Fofana PHD
Fofana PHD
Fofana PHD
DOCTOR OF PHILOSOPHY
Fofana, Adama
Award date:
2016
Awarding institution:
Coventry University
Link to publication
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners
and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.
• Users may download and print one copy of this thesis for personal non-commercial research or study
• This thesis cannot be reproduced or quoted extensively from without first obtaining permission from the copyright holder(s)
• You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal
Take down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately
and investigate your claim.
By
Adama Fofana
BEng European Engineering
MSc Control Engineering
January 2016
Contents
1 Introduction 1
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Context and problem statement . . . . . . . . . . . . . . . . . . . 1
1.3 Aim & objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Research methodologies . . . . . . . . . . . . . . . . . . . . . . . 4
1.5 Contributions and deliverable . . . . . . . . . . . . . . . . . . . . 5
1.6 Outline of the thesis . . . . . . . . . . . . . . . . . . . . . . . . . 7
I
2.4.7 Motivation to use Evolutionary Algorithm & Swarm Intel-
ligence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.5 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3 Powertrain modelling 37
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.2 Vehicle model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.2.1 Description of the vehicle . . . . . . . . . . . . . . . . . . . 38
3.2.2 Model validation . . . . . . . . . . . . . . . . . . . . . . . 40
3.2.3 Fuel consumption and CO2 emissions calculation . . . . . 42
3.3 Gear shift schedule design . . . . . . . . . . . . . . . . . . . . . . 43
3.3.1 Equation of motion . . . . . . . . . . . . . . . . . . . . . . 43
3.3.2 Design principle based acceleration . . . . . . . . . . . . . 46
3.3.3 Design principle based traction force . . . . . . . . . . . . 48
3.3.4 Design principle based minimum fuel, BSFC . . . . . . . . 49
3.3.5 Criteria of gear shift map . . . . . . . . . . . . . . . . . . 51
3.3.6 Final gear shift map design . . . . . . . . . . . . . . . . . 53
3.4 Gear ratio design . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.4.1 Main formulas for gear ratio design . . . . . . . . . . . . . 54
3.4.2 The largest gear ratio selection . . . . . . . . . . . . . . . 56
3.4.3 The smallest gear ratio selection . . . . . . . . . . . . . . . 56
3.4.4 The intermediate gear selection . . . . . . . . . . . . . . . 59
3.4.5 Criteria of gear ratio layout . . . . . . . . . . . . . . . . . 62
3.5 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4 Problem formulation 69
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.2 Gear shift map problem formulation . . . . . . . . . . . . . . . . . 70
4.2.1 Objective formulation . . . . . . . . . . . . . . . . . . . . 70
4.2.2 Constraint formulation . . . . . . . . . . . . . . . . . . . . 74
4.2.3 Design variables . . . . . . . . . . . . . . . . . . . . . . . . 76
4.2.4 Variable bounds . . . . . . . . . . . . . . . . . . . . . . . . 78
4.3 Gear ratio problem formulation . . . . . . . . . . . . . . . . . . . 79
4.3.1 Intermediate gear . . . . . . . . . . . . . . . . . . . . . . . 79
4.3.2 Last gear ratio . . . . . . . . . . . . . . . . . . . . . . . . 82
4.4 Handling of the objective functions . . . . . . . . . . . . . . . . . 83
4.4.1 Gear shift map with multi-objective functions . . . . . . . 83
4.4.2 Gear ratio with multi-objective functions . . . . . . . . . . 85
II
4.5 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 85
III
7.7 MOCS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
7.7.1 Results based on weighting set 1 . . . . . . . . . . . . . . . 159
7.7.2 Results based on weighting set 2 . . . . . . . . . . . . . . . 163
7.8 Combined gear ratio & shift map optimisation . . . . . . . . . . . 165
7.9 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 172
Index 203
References 205
IV
Contents
E Leading solution for CO2 emissions for Pareto and weighting sum229
E.1 Leading Pareto solution for set 1 . . . . . . . . . . . . . . . . . . 229
E.2 Leading Pareto solution for set 2 . . . . . . . . . . . . . . . . . . 230
E.3 Leading Pareto solution for set 3 . . . . . . . . . . . . . . . . . . 231
E.4 Leading weighting sum solution for set 1 . . . . . . . . . . . . . . 231
E.5 Leading weighting sum solution for set 2 . . . . . . . . . . . . . . 232
E.6 Leading weighting sum solution for set 3 . . . . . . . . . . . . . . 233
V
List of Figures
VI
Contents
3.13 BSFC curves with shift point between two adjacent gears compare
with standard and traction BSFC shift map . . . . . . . . . . . . 52
3.14 Initial shift map study . . . . . . . . . . . . . . . . . . . . . . . . 53
3.15 Performance power curves, where P Ex1, P Ex2 and P Ex3 are the
excess power of Over-revving (racing car), optimality and Under-
revving (passenger car) with their respective maximum vehicle
speed Vmax1, Vmax2 and Vmax3 . . . . . . . . . . . . . . . . . . 57
3.16 Gradeability performance . . . . . . . . . . . . . . . . . . . . . . . 63
3.17 Acceleration curves performance . . . . . . . . . . . . . . . . . . . 64
3.18 Traction curves performance . . . . . . . . . . . . . . . . . . . . . 64
3.19 Gear ratio, step and mean value . . . . . . . . . . . . . . . . . . . 65
3.20 Acceleration curves performance . . . . . . . . . . . . . . . . . . . 66
3.21 Saw curves performance . . . . . . . . . . . . . . . . . . . . . . . 67
3.22 Fuel consumption at 120 km/h . . . . . . . . . . . . . . . . . . . 67
4.1 The distance (dtk ) between the reference point O(wref , Tref ), and
the Upshift 1 (Up1 ) and Upshift 2 (Up2 ) at 0%, 40% and 100%
throttle positions, respectively. The grey dotted line represents the
engine maximum torque. The engine speed varies from the mini-
mum stable speed, stalling speed, to the maximum engine speed.
