steering wheel in
a certain transient way, e.g. a step or single sine wave at a certain friction level. From Reference [ (Kati, 2013)].
Figure 4-85 illustrates Yaw Damping. It is the ratio of two subsequent peaks, 𝐷 | ̂ ⁄ ̂ + | where
=ordinal of coupling and 𝑗=ordinal of “half period” after steering disturbance has ended. Assuming
linear model, it should not matter which two half periods one uses and, then it can give less sensitive
definition to use several peaks, e.g. 𝐷 | ̂ ⁄ ̂ +5 |/5 . Sometimes other qualtities than articulation
angle in coupling points, such as yaw velocity of units, can be used.

321
 Lateral Dynamics

𝐷 2

Figure 4-85: Yaw damping, 𝐷 denotes damping ratio of the articulation joint. From (Kharrazi ,
2012).

4.5.6.4 Zero-Damping Speed *

Function definition: Zero-damping speed for long heavy combination vehicles is the longitudinal speed at which a
yaw disturbance leads to undamped response.
The measure can be calculated from a linear model as the longitudinal speed when the damping coeffi-
cient, for any mode involving yaw velocity, equals zero. If the measure is applied on a two-axle vehicle
it is same as critical speed.

4.6 Lateral Control Functions

Some control functions involving lateral vehicle dynamics will be presented briefly. There are more,
but the following are among the most well-established ones. But initially, some general aspects of lat-
eral control are given.

4.6.1 Lateral Control

Sensors available in production vehicles and used for lateral control are, generally those mentioned as
available for Longitudinal Control, see 3.5 plus some more:
Steering wheel sensors gives at least steering wheel angle, if the vehicle is equipped with ESC (which is
a legal requirement on many markets). Additionally, if the steering assistance is electrical, the steering
wheel torque can be sensed.
High specification modern vehicles also have environment sensors (camera, radar, etc) that can give
laterally interesting information, such as: Subject vehicle lateral position versus lane markers ahead
and other vehicles to the side or rear of subject vehicle.
As general considerations for actuators, one can mention that interventions with friction brake nor-
mally have to have thresholds, because interventions are noticed by driver and also generate energy
loss. Interventions with steering are less sensitive and can be designed without thresholds.

4.6.2 Lateral Control Functions

4.6.2.1 Electronic Stability Control, ESC *
Function definition: Electronic Stability Control directs the vehicle to match a desired yaw behaviour, when
the deviation from desired behaviour becomes above certain thresholds. ESC typically monitors vehicle speed, steer
angle and yaw velocity to calculate a yaw velocity error and uses friction brakes as actuator to reduce it.
There are 3 parts of ESC: Over-steer control, Under-steer control, Over-speed control. The actual con-
trol error that the vehicle reacts on is typically the yaw velocity error between a desired yaw velocity
and the sensed yaw velocity. Desired yaw velocity is calculated from a so-called reference model. Some
of today’s advanced ESC also intervenes on difference between desired and estimated side slip.

322
 Lateral Dynamics

Desired yaw velocity and side-slip is calculated using a reference model and a closed loop control on
the reference, see Figure 4-86. The reference model requires at least steer angle and longitudinal ve-
locity as input. The reference model can be either of steady state type (approximately as Eq [4.20] or
the [4.15]) or transient (approximately as Eq [4.50]). The vehicle modelled by the reference model
should rather be a desired vehicle than the controlled vehicle. Figure 4-86 does NOT show: Slide slip
control, Reduction of 𝑓 due to low friction detection, Coordination with Engine/Steering interven-
tions, Arbitration with Pedal/ACC/ABS braking. (For single unit vehicles, can be seen as a yaw
moment.)

Driver ReferenceModel, e.g. steady state:

𝑓 𝐿 +𝐾 𝑛

or transient: ClosedLoop
Brake
+ Controller, 𝑇𝑠 Brake
𝑓 Coordinator 𝑇𝑠 Actuator
𝑓 conceptually: Brake
Actuator Controlled
𝑓 𝑓 (select wheel) 𝑇𝑠
Actuator Vehicle
𝑓 𝑓 𝑛 𝑠𝑓
𝑡 𝑛
𝑛 𝑠
𝑠𝑓 + 𝑓 ⁄ 𝑓
𝑠 ⁄
Environ-ent

Figure 4-86: Concept for ESC.

The manoeuvre in Figure 4-23 shows an example of ESC interventions. But, in order to avoid too much
friction brake interventions; the reference model cannot be too different. Also, in order to avoid that
vehicle yaws more than its path curvature; the reference model cannot be much less understeered
than the controlled vehicle, which typically can be arranged by saturating lateral tyre forces on the
front axle in the reference model. This requires some kind of friction estimation, especially for low-mu
driving.
When controlling yaw via wheel torques, one can identify some different concepts such as direct and
in-direct yaw moment, see 4.3.7.6. For ESC there is also a “pre-cautious yaw control” which aims at re-
ducing speed, see 4.6.2.1.3. The coordination of wheel torques handles these aspects, which are far
from trivial. Often, a dual-request concept for each wheel is used, see Figure 4-86 and 4.6.2.1.4. A sim-
plest possible coordinator can be to request braking only on one wheel at the time:
4.6.2.1.1 Over-Steer Control
Over-steer control was the first and most efficient concept in ESC. When a vehicle over-steers, ESC will
typically coordinate wheel torques so that outer front wheel is braked. It can brake to deep slip levels
(typically -50%) since losing side grip on front axle is desired in an over-steer situation. More ad-
vanced ESC variants also brake outer rear, but less and not to same deep slip level.
For combination vehicles with trailers that have controllable brakes, also the trailer is braked to avoid
jack-knife effect, see upper part of Figure 4-87, or swing-out of the towed units.
4.6.2.1.2 Under-Steer Control
Under-steer control means that inner rear is braked when vehicle under-steers. This helps the vehicle
turn-in. This intervention is most efficient on low mu, because on high mu the inner rear wheel nor-
mally has very little normal load. Also, the slip levels are not usually as deep as corresponding over-
steer intervention, but rather -10%. This is because there is always a danger in braking too much on
rear axle, since it can cause over-steering. More advanced ESC variants also brake inner front. The
more wheels that are braked, the more similar the understeer intervention becomes with the function
in 4.6.2.1.3.
Reduce Oversteer dominating Reduce Understeer
dominating side slip
side slip

323
 Lateral Dynamics

Figure 4-87: ESC brake interventions when oversteer and understeer, on a tractor with trailer.
4.6.2.1.3 Over-Speed Control
Over-speed control is not always recognised as a separate concept, but as a part of under-steer control.
The actuation is that propulsion is reduced, or more than just inner rear wheels are braked. In this
text, we identify this as done to decrease speed, which has a positive effect later in the curve.
4.6.2.1.4 Wheel-Level Control
A pre-requisite for all controls mentioned above in 4.6.2.1 is that the wheel torque actuator primarily
responds to a torque request. However, one need to have another request channel to adjust the lateral
force margin; normally one uses a longitudinal slip request, 𝑠 in Figure 4-86 and Figure 4-88. The
slip request is generally used as a “safety net” to avoid lock-up the wheel too much; so, it is a “max |𝑠 |
request”. Typically, 𝑠 is 0. . . 0. , for braking, but at RSC interventions (see 4.6.2.2) the lateral grip
should be braked away, so a deeper slip request is then used, typically 50-70%.
TorqueRequest,
𝑇 Torque
controller
ESC & RSC piston ℎ ℎ
Actuator 𝑭𝒂𝒙 caliper
controllers Min 𝑭𝒂𝒙
𝑇𝑠

Control
𝑜 (or large 𝑠 )
Closed loop Wheel rotational
brake pads
wheel
attaches
LongSlipRequest,
𝑠 slip controller speed sensor rotor=disk
here

𝑠
hub 𝑠
vehicle state Calculate
estimator slip normal request RSC request
𝑠 0. . . 0. % 𝑠 0.6. . 0.5%

”Smart actuator”
Figure 4-88: Individual wheel control by friction brakes for ESC-type functions. What is a “smart
actuator” can depend on which function architecture that vehicle manufacturer and brake
supplier has agreed.
4.6.2.1.5 Other Intervention than Individual Wheel Brakes
Balancing with Propulsion per Axle
For vehicles with controllable distribution of propulsion torque between the axles, ESC can intervene
also with a request for redistribution of the propulsion torque. If over-steering, the propulsion should
be redistributed towards front and opposite for understeering.
Torque Vectoring
For vehicles with controllable distribution of propulsion torque between the left and right, ESC can in-
tervene also with a request for redistribution of the propulsion torque. If over-steering, the propulsion
should be redistributed towards inner side and opposite for understeering.
Steering Guidance
For vehicles with controllable steering wheel torque, ESC can intervene also with a request for addi-
tional steering wheel torque. The most obvious function is to guide driver to open up steering (coun-
ter-steer) when the vehicle over-steers. Such functions are on market in passenger cars today. Less ob-
vious is how to guide the driver when vehicle is under-steering.
4.6.2.1.6 ESC using Environment Information / ESC for the Virtual Driver
A prognosis of the future development of ESC like functions is that environment sensors can be used to
better predict what driver tries to do; presently ESC can only look at steering wheel angle.
Related to this, but still somewhat different, would be to utilize the automated driving development by
utilizing that a “virtual driver” can be much better predicted than a “manual driver”. So, a predictive
ESC control is more possible.

324
 Lateral Dynamics

4.6.2.2 Roll Stability Control, RSC *

Function definition: Roll Stability Control, RSC, prohibits vehicle to roll-over due to lateral wheel forces from
road friction. RSC uses friction brake as actuator.
The purpose of RSC is to avoid un-tripped roll-overs. The actuator used is the friction brake system.
When roll-over risk is detected, via lateral acceleration sensor (or in some advanced RSC implementa-
tions, also roll velocity sensor), the outer front wheel is braked. RSC can brake to deep slip levels (typi-
cally -70%..-50%) since losing side grip on front axle is positive in this situation. To lock the wheels
(slip ) is undesired since wheel rotational inertia makes it difficult to quickly regain lateral grip
when needed after the intervention.
On heavy vehicles, RSC intervenes earlier and similar to function described in 4.6.2.1.3 Over-Speed
Control. Future RSC might be developed towards also using steering, and potentially counteract also
some tripped roll-overs.

4.6.2.3 Lane Keeping Aid, LKA *

Function definition: Lane Keeping Aid steers the vehicle without driver having to steer, when probability for
lane departure is predicted as high. It is normally actuated as an additional steering wheel torque. Conceptually, it
can also be actuated as a steering wheel angle offset.
Lane Keeping Aid (or Lane Departure Prevention) has the purpose to guide the driver to keep in the
lane. Given the lane position from a camera, the function detects whether vehicle tends to leave the
lane. If so, the function requests a mild steering wheel torque (typically 1..2 Nm) in appropriate direc-
tion. Driver can easily overcome the additional torque. Function does not intervene if too low speed or
turning indicator (blinker) is used. There are different concepts whether the function continuously
should aim at keeping the vehicle in centre of lane, or just intervene when close to leaving the lane, see
Figure 4-89.
steering wheel
added torque on

Two concepts:
• Keep in centre of lane
• Intervene when about
to leave lane

lateral position
in lane
Figure 4-89: Two concepts for Lane Keeping Aid.

4.6.2.4 Lateral Collision Avoidance/Automatic Emergency

Steering, LCA/AES *
Function definition: Lateral Collision Avoidance/Automatic Emergency Steering, supports the
driver when he has to do late lateral obstacle avoidance, when probability for forward collision is predicted as high.
There are systems on the market for Automatic Emergency Brake, see 3.5. These do Longitudinal Colli-
sion Avoidance. Automotive industry also aims at AES/LCA functions, which would automatically steer
away laterally from an obstacle ahead of subject vehicle. The market introduction is cautious, since
many things can go wrong with such functions. However, the first systems on market triggers only if
driver initiates steering. Another would be to trigger on a first collision impact, see Reference (Yang,
2013), when driver is less capable of steering by himself. A future similar situation could be AES/LCA
functions active only during automated driving, when driver also is less likely to steer.

4.6.2.5 Automated Driving (AD)

Combining longitudinal control (such as ACC, in 3.5.2.2) with a lateral control (such as LKA, see
4.6.2.1) results in functionality which clearly approaches automated driving (AD). AD is a very general
expression but are sometimes interpreted as more specific, but specific in different ways depending on

325
 Lateral Dynamics

context. In a way, AD is already reality since there are vehicles on the road which can have ACC and
LKA active at the same time. On the other hand, AD can be seen as very futuristic, since completely
driverless vehicle which can operate in all situations is far from mass-production.
It is not obvious if AD will mean higher or lower requirements on vehicle dynamics. Some (of many
more!) examples of changes, relevant for vehicle dynamics are:
• The vehicle control can better predict the next few seconds of a virtual driver (AD algorithms)
than of a (human) driver. This can facilitate loss-of-grip functions, such as ABS & ESC.
• There will be new requirements on vehicle response on requests from the virtual driver, in
parallel with requirements on response on human drivers pedal and steering wheel operation.
• There will be new requirements on vehicle relative motion, relative to surrounding road and
traffic, such as lane edges and other road users. These will be in synergy or conflict to corre-
sponding requirement for absolute motion response.
• The motion actuation will have to be more redundant, since driver is less likely to take back
control quickly. Emergency functions to reach safe stop will need to work with partly faulty
sensors and actuators. Failures needs to be designed according to (ISO 26262, 2011-2012)
• The maximum speed for which the vehicle is designed can possibly be lower, since reduced
transport efficiency could be accepted if driver can do something else or is not needed at all.
• Estimation of Road friction, Controlling to Safe stop, Self-Diagnose, etc.

326
 Vertical Dynamics

5 VERTICAL DYNAMICS
5.1 Introduction
The vertical dynamics are needed since vehicles are operated on real roads, and real roads are not per-
fectly smooth. Also, vehicle can be operated off-road, where the ground unevenness is even larger.
The irregularities of the road can be categorized. A transient disturbance, such as a pothole or bump,
can be represented as a step input or ramp. Undulating surfaces like grooves across the road may be a
type of sinusoidal or other stationary oscillating (or periodic) input. More natural input like the ran-
dom surface texture of the road may be a random noise distribution. In all cases, the same mechanical
system must react when the vehicle travels over the road at varying speeds including doing manoeu-
vres in longitudinal and lateral directions.
The chapter is organised around the 3 complete vehicle functions: 5.5.1 Ride Comfort *, 5.5.2 Fatigue
Life *, and 5.5.3 Road Grip *. It is, to a larger extent than Chapters 3 and 4, organised with mathemati-
cal theory first followed by the vehicle functions. In Figure 5-1 shows the 3 main functions. It explains
the importance of the vehicle’s dynamic structure. The vehicle’s dynamic structure calls for a pretty
extensive theory base, described mainly in 5.2.
𝒛̈ Human
Vibrations of Ride
perception of
human
Road surface
Vehicle’s vibrations Comfort
𝒛 irregularities 𝒛𝒓 𝒛
speed Vehicle’s Stresses in 𝒛 𝒛
𝒓 Material Fatigue
𝒛𝒓 dynamic vehicle
Vehicle structure structure
fatigue Life
𝑭𝒓𝒛 created Contact
disturbances Compression 𝑭𝒓𝒛 Road
between tyres
of tyre Grip
and road
Figure 5-1: Different types of knowledge and functions in the area of vertical vehicle dynamics,
organised around the vehicle’s dynamic structure.
Models in this chapter focus the disturbance from vertical irregularities from the road, i.e. only the ver-
tical forces on the tyre from the road and not the forces in road plane. This enables the use of simple
models which are independent of exact wheel and axle suspension, such as pivot axes and roll centres.
Only the wheel stiffness rate (effective stiffness) and wheel damping rate (effective damping), see Fig-
ure 2-61, influence. This has the benefit that the chapter becomes relatively independent of previous
chapters, but it has the drawback that the presented models are not really suitable for studies of steep
road irregularities (which have longitudinal components) and sudden changes in wheel torque or tyre
side forces. Also, noise (>≈ 5𝐻 ) is not covered in this compendium.

5.1.1 References for this Chapter

• 2.3 Suspension System and “Chapter 21 Suspension Systems” in Ref (Ploechl, 2013)
• “Chapter 29 Ride Comfort and Road Holding” in Ref (Ploechl, 2013)

5.2 Stationary Oscillations Theory

Many vehicle functions in this chapter will be studied using stationary oscillations (cyclic repeating),
as opposed to transiently varying. An example of transiently varying quantity is a single step function
or single square pulse. A stationary oscillation can be as a sum of several harmonic terms, a multiple
frequency harmonic stationary oscillation. The special case with only one frequency is called a single
frequency harmonic stationary oscillation. See Figure 5-2 and Equation [5.1].

327
 Vertical Dynamics

𝑯𝒂𝒓 𝒐𝒏𝒊𝒄 𝒕𝒂𝒕𝒊𝒐𝒏𝒂𝒓𝒚 𝒐 𝒄𝒊𝒍𝒍𝒂𝒕𝒊𝒐𝒏 :

𝑆 𝑛 𝑒 𝑓𝑟𝑒𝑞𝑢𝑒𝑛 𝑦 ∶ (𝜉 ) ̂ ∙ cos( ∙ 𝜉 + )

[5.1]
𝑁
𝑀𝑢 𝑡 𝑝 𝑒 𝑓𝑟𝑒𝑞𝑢𝑒𝑛 𝑒𝑠 ∶ (𝜉 ) ∑ ̂ ∙ cos( ∙𝜉+ )
=
ℎ𝑒𝑟𝑒 𝜉 𝑠 𝑡ℎ𝑒 𝑛 𝑒𝑝𝑒𝑛 𝑒𝑛𝑡 𝑟 𝑒.
m
zz1 zz1 zz1 zz1
c

t t t t
F1
Step, as one example of Harmonic None-harmonic Random noise
transient (none- stationary stationary oscillation,
stationary oscillating) oscillation, single single frequency (or multiple
frequency random-frequency
other examples can be (or multiple-frequency harmonic
step down, square pulse, harmonic stationary stationary
ramp etc. oscillation) oscillation)
Figure 5-2: Different types of variables, both transient and stationary oscillating. The independent
variable 𝜉 can, typically, be either time or distance.
The most intuitive is probably to think of time as the independent variable, i.e. that the variation takes
place as function of time and that 𝜉 𝑡 in Equation [5.1]. However, for one specific road, the vertical
displacement varies with longitudinal position, rather than with time. This is why we can either do
analysis in time domain (𝜉 𝑡) and space domain (𝜉 𝑥).
Since the same oscillation can be described either as a function of 𝜉 ( (𝜉 )) or as a function of fre-
quency ( ̂ ̂ ( ) ), we can do analysis either in the independent variable domain (𝜉) or in fre-
quency domain ( ).
The four combinations of domains are shown in Figure 5-3.

time Fourier Transform (time) frequency

domain, t domain, 𝒐𝒓 𝒇

𝒙 𝒗𝒙 𝒕 𝒇 𝒇⁄𝒗𝒙
𝝎 𝝎⁄𝒗𝒙

space Fourier Transform spatial frequency

domain, x domain, 𝝎 𝒐𝒓 𝒇

Figure 5-3: Four domains and transformations between them.

Time and space domains are treated in 5.2.1 and 5.2.2. In addition to the domains, we also need to dif-
fer between discrete and continuous representations in both domains.

5.2.1 Time as Independent Variable

With time as independent variable, the frequency has the meaning of “how often per time”. Even so,
there are two relevant ways to measure frequency: angular (time) frequency, and (time) frequency.