Z1 represents the zone with the most efficient operating point, Z2
and Z3 are zones with higher fuel consumption . . . . . . . . . . . 72
4.2 Conversion of a Downshift (∆V Dwg,g−1,tk ) and its corresponding
Upshift (∆V U pg−1,g,tk ) into design variables . . . . . . . . . . . . 76
VII
Contents
5.8 Multi-Objective Cuckoo with hybrid operators for gear ratio opti-
misation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
5.9 Firefly operator for gear ratio optimisation . . . . . . . . . . . . . 100
5.10 Combined gear ratio & shift map optimisation . . . . . . . . . . . 103
5.11 Semi-log plot results representing the spread of offspring based
on two parents P1 and P2. The offspring are generated using
GA crossover operators: direct, intermediate and extended line
recombination, also with hybrid MOCS operators: Levy Flight,
Bat, Firefly and Flower pollination . . . . . . . . . . . . . . . . . 105
5.12 Semi-log plot results representing the spread of offspring based on
two parents P1 and P2. The offspring are generated using GA
crossover operators: direct, intermediate and extended line recom-
bination, also with hybrid MOCS operators: Levy Flight, Bat,
Firefly and Flower pollination. This figure illustrates a zooming
view around the two parents. . . . . . . . . . . . . . . . . . . . . 106
VIII
Contents
6.5 Objective functions representing CO2 and IRP versus zone 1, zone
2 and zone 3. Marker with blue ’point’, red ’circle’ and magenta
’plus sign’ present setting 1, setting 2 and setting 3, respectively. . 115
6.6 Set of Optimised shift map obtained from Pareto optimal solution
denoted by A1, A2, A3, and A4, respectively (see Figure 6.1) . . . 125
6.7 Set of Optimised shift map obtained from Pareto optimal solution
denoted by A1, A2, A3, and A4, respectively (see Figure 6.1). The
NEDC was used to calculate the engine operating point (EOP).
The EOP of standard shift map is presented with red mark (+),
and optimised shift map presented with blue mark (o). . . . . . . 126
6.8 Gear change position based on 4 optimised shift maps (A1, A2, A3
and A4 (see Figure 6.1)) compared with initial shift map. . . . . . 127
6.9 Radar plot of objectives functions (CO2 emissions, IRP, Distances,
Gear change frequency, time spent on higher gear, different zone on
BSFC map defined as Zone 1, Zone 2, Zone 3) based on 4 optimised
shift maps (A1, A2, A3 and A4 (see Figure 6.1)) compared with
initial shift map. . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
6.10 Set of Optimised shift map obtained from Pareto optimal solution
denoted by B1, B2, B3, and B4, respectively (see Figure 6.1) . . . 129
6.11 Set of Optimised shift map obtained from Pareto optimal solution
denoted by B1, B2, B3, and B4 respectively (see Figure 6.1). The
NEDC was used to calculate the engine operating point (EOP).
The EOP of standard shift map is presented with red marked (+),
and optimised shift map presented with blue marked (o). . . . . . 130
6.12 Gear change position based on 4 optimised shift maps (B1, B2, B3
and B4 (see Figure 6.1)) compared with initial shift map. . . . . . 131
6.13 Radar plot of objectives functions (CO2 emissions, IRP, Distances,
Gear change frequency, time spent on higher gear, different zone on
BSFC map defined as Zone 1, Zone 2, Zone 3) based on 4 optimised
shift maps (B1, B2, B3 and B4 (see Figure 6.1)) compared with
initial shift map. . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
6.14 Set of Optimised shift map obtained from Pareto optimal solution
denoted by C1, C2, C3, and C4 respectively (see Figure 6.1) . . . 133
IX
Contents
6.15 Set of Optimised shift map obtained from Pareto optimal solution
denoted by C1, C2, C3, and C4 respectively (see Figure 6.1). The
NEDC was used to calculate the engine operating point (EOP).
The EOP of standard shift map is presented with red mark (+),
and optimised shift map presented with blue mark (o). . . . . . . 134
6.16 Gear change position based on four optimised shift maps (C1, C2,
C3 and C4 (see Figure 6.1)) compared with initial shift map. . . . 135
6.17 Radar plot of objectives functions (CO2 emissions, IRP, Distances,
Gear change frequency, time spent on higher gear, different zone
on BSFC map defined as Zone 1, Zone 2, Zone 3) based on four op-
timised shift maps (C1, C2, C3 and C4 (see Figure 6.1)) compared
with initial shift map. . . . . . . . . . . . . . . . . . . . . . . . . . 136
7.1 Optimised gear shift with reserve power under 3.2 kW . . . . . . . 142
7.2 Optimised gear shift results over the NEDC with reserve power
under 3.2 kW for one period of urban driving cycle . . . . . . . . 143
7.3 Optimised gear shift results over the NEDC with reserve power
under 3.2 kW for extra-urban driving cycle . . . . . . . . . . . . . 143
7.4 Optimised gear shift map repaired to enforce minimum reserve
power above 3.2 kW . . . . . . . . . . . . . . . . . . . . . . . . . 144
7.5 Simulation with a repaired gear shift map showing that the mini-
mum reserve power is met at all time over the whole NEDC. This
illustration shows one period of urban driving cycle . . . . . . . . 145
7.6 Simulation with a repaired gear shift map showing that the mini-
mum reserve power is met at all time over the whole NEDC. This
illustration shows one period of extra-urban driving cycle . . . . . 146
7.7 Illustration of GES. The full red, blue and green lines denote the
three solutions produced by the GES . . . . . . . . . . . . . . . . 147
7.8 Convergence of MOGA and MOGAOp based on Pareto under dif-
ferent objective combinations denoted set 1, set 2, and set 3 . . . 152
7.9 Convergence of MOGA, MOGAAg , MOGAOp and MOGAOp/Ag
based on Pareto under different objective combinations denoted
set 1, set 2, and set 3 . . . . . . . . . . . . . . . . . . . . . . . . . 153
7.10 Convergence of MOGA and MOGAOp based on Pareto under dif-
ferent objective combinations denoted set 1, set 2, and set 3 . . . 156
7.11 Convergence of MOGA and MOGAOp based on Pareto under dif-
ferent objective combinations denoted set 1, set 2, and set 3 . . . 157
X
Contents
7.12 Trade-off between CO2 against IRP, bandwidth and overall weighted
combination cost based on random selection of operators: Levy
Flight, Firefly, Bat and Flower pollination. The blue, red and
green circles denote the solutions trade-off after 10, 20 and 30 gen-
erations, respectively. . . . . . . . . . . . . . . . . . . . . . . . . . 161
7.13 Performance of gear ratio optimisation based on random selection
of operators: Levy Flight, Firefly, Bat and Flower pollination . . . 162
7.14 Performance of gear ratio optimisation based on Firefly . . . . . . 164
7.15 Trade-off with Firefly between CO2 against IRP, bandwidth and
overall weighted combination cost. The blue, red and green cir-
cles denote for solutions trade-off after 10, 20 and 30 generations
respectively . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
7.16 Illustration of a set of combination representing, a set of gear ratio,
with its initial and optimised gear shift maps respectively . . . . . 166
7.17 Ideal optimised gear shift map in terms of CO2 based on combSet1 167
7.18 Gear shift map results based on combSet1 of the most suitable re-
sults. The plots are: optimised gear shift map, reserve power, gear
selection and speed range each gear ratio compared to the original
shift map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
7.19 Gear shift map results based on combSet1 of the most suitable re-
sults. The plot presents the vehicle speed of the NEDC reflected
on the shift map. . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
7.20 Selection of optimised gear shift maps at early generation . . . . . 170
7.21 Gear shift map results based on combSet1 at the earlier generation.
The plots are: optimised gear shift map, reserve power, gear se-
lection and speed range for each gear ratio to the original shift
map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
7.22 Gear shift map results based on combSet1 at the earlier generation.
The plot presents the vehicle speed of the NEDC reflected on the
shift map. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
8.1 The New European Driving cycle (NEDC), based one urban part,
composed of four repeated ECE-15 driving cycles, and one extra-
urban driving cycle . . . . . . . . . . . . . . . . . . . . . . . . . . 177
8.2 Best selected optimised gear shift map 1 . . . . . . . . . . . . . . 180
8.3 Best selected optimised gear shift map 2 . . . . . . . . . . . . . . 181
8.4 Best selected optimised gear shift map 3 . . . . . . . . . . . . . . 182
XI
Contents
8.5 Best selected optimised gear shift map 1, gear position . . . . . . 182
8.6 Best selected optimised gear shift map 2, gear position . . . . . . 183
8.7 Best selected optimised gear shift map 3, gear position . . . . . . 183
8.8 Best selected optimised gear shift map 1, reserve power . . . . . . 184
8.9 Ideal selected optimised gear shift map 2, reserve power . . . . . . 184
8.10 Ideal selected optimised gear shift map 3, reserve power . . . . . . 185
8.11 Driveability test of Optimised shift map before loading on it in
ECU for rolling road test . . . . . . . . . . . . . . . . . . . . . . . 187
8.12 Engine operating point of standard shift map and optimised shift
map 1 from rolling road . . . . . . . . . . . . . . . . . . . . . . . 188
8.13 Engine operating of standard shift map and optimised shift map 2
from rolling road . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
8.14 This figure illustrates the first optimised shift map, engine maxi-
mum torque (estimate), actual torque and reserve torque. Engine
speed and estimated reserve power . . . . . . . . . . . . . . . . . 190
8.15 This figure illustrates the second optimised shift map, engine max-
imum torque (estimate), actual torque and reserve torque. Engine
speed and estimated reserve power . . . . . . . . . . . . . . . . . 190
8.16 Rolling road vehicle speed and gear shift position under initial shift
map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