328
 Vertical Dynamics

∙𝜋∙𝑓

[5.2]
ℎ𝑒𝑟𝑒 [𝑟 ⁄𝑠 ] 𝑛 𝑢 𝑟 (𝑡 𝑒) 𝑓𝑟𝑒𝑞𝑢𝑒𝑛 𝑦
𝑛 𝑓[ 𝑠 ⁄ 𝑜𝑠 𝑡 𝑜𝑛𝑠⁄𝑠] (𝑡 𝑒) 𝑓𝑟𝑒𝑞𝑢𝑒𝑛 𝑦

The time for one oscillation is called the period time. It is denoted 𝑇:

[5.3]
𝑇 ⁄𝑓 ∙ 𝜋⁄

5.2.1.1 Mean Square (MS) and Root Mean Square (RMS)

For a variable, z, we can define MS and RMS values as follows:
𝑉 𝑟 𝑒: (𝑡)
2
∫ ∙ 𝑡
𝑀𝑒 𝑛𝑆𝑞𝑢 𝑟𝑒: 𝑀𝑆 ( )

[5.4]
𝑡 𝑛

∫ 2 ∙ 𝑡
𝑅𝑜𝑜𝑡𝑀𝑒 𝑛𝑆𝑞𝑢 𝑟𝑒: 𝑅𝑀𝑆( ) √
𝑡 𝑛

If the variable is written as a single frequency harmonic stationary oscillation:

𝑉 𝑟 𝑒: ̂ ∙ cos( ∙ 𝑡 + )
2 ( ̂ ∙ cos( ∙ 𝑡 + ))2 ∙ 𝑡
∫ ∙ 𝑡 ∫
𝑀𝑒 𝑛𝑆𝑞𝑢 𝑟𝑒: 𝑀𝑆 ( )
𝑡𝑛 𝑡𝑛
2 [𝑡 sin( ∙ ∙ 𝑡) = 𝑡𝑛 sin( ∙ ∙ 𝑡 𝑛 )

[5.5]
] = 2
̂ ∙ + 4∙ ̂ ∙( + 4∙ ) ̂2
→
𝑡𝑛 𝑡𝑛 →∞

∫ 2 ∙ 𝑡 | ̂|
𝑅𝑜𝑜𝑡𝑀𝑒 𝑛𝑆𝑞𝑢 𝑟𝑒: 𝑅𝑀𝑆( ) √ √𝑀𝑆( )
𝑡 𝑛 √
If the variable is written as a multiple frequency harmonic stationary oscillation:
𝑁 𝑁

𝑉 𝑟 𝑒: ∑ ∑ ̂ ∙ cos( ∙𝑡+ )
= =
2
∫ 2
∙ 𝑡 ∫ (∑𝑁= ) ∙ 𝑡
𝑀𝑒 𝑛𝑆𝑞𝑢 𝑟𝑒: 𝑀𝑆 ( )
𝑡 𝑛 𝑡 𝑛
2
∫ (∑𝑁= ̂ ∙ cos( ∙𝑡+ )) ∙ 𝑡
⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗
𝑡𝑛 →
[5.6]

𝑡 𝑛
2 𝑁 𝑁
∫ ∑𝑁= ̂ ∙ (cos( ∙𝑡+ ))2 ∙ 𝑡 ̂ 2
⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗
𝑡𝑛 → ∑ 𝑀𝑆( ) ∑
𝑡 𝑛
= =
𝑁 2 𝑁 𝑁
̂ 2
𝑅𝑜𝑜𝑡𝑀𝑒 𝑛𝑆𝑞𝑢 𝑟𝑒: 𝑅𝑀𝑆( ) √𝑀𝑆( ) √∑ √∑ 𝑀𝑆 ( ) √∑(𝑅𝑀𝑆( ))
= = =

5.2.1.2 Power Spectral Density and Frequency Bands

So far, the frequency has been a discrete number of frequencies, 2 ⋯ 𝑁 . There are reasons to
treat the frequency as a continuous variable instead. The discrete amplitudes, ̂ ̂2 ⋯ ̂𝑁 , should then

329
 Vertical Dynamics

be thought of as integrals of a “continuous amplitude curve”, ̂ , where the integration is done over a
small frequency interval, centred around a mid-frequency, :
𝜔𝑖+𝜔𝑖+1
2
̂

[5.7]
+
̂ ∫ ̂ ∙ ̂ ( )∙ ̂ ( )∙ ⇒ ̂ ( )
𝜔 +𝜔𝑖
𝜔= 𝑖 1
2

We realize that the unit of ̂ has to be same as for , but per [rad/s]. So, if z is a displacement in [m], ̂
has the unit [m/(rad/s)]. Now, ̂ is a way to understand the concept of a spectral density. A similar
value, but more used, is the Power Spectral Density, PSD (also called Mean Square Spectral Density).
𝑆𝐷( ) which is a continuous function, while ̂ is a discrete function. That means that 𝑆𝐷( ) is fully
determined by a certain measured or calculated variable (𝑡), while ̂ depends on which discretiza-
tion (which or which ) that is chosen.

𝑀𝑆(𝑓 𝑡𝑒𝑟( (𝑡) ))

[5.8]
𝑆𝐷( (𝑡) ) ( )
ℎ𝑒𝑟𝑒 filter 𝑠 𝑛 𝑝 𝑠𝑠 𝑓 𝑡𝑒𝑟 𝑒𝑛𝑡𝑒𝑟𝑒 𝑟𝑜𝑢𝑛 𝑛 𝑡ℎ 𝑛 𝑡ℎ

PSD can also be defined with band width in time frequency instead of angular frequency. Eq [5.8] is the
same but replacing with 𝑓.
When the variable to study (z) is known and the band width is known, one often writes simply 𝑆𝐷( )
or ( ). G has the same unit as 2 , but per [rad/s] or per [oscillations/s]. So, if z is a displacement in
[m], G has the unit [ 2⁄(𝑟 ⁄𝑠)] or [ 2⁄( ⁄𝑠) 2
∙ 𝑠].
RMS is square root of the area under the PSD curve:
𝑁 𝑁 ∞

[5.9]
𝑅𝑀𝑆 ( ) √∑ 𝑀𝑆( ) √∑ ( )∙ √ ∫ ( )∙
= = 𝜔=

5.2.1.2.1 Differentiation of PSD

Knowing the PSD of a variable, we can easily obtain the PSD for the derivative of the same variable:

( ) 2
∙ ( ) [5.10]

5.2.1.3 Transfer Function

In a minimum model for vertical dynamics there is at least one excitation, often road vertical displace-
ment, , and one response, e.g. vertical displacement of sprung mass (=vehicle body), . A Transfer
function, 𝐻 𝐻 (𝑗 ∙ ), is the function which we can use to find the response, given the excitation:
𝑍( ) 𝐻( ) ∙ 𝑍 ( ) ⟺ ℱ( (𝑡)) 𝐻 ( ) ∙ ℱ( (𝑡))
[5.11]

∞
∙𝜔∙
ℎ𝑒𝑟𝑒 ℱ 𝑠 𝑡ℎ𝑒 𝑜𝑢𝑟 𝑒𝑟 𝑜𝑝𝑒𝑟 𝑡𝑜𝑟: 𝑍( ) ℱ( (𝑡)) ∫𝑒 ∙ (𝑡) ∙ 𝑡

𝐻 is complex, with magnitude, |𝐻| √( e(𝐻))2 + (Im(𝐻))2 , and phase, arg(𝐻( ))

arctan(Im(𝐻)⁄ e(𝐻)).
𝐴 𝑝 𝑡𝑢 𝑒: ̂ ( ) |𝐻( )| ∙ ̂ ( )
ℎ 𝑠𝑒: ( ) ( ) arg(𝐻 ( ))
[5.12]

ℎ𝑒𝑟𝑒 ∑ ̂( ) ∙ cos( ∙𝑡+ )

=

330
 Vertical Dynamics

Since there can be different excitations and responses in a system, there are several transfer functions.
To distinguish between those, a subscripting of 𝐻 is often used: 𝐻 𝑛→ 𝑝 𝑛 , which would be
𝐻 𝑟→ 𝐻 𝑝𝑙 𝑚 𝑛 → 𝑝 𝑛 𝑚 𝑝𝑙 𝑚 𝑛 in the example above. Other examples of relevant
transfer functions in vertical vehicle dynamics are:
• 𝐻 𝑝𝑙 𝑚 𝑛 → 𝑝 𝑛 𝑚 𝑙 [( ⁄𝑠 2 )⁄ ], see 5.5
𝑛
• 𝐻 𝑝𝑙 𝑚 𝑛 → 𝑝 𝑛 𝑛 𝑓 𝑚 𝑛 [ ⁄ ] , see 5.5.2
• 𝐻 𝑝𝑙 𝑚 𝑛 → 𝑓 [𝑁 ] , see 5.5.2
⁄
When transfer function for one derivative is found, it is often easy to convert it to another:

𝐻 1→ 𝑗∙ ∙ 𝐻 1→ 2

[5.13]
2
2
𝐻 1→ ̈2 𝑗∙ ∙ 𝑗 ∙ ∙ 𝐻 1→ 2 ∙𝐻 1→ 2
𝐻 1→ 2 3
𝐻 1→ 2 𝐻 1 → 3
The usage of the transfer function is, primarily, to easily obtain the response from the excitation, as
shown in Equation [5.12]. Also, the transfer function can operate on the Power Spectral Density,
PSD=G, as shown in the following:
2 2
𝑀𝑆( (𝑡) ) ( ̂ ( )) ⁄ (|𝐻( )| ∙ ̂ ( )) ⁄
( )

[5.14]
2
2 ( ̂ ( )) ⁄ 2
|𝐻 𝑟 → ( )| ∙ |𝐻 𝑟 → ( )| ∙ 𝑟
( )

Using Equation [5.9], we can then express 𝑅𝑀𝑆( ) (sprung mass), from knowing 𝑟
( ) (road):

[5.15]
∞ 2
𝑅𝑀𝑆( ) √∫𝜔= |𝐻 𝑟 → ( )| ∙ 𝑟
( )∙

5.2.2 Space as Independent Variable

All transformations, in this compendium, between time domain and space domain requires a constant
longitudinal speed, , so that:

[5.16]
𝑥 ∙𝑡+𝑥

The offset (𝑥 ) is the phase (spatial) offset (𝑥 ) is the correspondence to the phase angle ( ).
The corresponding formulas as given in Equations [5.2]..[5.13] can be formulated when changing to
space domain, or spatial domain. It is generally a good idea to use a separate set of notations for the
spatial domain. Hence the formulas are repeated with new notations, which is basically what will be
done in present section.
In space domain, the frequency has the common understanding of “how often per distance”. Even so,
there are two relevant ways to measure frequency: spatial angular frequency and spatial frequency.
∙𝜋∙𝑓
[5.17]

ℎ𝑒𝑟𝑒 [𝑟 ⁄ ] 𝑛 𝑢 𝑟 𝑠𝑝 𝑡 𝑓𝑟𝑒𝑞𝑢𝑒𝑛 𝑦
𝑛 𝑓 [ ⁄ 𝑜𝑠 𝑡 𝑜𝑛𝑠⁄ ] 𝑠𝑝 𝑡 𝑓𝑟𝑒𝑞𝑢𝑒𝑛 𝑦
The correspondence to period time is wavelength, denoted 𝜆:
[5.18]

𝜆[ ] ⁄𝑓 ∙ 𝜋⁄

Now, the basic assumption in Equation [5.16] and definitions of frequencies gives:

331
 Vertical Dynamics

[5.19]
∙ 𝑛 𝑓 ∙𝑓

The relation between the phase (spatial) offset (𝑥 ) and the phase angle ( ) is:

[5.20]
𝜆∙
𝑥
∙𝜋

5.2.2.1 Spatial Mean Square and Spatial Root Mean Square

In space domain, a variable, z, varies with 𝑉 𝑟 𝑒: (𝑥)
distance, x. We can define Mean Square ∫ 2
∙ 𝑥
and Root Mean Square values also in space 𝑀𝑒 𝑛𝑆𝑞𝑢 𝑟𝑒: 𝑀𝑆 ( )

[5.21]
𝑥 𝑛
domain. We subscript these with s for
space. ∫ 2 ∙ 𝑥
𝑅𝑜𝑜𝑡𝑀𝑒 𝑛𝑆𝑞𝑢 𝑟𝑒: 𝑅𝑀𝑆 ( ) √
𝑥 𝑛

Because 𝑥 is constant, the Mean Square and Root Mean Square will be the same in time and space
domain. If the variable is written as a single frequency harmonic stationary oscillation, these values
becomes as follows:
𝑉 𝑟 𝑒: ̂ ∙ cos( ∙ 𝑥 + 𝑥 )
̂2

[5.22]
𝑀𝑒 𝑛𝑆𝑞𝑢 𝑟𝑒: 𝑀𝑆 ( ) ⋯ 𝑀𝑆( )
| ̂|
𝑅𝑜𝑜𝑡𝑀𝑒 𝑛𝑆𝑞𝑢 𝑟𝑒: 𝑅𝑀𝑆 ( ) ⋯ 𝑅𝑀𝑆( )
√
If the variable is written as a multiple frequency harmonic stationary oscillation:
𝑁 𝑁

𝑉 𝑟 𝑒: ∑ ∑ ̂ ∙ cos( ∙𝑥+𝑥 )
= =
𝑁 2
̂
[5.23]
𝑀𝑒 𝑛𝑆𝑞𝑢 𝑟𝑒: 𝑀𝑆 ( ) ∑ 𝑀𝑆 ( )
=
𝑁
2
𝑅𝑜𝑜𝑡𝑀𝑒 𝑛𝑆𝑞𝑢 𝑟𝑒: 𝑅𝑀𝑆 ( ) √𝑀𝑆 ( ) √∑(𝑅𝑀𝑆( )) 𝑅𝑀𝑆 ( )
=

5.2.2.2 Spatial Power Spectral Density and Frequency Bands

A correspondence to Power Spectral Density in space domain is denoted 𝑆𝐷 in the following:
𝑀 (𝑓 𝑙 ( ( ) Ω 𝜆))
[5.24]

𝑆𝐷 ( (𝑥) 𝜆) 𝜆
Φ( )
where "filter" is a band pass filter centred around ω and with band width 𝑓

When the variable to study z is known and the band width is known, one often writes simply 𝑆𝐷 ( )
or Φ( ). The Φ has the same unit as 2 , but per [rad/m] or per [oscillations/m]. So, if z is a displace-
𝑚2 𝑚3 𝑚2
ment in [m], Φ has the unit [ ⁄𝑚
] or [ ⁄𝑚
3
].

332
 Vertical Dynamics

5.3 Road Models

In general, a road model can include ground properties such as coefficient of friction, damping/elastic-
ity of ground and vertical position. The independent variable is either one, along an assumed path, or
generally two, x and y in ground plane. In vertical dynamics in this compendium, we only assume ver-
tical displacement as function of a path. We use x as independent variable along the path, meaning that
the road model is: (𝑥). The function (𝑥) can be either of the types in
Figure 5-2. We will concentrate on stationary oscillations, which by Fourier series, always can be ex-
pressed as multiple (spatial) frequency harmonic stationary oscillation. This can be specialized to ei-
ther single (spatial) frequency or random (spatial) frequency. Hence, the general form of the road
model is multiple (spatial) frequencies:
𝑁

[5.25]
(𝑥) ∑ ̂ ∙ cos( ∙𝑥+𝑥 )
=

5.3.1 One Frequency Road Model

For certain roads, such as roads built with concrete blocks, a single (spatial) frequency can be a rele-
vant approximation to study a certain single wavelength. Also, the single (spatial) frequency road
model is good for learning the different concepts. A single (spatial) frequency model is the same as a
single wave length model (𝜆 ∙ 𝜋⁄ , from Equation [5.17]) and it can be described as:

[5.26]
(𝑥) ̂ ∙ cos( ∙ 𝑥 + 𝑥 )

5.3.2 Multiple Frequency Road Models

Based on the general format in Equation [5.25], we will now specialise to models for different road
qualities. In Figure 5-4, there are 4 types of road types defined. The 3 upper ones of those are also de-
fined as PSD-plots in Figure 5-5. The mathematical formula is given in Equation [5.27] and numerical
parameter values are given in Equation [5.28].
𝑤
𝛺 𝑀𝑆 ( 𝛺)
𝛷 𝛷(𝛺) 𝛷 ∙( )
𝛺 𝛺
m2
[5.27]
ℎ𝑒𝑟𝑒 𝛷 road severity [ ]
rad⁄m
w road waviness [ ]
𝛺 spatial angular frequency [rad⁄m]
𝛺 [rad/m]
Typical values are:
m2
6
ery good road: 𝛷 ∙ 0 []
rad⁄m
m2
Bad road ∶ 𝛷 0∙ 0 6 [ ]
[5.28]

rad⁄m
m2
ery bad road ∶ 𝛷 00 ∙ 0 6 [ ]
rad⁄m
The waviness is normally in the range of .. [ ]
where smooth roads have larger waviness than bad roads.
The decreasing amplitude for higher (spatial) frequencies (i.e. for smaller wavelength) can be ex-
plained by that height variation over a short distance requires large gradients. On micro-level, in the
granular level in the asphalt, there can of course be steep slopes on each small stone in the asphalt.

333
 Vertical Dynamics

These are of less interest in vehicle vertical dynamics, since the wheel dimensions filter out wave
length << tyre contact length, see Figure 2-58. Reference (ISO 8608) uses road waviness, w=2 for all
roads. Figure 5-5 is based on measurements on real roads, which shows that waviness actually varies
with road severity, 𝛷 . A certain road can be described with:
• 𝛺 ⋯ 𝛺𝑁
• ̂ ⋯ ̂𝑁
• 𝑥 ⋯ 𝑥 𝑁

Smooth Very Rough

The road condition is assessed as Smooth if The road condition is assessed as Very
at least 95% of the total distance is covered Rough if more than 5% of the total
on properly surfaced roads of good quality. distance is covered on extremely poor
Shorter distances on poor surfaces may roads or off-road.
arise.

Rough
The road condition is assessed as Rough if Cross Country
the road surface is of poor quality or if the The road condition is assessed as Cross
road is not properly maintained. Up to 5% of Country if a considerable amount of
the total distance may be covered on driving occurs in severe off-road
extremely poor roads or off-road. conditions.

Figure 5-4: Four typical road types. From (AB Volvo, 2011).

-2
Φ Φ ⁄
PHI=PHI0.*((OMEGA/OMEGA0).^(-waviness))
10
very rough very bad road
-3
rough bad road
10 smooth very good road
⁄𝑚
]

-4
𝑚2

10
𝑆𝐷 PHI /[m*m/(rad/m)]
in [

-5
10
Φ

-6
10

-7
10
m2 𝟔
very rough 𝟎 𝟏𝟎𝟎 𝟏𝟎 𝒘 𝟐 𝟏;
ad⁄m
m2
-8 rough 𝟎 𝟏𝟎 𝟏𝟎 𝟔 ad⁄m 𝒘 𝟐. 𝟏
10 m2
smooth 𝟎 𝟏 𝟏𝟎 𝟔 ad⁄m 𝒘 𝟏
𝟎 𝟏 rad⁄m for all;
-9
10
-1 0 1
10 10 in [𝑟 ⁄ ]→ 10
OMEGA /[rad/m]
𝜆 50 0 0 5

Figure 5-5: PSD spectra for the three typical roads in Figure 5-4.