8.17 Rolling road vehicle speed and gear shift position under optimised
shift map 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
8.18 Rolling road vehicle speed and gear shift position under optimised
shift map 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
B.1 This figure represents the initial calibration shift map of SAIC 6
speed DCT. The Upshift 1-2, Upshift 2-3, Upshift 3-4, Upshift 4-5
and Upshift 5-6 are represented by solid black, blue, red, magenta
and green lines, respectively. The Downshift 2-1, Downshift 3-2,
Downshift 4-3, Downshift 5-4 and Downshift 6-5 are described by
dotted black, blue, red, magenta and green lines, respectively. . . 217
C.1 Vehicle ROEWE 950 coast down data. These data represents ve-
hicle speed, time and distance recorded during vehicle deceleration
from 130 km/h to 5 km/h . . . . . . . . . . . . . . . . . . . . . . 219
C.2 Actual coastdown test performed on chassis dynamometer of ROEWE
950 for gear shift map fuel and emissions assessment . . . . . . . . 220
XII
Figure
E.1 Leading Pareto shift map for minimum CO2 emissions under set 1
condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
E.2 Leading Pareto shift map for minimum CO2 emissions under set 2
condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
E.3 Leading Pareto shift map for minimum CO2 emissions under set 3
condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
E.4 Leading weighting sum shift map for minimum CO2 emissions un-
der set 1 condition . . . . . . . . . . . . . . . . . . . . . . . . . . 232
E.5 Leading weighting sum shift map for minimum CO2 emissions un-
der set 2 condition . . . . . . . . . . . . . . . . . . . . . . . . . . 232
E.6 Leading weighting sum shift map for minimum CO2 emissions un-
der set 3 condition . . . . . . . . . . . . . . . . . . . . . . . . . . 233
F.1 Gear shift map, engine reserve power, gear selection and operating
speed range for each gear ratio for combSet1 . . . . . . . . . . . . . 234
F.2 Gear shift map, engine reserve power, gear selection and operating
speed range for each gear ratio for combSet2 . . . . . . . . . . . . . 235
F.3 Gear shift map, engine reserve power, gear selection and operating
speed range for each gear ratio for combSet2 . . . . . . . . . . . . . 235
F.4 Gear shift map, engine reserve power, gear selection and operating
speed range for each gear ratio for combSet1 , combSet2 and combSet3 236
F.5 Gear shift map, engine reserve power, gear selection and operating
speed range for each gear ratio for combSet2 . . . . . . . . . . . . . 237
F.6 Gear shift map, engine reserve power, gear selection and operating
speed range for each gear ratio for combSet3 . . . . . . . . . . . . . 237
XIII
Summary
In the West, the current political roadmap aims to move to a low carbon economy
and, in particular, to reduce pollution associated with transportation systems.
This has resulted in increased pressure on manufacturers to reduce their vehicle
fleets harmful emissions. This work focuses on the use of advanced control and
optimisation to reduce vehicle emissions whilst taking into account the desire for
an enjoyable driving experience. The vehicle systems considered are the gear
ratio and the gear shift map.
A new efficient and effective problem formulation has been developed to opti-
mise gear ratio and gear selection, first independently and then in combination.
Traditional as well as two novel objectives have been developed to capture en-
gineering requirements such as reducing emission, maintaining or improving the
vehicle driveability, promoting the durability of transmission components whilst
simultaneously meeting problem specific constraints. The first novel objective
formulation rewards fuel efficient engine operating points and the second objec-
tive rewards the time spent in higher gears to reduce fuel consumption. A Pareto
based multi objective optimisation strategy has been adopted to identify the rel-
ative trade-off between the different objectives.
A new problem specific operator was designed, to reduce CO2 emissions by
shifting, towards the left side to promote rapid gear shifting.
Three nature inspired optimisation algorithms have been developed and crit-
ically evaluated against the Interior-Point Optimization (F mincon), and the
Multi-Objective Genetic Algorithm (MOGA) from the MATLAB toolbox. Multi-
Objective hybrid Cuckoo Search (MOCS) is used to optimise gear ratio. MOGA
combined with the new problem specific operator and constraint handling opti-
mised gear shift map. Finally MOGA was combined with MOCS operator for
gear shift map optimisation. Optimised gear shift maps were implemented on a
vehicle and tested on a rolling road, following an NEDC cycle. The benefit of
the optimisation procedure being developed was demonstrated and resulted in
reduction of CO2 emissions by 2.5%.
XIV
Acknowledgement
All the praises and gratefulness due to Allah Subhanahu wa Taala and peace be
upon the beloved final Prophet Muhammad Salallahu alaihi wa salam. I thank
Allah SWT for His guidance and mercy that are always given to me along my life.
Only because of His help and blessing that have given me strength and taught
me patience.
I would like to thank and express my gratitude to Prof. Keith J. Burnham,
Dr. Olivier Haas and Dr. Vincent Ersanilli for supervising me throughout this
enjoyable research project and the overall PhD candidature. Prof. Keith J.
Burnham introduced me to this project, while Dr. Olivier Haas provided me
with optimisation expertise, patient assistance and guidance, and Dr. Vincent
Ersanilli guided me with the automotive expertise.
Additionally thanks to my other supervisors Dr. Karen Vithanage for her
support and the competent discussions to guide my studies.
Further thanks are due to Mr. Chris Woolley, Mr. Joe Mahtani and Mr. Ian
Entwistle for their technical support, which added valuable information to my
project. Additionally, I want to thank Dr. Dina Shona Laila for her support
during my write up.
I am grateful for the financial support from the Knowledge Transfer Partner-
ships (KTP) funding through CTAC that has sponsored my research. I want to
also thank Prof. Elena Gaura for giving a year bursary for the completion of my
PhD.
I wish to thank my fellow students, colleagues and friends (Othman, Toheed,
Oluwaleke, Susan, Jenifer, Wallid, George, Stephane) at Coventry University and
Control Theory and Application Centre (CTAC) for the friendly and pleasant
working atmosphere.
I would like to thank Andreea whose patience enabled the completion of this
work.
Last but not least I would like to thank my family (Maman (Bintou), Papa
(Ibrahima), Tonton (Yacouba), Amadou, Mina, Fati, Anita, Ba-doussou) for their
continuous support throughout my journey of life that has made it possible for
me to reach this achievement.
XV
Nomenclature
Nomenclature
Abbreviations
XVI
Nomenclature
Variables
The following table gives an overview of the used variables:
XVII
Nomenclature
XVIII
Nomenclature
XIX
Chapter 1
Introduction
1.1 Introduction
This chapter introduces an overview of the research and defines its aim and
objectives. It then describes the research methodology, outlines the novelties
and finally, gives a high level description of the chapters and thesis organisation.
Government initiatives, legislation (Siskos et al. 2015) as well as the socio political
(Yao et al. 2015) prompted vehicles original equipment manufacturers (OEM) and
tier one suppliers to make significant investment in research and technology to
reduce vehicle emissions.
In the European market, vehicles are tested for fuel consumption and hence
emissions production according to the United Nations Economic Community for
Europe (UNECE) regulation 101 (Mahlia et al. 2012), (Bielaczyc et al. 2014).
Therefore, whilst the approach developed in this thesis is applicable to any drive
cycle, the implementation and practical verification employs only the New Euro-
pean Drive Cycle (NEDC) (Barlow et al. 2009a), (Tzirakis et al. 2006).
1
Introduction
The most significant CO2 emissions saving at the point of use arises from the
adoption of hybrid technologies (25%). However, significant savings can also be
gained from novel hardware design and software solutions applied to conventional
engine and transmission systems. Improved fuel economy can be achieved by
moving the engine towards its most efficient regions on the Brake Specific Fuel
consumption (BSFC) map in terms of both emission and performance through
optimised gear ratio and gear shift map (B. Mashadi 2012).
Applying a control strategy on gear shift map can achieve an improved per-
formance, especially an Automated Manual Transmission (AMT) (Lucente et al.
2007) and Dual Clutch Transmission (DCT) (Henrique et al. 2006) are both
semi-automatic transmissions, as they have the advantage of manual transmis-
sion, however their clutches and gearbox mechanism are electronically controlled.
Therefore, improving clutch and gear shift software control can results in better
fuel economy. André & Hugot (2003) has investigated the impact of gear shift
strategy on emissions test, which is mainly influenced by the driver, vehicle type
and driving conditions. Ivarsson et al. (2013) designed an optimal gear shift con-
trol to minimise fuel consumption for an AMT, Qin et al. (2004) has proposed a
gear shift indicator to improve fuel economy based on the environment and driver
intention. Fuel economy can be realised through software modification, especially
gear shifting strategies from 0.5% to 2% (Sovran & Blaser 2003).
This PhD research was initially started as part of a Knowledge Transfer Part-
nership (KTP1 ) project, a collaboration research between SAIC Motor UK Tech-
nical Centre (SMTC UK) at Longbridge, and Control Theory and Applications
Centre (CTAC), Coventry University. This research programme was targeted to
be implemented on their new vehicle programme ROEWE 950.
1
Knowledge Transfer Partnership (KTP) is European leading programme helping businesses
to improve their competitiveness and productivity through the use of knowledge, technology
and skills that reside within the UK knowledge base. Through KTP, academics can develop
business relevant teaching and research, apply knowledge and expertise to important business
challenges to identify and develop new research themes & student projects.
2
Introduction
The fundamental aim of this project was to reduce CO2 emissions while main-
taining a good dynamic response of the vehicle by optimising gear ratio and gear
shift map. To achieve this aim, the following objectives were addressed:
• Evaluate and analyse the results obtained using the validated simulation
model.
3
Introduction
Following an initial state of the art survey identifying potential means to save
CO2 it was decided to focus research on software means to achieve CO2 emissions
through gear shift map optimisation. Subsequently, means to achieve further CO2
savings through hardware optimisation in terms of gear ratio were investigated.
To determine optimal or near optimal solutions to a particular problem it is
necessary to be able to evaluate alternative solutions in a safe, cost effective and
controlled environment. This is usually realised using a computer simulation
of the actual system to be optimised. Therefore, a comprehensive proprietary
simulation model of a ROEWE 950 prototype vehicle, equipped with a Dual
Clutch transmission, was tuned and initially validated against rolling road data.
To carry out the optimisation, a number of criteria were developed. These
criteria included CO2 , as well as problem specific criteria to evaluate the quality
of the solutions. All criteria considered in the optimisation problem were nor-
malised to obtain values of the same magnitude for each criterion. The review
of normalisation techniques is given in Section 2.3.5, in Chapter 2. In this work,
the normalisation was carried out with respect to the criteria values obtained
from the manufacturer’s current gear shift map and gear ratios. The relative
importance of each criterion was then realised by associating a weighting to each
normalised criterion.
The model validation provided the necessary confidence upon which to evalu-
ate alternative gear shift mappings and gear ratios. The model was realised using
the commercial software environment MATLAB/Simulink, which is the tool of
choice in the automotive control sector, and provided a convenient platform to
evaluate the alternative optimisation strategies developed in this work. Model
4
Introduction
based optimisation minimises the requirement for vehicle testing and calibration.