334
 Vertical Dynamics

Number of frequency components, N, to select is a matter of accuracy or experience. The offsets,

𝑥 ⋯ 𝑥 𝑁 , can often be assumed to be zero. If phase is to be studied, as in Figure 5-5, a random gen-
eration of offsets is suitable. See also Reference (ISO 8608).
We can generate (𝑥) curves for the 3 road types in Figure 5-5 as shown in Figure 5-6. To generate
those plots, we have assumed different number of harmonic components (N in Equation [5.25]) and
also randomly generate the phase for each component (each 𝑥 ).
very rough
RoadQuality=1
0.05
N=10
zr /[m]

N=20
0 N=100

-0.05
0 1 2 3 4 5 6 7 8 9 10
x/[m]

rough
RoadQuality=2
0.05
N=10
zr /[m]

N=20
0 N=100

-0.05
0 1 2 3 4 5 6 7 8 9 10
x/[m]

smooth
RoadQuality=3
0.05
N=10
zr /[m]

N=20
0 N=100

-0.05
0 1 2 3 4 5 6 7 8 9 10
x/[m]

Figure 5-6: Road profiles, (𝑥), for the three typical roads in Figure 5-4.

5.3.2.1 Transfer Function from Road Spectrum in Spatial Do-

main to System Response in Time Domain
Since we assume constant longitudinal velocity, , the road spectrum can be transformed to the time-
frequency domain:
𝑀𝑆( ) 𝑀𝑆 ( 𝛺)
𝑟
( ) {𝑢𝑠𝑒: ∙ 𝛺}
∙ 𝛺
𝛺 𝑤
𝛺 𝑤 𝑀𝑆 ( 𝛺) 𝛷 ∙ (𝛺 )
[5.29]

{𝑢𝑠𝑒: 𝛷 ∙ ( ) }
𝛺 𝛺
𝑤
𝛷 ∙𝛺 𝑤
𝛷 ( ) 𝛷 𝑤 𝑤
∙ ∙ ∙
𝛺 𝑤∙ 𝛺 𝑤
𝛺 𝑤

Then, we can use Equation [5.14] to obtain the response :

𝛷0
[5.30]

2 2 𝑤 𝑤
( ) |𝐻 𝑟→
( )| ∙ ( ) |𝐻 𝑟 → ( )| ∙ 𝑤 ∙ ∙
𝑟
𝛺

Then we can use Equation [5.9] to obtain the RMS of the response :

335
 Vertical Dynamics

𝑁 𝑁
𝛷 𝑤
2
𝑤
𝑅𝑀𝑆 ( ) √∑ ( )∙ √ 𝑤 ∙ ∙ ∑|𝐻 𝑟 → ( )| ∙ ∙
𝛺
= =

[5.31]
or
∞ ∞
𝛷 𝑤
2
𝑤
𝑅𝑀𝑆( ) √ ∫ ( )∙ √ 𝑤 ∙ ∙ ∫ |𝐻 𝑟 → ( )| ∙ ∙
𝛺
𝜔= 𝜔=

5.4 1D Vehicle Models

“One-dimensional” refers to pure vertical motion, i.e. that the vehicle heaves without pitch and without
roll. The tyre is stiff and massless.
This can be seen as that the whole vehicle mass, m, is modelled as suspended by the sum of all wheels’
vertical forces, 𝑓𝑙 + 𝑓 + 𝑙 + . However, the model can sometimes be referred to as a
“quarter-car-model”. That is because one can see the model as a quarter of the vehicle mass, /4,
which is suspended by one of the wheel’s vertical force, . The exact physical interpolation of a quar-
ter car is less obvious, since one can argue whether the fraction ¼ of the vehicle mass is the proper
fraction or from which point of view it is proper. Using the fraction ¼ is as least debatable if the vehi-
cle is completely symmetrical, both left/right and front/rear.

5.4.1 1D Model without Dynamic dofs

“Without dynamic degree of freedom” refers to that the (axle) suspension is modelled as ideally stiff.
The model can be visualised as in Figure 5-7.

real vehicle: model: ̈

+ 𝑥

Figure 5-7: One-dimensional model without dynamic degree of freedom

The equations could be set up directly ( ∙ ̈ 𝑛 (𝑡) ), but the following equations
gives a formalism which will be useful for the more complex models in 5.4.2 and 5.4.3.

𝑬𝒒 𝒊𝒍𝒊𝒃𝒓𝒊 : ∙ ̈ + ∙
[5.32]

𝑪𝒐 𝒑𝒂𝒕𝒊𝒃𝒊𝒍𝒊𝒕𝒚: 𝑛
𝑬𝒙𝒄𝒊𝒕𝒂𝒕𝒊𝒐𝒏: (𝑡)

5.4.1.1 Response to a Single Frequency Excitation

Assume that the road has only one (spatial) frequency, i.e. one wavelength. Then the excitation is as
follows:

336
 Vertical Dynamics

∙π
(𝑥) ̂ ∙ cos( ∙ 𝑥 + 𝑥 ) ̂ ∙ cos ( ∙𝑥+𝑥 )
{ 𝜆 }⇒
𝑥 ∙𝑡
𝐴𝑠𝑠𝑢 𝑒 𝑥 0

[5.33]
∙π∙
⇒ (𝑡) ̂ ∙ cos ( ∙ 𝑡) ̂ ∙ cos(ω ∙ 𝑡) ⇒
𝜆
∙π∙ ∙π∙
⇒ (𝑡) ∙ ̂ ∙ sin ( ∙ 𝑡) ω ∙ ̂ ∙ sin(ω ∙ 𝑡) ⇒
𝜆 𝜆
2
∙π∙ ∙π∙
⇒ ̈ (𝑡) ( ) ∙ ̂ ∙ cos ( ∙ 𝑡) ω2 ∙ ̂ ∙ cos(ω ∙ 𝑡)
𝜆 𝜆
Insertion in the model in Equation [5.32] (with eliminated ) gives directly the solution:

(𝑡) ∙ + (𝑡) ∙ + Δ̂ ∙ cos(ω ∙ 𝑡)

[5.34]
{
(𝑡) (𝑡) ̂ ∙ cos (ω ∙ 𝑡) 𝑛 ̈ (𝑡) ̈ (𝑡) ̂ ∙ cos(ω ∙ 𝑡)
ℎ𝑒𝑟𝑒 ̂ 2
∙ω ∙ ̂ 𝑛 ̂ 2
ω ∙ ̂

We can identify the magnitude of the transfer functions 𝐻. The negative sign in Equation [5.35] means
180 degrees phase shift:
ℱ( )
𝐻 𝑟→ {𝐻 𝑟 → } { (𝑡) (𝑡)} +𝑗∙0
ℱ( )

[5.35]
𝐻 𝑟→ 𝑟 {𝐻 𝑟 → 𝑟 𝐻 𝑟→ 𝐻 → 𝐻 𝑟→ } 0+𝑗∙0
𝐻 𝑟→ ̈ {𝐻 𝑟 → ̈ (𝑗 ∙ )2 ∙ 𝐻 𝑟 → ω2 ∙ 𝐻 𝑟 → } ω2 + 𝑗 ∙ 0
𝐻 𝑟 → 𝑟𝑧 {𝐻 𝑟 → 𝑟𝑧 ∙ 𝐻 𝑟→ ̈ } ∙ ω2 + 𝑗 ∙ 0
The motivation to choose exactly those transfer functions is revealed later, in 5.5, 5.5.2 and 5.5.2. For
now, we simply conclude that various transfer functions can be identified and plotted. The plots are
found in Figure 5-8. Numerical values for m and 𝜆 are chosen.
5.4.1.1.1 Example Analysis
An example of how to use Figure 5-8 is: A certain road has amplitude of 1 cm ( ̂ 0.0 ). The vehi-
cle drives on it with a longitudinal velocity of 50 km/h ( ≈ 4 /𝑠 ̂ ≈ .8𝐻 ):
• |𝐻 𝑟 → ̈ ( )| ≈ 05 ⇒ | ̂| 05 ∙ ̂𝑟 05 ∙ 0.0 .05 ⁄𝑠 2 . From this we can calcu-
late 𝑅𝑀𝑆( ̈ 𝑠 ) | .05| ⁄√ ≈ . 6 ⁄𝑠 2 . The RMS value of acceleration will later be related to
ride comfort, see 5.5.
• |𝐻 𝑟 → 𝑢 ( )| 0 is the transfer function to deformation of suspension, which later
will be related to fatigue life, see 5.5.2. The model in 5.4.1 is not good for measuring fa-
tigue, since the ̂ ̂ is intrinsically zero because of no compliance between un-
sprung and sprung mass.
• |𝐻 𝑟 → 𝑟 ( )| ≈ 487000 ⇒ Δ̂ 𝑟 487000 ∙ ̂𝑟 487000 ∙ 0.0 4870 𝑁 . If Δ̂ > ∙
≈ 6000 𝑁, the model is outside its validity region, because it would require pulling forces
between tyre and road. If changing to ̂ 0. , this limit is defined by |𝐻 𝑟 → 𝑟 ( )| >≈
6
.6 05 [𝑁⁄ ], which is used to examplify the validity limit in Figure 5-8;
.
model becomes invalid for >≈ 00 ⁄𝑠. The variation in tyre road contact force will be re-
lated to road grip, see 5.5.2.
The phases for the studied variables can be found in Equation [5.35]. With this model, the phases be-
come constant and ±90 𝑒 .

337
 Vertical Dynamics

7
Driving at road with wave-length, lambda = 5 [m].
10
invalid if ̂ 0. [ ]

6
10
𝐻 𝑟→ 𝑟𝑧
.6 05
5
10

4
10
H_zr_zs (==1)
abs(H)

3
10 H_zr_zr-zs (==0)
H_zr_derderzs
2 H_zr_Frz
10
m = ms+mu = 1600[kg];
1
10 f=1 Hz at vx=5[m/s]; f=10 Hz at vx=50[m/s]

0
10

-1
10
0 1 10 Hz 2 3
10 1 Hz 10 10 10
vx [m/s]
Figure 5-8: Transfer functions from model in Figure 5-7, excited with single frequencies.

5.4.1.2 Response to a Multiple Frequency Excitation

Using Eq [5.31], Eq [5.42] and values for road type “rough” in Figure 5-5, we can conclude:
𝛷0 ∞ 2
𝑅𝑀𝑆( ) √𝛺 ∙ 𝑤 ∙ ∫𝜔= |𝐻 𝑟 → ( )| ∙ 𝑤 ∙

[5.36]
𝑤
0

∙ 6 ∞ 2
√ ∙ 2.5 ∙ ∫𝜔= |𝐻 𝑟 → ( )| ∙ 2.5 ∙

For now, we simply note that it is possible to calculate this (scalar) RMS value for each vehicle speed
over the assumed road. In corresponding way, an RMS value can be calculated for any of the oscillating
variables, such as ̈ , and . We will come back to Equation [5.36] in 5.5.1.2.

5.4.2 1D Model with 1 Dynamic dof

“With 1 dynamic dof” refers to that the axle suspension is modelled as a linear spring and linear (vis-
cous) damper in parallel. Compared to the model in 5.4.1, the tyre is still stiff, but the unsprung parts
are assigned a mass . The model can be visualised as in Figure 5-9.

real vehicle model

̈

z
̈

+ x
c d
py
̈

̈ ̈
Figure 5-9: One-dimensional model with 1 dynamic degree of freedom

338
 Vertical Dynamics

The mathematical model becomes as follows:

Equilibrium sprung mass: ∙ ̈ ∙ 0
Equilibrium unsprung mass: ∙ ̈ ∙ 0

[5.37]
Constitution: ∙( )+ ∙( )+ ∙
Compatibility:
Excitation: (𝑡)

5.4.2.1 Response to a Single Frequency Excitation

Eliminating , and gives:
∙ ̈ ∙ ( (𝑡) ) + ∙ ( (𝑡) )

[5.38]
⏟ ( + )∙ ∙ ( (𝑡) ) + ∙ ( (𝑡) )+ ∙ ̈ (𝑡)
Δ 𝑟𝑧

Note that since we measure and from the static equilibrium, the static load, ∙ , disappears
when constitution is inserted in equilibrium. The Δ denotes the variation from static contact force
between road and tyre.
Assume that the road has only one (spatial) frequency, i.e. one wavelength. Then the excitation is as in
Equation [5.26], in which we assume 𝑥 0. So, we can insert (𝑡) ̂ ∙ cos(ω ∙ 𝑡) ⇒ (𝑡)
̂ ∙ sin(ω ∙ 𝑡) ⇒ ̈ (𝑡) 2
̂ ∙ cos(ω ∙ 𝑡) in Equation [5.38] and solve it for (𝑡) and Δ (𝑡) with
trigonometry or Fourier transform.
In 4.4.3.1.1, we applied Fourier transform on the linear explicit form model. To show a slightly other
way, we do not rewrite to explicit form, but apply Fourier transform on Eq [5.38] directly:

∙ ( ω2 ∙ ℱ( ))

[5.39]
∙ (ℱ( ) ℱ( )) + ∙𝑗∙ ∙ (ℱ( ) ℱ( ))
ℱ( 𝑟 ) ∙ (ℱ( 𝑟 ) ℱ( 𝑠 )) + ∙𝑗∙ ∙ (ℱ( 𝑟 ) ℱ( 𝑠 )) 𝑢 ω ∙ ℱ( 𝑟 )

From this, we can then solve for the transfer functions:

ℱ( ) +𝑗∙ ∙
𝐻 𝑟→
ℱ( ) +𝑗∙ ∙ ω2

[5.40]
ℱ( ) ℱ( )
𝐻 𝑟 → 𝑟𝑧 ( +𝑗∙ ω2 ) ( + 𝑗 ∙ )∙
ℱ( ) ℱ( )
( +𝑗∙ 2
ω ) ( +𝑗∙ ) ∙ 𝐻 𝑟→
We can elaborate further with Eq [5.40]:
̂ + ∙ ∙ω | 1| √ + ∙ω
Amplitude: |𝐻 𝑟 → | |( | {| 1 | }
̂𝑟 𝑚∙ω2 )+ ∙ ∙ω 2 | 2| √( ∙ω ) + ∙ω
[5.41]
+ ∙ ∙ω
Phase: ( ) ( ) arg ( 𝑚∙ω2 + + ∙ ∙ω
) {arg ( ) 1
arg( ) arg( 2 )}
2
∙ ∙
arg( + 𝑗 ∙ ∙ ω) arg( ∙ ω2 + + 𝑗 ∙ ∙ ω) ⋯ arctan (
∙ ∙ + ∙
)
Equation [5.13] now allows us to get the magnitudes of the other transfer functions as well:
𝐻 𝑟→ from Equation [5.40]
𝐻 𝑟→ 𝑟
𝐻 𝑟→ 𝑟 𝐻 𝑟→ 𝐻 𝑟→
2
𝐻 𝑟→ ̈ ∙ 𝐻 𝑟→
̈ + ̈ +
[5.42]

𝐻 𝑟→ { }
𝑟𝑧 ̈ + ∙( )+ ∙( )
𝐻 𝑟→ ̈𝑟 + ∙ (𝐻 𝑟 → 𝑟
𝐻 𝑟→ ) + ∙ (𝐻 𝑟 → 𝑟
𝐻 𝑟→ )
(𝑗 ∙ )2 + ( + 𝑗 ∙ ∙ ) ∙ (𝐻 𝑟 → 𝑟
𝐻 𝑟→ )
2
+( +𝑗∙ ∙ )∙( 𝐻 𝑟→ )
The motivation to choose exactly those transfer functions is revealed later, in 5.5. Some of those mag-
nitudes are easily expressed in reel (non-complex) mathematics using Equation [5.41]:

339
 Vertical Dynamics

+ ∙ω
|𝐻 𝑟 → | √
( ∙ω ) + ∙ω

[5.43]
2
+ ∙ω
|𝐻 𝑟 → ̈ | ∙√
( ∙ω ) + ∙ω

5.4.2.1.1 Example Analysis

The transfer functions in Equation [5.41] are plotted in Figure 5-10 and Figure 5-11. Numerical values
for m and 𝜆 have been chosen. If we use Figure 5-10 as the example in 5.4.1:
• Ride comfort related: |𝐻 𝑟 → ̈ ( )| ≈ 0 ⇒ | ̂| 0 ∙ ̂𝑟 0 ∙ 0.0 . 0 ⁄𝑠 2 .
From this we can calculate 𝑅𝑀𝑆( ̈ 𝑠 ) | . 0| ⁄√ ≈ 0.8485 ⁄𝑠 2 .
• Fatigue life related: |𝐻 𝑟 → 𝑟 ( )| ≈ . ⇒ | ̂ ̂ | . ∙ ̂𝑟 . ∙ 0.0
0.0 . .
• Road grip related: |𝐻 𝑟 → 𝑟 ( )| ≈ 59795 ⇒ | 𝑟 | 59795 ∙ ̂𝑟 59795 ∙ 0.0
598 𝑁 .

1 Hz 10 Hz 100 Hz
/[ /𝑠]
Figure 5-10: Transfer functions for amplitudes from model in Figure 5-9, excited with single
frequencies. Thin lines are without damping. Notation: 𝐻 → is denoted H_a_b.

340
 Vertical Dynamics

/[ /𝑠]

Figure 5-11: Transfer functions for phase delays from model in Figure 5-9, excited with single
frequencies.
We compare these numbers with the corresponding numbers for the simpler model in 5.4.1. The com-
fort is better. The fatigue life and road grip have become more realistic.
Figure 5-10 also shows the curves for the undamped system (d=0). The highest peaks appear at ap-
proximately 5. .6 m/s. This corresponds to the speed where the natural (=undamped) eigen fre-
quency appears ( 𝜆∙𝑓 𝜆∙ ⁄( ∙ 𝜋) 𝜆 ∙ √ ⁄ ⁄( ∙ 𝜋) ≈ 5.5 /𝑠).
Figure 5-11 shows the phase angles for the different responses.
5.4.2.1.2 § Solution with Trigonometry
The purpose with 5.4.2.1.2 is that it might help some readers to better/intuitively understand what the
previously used, more efficient, Fourier transform method does.
One way to solve the mathematical model in Eq [5.37] is to assume a real solution, insert the assump-
tion and find expressions for the coefficients in the assumption. Assume such solution (and that 𝑥 0
in Equation [5.26]): ̂ cos( ∙ 𝑡). Insertion in 1st scalar equation in Eq [5.37] gives:
∙ ̈ + ∙ + ∙ ̂ ∙ ( ∙ cos( ∙ 𝑡) ∙ ∙ sin( ∙ 𝑡))
Assumed solution:
̂ ∙ cos( ∙ 𝑡 ) ⇒
⇒ {𝑢𝑠𝑒: cos( ) cos ∙ cos + sin ∙ sin } ⇒
⇒ ̂ ∙ [cos( ∙ 𝑡) ∙ cos + sin( ∙ 𝑡) ∙ sin ] ⇒
⇒ ̂ ∙ ∙ [ sin( ∙ 𝑡) ∙ cos + cos( ∙ 𝑡) ∙ sin ] ⇒
⇒ ̈ ̂ ∙ 2 ∙ [cos( ∙ 𝑡) ∙ cos + sin( ∙ 𝑡) ∙ sin ]
Insertion:
∙ ̂ ∙ 2 ∙ [cos( ∙ 𝑡) ∙ cos + sin( ∙ 𝑡) ∙ sin ] +
+ ∙ ̂ ∙ ∙ [ sin( ∙ 𝑡) ∙ cos + cos( ∙ 𝑡) ∙ sin ] +
+ ∙ ̂ ∙ [cos( ∙ 𝑡) ∙ cos + sin( ∙ 𝑡) ∙ sin ]
̂ ∙ ( ∙ cos( ∙ 𝑡) ∙ ∙ sin( ∙ 𝑡)) ⇒
𝒄𝒐 𝒕𝒆𝒓 : ∙ ̂ ∙ 2 ∙ cos + ∙ ̂ ∙ ∙ sin + ∙ ̂ ∙ cos ̂ ∙
⇒ { ⇒
𝒊𝒏 𝒕𝒆𝒓 : ∙ ̂ ∙ 2 ∙ sin ∙ ̂ ∙ ∙ cos + ∙ ̂ ∙ sin ̂ ∙ ∙

341
 Vertical Dynamics

∙ ∙ 3
arctan ( 2
)
∙ ∙ 2+ 2 ∙ 2
⇒
̂ ∙
2
|𝐻 𝑟→ 𝑠
|
{ ̂ ( ∙ ) ∙ sin + ∙ ∙ cos
∙π∙ 𝑥
ℎ𝑒𝑟𝑒
𝜆
We have identified |𝐻 𝑟 → ( )|, which can be compared to |𝐻 𝑟 → ( )| in Eq [5.35]. The other trans-
fer functions in Eq [5.35] are more difficult to derive using the method with trigonometry.