However, to determine the viability and appropriateness of the solutions devel-
oped, the author liaised with SMTC China for the testing of various optimised
gear shift maps on a rolling road. The latter provided useful qualitative informa-
tion as to the nature of a good gear shift map and demonstrated the effectiveness
of the approach with quantifiable improvement compared to the standard gear
shift map used in the ROEWE 950.
The work carried out during this project has led to a number of contributions
and adaptations of existing ideas. These are ranked in terms of significance:
• Problem specific design variables formulations for both gear shift map and
gear ratio optimisation. These formulations are applicable to any optimi-
sation technique. The proposed variable formulation maps a set of inde-
pendent variables to a set of relative increments. Such formulation for the
gear shift map enforces the following engineering constraints: i) prevent
crossing between downshift and upshift, ii) maintain a minimum hystere-
sis between downshift and upshift to avoid frequent gear changes for small
velocity variations.
• Similarly the gear ratio problem specific design variables formulation are
5
Introduction
defined to facilitate the optimiser to find the optimal design variables from
which the gear ratio can be reconstructed and simulate their effect. Addi-
tionally, this formulation impose the following engineering constraints: i)
keep a minimum spacing between two adjacent gear ratios, ii) define a set
of gear ratio in descending order.
• A gear shift map problem specific local search strategy, named gear early
shifting (GES) operator, was developed and combined with the aforemen-
tioned hybrid GA. It reduces the CO2 emissions by promoting an earlier
upshift gear change.
• Developed a hybrid Cuckoo search algorithm which includes the Levy flight
operator as well operators used in the Bat, Firefly and Flower pollination
algorithms for gear ratio optimisation to improve the algorithm exploration
and convergence.
(ii) time spent within the most efficient zones within the engine fuel map
(Brake Specific Fuel Consumption (BSFC)).
6
Introduction
(iii) distance between a key reference point within the BSFC map and each
engine operating point.
• Developed a combined Gear ratio and gear shift map optimisation strategy
based on genetic algorithm and Cuckoo search.
• The work has led to one publication presented at the 14-th IFAC Sympo-
sium on Control in Transportation Systems (Fofana et al. 2016).
Fofana, A., Haas, O., Ersanilli, V., Burnham, K., Mahtani, J., Woolley, C.,
and Vithanage, K. (2016) Multi-Objective Genetic Algorithm for Automatic
Transmission Gear Shift Map Optimisation. 14-th IFAC Symposium on
Control in Transportation Systems, May 18-20, 2016, Istanbul Technical
University, Taksim, Istanbul, Turkey.
7
Introduction
Figure 1.1: Outline schematic flow of logical connection between different chapters
of this thesis
Chapter 4 presents the problem formulation, which is one of the major novelty
of the work. It converts the engineering requirements for the gear shift map and
the gear ratio design into a mathematical framework.
Chapter 5 describes Evolutionary Algorithm (EA) and swarm intelligence with
multiple objective functions combined with problem specific operator. Addition-
ally a repair mechanism application is proposed to correct an optimised gear
shift map with a given limited reserve power. Chapter 6 describes preliminary
simulation studies with a description of parameters selection, weighted coeffi-
8
Introduction
9
Chapter 2
2.1 Introduction
The main focus of the thesis is on automating the determination of the most
appropriate set of gear ratios and accompanying gear shift map. This chapter
presents a critical review of the state of the art in both the theoretical/technical
domain and the application domain. The first part focuses on the application
domain, namely the powertrain description with emphasis on the gear shift and
gear ratio design. The second part describes the key performance indicators or
objectives that have been used to evaluate the performance of alternative gear
shift and gear ratio designs. The third part reviews current approaches in numer-
ical optimization, with particular attention given to nature inspired constrained
multi-objective optimization algorithms and associated methods, to formulate the
engineering problem to be solved. The chapter concludes with a justification of
the methods and algorithms adopted in this work.
10
Background and literature research
This section describes powertrain system (see Figure 2.1). A powertrain system
is composed of an internal combustion engine and a transmission. A connector
is used to join the crankshaft of the engine to the input shaft. The connector
can either be a torque converter or a frictional clutch. A differential unit and an
ultimate drive gear is used to connect the output shaft of the transmission to the
wheels of the vehicle. A transmission has several speed, torque and gear ratios.
Speed ratio is being calculated following the similar way ratio of input speed
and transmission output speed are calculated, whereas torque ratio is measured
following the ratio of input torque and transmission output torque respectively.
11
Background and literature research
minimise f(x)
Subject to gj (x) ≥ 0, j = 1, ..., J (2.1)
hk (x) = 0, k = 1, ..., K
where,
f(x) (f1 (x), ..., fn (x)) f or i = 1, 2, ..., n.
gj (x) j-th inequality constraint evaluate at x.
hk (x) k-th equality constraint evaluate at x.
fi (x) i-th objective function evaluate at x.
x x1 , ..., xp is a vector of decision variables.
n number of objectives or criteria to be optimised.
p number of decision variables.
The constraints define the feasible region x, where any point x ∈ Rn presents
a feasible solution. The vector f(x) presents the values of objective functions to
be minimised or maximised. Note that in this thesis, only inequality constraints
are considered.