5.4.3 1D Model with 2 Dynamic dofs

The expression “2 dynamic dofs” refers to that both elasticity between road and wheel (unsprung
mass) as well as between wheel (unsprung mass) and sprung mass is modelled. The model can be vis-
ualised as in Figure 5-12. No damping is modelled in tyre (in parallel with elasticity ) because it is
generally relatively low.

real vehicle model

̈
zs
Fsz Fsz
z zs
cs ds
+ x Fsz
zu
py zu
̈ Fsz
ct dt
Frz Frz
zr zu

zr
Frz
Figure 5-12: One-dimensional model with two dynamic degrees of freedom
The corresponding mathematical model becomes as follows:
Equilibrium:
∙ ̈ ∙
∙ ̈ ∙
[5.44]

Constitution (displacements counted from static equilibrium):

∙( )+ ∙( )+ ∙
∙( )+ ∙( )+( + )∙
Excitation: (𝑡 )
The same can be formulated with matrices and Fourier transforms:

342
 Vertical Dynamics

0 ̈ 0 0
[ ]∙[ ]+[ ]∙[ ]+[ + ]∙[ ] [ ]∙ +[ ]∙
0 ̈ +
⇒
̈
⇒ 𝑴 ∙ [ ] + 𝑫 ∙ [ ] + 𝑪 ∙ [ ] 𝑫 𝒓 ∙ + 𝑪𝒓 ∙ ⇒
̈

[5.45]
ℱ( ) ℱ( ) ℱ( )
⇒ 𝑴∙( 2 ∙[ ]) + 𝑫 ∙ (𝑗 ∙ ∙ [ ]) + 𝑪 ∙ [ ]
ℱ( ) ℱ( ) ℱ( )
𝑫𝒓 ∙ (𝑗 ∙ ∙ ℱ ( )) + 𝑪𝒓 ∙ ℱ ( ) ⇒
ℱ( )
⇒ ( 2 ∙ 𝑴 + 𝑗 ∙ ∙ 𝑫 + 𝑪) ∙ [ ] (𝑗 ∙ ∙ 𝑫𝒓 + 𝑪𝒓 ) ∙ ℱ ( )
ℱ( )

5.4.3.1 Response to a Single Frequency Excitation

We can find the transfer functions via Fourier transform, starting from Eq [4.56]:

[5.46]
𝐻 → ℱ( ) 2
[ 𝑟 ] [ ]∙ ( ∙𝑴+𝑗∙ ∙ 𝑫 + 𝑪) ∙ (𝑗 ∙ ∙ 𝑫 𝒓 + 𝑪𝒓 )
𝐻 𝑟→ 𝑢 ℱ( ) ℱ( )

This format is very compact, since it includes both transfer functions for amplitude and phase. For nu-
merical analyses, the expression in Eq [5.46] is explicit enough, since there are tools, e.g. Matlab, which
do numerical matrix inversion and complex mathematics. Symbolic solution is rather lengthy, but one
can use symbolic tools, e.g. Mathematica or Matlab Symbolic Toolbox.
Expression in real (without phase information) can be derived, see Eq [5.47].
2
√( ∙ ∙ ∙ 2 )2 +( ∙( ∙ + ∙ ))
|𝐻 𝑟 → ̈ | 2
∙
√𝐴2 + 𝐵2
∙ √( ∙ 2 )2 + ( ∙ 3 )2

[5.47]
|𝐻 𝑟 → 𝑢
|
√𝐴2 + 𝐵2
√( ∙ ∙ 4+ 2 ∙( + )∙ )2 + ( 3 ∙( + )∙ )2
|𝐻 𝑟 → 𝑟 𝑢
|
√𝐴2 + 𝐵2
4 2
𝐴 ∙ ∙ ∙( ∙ + ∙ + ∙ + ∙ )+ ∙
3
𝐵 ∙( ∙ + ∙ + ∙ ) ∙( ∙ + ∙ )
With 0, the solutions (with phase information, i.e. complex) becomes as follows:
( ∙ ω2 + 𝑗 ∙ ∙ω+ + )∙
( + 𝑗 ∙ ∙ ω)2
∙ ω2 + 𝑗 ∙ ∙ω+ +
ℱ( ) ∙ ω2 + 𝑗 ∙ ∙ ω +
𝐻 𝑟→
ℱ( ) +𝑗∙ ∙ω
[5.48]

ℱ( )
𝐻 𝑟→ 𝑢
ℱ( ) ( + 𝑗 ∙ ∙ ω)2
∙ ω2 + 𝑗 ∙ ∙ω+ +
∙ ω2 + 𝑗 ∙ ∙ ω +
∙π∙ 𝑥
where 𝜆

Equation [5.13] now allows us to get the magnitudes of the other transfer functions as well:

343
 Vertical Dynamics

𝐻 𝑟→ {use Eq [5.48]}
𝐻 𝑟→ 𝑢 {use Eq [5.48]}
𝐻 𝑟→ 𝑟 𝑢
𝐻 𝑟→ 𝑟 𝐻 𝑟→ 𝑢
𝐻 𝑟→ 𝑢
;
𝐻 𝑟→ 𝑢 𝐻 𝑟→ 𝑢 𝐻 𝑟→ 𝑠
;

[5.49]
𝐻 𝑟→ ̈ ∙ 𝐻 𝑟→ 𝑠 ;
𝐻 𝑟 →Δ 𝑧
{Δ ∙( )+ ∙( )}
∙ (𝐻 𝑟 → 𝑢 𝐻 𝑟 → ) + ∙ 𝑗 ∙ ∙ (𝐻 𝑟 → 𝑢 𝐻 𝑟 → )
( +𝑗∙ ∙ ) ∙ (𝐻 𝑟 → 𝑢 𝐻 𝑟 → );
𝐻 𝑟 →Δ 𝑟𝑧
{Δ ∙( )} ∙ (𝐻 𝑟 → 𝑟 𝐻 𝑟 → 𝑢 ) ∙( 𝐻 𝑟→ 𝑢
);
The transfer functions in Equation [5.49] are plotted in Figure 5-13.
5.4.3.1.1 Example Analysis
If we use Figure 5-13 as the example in 5.4.1:
• Ride comfort related: |𝐻 𝑟 → ̈ ( )| ≈ ⇒ | ̂| ∙ ̂𝑟 ∙ 0.0 . ⁄𝑠 2 .
From this we can calculate 𝑅𝑀𝑆( ̈ 𝑠 ) | . | ⁄√ ≈ 0.8697 ⁄𝑠 2 .
• Fatigue life related: |𝐻 𝑟 → 𝑢 ( )| ≈ . 4 ⇒ | ̂ ̂ | . 4 ∙ ̂𝑟 . 4 ∙ 0.0
0.0 4 . 4 .
• Road grip related: |𝐻 𝑟 →Δ 𝑟 ( )| ≈ 77470 ⇒ | 𝑟 | 77470 ∙ ̂𝑟 77470 ∙ 0.0
775 𝑁 .
7
Driving at road with wave-length, lambda = 5 [m].
10

6
10

5
10
ms = 1415[kg]; cs = 68 [kN/m]; ds = 8[kN/(m/s)];
4
10 mu = 185[kg]; ct = 676 [kN/m];
f=1 Hz at vx=5[m/s]; f=10 Hz at vx=50[m/s]
abs(H)

3
10

2
10

1
10 H_zr_zs
H_zr_zu-zs
0
10 H_zr_derderzs
H_zr_Frz
-1
10
0 1 10 Hz 2 3
10 10 10 10
1 Hz vx [m/s]
Figure 5-13: Transfer functions for amplitudes from model in Figure 5-12, excited with single
frequencies.
This analysis can be compared with the analysis in 5.4.2.1.1. Ride comfort and fatigue does not change
a lot, but road grip does. This indicates that the more advanced model is only needed for road grip
evaluation.
Figure 5-14 shows the phase angles for the different responses.

344
 Vertical Dynamics

200

150

100

50

arg(H)
0

-50

-100
H_zr_zs
H_zr_zu-zs
-150
H_zr_derderzs
H_zr_Frz
-200
0 1 2 3
10 10 10 10
vx [m/s]
Figure 5-14: Transfer functions for phase delays. Same model and data as in Figure 5-13.
Figure 5-15, shows the amplitude gains for the corresponding un-damped system. Natural frequencies
are around 5 m/s and 50 m/s. These two speeds correspond to frequencies /𝜆, i.e. approxi-
mately 1 Hz and 10 Hz. The 1Hz frequency is an oscillation mode where the both masses move in
phase with each other, the so called “heave mode” or “bounce mode”. The 10 Hz frequency comes from
the mode where the masses are in counter-phase, the so called “wheel hop mode”. In the wheel hop
mode, the sprung mass is almost not moving at all. We will come back to these modes in 5.4.4.
7
10

6
10

5
10

4
10 H_zr_zs
H_zr_zu-zs
abs(H)

3
10 H_zr_derderzs
H_zr_Frz
2
10

1
10

0
10

-1
10
0 1 2 3
10 10 10 10
vx [m/s]
Figure 5-15: Un-damped transfer functions. Same model and data as in Figure 5-13, except 0.

5.4.3.2 § Active Damping using Skyhook Control

Contribution from Waltteri Koskinen, student at Tampere University, Finland

A traditional damper follows 𝑝 𝑝 , where subscript cpr means compression. Since

comes from flow losses of oil in the damper, will always be > 0. Taking into account that pressure
losses are proportional to hydraulic flow in square, a better model is 𝑝 2 sign( 𝑝 )
2
𝑝

345
 Vertical Dynamics

2 𝑝 | 𝑝 | . But, regardless of the exact curve form, we can conclude that a traditional damper
can only dissipate energy. A step towards controllable suspension is to use a semi-active damper,
which has varying hydraulic orifices. It can still only dissipate energy, but the curve shape is variable.
If adding energy via a hydraulic pump, one can take one more step, called a “active damper”. In princi-
ple, an active damper can generate both positive and negative force (subscript for active
damper) for any sign on 𝑝 . A common way to design control algorithms for such dampers is to use
the to mimick a virtual “Skyhook damper” connected between vehicle unsprung parts and “the
sky”, i.e. a fix inertial system.

Ctrl
̈

Control design model is a damper

between sprung body and “sky”. The
active dampers shall lift body as much
as this damper would have done.

Figure 5-16: § One-dimensional model with two dynamic degrees of freedom and with the
traditional dampers replaced by a Skyhook controlled active damper (ideal sensing, computation
and actuation). Tyre damping neglected.
The corresponding mathematical model becomes as follows:
Equilibrium:
∙ ̈ + ∙
∙ ̈ ∙
Constitution (displacements counted from static equilibrium):
∙( )+ ∙
∙( )+( + )∙
Control and actuation: ∙ ∙
Excitation: (𝑡 )
The same can be formulated with matrices and Fourier transforms:
0 ̈ 0 0
[ ]∙[ ]+[ ]∙[ ]+[ + ]∙[ ] [ ]∙
0 ̈ 0
Eq [5.46] is still valid with same matrix names, but the definition of matrix 𝑫 is different. No transfer
functions are plotted here, but in 5.6.5.

5.4.4 One-Mode Models

The sprung mass is typically around 10 times larger than the unsprung mass and the suspension
spring is usually around 10 times lower than the tyre stiffness. Hence, there are the 2 distinguished
modes, identified in 0. If only interested in a certain frequency range around one of the eigenfrequen-
cies, one can split the model in 2 models, which explains one mode each, see Figure 5-17.

346
 Vertical Dynamics

Modes Models
(Arrows marks displacement amplitudes) heave (or bounce) mode wheel hop mode

𝒄 𝒅
𝒄 𝒅

heave (or bounce) wheel hop 𝒄𝒕 𝒄𝒕

(sprung & unsprung (sprung & unsprung
in phase) out of phase)
Figure 5-17: Modes and approximate models.
We will now derive the natural frequencies for the two models and compared with the natural fre-
quencies ( 0) found for the combined model, in Figure 5-15. Both models are one degree of free-
dom models with mass and spring, why the eigenfrequency is √stiffness/mass.
For the heave model, the mass is . Stiffnesses and are series connected, which means that the
total stiffness ⁄(( ⁄ ) + ( ⁄ )).
For the wheel hop model, the mass is . Stiffnesses and are parallel connected, which means
that the total stiffness + .

[5.50]
1 1
⁄( + ) ad + ad
√ 6.6 .05 H √ 6 .4 0. H
𝑛 𝑚 𝑊 𝑙𝐻 𝑝 𝑚𝑢

Numerical values from 5.4.3 is used ( 4 5𝑘 85 𝑘 68 𝑘𝑁⁄ 676 𝑘𝑁⁄ )

and eigenfrequencies coincide well with Figure 5-15.
Heave (or Bounce) refers to the mode where the sprung mass has the greatest amplitude and wheel
hop is related to the case when the unsprung mass exhibits the greatest amplitude. For a passenger
car, the spring mass has the lowest frequency, typically around 1 Hz while tyre hop is more prevalent
at frequencies around 10 Hz.

5.5 Functions for Stationary Oscillations

5.5.1 Ride Comfort *
Function definition: (Stationary) Ride comfort is the comfort that vehicle occupants experience from station-
ary oscillations when the vehicle travels over a road with certain vertical irregularity in a certain speed. The measure
is defined at least including driver (or driver seat) vertical acceleration amplitudes.
Ride comfort is sometimes divided into:
• Primary Ride – the vehicle body on its suspension. Heave (Bounce), Pitch and Roll 0..4 Hz
• Secondary Ride – same but above body natural frequencies, i.e. 4..25 Hz

5.5.1.1 Single Frequency

It is generally accepted for stationary vibrations, that humans are sensitive to the RMS value of the ac-
celeration. However, the sensitivity is frequency dependent, so that highest discomfort appears for a
certain range of frequencies. Some human tolerance curves are shown in Figure 5-18 and Figure 5-19.
The curves can be considered a threshold for acceptance where everything above the line is unac-
ceptable and points below the curve are acceptable. Discomfort is a subjective measure, and this is

347
 Vertical Dynamics

why the different diagrams cannot be directly compared to each other. The SAE has suggested that fre-
quencies from 4 to 8 Hz are the most sensitive and the accepted accelerations for these are no higher
than 0.025 g (RMS).
The curves in Figure 5-18 mostly represent an extended exposure to the vibration. As one can expect, a
human can endure exposure to more severe conditions for short periods of time. The SAE limits pre-
sented are indicative of 8 hours of continuous exposure. Curves for different exposure times can also
be obtained from ISO, (ISO 2631). The ISO curves are from the first version of ISO 2631 and were later
modified, see Figure 5-19.

How to use the diagram

RMS not
accele- OK
ration
OK

4 Hz 8 Hz Frequency

Figure 5-18: Various Human Tolerance Curves to Vertical Vibration, (Gillespie, 1992)

348
 Vertical Dynamics

Figure 5-19: ISO 2631 Human Tolerance Curves

5.5.1.2 Multiple Frequencies

The curves in Figure 5-18 and Figure 5-19 can be interpreted as a filter, where the response of the hu-
man is influenced by the frequencies they are exposed to. This leads to the concept of a Human Filter
Function 𝑊𝑘 (𝑓). (𝑊𝑘 refers to vertical whole human body vibration sensitivity, while there are other
for sensitivities for other directions and human parts.) This can be seen as a transfer function from
driver seat to somewhere inside the driver’s body or brain, where discomfort is perceived.
#2018: Ska jag lägga till Wd curvan i vänstra diagrammet, så att man ser att
1
ISO2631 Asymptotic approximations of ISO 2631 weighting curves
10
)

0
/2

𝑊
𝑊𝑘
𝑚

-5
0
𝑊
0𝑛

10

𝑊𝑘 𝑊𝑘
-10
𝑊𝑘
⇒ 𝑊𝑘
𝑊𝑘 (dB)

𝑊
W in dB
[1] ]

-1 -15
Wk [

10
(number= 0 log Gain
𝑊

𝑊
-20
-2
10 -25 Wd (horizontal)
Wk (vertical)
-30
-3
10 0 1
-1
10 10
0 1
10 10
2
10
3 10 10
f [Hz] Frequency (Hz)
Figure 5-20: Human Sensitivity Filter Function. From (ISO 2631). Right: Asymptotic
approximation

349
 Vertical Dynamics

Figure 5-21: Human Filter Function for vertical vibrations. Table from (ISO 2631).
With formulas from earlier in this chapter we can calculate an RMS value of a signal with multiple fre-
quencies, see Equation [5.6]. Consequently, we can calculate RMS of multiple frequency acceleration.
Since humans are sensitive to acceleration, it would give one measure of human discomfort. However,
to get a measure which is useful for comparing accelerations with different frequency content, the
measure has to take the human filter function into account. The Weighted RMS Acceleration, a w, in the
following formula is such measure:

𝑁 2 𝑁 2
̂̈ (𝑊𝑘 ( ) ∙ ̂̈ )
𝑤 𝑤 ( ̈ (𝑡)) 𝑢𝑠𝑒: 𝑅𝑀𝑆( ̈ (𝑡)) √∑ √∑ or
= =

[5.51]
{ }
∞ ∞
2
𝑤 𝑤( ̈ (𝑡)) {𝑢𝑠𝑒: 𝑅𝑀𝑆( ̈ (𝑡)) √ ∫ ̈( )∙ } √ ∫ (𝑊𝑘 ( )) ∙ ̈( )∙
𝜔= 𝜔=

Equation [5.51] is written for a case with only vertical vibrations, hence 𝑊𝑘 and ̈ . If vibrations in sev-
eral directions, a total 𝑤 can still be calculated, see (ISO 2631).
In (ISO 2631) one can also find the following equation, which
[5.52]

weights together several time periods, with different vibrations ∑ 𝑤

2 ∙𝑇
spectra. Time averaged whole-body vibration exposure value is 𝑤 𝑣 √
∑ 𝑇
denoted 𝑤 𝑣 .
The 𝑤 in Eq [5.52] is used both for vehicle customer requirement setting at OEMs and governmental
legislation. One example of legislation is (DIRECTIVE 2002/44/EC, 2002). This directive stipulates that
𝑤 in Eq [5.52] in any direction, normalized to 8 hours, may not exceed 1.15 m/s , and if the value ex-
2

ceeds 0.5 m/s2 action must be taken.

5.5.1.2.1 Certain Combination of Road, Vehicle and Speed
Now we can use Equation [5.36] without assuming road type. However, we have to identify ̂̈ and
multiply it with 𝑊𝑘 ( ), according to Equation [5.51]. Then we get [5.53].
Using Equation [5.53], we can calculate the weighted RMS value for the different models in 5.4.1, 5.4.2
and 5.4.3. For each model, it will vary with speed, . A plot, assuming a certain road type (“Rough”
from Figure 5-5) is shown in Figure 5-30. We can see that the simplest model “stiff tyre, no unsprung

350
 Vertical Dynamics

mass” gives much different comfort value than the two others, so the simplest is not goof to estimate
comfort. However, the two other models give approximately same result, which indicates that the me-
dium model, “stiff tyre, no unsprung mass”, is enough for comfort evaluation. This is no general truth
but an indication that the most advanced model, “two masses, elastic tyre”, is not needed for comfort on
normal roads. The advanced model is more needed for road grip.
We can also see that the comfort decreases, the faster the vehicle drives. If we read out at which speed
we reach 𝑤 ⁄𝑠 2 (which is a reasonable value for long time exposure) we get around ≈
70 ⁄𝑠 ≈ 50 𝑘 /ℎ on this road type (“Rough”) with the medium (and advanced) model. With the
simplest model, we get ≈ ⁄𝑠 ≈ . .4 𝑘 /ℎ.

∞
Eq 𝛷 2 𝐻 𝑟→ ̈
𝑅𝑀𝑆( ̈ ) { } √ ∙ 𝑤 ∙ ∫ |𝐻 𝑟 → ̈ ( )| ∙ 𝑤 ∙ { }
[5. ] 𝑤 2
𝛺 ∙ 𝐻 𝑟→
𝜔=

∞
𝛷 𝑤 2
2
𝑤
√ 𝑤 ∙ ∙ ∫ | ∙ 𝐻 𝑟 → ( )| ∙ ∙
𝛺

[5.53]
𝜔=

∞
𝛷 𝑤
2
4 𝑤
Eq
√ 𝑤 ∙ ∙ ∫ |𝐻 𝑟 → ( )| ∙ ∙ ⇒{ }⇒
𝛺 [5.5 ]
𝜔=

∞
𝛷 𝑤
2 2
4 𝑤
⇒ 𝑤 √ 𝑤 ∙ ∙ ∫ (𝑊𝑘 ( )) ∙ |𝐻 𝑟 → ( )| ∙ ∙
𝛺
𝜔=

5.5.1.3 Other Excitation Sources

Present chapter focuses on the influence of excitation from vertical displacement of the road. Exam-
ples of other, but often co-operating, excitation sources are:
• Powertrain vibrations, non-uniform rotation in engine. Frequencies will be proportional to en-
gine speed
• Wheel vibrations, e.g. due to non-round wheels or otherwise unbalanced wheels. Frequencies
will be proportional to vehicle speed.
• Special machineries mounted on vehicles (e.g. climate systems or concrete mixers)

351
 Vertical Dynamics

2
Ride Comfort. For road type "Rough"
10

1
10
Weigthed RMS value, aw [m/(s*s)]

0
10

-1
10

-2
10

-3
10
-2 -1 0 1 2
10 10 10 10 10
vx [m/s]

Figure 5-22: Weighted RMS values for road type “Smooth” from Figure 5-5. The 3 curves show 3
different models: Simplest (from 1.5.1), Medium (from 1.5.2) and Most advanced (from 1.5.3).

5.5.2 Fatigue Life *

Function definition: (Vehicle) Fatigue life is the life that the vehicle, mainly suspension, can reach due to sta-
tionary oscillations when vehicle travels over a road with certain vertical irregularity in a certain speed. One meas-
ure is the suspension vertical deformation amplitude.
Beside human comfort, the fatigue of the vehicle structure itself is one issue to consider in vertical ve-
hicle dynamics.

5.5.2.1 Single Frequency

5.5.2.1.1 Loads on Suspension Spring
In particular, the suspension spring may be subject to fatigue. The variation in spring material stress is
dimensioning, which is why the force variation or amplitude in the springs should be under observa-
tion. Since spring force is proportional to deformation, the suspension deformation amplitude is pro-
posed as a good measure (at least if spring design is not varied). This is the explanation to why the am-
plitude of is plotted in Figure 5-10.
Beside fatigue loads, is also relevant for judging whether suspension bump-stops become en-
gaged or not. At normal driving, that limit should be far from reached, except possibly at high loads
(many persons/much payload).
It can be noted that represents the variation in material stress only if spring is not changed. So,
if different spring designs are compared, it is not sufficient to study only .
5.5.2.1.2 Fatigue of Other Components
Fatigue of other parts may require other amplitudes.
One other example can be the damper fatigue. Damper fatigue would be more relevant to judge from
amplitude of , which determines the force level and hence the stress level.
Another example is to judge the force amplitude in parts that carry both spring and damper forces. For
those it is motivated to consider Δ ∙( )+ ∙( ) as in Eq [5.49].
Yet another example is the load of the road itself. For heavy trucks, it is relevant to consider how much
they wear the road. At some roads with legislated maximum (static) axle load, one can be allowed to
exceed that limit if the vehicle has especially road friendly suspensions. For these judgements, it is the
contact force between tyre and road, ̂ , which is important. These considerations are primarily for
road authorities but becomes aspects for vehicle developers as legal requirements.

352
 Vertical Dynamics

5.5.2.2 Multiple Frequencies

If the excitation is of one single frequency, the stress amplitude can be used when comparing two de-
signs. However, for spectra of multiple frequencies, one cannot look at amplitudes solely,
[ ̂ 2̂ ⋯ 𝑁 ̂ ], because the amplitudes will depend on how the discretization is done, i.e. the number N.
Some kind of integral of a spectral density is more reasonable. In this compendium, it is proposed that
a very approximate measure of fatigue load is calculated as follows, exemplified for the case of fatigue
of the spring:
𝑀𝑒 𝑠𝑢𝑟𝑒 𝑓𝑜𝑟 𝑠𝑝𝑟 𝑛 𝑓 𝑡 𝑢𝑒 𝑓𝑒 𝑅𝑀𝑆( 𝑢 (𝑡) 𝑠 (𝑡))
𝛷0 2 𝑢𝑠𝑒: 𝐻 𝑟→ 𝑢

[5.54]
√𝛺 𝑤 ∙ 𝑥 ∙ ∫ 0|𝐻 𝑟→ 𝑢 𝑠
( )| ∙ ∙ {
𝐻 𝐻 𝑟→
𝑠
}
0 𝑟→ 𝑢 𝑠
𝛷0 2
√𝛺 𝑤 ∙ 𝑥 ∙ ∫ 0|𝐻 𝑟→ 𝑢 ( ) 𝐻 𝑟→ 𝑠 ( )| ∙ ∙
0

Equation is written for application to a known road spectra (𝛷 ) and vehicle dynamic structure
(𝐻 𝑟 → 𝑢 𝐻 𝑟 → 𝑠 ), but the first expression (𝑅𝑀𝑆( 𝑢 (𝑡) 𝑠 ( 𝑡))) is applicable on a measured or simu-
lated time domain solution.

5.5.3 Road Grip *

Function definition: Road grip (on undulated roads) is how well the longitudinal and lateral grip between
tyres and road is retained due to stationary oscillations when the vehicle travels over a road with certain vertical ir-
regularity in a certain speed.
In 3.3 and 0, the brush model explain how the tyre forces in the ground plane appears. It is a physical
model where the contact length influences how stiff the tyre is for longitudinal and lateral slip. There
is also a brief description of relaxation models for tyres. This together motivates that a tyre has more
difficult to build up forces in ground plane if the vertical force varies. We can understand it as when
contact length varies, the shear stress builds up has to start all over again. As an average effect, the
tyre will lose more and more grip, the more the vertical force varies.

5.5.3.1 Multiple Frequencies

If the excitation is of one single frequency, the force amplitude can be used when comparing two de-
signs. However, for spectra of multiple frequencies, one cannot look at amplitudes solely,
[ ̂ ̂2 ⋯ ̂𝑁 ], because the amplitudes will depend on how the discretization is done, i.e. the number N.
Some kind of integral of a spectral density is more reasonable. In this compendium, it is proposed that
a very approximate measure of road grip is calculated as follows:
𝑀𝑒 𝑠𝑢𝑟𝑒 𝑓𝑜𝑟 ( )𝑟𝑜 𝑟𝑝 𝑅𝑀𝑆(Δ 𝑟 (𝑡))
𝛷0 2
√𝛺 ∙ 𝑥 ∙ ∫ 0|𝐻 𝑟→Δ 𝑟 ( )| ∙ ∙
[5.55]

𝑤
0

{𝑢𝑠𝑒: 𝐻 𝑟 → 𝑢 𝑠 𝑡 ∙( 𝐻 𝑟 → 𝑢 )}
𝛷0 2
√𝛺 𝑤 ∙ 𝑥 ∙ ∫ 0| 𝑡 ∙( 𝐻 𝑟→ 𝑢
)| ∙ ∙
0

Equation is written for application to a known road spectrum (𝛷 ) and vehicle dynamic structure
(𝐻 𝑟 → 𝑢 ), but the first expression (𝑅𝑀𝑆(Δ 𝑟 𝑡 ) is applicable on a measured or simulated time do-
( ))
main solution.

5.5.4 Other Functions

Present chapter focuses on the functions, (vertical) ride comfort, fatigue and road grip. Examples of
other functions are:

353
 Vertical Dynamics

• An area of functions that encompasses the vertical dynamics is Noise, Vibration, and Harsh-
ness – NVH. It is similar to ride comfort, but the frequencies are higher, stretching up to sound
which is heard by humans.
• Ground clearance (static and dynamic) between vehicle body and ground. Typically, im-
portant for off-road situations.
• Longitudinal comfort, due to drive line oscillations and/or vertical road displacements. Espe-
cially critical when driver cabin is separately suspended to the body. This is the case for heavy
trucks.
• Disturbances in steering wheel feel, due to one-sided bumps. Especially critical for rigid
steered axles. This is often the design of the front axle in heavy trucks.
• There are of course an infinite number of combined manoeuvres, in which functions with re-
quirements can be found. Examples can be bump during strong cornering (possibly destabi-
lizing vehicle) or one-sided bump (exciting both heave=bounce, pitch and roll modes). When
studying such transients, the vertical dynamics is not enough to capture the comfort, but one
often need to involve also longitudinal dynamics; the linkage with ant-dive/anti-squat geome-
try from Chapter 3 becomes important as well as tyre vertical (radial) deflection characteris-
tics.
• Energy is dissipated in suspension dampers, which influence energy consumption for the ve-
hicle. This energy loss is much related, but not same as, to (tyre) rolling resistance. Suspension
characteristics do influence this energy loss, but it is normally negligible, unless driving very
fast on very uneven road.

5.6 Variation of Suspension Design

The influence of design parameters on vehicle functions Ride comfort, Suspension fatigue and Road
grip can now be made. E.g., it is important to not only use the transfer function, but also take the road
and human sensitivity into account, which calls for different weighting for different frequencies.
Transfer function for the model in 5.4.3 is shown as dashed lines in
Figure 5-23. Same figure also shows the Road- and Human-weighted versions. Studying how these
curves change with design parameters gives a quantitative understanding of how different suspension
design parameters influence. Such variations will be done in 5.6.1 to 5.6.4.
There are two particular frequency intervals of the graphs to observe. These are the 2 peaks around
the two the natural frequencies of the sprung and unsprung masses, the peak at lower frequency is
mainly a resonance in heave mode, while the higher one is in wheel hop mode.

354
 Vertical Dynamics
3
10

2
10

1
10

amplification /[(m/(s*s))/(m)]
0
10

-1
10

-2
10

-3
10
1

Figure 5-23: For a passenger car with 600 [𝑘 ] 00 [𝑘 ] 76 [𝑘𝑁 ⁄ ]

9 [𝑘𝑁⁄( ⁄𝑠)] 764 [𝑘𝑁 ⁄ ] 0. Left is vertical acceleration (amplitude) of sprung mass
for Ride Comfort. Middle is relative displacement (amplitude) between sprung and unsprung mass
for Suspension Fatigue. Right is deformation (amplitude) of tyre spring for Road Grip. Weightings
for typical road and for human sensitivity is shown.

5.6.1 Varying Suspension Stiffness

In Figure 5-24 the benefits of the low suspension stiffness (1 Hz) is seen for suspension travel and
comfort without much change in the road grip performance.

Figure 5-24: Variation of suspension stiffness,

355
 Vertical Dynamics

Regarding Figure 5-24 and Figure 5-25 we see that there is a large influence of the acceleration gain at
low frequencies with little change at the wheel hop and higher frequencies. The suspension stiffness
and damping were seen to have little influence on the ride comfort / road grip response around 10 Hz.

5.6.2 Varying Suspension Damping

In Figure 5-25, we see that the changes in suspension damping have opposite effects for the heave and
wheel hop frequency responses. High damping is good for reducing heave, but not so effective for
wheel hop.
2
Comfort
10

Road- and Human-weighted amplification /[(m/(s*s))/(m)]

1
10

0
10

-1
10

-2
10

-3
10
-1 0
10 10

Figure 5-25: Variation of suspension damping,

5.6.3 Varying Unsprung Mass

In Figure 5-26, we see that if the response around the wheel hop frequency is to be changed, the un-
sprung mass is one of the most influential parameters. The unsprung mass is usually in the range of
10% of the sprung mass. Opposite to the suspension parameters, the unsprung mass influences fre-
quencies around the wheel hop frequency with little influence around the heave frequency.
In Figure 5-26, the case with 0 is added. This is to demonstrate what a model with neglected
mass gives and can be nearly compared with the model in 5.4.2.

356
 Vertical Dynamics

2
10

Road- and Human-weighted amplification /[(m/(s*s))/(m)]

1
10

0
10

-1
10

-2
10

-3
10
-1
10

Figure 5-26: Variation of unsprung mass,

5.6.4 Varying Tyre Stiffness

In Figure 5-27 a general observation is that low sprung mass natural frequencies are preferred for
comfort considerations. Another parameter that has a strong affect near the wheel hop frequency is
the tyre stiffness. The strongest response is noticed for the road grip function. (Note that, since is
now varying, we have to express road grip as ∙ ( ); only does not give a fair compari-
son.)

2
Com
10

Road- and Human-weighted amplification /[(m/(s*s))/(m)]

1
10

0
10

-1
10

-2
10

-3
10
-1
10

Figure 5-27: Variation of Tyre Stiffness,

357
 Vertical Dynamics

5.6.5 § Varying Skyhook Damping

Contribution from Waltteri Koskinen, student at Tampere University, Finland

In figure below, the suspension dampers are replaced by ideal active dampers (ideal sensing, control
algorithm computation and actuation). One can compare with how variation of (traditional) damper in
Figure 5-25. The active dampers improve comfort and road grip, except for at wheel hop frequency
around 10 Hz. The fatigue is generally worse. The transfer functions are plotted using the model in
5.4.3.2.

Figure 5-28: § Variation of Skyhook damping,

5.7 Two Dimensional Oscillations

The one-dimensional model is useful for analysing the response of one wheel/suspension assembly.
Some phenomena do connect other vehicle body motions than the vertical translation, especially pitch
and roll. Here, other models are needed, such as Figure 5-30 and Figure 5-29.

5.7.1 Heave and Roll

A model like in Figure 5-29 is proposed. We have been studying Heave (bounce) and pitch before, in
4.3.10 and 4.5.2.3. In Chapter 4, the excitation was lateral tyre/axle forces, while the vertical displace-
ment of the road was assumed to be zero. In vertical vehicle dynamics, it is the opposite. That means
that the linkage geometry (roll centre or wheel pivot points) is not so relevant here. So, the model can
be somewhat simpler.

358
 Vertical Dynamics

m*az m*g
z
zl zr J*der(wx)
px

zrl zrr
zrl
zrr Flz Frrz
x x
Figure 5-29: Heave and roll model. Anti-roll bar not drawn but can be included in equations in
matrix 𝑪.
No equations are formulated for this model in this compendium, but a model will typically show two
different modes, the heave and roll. Heave Eigen frequency is typically 1-1.5 Hz for a passenger car, as
mentioned before. The roll frequency is similar or somewhat higher.
If modelling unsprung masses without inertia, we still get 2 state variables, heave and roll . Using
same mathematical form of equations as in Eq [5.45] we get this model (subscripts 𝑟 for “road left”
and 𝑟𝑟 for “road right”):
𝑴 ∙ 𝒛̈ + 𝑫 ∙ 𝒛 + 𝑪 ∙ 𝒛 𝑫 𝒓 ∙ 𝒛 𝒓 + 𝑪𝒓 ∙ 𝒛 𝒓
𝑙
ℎ𝑒𝑟𝑒 𝒛 [ ] 𝑛 𝒛𝒓 [ ]

[5.56]
The disturbances from the road are two independent ones, so the
transfer functions will be a × matrix:
ℱ( ) ℱ( 𝑙 ) 𝐻 → 𝐻 𝑟𝑟 → ℱ( 𝑙 )
[ ] 𝑯 [ ] [ 𝑟 ] [ ]
ℱ( ) ℱ( ) 𝐻 𝑟 →𝜑 𝐻 𝑟𝑟 →𝜑 ℱ( )
Note that the restoring matrix might need to include both elastic restoring (wheel springs and anti-
roll-bars) and (roll) pendulum effects, see 4.3.10.2 and Reference (Mägi, The Significance of System
Pre-Load at Modal Analysis of Low-Resonant Mechanical Systems, 1988). For high-loaded trucks, the
pendulum effect is really relevant, while it often can be omitted for a low sportscar.

5.7.2 Heave and Pitch

A model like in Figure 5-30 is proposed. We have been studying heave (or bounce) and pitch before, in
3.4.5.2. Hence compare with corresponding model in Figure 3-32. In Chapter 3, the excitation was lon-
gitudinal tyre forces, while the vertical displacement of the road was assumed to be zero. In vertical
vehicle dynamics, it is the opposite. The importance of model with linkage geometry (pitch centre or
axle pivot points above ground level) is that tyre forces are transferred correctly to the body. That
means that the linkage geometry is not so relevant for vertical vehicle dynamics in Chapter 5. So, the
model can be somewhat simpler.
A mathematical model would typically show two different modes, see Figure 5-31. The heave eigen-
frequency is typically 1-1.5 Hz for a passenger car. The pitch frequency is somewhat higher.
We should reflect on that the models in 5.4 and 5.7.1 refer to the same bounce mode. But the models
will most likely give different numbers of, e.g., Eigen frequency. A total model, with all degrees of free-
dom, would align those values, but the larger a model is the more data it produces which often leads to
less easy design decisions.

359
 Vertical Dynamics

Heave (or Bounce) Pitch Motion

𝑓
mode mode node for
Pitch mode

ℎ 𝑓 𝑓
Motion node
for Heave mode

𝑓
𝑓
𝑓

Figure 5-30: Heave and pitch physical model. Figure 5-31: Oscillation modes of a
Heave and Pitch model.

5.7.2.1 Wheelbase Filtering

See Figure 5-32. If the wheelbase is an integer multiple of the wavelength, only heave (bounce) will be
excited. If wavelength is in the middle between those, only pitch will be excited. This phenomenon is
called “wheelbase filtering”.

Low heave
excitation

Low pitch
excitation

Figure 5-32: Wheelbase Filtering, as compared to “uncorrelated”. Response |𝐻 𝑓→𝑣 | and

|𝐻 𝑓→𝜔 |. Two frequencies, 0.5 and 5 Hz. Varying Road Wavelengths, so also has to vary.