12
Background and literature research
The main goal of optimising is that there are some merit functions which need
to be minimised or maximised, and can be used as quantitative criteria to assess
the effectiveness of each design variable. One of the main driving objectives in
this thesis is CO2 emissions. However, minimising CO2 emissions can deteriorate
the car dynamics response. The main objective functions used for powertrain
optimisation can be classified as follows:
The primary objective function is the fuel consumption (Mammetti et al. 2013),
where emissions are a by-product of fuel. CO2 , carbon monoxide (CO), nitrogen
oxides (NOx ), hydrocarbons (HC) emissions Wallington et al. (2008) are mainly
13
Background and literature research
Driveability
Reducing fuel consumption or CO2 emissions can also results the car to poor
driving condition, therefore a line must be drawn to limit the reduction of fuel
or emissions. A poor driving condition can be defined as driveability, Ngo, Colin
Navarrete, Hofman, Steinbuch & Serrarens (2013) have defined the driveability by
the car responsiveness, operating smoothness. Therefore, it was expressed that
the driveability was the remainder acceleration capability after a certain gear
shift, which can be represented by the engine reserve power defined as follows:
where ωe denotes the engine speed, Te,max the maximum torque and Te the engine
torque defined as:
Rw
Te = Ftrac (2.3)
ig η
where ig and Rw are the gear ratio and wheel radius, respectively. The engine
speed ωe , is defined as follows (Liu et al. 2009):
ig 60
ωe = v (2.4)
2π3.6Rw
14
Background and literature research
2.3.3 Constraint
2.3.4 Bounds
In practice, the design variables are in most cases constrained by the physical
limit of the system, therefore considering continuous variables, a minimum and
maximum bound must be set on each design variables. In this condition the
research space is restricted.
Most engineering designs have either one single objective function or more than
one objective functions. In general, a single objective function, is often repre-
sented by a single scalar, where the optimiser will have only one objective to
focus on. However for multi-objective functions, the task becomes more complex,
15
Background and literature research
In this method, multiple objective functions are converted into one single objec-
tive using a weighted sum:
n
X
F (x) = wk fk (x) (2.6)
k=1
where x, F(x), k denotes a set of design variables, single objective function after
conversion and number of objective functions, respectively. wk , k = 1...n, are the
fractional weighting coefficients where the solutions selected will forcibly depend
on them. The weighting coefficients must be positive and satisfy:
k
X
wi = 1, wi ∈ (0, 1). (2.7)
i=1
The method has the advantage of combining all objective functions, i.e Yin
et al. (2013) used a weighted sum to combine power, fuel and emissions in one ob-
jective function in order to optimise a gear shift using GA. However the drawback
is that weighting for each objective function must be known in advance (Vachhani
et al. 2015). Also in order to obtain several solutions, the algorithm must be run
multiple times, which is time consuming (Odu & Charles-Owaba 2013).
e-Constraint
In this method, only one objective function is considered, presumably the most
significant on and considering the other objective functions as constraints bounds.
It is very easy to implement, however it might require high computational cost
16
Background and literature research
(Mavrotas 2009).
Pareto Optimality
1
ζk = (2.8)
fk (x0 )
17
Background and literature research
One of the best method is to normalise the objectives based on Nadir and
Utopia points. The Nadir and Utopia represent the length of the intervals
where the optimal objective functions vary within the Pareto optimal set.
Nadir and Utopia points are described as follows: The Utopia point f U topia
is defined as,
h i
fiU topia (1) (2) (k)
= f1 (x ) f2 (x ) ... fk (x ) (2.10)
where x(k∗) is the optimal point solution vector for the single objective
function of the k−th objective function fk , and the Nadir point fiN adir for
each component is defined as,
h i
fiN adir = max f1 (x(1∗) ) f2 (x(2∗) ) ... fk (x(k∗) ) (2.11)
h i
fi∗ = fi (x(1∗) ) fi (x(2∗) ) ... fi (x(k∗) ) (2.12)
fi − fiU topia
fi = N adir . (2.13)
fi − fiU topia
The first method was selected for convenience as it is easier to compare so-
lutions with the baseline. Investigation of alternative normalisation methods is
18
Background and literature research
The Most important inventions were designed by observing the nature, i.e, sub-
marines have been designed by observing fish, or radar system by adopting bats
behaviour. These algorithms can be classified by Evolutionary Algorithm and
Swarm intelligence (Binitha & Sathya 2012). This section describes and gives an
insight definition on them, also a more detail survey can be find in (Nanda &
Panda 2014).
Definition
Selection method
Selection method mimics the natural selection of species, where the fittest indi-
viduals have a better chance to be selected as parents for reproduction. The most
known are the Roulette Wheel Selection and Stochastic Universal Selection (SUS)
19
Background and literature research
(Baker 1987), meaning that the fittest individuals are not necessarily guaranteed
to be selected, however they have a better probability to be selected (Reeves
2010). Various selection methods can be found in (Goldberg & Deb 1991). Jebari
& Madiafi (2013) have implemented and compared the performance of different
selection methods and concluded that RWS and SUS maintain a good diversity
among the population and prevent a premature convergence to the local optimal.
A solution can be represented ether by binary code or real code. GA was originally
created using binary code as it was to represent the biological gene. The binary
can be used to represent small real values and cannot represent exact real number,
which means a scaling factor must be considered (Wright 1990). A lot of research
has proven that the application of real code outperforms the binary code (Tsutsui
& Ghosh 1998), (Raghuwanshi & Kakde 2007). In this thesis real code has been
preferred to binary because of the difficulty and exact conversion from binary to
real numbers.