5.7.2.2 Mathematical Model

From Figure 5-30 we can derive the following mathematical models:
Equilibrium: 𝑓 + 0 𝑛 𝑓 𝑓+ 0
Compatibility: 𝑓 𝑓 𝑛 +

360
 Vertical Dynamics

Constitution: 𝑓 𝑓 + 𝑓 ( 𝑓 𝑓 ) 𝑛 𝑓 𝑓 ( 𝑓 𝑓 ) 𝑛
+ ( ) 𝑛 ( )
Above model is formulated on first order form ( ⋯ and (⋯ ) ) to show an alternative to
second order differential equation form ( ̈ ⋯ ) used in 5.4.
We need to express in differentiated variables. Then we can either assume [ ] [ ] (ve-
locity components in ground fixed directions) or [ ] [ 𝑣 𝑣 ] (vehicle fixed directions). Both
are correct, in similar way as for the yaw rotation, see 4.4.2.3.3. If ground fix: [ ] [ ] . If
ground fix: [ 𝑣 𝑣 ] [ 𝑣 + 𝑣 ≈ 𝑣 𝑣 𝑣 ] . The intention for 5.7.2 is to study con-
stant speed over ground, so we know 0 . Therefore it is easiest to use ground fix directions. We
can compare with the pitching model in 3.4.5, which is typically used for longitudinal acceleration and
braking. Then it is most natural to use vehicle fix and solve 𝑣 and 𝑣 as state variables in an ode.
We formulate the matrix form of the model in ground fix direction :
0 0 0
0 0 0
0 0 ⁄𝑓 0 𝑓
[⏟0 0 0 ⁄ ] [ ]
𝑴𝑪
𝑓 𝑓 𝑓 𝑓
2 2
𝑓 𝑓 𝑓 𝑓 𝑓 + 0 𝑓 𝑓 𝑓
[ ]+[ ]+[ ] [ ]
+𝑓 0 0 𝑓 0 0 ⏟
[ 0 0 ] ⏟ ⏟ 0 ⏟0 𝒛𝒗𝒓
⏟
𝒛𝒗𝑭 𝒈 𝑫𝒓
𝑫

Variable substitution 𝒛𝒗𝑭 𝒛𝒗𝑭𝟎 𝑫 𝟏 𝒈 gives: 𝑴𝑪 𝒛𝒗𝑭𝟎 𝑫 𝒛𝒗𝑭𝟎 + 𝑫𝒓 𝒛𝒗𝒓𝒛 .

Fourier transform gives: 𝑗 𝑴𝑪 𝓕(𝒛𝒗𝑭𝟎 ) 𝑫 𝓕(𝒛𝒗𝑭𝟎 ) + 𝑫𝒓 𝓕(𝒛𝒗𝒓𝒛 ) The Transfer functions in
Figure 5-32 can now be plotted. Note that the driver comfort, earlier measured in vertical accelera-
tion amplitude ̂ amplitude( ) , now has two possibilities: ̂ amplitude( ) ̂ or ̂ 𝑣
amplitude( 𝑣 𝑣 ) . The latter one makes most sense, since seat and driver rotates with the
vehicle. With this measure of driver comfort, the transfer function for in Figure 5-32, |𝐻 𝑓→𝑣 |
|𝐻 𝑓→𝑣 |, should be adjusted to |𝐻 𝑓→𝑣 𝑣 |, for driver comfort measure.
If eliminating the forces, in the model in vehicle directions, we get the 2nd order differential equation:
0 ̈ 𝑓 𝑓 𝑓 𝑓 0
[0 ] [ ] [ 2 2 ] [ ]+[ ] [ ]
̈ 𝑓 𝑓 𝑓 𝑓 𝑓 0
𝑓 𝑓 𝑓 𝑓
([ ] [ ]+[ ]) + [ ] [ ]
𝑓 𝑓

5.7.2.2.1 Correlated or Wheelbase Filtered

Front and rear are excited with same frequencies, but delayed at the rear:
ℱ( 𝑓 ) 𝑓 ̂ cos( 𝑡)
𝓕(𝒛𝒗𝒓𝒛 ) [ ] 𝓕 ( [ ]) 𝓕 ( [ ])
ℱ( ) 𝑡 𝑡 ̂ cos ( 𝑡 𝜋 𝜆 ⁄𝐿)
[ ( ] ℱ( ̂ cos( 𝑡))
cos 𝜋 𝜆 ⁄𝐿) 𝑗 sin( 𝜋 𝜆 ⁄𝐿)
[ ( ] ℱ( 𝑓 )
cos 𝜋 𝜆 ⁄𝐿) 𝑗 sin( 𝜋 𝜆 ⁄𝐿)
𝑗 [ ( ] ℱ( 𝑓 ) 𝑗 𝒅𝝀 ℱ( 𝑓)
cos𝜋 𝜆⁄𝐿) 𝑗 sin( 𝜋 𝜆⁄𝐿)
Insertion gives: 𝑗 𝑴𝑪 𝓕(𝒛𝒗𝑭𝟎 ) 𝑫 𝓕(𝒛𝒗𝑭𝟎 ) + 𝑫𝒓 𝑗 𝒅𝝀 ℱ( 𝑓 ) ⇒

361
 Vertical Dynamics

𝐻 𝑟𝑓 →𝑣𝑧
𝐻 𝑟𝑓 →𝜔𝑦
⇒ 𝓕(𝒛𝒗𝑭𝟎 ) (𝑗 𝑴𝑪 𝑫) 𝑫𝒓 𝑗 𝒅𝝀 ℱ( 𝑓) ℱ( 𝑓) 𝑯𝒛𝒓𝒇 →𝒗𝒛 𝒄𝒐𝒓𝒓 ℱ( 𝑓)
𝐻 𝑟𝑓 → 𝑓
[ 𝐻 𝑟𝑓 → 𝑟 ]
All 𝐻 depend on (or 𝜆) and .
5.7.2.2.2 Uncorrelated
We now assume that front and rear are excited “uncorrelated”. This is wrong if driving on a road
where rear axle follows front axle, but it is correct for a vehicle with independent excitation under
each axle, which can be achieved e.g. in a shake rig.
ℱ( 𝑓 ) ℱ( 𝑓) ℱ( 𝑓)
𝓕(𝒛𝒗𝒓 ) [ ] [ ] 𝑗 [ ]
ℱ( ) ℱ( ) ℱ( )
𝐻 𝑟𝑓 →𝑣𝑧 𝑛 𝐻 𝑟𝑟 →𝑣𝑧 𝑛

ℱ( 𝑓)
𝐻 𝑟𝑓 →𝜔𝑦 𝑛 𝐻 𝑟𝑟 →𝜔𝑦 𝑛 ℱ( 𝑓)
𝓕(𝒛𝒗𝑭𝟎 ) (𝑗 𝑴𝑪 𝑫) 𝑫𝒓 𝑗 [ ] [ ]
ℱ( ) 𝐻 𝑟𝑓 → 𝑓 𝑛 𝐻 𝑟𝑟 → 𝑓 𝑛 ℱ( )
][ 𝐻 𝑟𝑓 → 𝑟 𝑛 𝐻 𝑟𝑟 → 𝑟 𝑛

If 𝑓 and 𝑓 have the same amplitude(frequency) content, and is called , we can write:
𝐻 𝑟𝑓 →𝑣𝑧 𝑛 𝐻 𝑟𝑟 →𝑣𝑧 𝑛
𝐻 𝑟𝑓 →𝜔𝑦 𝑛 𝐻 𝑟𝑟 →𝜔𝑦 𝑛
𝓕(𝒛𝒗𝑭𝟎 ) [ ] ℱ ( ) 𝑯𝒛𝒓 →𝒗𝒛 𝒏𝒄𝒐𝒓𝒓 ℱ ( )
𝐻 𝑟𝑓 → 𝑓 𝑛 𝐻 𝑟𝑟 → 𝑓 𝑛
[ 𝐻 𝑟𝑓 → 𝑟 𝑛 𝐻 𝑟𝑟 → 𝑟 𝑛 ]
The elements in 𝐻 depend on (or 𝜆) and .

5.8 Three Dimensional Oscillations

A real road generates motion in all out-of-road-plane dimensions: heave , roll and pitch . One
can, for instance, use [5.27] to generate a vertical profile for each side of the vehicle, but then ran-
domly generate different phases for each side. Another way is to record a certain piece of a road, typi-
cally at the vehicle manufacturer’s test track. One then gets wheelbase filtering one each side and a roll
influence due to that left and right side are not in phase with each other.
If modelling unsprung masses without inertia, we still get 3 state variables: heave , pitch and roll
. Using same form of equations as in Eq [5.45] we get this model (subscripts 𝑟𝑓 for “road front left”
and so on):
𝑴 ∙ 𝒛̈ + 𝑫 ∙ 𝒛 + 𝑪 ∙ 𝒛 𝑫 𝒓 ∙ 𝒛 𝒓 + 𝑪𝒓 ∙ 𝒛 𝒓

ℎ𝑒𝑟𝑒 𝒛 [ ]

𝑓𝑙 0
𝑓 0 𝑓𝑙
𝑛 𝒛𝒓 [ ] [ ( )
cos + 𝑗 sin ( ) 0
] [ ]
𝑙 𝑓
[5.57]

0 cos( ) + 𝑗 sin( )
ℎ𝑒𝑟𝑒 𝜋 𝐿⁄
The disturbances from the road are two independent ones, so the transfer
functions will be a × matrix:
ℱ( ) 𝐻 𝑟𝑓 → 𝐻 𝑟𝑓𝑟 →
ℱ( 𝑓𝑙 ) ℱ( 𝑙 )
[ℱ( ) ] 𝑯 [ ] [𝐻 𝑟𝑓 →𝜑 𝐻 𝑟𝑓𝑟 →𝜑 ] [ ]
ℱ( 𝑓 ) ℱ( )
ℱ( ) 𝐻 𝑟𝑓 →𝜑𝑦 𝐻 𝑟𝑓𝑟 →𝜑𝑦

362
 Vertical Dynamics

5.9 Transient Vertical Dynamics

The majority of the chapter you read now, considers driving for during long time periods on roads
with repetitive unevenness. This is one relevant use case and the functions are then suitably analysed
using frequency analysis, since the quantities vary as stationary oscillations.
However, vertical vehicle dynamics also have transient disturbances to consider. Test cases can be
one-sided or two sides road bumps or potholes. Two-sided bump is envisioned in Figure 5-33. It can
represent driving over a speed bump or a low obstacle.
Models from earlier in this chapter are all relevant for two-sided bumps or holes, but one might need
to consider non-linearities such as bump stops or wheel lift as well as different damping in compres-
sion and rebound. For one-sided bumps/holes, the models from earlier in this chapter are generally
not enough. The computation is rather time simulation than frequency analysis. The function
measures (and requirements) should be shifted somewhat:
• Human comfort for transients is often better described as time derivative of acceleration
(called “jerk”). Peak-to-peak values of the variables can be used.
• The material loads are more of maximum load type than fatigue life dimensioning, i.e. higher
material stress but fewer load cycles during vehicle lifetime.
• Road grip studies over road bumps and potholes are challenging. Qualitatively, the tyre mod-
els often must include relaxation, because that is the mechanism which reduces road grip when
vertical load shifts. To get quantitatively correct tyre models is beyond the goal of the compen-
dium you presently read.
• Roll-over can be tripped by large one-sided bumps. This kind of roll-overs is unusual and re-
quires complex models.

Figure 5-33: Response of Vehicle for Front and Rear Axle Impulses, (Gillespie, 1992)
Models for studying transient vertical dynamics can, in general be categorized as the stationary oscilla-
tion models, 1D, 2D and 3D. But they cannot generally be linear, so they require simulation, not fre-
quency analysis. One typically need to add inertia of unsprung parts and vertical elasticities in each
tyre. And “trivial linkage” suspension is generally not enough if sharp road unevenness, but instead
one might identify the pivot axis in space for each wheel linkage.
A 3D model according to these concepts gets the states 𝒛 containing 𝑓𝑙 𝑓 𝑙 𝑛 if
modelled with a second order differential equation (𝒇 𝒛̈ 𝒛 𝒛 𝑡
( ) 0 ). If modelled with first order
differential equations (𝒇(𝒛 𝒛 𝑡) 0 ) and the concept of using forces in elasticities as states, see
1.5.2.1.2, the states will instead contain 𝑓𝑙 𝑓 𝑙 𝑓𝑙 𝑓 𝑙 𝑛 ,
where is vertical velocity of unsprung mass in wheel 𝑗 and is elastic part of vertical force un-
der wheel 𝑗. The inputs (disturbance) 𝒛𝒓 will contain 𝑓𝑙 𝑓 𝑙 𝑛 .

5.10 Control Functions

This section about Control functions for vertical dynamics is much smaller than corresponding for
other direction, 3.5 and 4.6. Vertical Control Functions do exist, e.g. levelling control, active anti-
rolbars and active damper control. However, they are less common and generally influence less than

363
 Vertical Dynamics

the stronger propulsion, brake and steering control functions. The compendium does not go deep into
this area, but a sample is seen in 5.4.3.2.

364
 Bibliography

BIBLIOGRAPHY
AB Volvo. (2011). Global Transport Application.
Andreasson, J. (2007). On Generic Road Vehicle Motion Modelling and Control. Stockholm: KTH.
Bakker, E. N. (1987). Tyre Modelling for use in Vehicle Dynamics Studies.
Barnard, R. (2010). Road Vehicle Aerodynamic Design. Mechaero Publishing.
Barnard, R. H. (u.d.). Road Vehicle Aerodynamic Design. Mechaero Publishing.
Besselink, I. J. (2011). Experiences with the TYDEX standard tyre interface and file format. Vehicle
System Dynamics. doi:https://doi.org/10.1080/00423110500109299
Boerboom, M. (2012). Electric Vehicle Blended Braking maximizing energy recovery while maintaining
vehicle stability and maneuverability. Göteborg, Sweden: Chalmers University of Technology.
Brian Bole, e. a. (2012). Energy Management Control of a Hybrid Electric Vehicle with Two-Mode
Electrically Variable Transmission. Los Angeles: 2012 Electric Vehicle Symposium, EVS26.
doi:10.13140/2.1.2827.6480
Bruzelius, F. (2004). Linear Parameter-Varying Systems - an approach to gain scheduling (Vol. Doctoral
thesis). Chalmers University of Technology.
Clark, S. (. (1971). Mechanics of Pneumatic Tires, Monograph 122. National Bureau of Standards, USA.
Cooper Tire & Rubber Co. (den 16 September 2007). Passenger radial tire cutaway. Online Art. (Cooper
Tire & Rubber Co.) Hämtat från http://deantires.com/us/en/information/info-
construction.asp
DIRECTIVE 2002/44/EC. (2002). DIRECTIVE 2002/44/EC OF THE EUROPEAN PARLIAMENT AND OF
THE COUNCIL.
Drenth, E. F. (1993). Brake Stability of Front Wheel Driven Cars at High Speed. Delft, Netherlands: Delft
University of Technology.
Dugoff, H., Fancher, P., & Segel, L. (1969). Tire performance characteristics affecting vehicle response to
steering and braking control inputs. Final report. Washington, US. Hämtat från
http://hdl.handle.net/2027.42/1387
Encyclopædia Britannica Online. (2007, September 16). belted tire: tire designs." Online Art.
Encyclopædia Britannica Online. Retrieved from http://www.britannica.com/eb/art-7786
Gillespie, T. (1992). Fundamentals of Vehicle Dynamics. Society of Automotive Engineers.
Grosch, K., & Schallamach, A. (1961, September–October). Tyre wear at controlled slip. Wear, Volume
4(Issue 5), pp. Pages 356–371.
Happian-Smith, J. (2002). An Introduction to Modern Vehicle Design. Butterwoth-Heinemann, ISBN 0-
7506-5044-3.
Hedman, A. (1994). Intermittent Engine Operation -- A Way to Reduce Vehicle Fuel Consumption. Beijing:
FISITA, XXV FISITA Congress, 17-21 October 1994.
Hirschberg, W., Rill, G., & Weinfurter, H. (2007). Tire model TMeasy. Vehicle System Dynamics, 45:1, pp.
101 — 119. doi:10.1080/00423110701776284
Hucho, W.-H. (1998). Aerodynamics of Road Vehicles. SAE International.
ISO 11026. (u.d.). ISO 11026 Heavy commercial vehicles and buses - Test method for roll stability -
Closing-curve test. International Organization for Standardization, Genève, Switzerland.
ISO 14791. (u.d.). ISO 14791 Road vehicles - Heavy commercial vehicle combinations and articulated
buses - Lateral stability test methods. International Organization for Standardization, Genève,
Switzerland.
ISO 14792. (u.d.). ISO 14792 Road vehicles - Heavy commercial vehicles and buses - Steady-state circular
tests. International Organization for Standardization, Genève, Switzerland.

365
 Bibliography

ISO 14793. (u.d.). ISO 14793 Road vehicles – Heavy commercial vehicles and buses – Lateral transient
response test methods. International Organization for Standardization, Genève, Switzerland.
ISO 14794. (2011). ISO 14794 Heavy commercial vehicles and buses - Braking in a turn - Open-loop test
methods. International Organization for Standardization, Genève, Switzerland.
ISO. (2006). Passenger cars – Braking in a turn – Open-loop test method. ISO.
ISO. (2011). ISO 14794 Heavy commercial vehicles and buses - Braking in a turn - Open-loop test
methods. ISO.
ISO 26262. (2011-2012). Road vehicles – Functional safety, 1-10. International Organization for
Standardization, Genève, Switzerland.
ISO 2631. (n.d.). ISO 2631 – Evaluation of Human Exposure to Whole-Body Vibration. International
Organization for Standardization, Genève, Switzerland.
ISO 3888. (u.d.). ISO 3888 Passenenger Cars -- Test track for severe lane change manouvre -- Part 2:
Obstacle Avoidance. International Organization for Standardization, Genève, Switzerland.
ISO 4130. (u.d.). ISO 4130: Road vehicles -- Three-dimensional reference system and fidicial marks --
Definitions, 1979.
ISO 4138. (u.d.). ISO 4138 Passenger cars – Steady-state circular driving behaviour - Open-loop test
methods. International Organization for Standardization, Genève, Switzerland.
ISO 7401. (u.d.). ISO 7401 Lateral transient response test methods - Open-loop test methods.
International Organization for Standardization, Genève, Switzerland.
ISO 7975. (2006). ISO 7975 Passenger cars – Braking in a turn – Open-loop test method. International
Organization for Standardization, Genève, Switzerland.
ISO 80000-3. (2013). ISO 80000 – Quantities and units – Part 3: Space and time. International
Organization for Standardization, Genève, Switzerland.
ISO 8608. (u.d.). ISO 8608 Mechanical vibration - Road surface profiles - Reporting of measured data.
International Organization for Standardization, Genève, Switzerland.
ISO 8855. (n.d.). ISO 8855 – Road vehicle – Vehicle dynamics and road holding ability – Vocabulary.
International Organization for Standardization, Genève, Switzerland.
ISO19377. (2017). ISO 19377 Heavy commercial vehicles and buses - Emergency braking on a defined
path - Test method for trajectory measurement. ISO.
ISO28580. (2018). Passenger car, truck and bus tyre rolling resitance measurement method -- single
point test and correlation of measurement results. ISO.
Jacobson, B. (1993). Gear Shifting with Retained Power Transfer. Göteborg, Sweden: Chalmers
Univeristy of Technology. Hämtat från https://research.chalmers.se/en/publication/1485
Jacobson, B. e. (2014). Vehicle Dynamics Compendium for Course MMF062. Göteborg, Sweden: Chalmers
University of Technology.
Jacobson, B., Sundström, P., Kharrazi, S., Fröjd, N., & Islam, M. (2017). An Open Assessment Tool for
Performance Based Standards of Long Combination Vehicles. Hämtat från
https://research.chalmers.se/en/publication/251269
Jonasson, M. (2009). Exploiting individual wheel actuators to enhance vehicle dynamics and safety in
electric vehicles (Vol. Doctoral thesis). Stockholm: Royal Insitute of Technology.
Jonsson, A., & Olsson, E. (2016). A Methodology for Identification of Magic Formula Tire Model
Parameters from In-Vehicle Measurements. Chalmers University of Technology. Hämtat från
http://studentarbeten.chalmers.se/publication/239258-a-methodology-for-identification-of-
magic-formula-tire-model-parameters-from-in-vehicle-measurements
Kati, M. S. (2013). Definitions of Performance Based Characteristics for Long Heavy Vehicle
Combinations. Signals and Systems. Chalmers University of Technology. Hämtat från
http://publications.lib.chalmers.se/records/fulltext/176464/176464.pdf
Kharrazi , S. (2012). Steering Based Lateral Performance Control of Long Heavy Vehicle Combinations.
Göteborg, Sweden: Chalmers University of Technology.

366
 Bibliography

Kiencke, U. a. (2005 ). Automotive Control Systems.

Lex, C. (2015). Estimation of the Maximum Coefficient of Friction between Tire and Road based on
Vehicle State measurements. Graz: Graz University of Technology.
Lidberg, M., & Alfredsson, J. (2009). Control of the Direction Sensitive Locking Differential (DSLD) for a
Front Wheel Drive Passenger Car in Transient Cornering. Proceedings of the IAVSD'09
Conference.
Ludwig, C., & Kim, C. (2017). Influence of testing surface on tire lateral force characteristics. München:
Springer Fachmedien.
Michelin. (2003). The tyre Rolling resistance and fuel savings. Michelin.
Mägi, M. (1988). The Significance of System Pre-Load at Modal Analysis of Low-Resonant Mechanical
Systems. Århus, Denmark: Modal Testing & FEM Seminar in Århus.
Mägi, M., Jacobson, B., & Resev, J. (1998). DYNAMIC ANALYSIS OF PLANETARY GEAR TRAINS. Tallin.
Hämtat från https://research.chalmers.se/en/publication/191074
NHTSA. (u.d.). Electronic Stability Control Systems. NHTSA.
Pacejka, H. (2005). Tyre and Vehicle Dynamics, 2nd ed. Elsevier.
Pettersson, P. (2017). On Numerical Descriptions of Road Transport Missions. Gothenburg, Sweden:
Chalmers University of Technology. Hämtat från
https://research.chalmers.se/en/publication/249525
Pettersson, P. (2019). Operation cycles representations for road vehicles. Göteborg: Chalmers University
of Technology.
Ploechl, M. (Red.). (2013). Road and Off-Road Vehicle System Dynamics Handbook. CRC Press.
Rajamani, R. (2012). Vehicle Dynamics and Control. Springer. Hämtat från
http://link.springer.com/book/10.1007%2F978-1-4614-1433-9
Rhyne, T. B. (2005). Development of a Vertical Stiffness Relationship for Belted Radial Tires. Tire Science
and Technology, TSTCA, Vol. 33, no. 3, July-September 2005.
Rill, G. (2006). FIRST ORDER TIRE DYNAMICS. Lisbon, Portugal: III European Conference on
Computational Mechanics Solids, Structures and Coupled Problems in Engineering, 5–8 June
2006. Hämtat från https://hps.hs-
regensburg.de/~rig39165/paper/First_Order_Tire_Dynamics.pdf
Robert Bosch GmbH. (2004). Bosch Automotive Handbook 6th Edition. Bentley Publishes.
Romano, L., Bruzelius, F., & Jacobson, B. J. (2020). Unsteady-state brush theory. Vehicle System
Dynamics. Hämtat från https://research.chalmers.se/publication/517277
Ross, I. M. (2015). A Primer on Pontryagin's Principle in Optimal Control.
Rösth, M. (2007). Hydraulic Power Steering System Design in Road Vehicles: Analysis, Testing and
Enhanced Functionality. Linköping: Linköping Univeristy, Department of Management and
Engineering, Fluid and Mechatronic Systems.
SAE J1441. (2016.). Standard J1441, Subjective Rating Scale for Vehicle Ride and Handling. SAE
International, .
SAE_J3016. (2016). SAE J3016 Taxonomy and Definitions for Terms Related toDriving Automation
Systems for On-Road Motor Vehicles. SAE International.
SAEJ670. (u.d.). SAE J670e - Vehicle Dynamics Terminology. Society of Automotive Engineers,
Warrendale, PA, USA.
Sander, U. (2018). Predicting Safety Benefits of Automated Emergency Braking in Intersections -- Virtual
simulations based on real-world accident data. Göteborg: Chalmers University of Technology,
PhD thesis.
Schuetz, T. (2015). Aerodynamics of Road Vehicles, Fifth Edition. SAE.
Sundström, P., Jacobson, B., & Laine, L. (2014). Vectorized single-track model in Modelica for
articulated vehicles with arbitrary number of units and axles. Lund, Sweden: Modelica

367
 Bibliography

conference 2014, March 10-12, 2014. Hämtat från

http://publications.lib.chalmers.se/publication/190501
Svendenius, J. (2007). Tire Modeling and Friction Estimation. Lund, Sweden: Department of Automatic
Control, Lund University.
Tagesson, K. (2017). Driver-centred Motion Control of Heavy Trucks. Gothenburg, Sweden: Chalmers
University of Technology.
Tagesson, K., & Cole, D. (2017). Advanced emergency braking under split friction conditions and the
influence of a destabilising steering wheel torque. Vehicle System Dynamics, 55.
doi:http://dx.doi.org/10.1080/00423114.2017.1278779
Tiller, M. (2019). Modelica by example. Hämtat från https://mbe.modelica.university/ den 4 June 2019
UN ECE 111. (2001). UNIFORM PROVISIONS CONCERNING THE APPROVAL OF TANK VEHICLES OF
CATEGORIES N AND O WITH REGARD TO ROLLOVER STABILITY. UNITED NATIONS. Hämtat från
http://www.unece.org/fileadmin/DAM/trans/main/wp29/wp29regs/r111e.pdf
Utbult, J., Jonasson, M., Yang, D., & Jacobson, B. (accepted 2020). Energy reduction by power loss
minimisation through wheel torque allocation in electric vehicles. Vehicle Systems Dynamics.
Weber, R. (1981). Reifen Führungskräfte bei schnellen Änderungen von Schräglauf und Schlupf.
Karlsruhe: University of Karlsruhe.
Wedlin, J., Tillback, L.-R., & Bane, O. (1992). Combining Properties for Driving Pleasure and Driving
Safety: A Challenge for the Chassis Engineer. SAE. doi:10.4271/921595
Wipke, K. B., Cuddy, M. R., & Burch, S. D. (1999). ADVISOR 2.1: A User-Friendly Advanced Powertrain
Simulation Using a Combined Backward/Forward Approach. IEEE TRANSACTIONS ON
VEHICULAR TECHNOLOGY, 48(6), ss. 1751-1761.
Wong, J. (2001). Theory of Ground Vehicles (3rd ed.). John Wiley and Sons, Inc., New York.
Yang, D. (2013). Vehicle Dynamics Control after Impacts in Multiple-Event Accidents. Göteborg:
Chalmers University of Technology. Hämtat från
http://publications.lib.chalmers.se/publication/182012-vehicle-dynamics-control-after-
impacts-in-multiple-event-accidents

368
 Modules in the Course
MMF062 Vehicle Dynamics

MODULES IN THE COURSE

MMF062 VEHICLE DYNAMICS
This Appendix is for the (MSc) course MMF062 Vehicle Dynamics Chalmers University of Technology,
Göteborg, Sweden.
It covers each Module, in terms of Learning Objectives and Preparation Reading. Complementing Prepa-
ration material is found on the course web for each Module: Preparation Questions including some Vid-
eos, beside the Classroom Session material: Classroom Lecture and Classroom Problem.

Modules on Chapter 1 INTRODUCTION

Module 1:1 The Subject of Vehicle Dynamics
Learning Objectives Reading
• The course teaching & learning consist of 26 Modules and 3 As-
signments. (One Module is 45 minutes Classroom Session with
preparations done by each course participant.)
• Course Memo
• The course examination consists of 1 Written Examination and 3
Assignment reports and 1 Assignment Quiz
• Most administrative information is found in Course Memo
• Know the major sub-systems by name.
• 1.1 Definition of Vehicle Dy-
• Explain and exemplify Attribute, Function and Requirement on
namics
complete vehicle level.
• 1.2 About this compendium
• Describe where Vehicle Dynamics is used in the vehicle develop-
• 1.3 Automotive engineering
ment: Requirement setting, Design/Re-design, Verification.
• 1.4 Requirement Setting
• Explain Virtual Verification and how it appears in the develop-
• Figure 1-2
ment “V process” for sub-systems and complete vehicle.
• Understand the difference between Design parameters and Oper-
• Figure 1-8
ation/Manoeuvre parameters.

Module 1:2 Modelling

Learning objectives Reading
• Explain and exemplify the 3 modelling stages: Physical (≈FBD), Mathematical
(≈DAE) and Explicit form (≈ODE) modelling.
• Explain how an Explicit form model is used to do a simulation with the Euler • 1.5.1.1
forward as example of integration method. Stages in
• Recognize Dataflow diagrams (or Block diagrams) as a graphical representation (Dynamic)
of an Explicit form model. Model Based
• Differ between parameters/constants and variables/signals in a dynamic model. Engineering
• Realize that causality is the information flow direction (or calculation order) in • Figure 1-9
an explicit form model and that causality of a submodel is dependent of the su-
perior model.

369
 Modules in the Course
MMF062 Vehicle Dynamics
• Construct FBDs for mechanical systems, so they support formulating equations.
• Know that fictive forces can be drawn in FBD (and in compendium they are
drawn with dashed arrows).
• Understand that “Equilibrium with fictive forces, 0 ” is an alterna- • 1.5.2 Me-
tive way to formulate “Equations of motion, ”. chanical/Ma-
• Identify origin of equations: Equilibrium, Compatibility, Constitution and “Algo- chine Engi-
rithms”. neering
• Understand that one can select “Force in elastic part” as state variables, as an • Figure 1-22
alternative to “Position of inertial body”. • Figure 1-23
• Understand the difference between the operating conditions static/low speed,
steady state, stationary oscillating and transient.
• Set up Mathematical models for small systems, given the Physical model.
• Recognize the terms algebraic loop and high index as difficulties that can appear • Figure 1-16
in modelling. • Figure 1-17

Module 1:3 Vehicle Coordinates, Types, Termi-

nology
Learning objectives Reading
• Know vehicle motion degrees of freedom and the ISO8855 • 1.6.1 Vehicle Motions and Co-
vehicle-fix coordinate system ordinate Systems
• Explain difference between models that captures in-road- • Figure 1-42
plane and out-of-road-plane motion • Figure 1-43
• Transform velocities and forces between vehicle-fix and • Figure 1-47
wheel-fix coordinate system • Eq[1.3]
• 1.6.2 Complete Vehicle Model-
• Explain difference between one-track and two-track models
ling Concepts
• Know that centrifugal acceleration appears due to vehicle fix
• Figure 1-52
(rotating) coordinate system.
• Eq [1.7]
• Recognize and explain terms in “1.6.3 Vehicle Dynamics
• 1.6.3 Vehicle Dynamics Terms
Terms”
• 1.6.4 Vehicle Architectures
• Recognize and explain the concept “Function architecture”
• Explain the major differences, relevant for vehicle dynamics,
1.7 Heavy Trucks
between passenger cars and heavy vehicles

370
 Modules in the Course
MMF062 Vehicle Dynamics

Module 1:4: Virtual Verification

Learning objectives Reading
• Understand the business value in virtual verification as alternative
• 1.6.5 Verification Meth-
to building pre-series vehicles
ods with Real Vehicle
• Name and describe differences and similarities between verifica-
• 1.6.6 Verification Meth-
tion methods: Off-line simulation, MIL, SIL, HIL simulation, Driving
ods with Virtual Vehicle
simulator, Test track, Testing/Logging in real traffic/operation
• 1.5.4 Tools
• Understand that different requirement on vehicle model for virtual
• Figure 1-36
verification of functions that actuate vehicle motion and those who
• Figure 1-37
doesn’t. Examples from the latter: Warning/information and Sup-
• Figure 1-38
porting functions, such as estimators.
• Understand opportunities and challenges with modularization and • Figure 1-39
reuse/exchange of submodels for a vehicle manufacturer and a ve- • 1.6.4.3 Virtual Vehicle
hicle supplier Architecture
• Name and describe the standard Modelica • 1.5.4.6 FMI Supporting
Simulation Tools
• Name and describe the standard FMI

Modules on Chapter 2 VEHICLE

INTERACTIONS AND VEHICLE
SUBSYSTEMS
Module 2:1 Tyre Design and Rolling
Learning objectives Reading
• Sketch how a tyre model can be interfaced to a model of the • 2.2 Wheels and Tyre
rest of the vehicle. • Figure 2-8
• Know how to explain rolling resistance (on rigid ground) from
• 2.2.1.6 The Wheel as a Trans-
offset in contact pressure
mission
• Know how rolling resistance comes in when relating wheel
• 2.2.1.7 Tyre Rolling and Radii
longitudinal force to actuated wheel torque 𝑇 (=shaft torque
• Eq [2.1]
+ torque from unsprung parts)
• Eq[2.7]
• Definition of longitudinal slip 𝑠 (𝑅 )⁄|𝑅 |
• Understand that contact length is an intermediate variable, not
a design parameter, e.g.: [𝐷𝑒𝑠 𝑛 𝑟 𝑒𝑡𝑒𝑟𝑠 ] ⇒ 𝐿 ⇒ • 2.2.1.8 Tyre Contact Length
[𝑅𝑅 ] • 2.2.2 Rolling Resistance of
• Know difference between free-rolling and pure rolling wheel Tyres
• Know typical numerical value of rolling resistance coefficient • Figure 2-15
for passenger cars and trucks

371
 Modules in the Course
MMF062 Vehicle Dynamics

Module 2:2 Tyre Longitudinal Force and Slip

Learning objectives Reading
• Explain similarities and differences between how friction force
varies with sliding speed for “sliding block” and for “rolling wheel
• Figure 2-21
of rubber”
• 2.2.3.1.1 Uniform Pressure
• Know that generation of longitudinal force requires longitudi-
Distribution and Known
nal slip 𝑠
Contact Length
• Know that longitudinal tyre slip models are only valid if lateral
• Figure 2-23
slip and forces are low
• Eq[2.9]
• Be able to follow a derivation of simplest possible brush model
that explains the gradual transition between low slip and high slip
• Know tyre model for low force (far below saturation): • 2.2.3.1.2 Longitudinal Tyre
≈ 𝑠 𝑠 Slip Stiffness
• Know tyre model for high force (saturated): • Eq[2.11]
≈ sign(𝑠 ) • Eq[2.12]
• 2.2.3.1.5 Brush Model with
• Know that pressure distribution assumption changes the (𝑠 )
Parabolic Pressure Distri-
curve.
bution
• Find and use the brush model with formula for (𝑠 )
• Eq[2.16]
• Know that there are many empirical tyre models
• 2.2.3.4 Curve Fit Tyre
• Find and use the “Magic tyre formula” for (𝑠 )
Models

Module 2:3 Tyre Lateral and Combined Slip

Learning objectives Reading
• Know that lateral force and slip relation is conceptually same as
longitudinal, but with (𝑠 ) instead of (𝑠 ). • 2.2.4.1.1 Model with Inde-
pendent Bristles
• Brush model concept can also be used for lateral
• Figure 2-36
• Far below saturation: 𝑠 𝑠
• Eq[2.29]
• Near or on saturation: ≈ sign(𝑠 )
• Eq[2.28]
• Lateral tyre models are only valid if longitudinal slip is low.
• Know the difference between lateral wheel slip 𝑠 𝑤 ⁄| | • 2.2.4.2 Lateral Tyre Slip
tan( ) and lateral tyre slip 𝑠 ⁄|𝑅 |. Stiffness
• Know the relation between 𝑠 𝑣 and slip angle : • Eq[2.33]
arctan ( ) or arctan(𝑠 𝑤 ) . • Eq[2.34]
• Know that a version of Magic Tyre Formula is for lateral slip • 2.2.4.3 Influence of Vertical
• Know that is degressively varying with Load
• 2.2.5.3.1 Using Scalar Force
Function of Combined Slip
• Explain combined slip with the two main principles:
2
+ 2<( )2 and ⁄ 𝑠 ⁄𝑠 • Eq[2.43]
• Figure 2-51
• and can be quasi-physically modelled as “downscaled due
• 2.2.5.3 Approximate Com-
to utilisation of , inspired by friction circle” bined Slip Model
• Know that a version of Magic Tyre Formula is for combined slip • Eq[2.39]
• Eq[2.45]

372
 Modules in the Course
MMF062 Vehicle Dynamics

Module 2:4 Tyre Vertical and other Vehicle In-

teractions
Learning objectives Reading
Explain tyre slip models categories:
• Longitudinal/Lateral/Combined • 2.2.6 Summary of Tyre Force vs
• Ideal/Linear to slip and vertical force/Saturated Slip Models
A tyre vertical constitution is relatively well modelled as lin- • 2.2.7 Vertical Properties of Tyres
ear around its static load: ≈ +
Browse:
• List the subsystems relevant for vehicle dynamics • 2.3 Suspension System
• Propose simplest possible models for the subsystems and • 2.4 Propulsion System
how they can be interfaced to the model of “vehicle body • 2.5 (Wheel) Braking System
with wheels” • 2.6 (Wheel) Steering System
• 2.8 Environment Sensing System
• 2.8 Vehicle Aerodynamics
• Use aerodynamic force models with given “ coefficients”:
2 • Figure 2-97
0.5 ∙ ∙ 𝜌 ∙ 𝐴 ∙ 𝑙
• Eq[2.62]
• Understand that a vehicle model needs “driver” and “envi-
• 2.9 Driving and Transport Appli-
ronment” models to form a simulateable model.
cation

Modules on Chapter 3 LONGITUDINAL

DYNAMICS
Module 3:1 Longitudinal: Steady State
Learning objectives Reading
• Describe propulsion system as prime mover, transmission
and sometimes energy buffer. It actuates torques on driven
wheels. • 2.4 Propulsion System
• Describe [𝑡𝑜𝑟𝑞𝑢𝑒 𝑠𝑝𝑒𝑒 ] characteristics for different prime • 2.4.2 Prime movers
movers: 𝑇 𝑓( ) • Figure 2-70
• Know transmissions transformation of [𝑇 ] to [ 𝑤 𝑤 ] • 2.4.3 Transmissions
with different ratios on different gears. • Eq[2.52]
• Know approximate forms on energy consumption charac- • Eq[2.53]
𝑇𝜔
teristics: 𝑛 ⁄ 𝑡 𝑛 𝑓 ( 𝑇 𝑒 𝑟) (or 𝜂 )
𝑜

• 3.2.1 Traction Diagram

• Construct traction diagram from given propulsion system, • Figure 3-1
vehicle and road. • 3.2.2 Power and Energy Losses
• Use a traction diagram to explain functions, such as Top • Figure 3-5
Speed and Gradeability. • 3.2.3 Functions After Start
• Figure 3-7
• 3.2.5 Steady State Vertical Force
• Calculate steady state vertical force distribution over axles. Distribution between Axles
• Calculate when road friction limits propulsion force on an • Eq[3.5]
axle (for steady state). • 3.2.6 Friction Limit, Eq[3.7]
• Calculate Startability and Gradeability. • 3.2.7 Start Functions
• Figure 3-10, Figure 3-11

373
 Modules in the Course
MMF062 Vehicle Dynamics

Module 3:2 Longitudinal: Long Cycles

Learning objectives Reading
• Know that (𝑡) is a traditional driving cycle, used for legislation and
vehicle development. • 3.3.1 Description For-
• Know that driving cycles (𝑡) or (𝑠) have shortcomings. Alterna- mats of Vehicle Opera-
tives are more physical models of the transport operations, e.g. de- tion
scribing legal speed, road gradient and curves, traffic and driver.
• Calculate influence on acceleration from rotating inertias on vehicle
longitudinal acceleration • 3.3.2 Rotating Inertia
• Understand that influence from rotating parts can be modelled with • Eq[3.10]
either a “k-factor”, slipping tyres or clutch or compliant shafts.
• Calculate cycle measures (such as energy consumption) for simple
• 3.3.4 Functions Over
and short cycles.
Cycles
• Understand how more complex and longer cycles can be analysed,
• Eq[3.12]
using a programming/simulation tool.
• 3.3.5 Load Transfer
• Calculate longitudinal load transfer due to and body forces (grav-
with Rigid Suspension
ity and aerodynamics) assuming rigid suspension,
• Figure 3-24, Eq[3.13]
• Calculate acceleration performance for vehicle models with rigid sus-
• 3.3.6 Acceleration
pension
• Eq[3.15]

Module 3:3 Longitudinal: Short Events

Learning objectives Reading
• Figure 2-89
• Know the principles of a brake system: Hydraulic or pneumatic pressure
• Figure 2-90
applied on a friction brake. Applied via Brake pedal or request from Con-
• 3.4.1 Typical Test
trol algorithms.
Manoeuvres
• List some complete vehicle manoeuvres which are relevant for brake sys-
• 3.4.2 Deceleration
tem verification
Performance
Front/rear axle locking:
• Know that locking (or spinning) rear axle tends to make a two-axle vehi-
cle yaw-unstable.
• Know that locking (or spinning) front axle tends to reduce steerability of a • 3.4.4 Brake Pro-
two-axle vehicle; vehicle ploughs straight ahead. portioning
Brake proportioning diagram: • Eq[3.16]
• Construct Brake proportioning diagram from brake system and vehicle. • Figure 3-28
• Use a Brake proportioning diagram to propose Brake Distribution when.
• Calculate for which deceleration or brake force one axle locks and
whether it is front or rear axle.

Module 3:4 Longitudinal: Control Functions and

Heavy Vehicle Brake System and ABS
Learning objectives Reading
• List sensors relevant for longitudinal control: WSS, , pedals, fwd Browse:
looking, estimated torques from propulsion and braking systems. • 3.5.1 Longitudinal Control
• Know that propulsion and braking system are main actuators for • 3.5.3 Longitudinal Motion
longitudinal control function. Function Architecture

374
 Modules in the Course
MMF062 Vehicle Dynamics

Learning objectives Reading

• Describe the difference between CC (speed control) and ACC
• 3.5.2.2 Cruise Control and
(adding distance control)
Adaptive Cruise Control (CC,
• Understand that arbitration is needed between CC (or ACC) and
ACC) *
Pedal driving
• Explain ABS control in a tyre force-vs-slip diagram
• 3.5.2.3 Anti-Lock Braking
• Understand that ABS has two purposes to be balanced: Large
System, ABS *
longitudinal brake forces and Maintain a margin for lateral tyre
• Figure 3-39
forces
• Describe differences between passenger car and heavy vehicle
• Figure 2-91
brake systems (media, pressure levels, “by-wire level”, …)
• 3.5.2.3 Anti-Lock Braking
• Describe differences in ABS function between heavy vehicles
System, ABS *
and passenger cars

Modules on Chapter 4 LATERAL

DYNAMICS
Module 4:1 Lateral: Linear One-Track Model
Learning objectives Reading
Explain main assumptions for these categories of lateral models:
• Low speed: Neglect all inertial forces ( + ), • 4.1.1 Lateral Model
Steady state at high speed cornering: Neglect , or Categorization
Transient: Include all ( + ) • Repeat: 1.6.2 Com-
• In road plane (irp) or Out-of-road plane (oorp) plete Vehicle Mod-
• One-track or Two-track Models elling Concepts
• Tyre models: Ideal-tracking, Linear, Saturated and Combined slip
• Derive the (transient, lateral dynamics) “linear one-track model”, at least
the simpler way in Figure 4-48 • 4.4.2 One-Track
• Write the “linear one-track model” on matrix form Model
• Derive and use the “linear one-track model” for simulation with known • Figure 4-46
(𝑡). • Eq [4.47]
• Discuss validity limits for “linear one-track model” above: level ground, 2 • Figure 4-48
axles, low CoG, tyre forces in ground plane far from saturation, small vari- • Eq[4.50]
ations in , no yaw torque from individual wheel torque actuation.

Module 4:2 Lateral: Low Speed, Ackermann

Learning objectives Reading
• Know that left and right (road) wheels are normally geomet-
Repeat:
rically constrained to each other and to steering (hand)
wheel • 1.6.1.6.1 Path with Orientation
• Explain and draw Ackermann steering geometry: between • 2.6.1 Chassis Steering Geometry
left and right wheel and between axles • Figure 2-92
• Recap that we can postprocess a solution ( ) (𝑡) • Figure 2-93
( )
to a path 𝑥 (𝑡) 𝑦(𝑡) with orientation (𝑡) • 2.2.6 Summary of Tyre Force vs
Slip Models
• Recap lateral tyre model: 𝑠

375
 Modules in the Course
MMF062 Vehicle Dynamics

Learning objectives Reading

• 4.2.1 Low Speed Model, Acker-
Derive and use low speed models without forces of a two-axle ve-
mann, without Forces
hicles.
• Eq[4.1]
• 4.2.2 Low Speed Functions
Explain the functions: Turning radius, Swept Path Width and Off-
• Figure 4-1
Tracking
• Figure 4-2
• 4.2.3 Low Speed Model, Acker-
Understand how forces can be included in low speed model. And, mann, with Forces
if Ackermann geometry, either as 𝑠 0 or 𝑠 . • Figure 4-4
• Eq[4.3]

Module 4:3 Lateral: Low Speed, Non-Acker-

mann, Articulation
Learning objectives Reading
• 4.2.4 Low Speed Model,
Non-Ackermann
• Understand how Non-Ackermann or manoeuvre with large forces
• Figure 4-6
requires/proposes to include forces with 𝑠 .
• Eq[4.5]
• Figure 4-8
• Understand that articulated vehicles have a transient (in articula-
tion angle) which distinguish between circle and transient low
speed driving.
• Construct instantaneous points of rotation for an articulated vehi- • 4.2.5 Articulated Vehicles
cle modelled without tyre forces. Derive and use such models.
• Reason about how non-zero forces (due to non-Ackermann geom-
etry or other) affect low speed manoeuvres of articulated vehicles

Module 4:4 Lateral: Steady State Cornering

without Load Transfer
Learning objectives Reading
• Understand that 3 scalar measures fully define a Steady State Corner-
ing operating condition for a general vehicle, e.g. 𝑅𝑝 .
• Understand that, with a certain vehicle, the driver is limited to 2 “re- • 4.3.1 Steady State
quest channels”: Pedals + SteeringWheel. Driving Manoeuvres
• So, with a certain vehicle, the driver can select radius 𝑅𝑝 with steering • Figure 4-13
and speed with pedals, while the side slip angle is a result of how
vehicle (and road surface) is built.
• 4.3.2 Steady State
• Derive and use model for steady state cornering. Derivation either
One-Track Model
from transient model “linear one-track” (by neglecting terms) or from
• Figure 4-17
Physical model (at least the simpler way in Figure 4-21).
• Figure 4-21
• Find and use the model for steady state for cornering
• Eq[4.9]

376
 Modules in the Course
MMF062 Vehicle Dynamics

Module 4:5 Lateral: Steady State Cornering

Functions
Learning objectives Reading
• Understeer gradient is a steady state cornering
measure, but maybe be defined differently: [ ⁄𝑁],
[ ] or [ ⁄( ⁄𝑠 2 )]
• Understeer gradient can be defined for straight
ahead driving (linearized around zero lateral ac-
• 4.3.3 Under-, Neutral- and Over-steering *
celeration) or linearized around a non-zero lateral
acceleration.
• Understeering is sometimes used as a momentane-
ous measure during an arbitrary transient ma-
noeuvre
• 4.3.4 Required Steer Angle
Sketch these steady state diagrams with as abscissa: • Eq[4.17]
• Steering Wheel Angle • Figure 4-25
• Yaw Velocity Gain • Eq[4.18]
• Lateral Acceleration Gain • 4.3.5 Critical and Characteristic Speed *
Know that Handling diagram is a way to visualize Figure 4-28
steady state cornering behaviour • Figure 4-30
• Browse: 4.3.9 Handling Diagram
Understand that understeering can be varied in many
ways, e.g.:
• Longitudinal load transfer due to tyre longitudinal
forces, acceleration and deceleration • 4.3.7 How Design Influences
• Roll stiffness distribution between axles (stiffer • Figure 4-26
front gives understeered) • Figure 4-32
• Side-force-steering and roll-steering • Browse: 4.3.7 How Design Influences
• Wheel torque distribution, left-right and front-rear Steady State Gains
(more wheel torque front gives understeered)
Understand that yaw responsiveness can be both too low
and too high

377
 Modules in the Course
MMF062 Vehicle Dynamics

Module 4:6 Lateral: Steady State Cornering

with Load Transfer
Learning objectives Reading
• Describe wheel/axle suspension as consisting of: Linkages, Elasticities
(springs, anti-roll bars, bushings, flexible links) and Dampers (shock ab- • 2.3 Suspension Sys-
sorbers, friction). tem
• Know that each wheel can be individually suspended, or each axle sus- • Figure 4-37
pended as one unit.
• Lateral load transfer on each axle depends on suspension linkage and
elasticities. Dampers does not influence in steady state.
• One can model suspension with 2 pivot points (wheel-individual pivot
• 4.3.10.3 Lateral
points or 2 points for a rigid axle) or with 1 roll centre height for each
Load Transfer
axle.
• Figure 4-38
• If roll centre was in ground plane, the load transfer appears first when a
• Eq [4.38]
roll angle appears; Force play would go "via elasticities".
• If roll centre was at CoG height, the load transfer happens directly with-
out a roll angle; Force play would go "via linkage".
• Know that Off-Tracking can be defined for also high speed (it was de- • 4.3.11 High Speed
fined for low speed before) Steady State Vehicle
• Know that wheel-lift appears at different lateral acceleration for differ- Functions
ent axles • Figure 4-40

Module 4:7 Lateral: Stationary Oscillating

Steering
Learning objectives Reading
• 4.4.1 Stationary Oscillating Steering
Tests
• Repetition: 4.4.2 One-Track Model
• Understand that stationary oscillating steering re-
• Eq [4.50]
quires a transient model, but the solutions are sta-
• Repetition (of Mathematics): Fourier and
tionary (constant amplitude, frequency and phase)
Laplace transform in 1.5.1.9 Mathematics
• Know how to use “linear one-track model” and Fou-
and Notation Conventions
rier-transform to find frequency responses
• 4.4.3.1.1 Solution with Fourier Trans-
form
• Eq [4.51]
• 4.4.3.3.2 Lateral Acceleration Frequency
• Sketch Yaw Rate amplitude response as function of Response
steering frequency for different • 4.4.3.4 Other Frequency Responses to
• Sketch Lateral Acceleration response as function of Oscillating Steering
steering frequency for different • Figure 4-52
• Figure 4-55

378
 Modules in the Course
MMF062 Vehicle Dynamics

Module 4:8 Lateral: Transient Manoeuvres,

incl. Step Steer
Learning objectives Reading
• Browse: 4.5.1 Transient Driving
• Transient driving is used as most arbitrary way of driving Manoeuvres *
• Expanding the “linear one-track model” to varying and sat- • 4.5.2 One-Track Models, with-
urated tyres gives a useful model for many transient driving out Lateral Load Transfer
manouvres • Figure 4-46
• Eq [4.59]
• Articulated vehicles:
Browse:
o Each articulation point is represented in mathematical
• 4.5.2.2 Articulated Vehicles
model: 2 equilibrium Eqs and 2 compatibility Eqs
• Figure 4-62
o The articulation points make the explicit form model
• 4.5.2.3 Cambering Vehicle
much harder to find, due to “high index”
Model
o For each added unit: 2 additional states appear: and
• Figure 4-68
• For cambering vehicles, the roll angle is important to model
The load transfer in transient models adds, compared to steady
state cornering and transient one-track model:
• Forces in dampers Browse:
• Roll dynamics • 4.5.3.2 Example of Explicit Form
• Heave and Pitch Dynamics depending on suspension models Model; Two-Axle Vehicle, Driver
Example model: and Environment
• There is an example of relatively complete model (driver, ve- • Figure 4-73
hicle environment) with proposed modularity and interfaces. • Figure 4-74
• The model shows how Control and Actuation can be mod-
elled as a separate module.
If tyres are far below grip limit, step steering response can be • 4.5.4.1 Mild Step Steering Re-
evaluated with the linear one-track model. Solutions can be on sponse
closed form. • Eq [4.68]
If tyres are closer to grip limit, step steering response requires
• 4.5.4.2 Violent Step Steering
more advanced models. Solutions generally requires time simu-
Response
lation.
Long articulated vehicles have special functions/measures de-
fined. Rearward amplification is one, which often reveals that • Browse: 4.5.6 Long Heavy Com-
yaw velocity is amplified to typically the double for the rear- bination Vehicles
most unit.

379
 Modules in the Course
MMF062 Vehicle Dynamics

Module 4:9 Lateral: Vehicle Simulator as Verifi-

cation Tool
Learning objectives Reading
• Understand the cons and pros with driver in the loop simulators (DIL)
compared to “office-simulations” (MIL, SIL, HIL)
• Understand the cons and pros with driver in the loop simulators (DIL)
compared to real test-vehicles • 1.6.6.1 Testing
• Understand the cons and pros with driver in the loop simulators with and with Real Driver
without motion platform • Figure 1-58
• Understand that Motion-queuing bridges the inevitable gap between simu- • 1.5.4.6FMI Sup-
lated vehicle motion and motion platform motion. porting Simula-
• Know state of the art for in-vehicle-programme use of simulators. How far tion Tools
have simulators come?
• Know some research questions on the research/development front for
simulators.

Module 4:10 Lateral: Control Functions

Learning objectives Reading
ESC is a function which corrects the yaw velocity when lateral grip is
lost on one axle (under- or oversteering situation).
• Outer front wheel is primary wheel to brake when detecting
• 4.6.2.1 Electronic Stability
over-steering.
Control, ESC *
• Inner rear wheel is primary wheel to brake when detecting un-
• 4.6.2.2 Roll Stability Con-
der-steering.
trol, RSC *
• RSC brakes away lateral grip on front axle.
• Brakes are main actuator for ESC and RSC, but development to-
wards using more actuators is on-going.
• 4.6.2.3 Lane Keeping Aid,
LKA, LCA/AES and AD are functions with automatic lateral interven-
LKA *
tions in increasing levels.
• 4.6.2.4 Lateral Collision
• LKA is a relatively stand-alone function.
Avoidance/Automatic
• LCA/AES needs to be designed together with (longitudinal) AEB.
Emergency Steering,
• AD is a very wide term. AD ranges from allowing LKA and (longi-
LCA/AES *
tudinal) ACC at the same time to full AD with selection of route
• 4.6.2.5 Automated Driving
and absent driver.
(AD)

380
 Modules in the Course
MMF062 Vehicle Dynamics

Modules on Chapter 5 VERTICAL

DYNAMICS
Module 5:1 Vertical: Suspension and Stationary
Oscillation theory
Learning objectives Reading
• Understand that vertical dynamics often is studied in frequency domain• 5.1 Introduction
• Know that 3 areas of functions are: Comfort, Road grip and Fatigue • Figure 5-1
• 5.2 Stationary Oscil-
• Two frequency domains: time frequency (𝑡) and spatial frequency (𝑥) lations Theory
• Definition and usage of Root Mean Square, 𝑅𝑀𝑆( ) [𝑢𝑛 𝑡: 𝑠 𝑒 𝑠 ] • Figure 5-3
• Definition and usage of Power Spectral Density, 𝑆𝐷: • Eq [5.4]
𝑚
o 𝑆𝐷( ) and 𝑆𝐷( ) [𝑢𝑛 𝑡: )] • Eq [5.8]
⁄𝑚
• Definition and usage of Transfer Function 𝐻 → ( ) ̂ ⁄ ̂ • Eq [5.11]
• Eq [5.13]

Module 5:2 Vertical: Road Models, 1D Vehicle

Models
Learning objectives Reading
• Typical roads have large amplitude [ ] for long waves (=low spatial fre- • 5.3 Road Models
quency) and small amplitudes for short waves (=high spatial frequency) • Eq [5.27]
• Calculate RMS value of a response variable if knowing: road (Φ ), • Figure 5-5
Vehicle speed and Transfer function 𝐻 𝑟 → , using Eq [5.31]. • Eq [5.31]
• 5.4.3 1D Model
with 2 Dynamic
• Derive, and use transfer functions, for “1D model with 2 dynamic dofs”
dofs
• Understand that different models used to study 1D (vertical) motion
• Figure 5-12
• Figure 5-13

Module 5:3 Vertical: Comfort and Grip

Learning objectives Reading
• Understand that human’s sensitivity to accelerations is different for dif-
ferent frequencies, with maximum sensitivity for 𝑓 4. .8 𝐻 • 5.5 Functions for
Stationary Oscilla-
• Calculate 𝑤 value of a response variable if knowing: road (Φ ),
tions
Vehicle speed and Transfer function 𝐻 𝑟 → , using Eq [5.53].
• Figure 5-20
• Calculate Time averaged whole-body vibration exposure value 𝑤 𝑣 if
• Eq [5.51]
knowing 𝑤 for the time periods using Eq [5.52] • Eq [5.52]
• Understand that Road grip can be measured with amplitude of vertical • Eq [5.53]
force in tyre contact patch Δ
• 5.6 Variation of Sus-
Plot and use frequency response diagrams to analyse for which frequencies pension Design
Comfort, Fatigue and Road grip becomes worse or better for variations of • Figure 5-23
suspension design parameters. • Figure 5-24
• Figure 5-25

381
 Modules in the Course
MMF062 Vehicle Dynamics

Module 5:4 Vertical: Heavy Vehicle Suspension

Learning objectives Reading
• Describe differences between passenger • Repeat/Browse: 2.3 Suspension System
car and heavy vehicle suspension systems • 3.4.5.2 Transient Load Transfer and Non-Trivial
(number of axles, variation in payload, …) Linkage
• Qualitatively explain roll-steering, ac- • 3.4.7 Pitch Functions at Transient Wheel Torques
tive/reactive suspension linkage, … • 4.3.10.3 Lateral Load Transfer

382