Crossover
After selecting individuals to produce new solutions, there are various opera-
tors which can be used to modify information of each individual. These process
is called crossover or recombination, which is a genetic operator that combines
two individuals (parents) to form a new individual (Chromosome). This are
many variants of crossover which can be found Kaya et al. (2011). For real code
crossover implementation, Single-point, Multipoint-point and Uniform crossover
are more suitable due to difficulty of binary conversion to real code, also in Kita
et al. (1999) the rational reason of using real code for crossover has been debated
.
20
Background and literature research
Mutation
Mutation can be considered as a random process, which aims to find new solutions
in search space by randomly modifying one or several values of the chromosome.
One of the interesting properties of mutation is that preserves diversity among
the population by searching in an unknown search space. A usual method is
the uniform mutation, where a random number is added to each value of the
individual. Deb & Goyal (1996) have discussed the binary coded over real coded
mutation, and has summarised that real coded in mutation was more efficient.
Elitist
Local search
In GA, the use of local search is often applied to solutions in order to improve
them. However, the use of local search must be taken carefully as it can lead to
a premature convergence (Pandey et al. 2014).
Constraints handling
The most simple way to handle constraint is to penalise invalid solutions by im-
posing a penalty function, where this penalty might prevent the faulty individual
contribution to the next generation. Coello Coello (1999) proposed a survey on
different constraints handling, where one of the most extreme is the death penalty,
which removes invalid solutions from the population. However, the major prob-
lem of this method, is assuming that at least one solution is valid among the
21
Background and literature research
population. The next penalty is static penalty, which gradually reduces the fit-
ness values of solutions disobeying the constraints, meaning the penalty is severed
for many constraints violations, and soft for low constraints violations. Another
penalty, increase the penalty function over generation referred to as dynamic
penalty, meaning that the constraints in earlier generations are less penalised
than the constraints in later generation, therefore, the drawback is, if the penalty
factor is badly selected, the solutions could converge to a non-optimal solution
space.
Furthermore, in the same survey, the author has also proposed a method
called Co-evolutionary penalty, where two penalty values are used to distinguish
the number of violated constraints, and the corresponding amount of violations.
They are also two sub-populations, where the first contains the individuals, while
the second implements the set of weighted combinations used to calculate the
fitness function and also contains the penalty factors. The drawback of this
method is the addition of extra-parameters, also the requirement of initialising
them. Additionally, Mani & Patvardhan (2009) has proposed an improved version
of Co-evolutionary, by using a self determining and regulating penalty factor,
however this method still required two sub-populations and it is expensive in
computational time.
Another interesting constraint handling, unlike the previous constraint, is
based on penalty functions. This method proposes to repair invalid solutions
in order to turn them into valid solutions. It means that the encoding design
variable is then modified to suit the constraint. Salcedo-Sanz (2009) proposed
a review of the main repair mechanism used to handle constraint, where he de-
scribed the procedure and applications of different repair mechanisms. One of
the interesting repair mechanisms mentioned, was applied to gene permutation
due to cross-over procedure, where additional operators were used to modify the
crossover operator referred to as partially mapped operator (Goldberg & Lingle
22
Background and literature research
1985) and tie breaking crossover. Mitchell et al. (2003) proposed a combined re-
pair operator named GeneRepair with crossover and mutation operators, where
it is based on two tasks, one is for fault detection and the second is for cor-
rection. Repair operators increase the valid search space, however they do not
necessarily improve the performance of the algorithm. In this thesis, a problem
specific repair mechanism was proposed to modify solutions obtained by the op-
timisation algorithm to produce new solutions with specific minimum value for
the reserve power (see Section 5.2.2, in Chapter 5). This repair mechanism can
also be applied during the design process to quickly obtain solutions that meet
specific performance based on a set of existing solutions.
The literature on EA is expanding (Zhou et al. 2011), (Khajehzadeh et al.
2011). In Vachhani et al. (2015) an excellent survey on EA handling multi-
objective problem is provided, where various optimisation algorithms are com-
pared more particularly in terms of diversity and convergence. The following
section will describe Swarm Intelligence algorithm.
23
Background and literature research
performance, and it only requires two parents (Pavlyukevich 2007), however its
disadvantage is to select the value of its step (Gopal Dhal et al. 2015). CS is
population based, where CS lays eggs in communal nests of other birds, with the
eggs considered to be the solution of the objective function.
There are hybrid CS, which combine different algorithms, i.e Kanagaraj et al.
(2013) have proposed CS integrated with genetic operator to solve the reliabil-
ity and redundancy allocation problem. It was confirmed based on experimental
tests, that the proposed method was efficient in terms of balance between explo-
ration and exploitation (Kanagaraj et al. 2013), as crossover maintain the parent
cuckoo birds identity and at the same time creating diversity in the search space,
while the mutation of CS is considered as local search by making small changes in
the design variables. Rani et al. (2012) proposed hybrid of modified CS by inte-
grating it with two evolutionary algorithms, Particle Swarm Optimisation (PSO)
and GA, where it was applied as multi-objective optimisation to the location of
amplitude and phase of symmetric linear array element. The results achieved
good improvement in comparison to original CS.
Bat algorithm (BA) was created by Yang (2011a). The main idea of the al-
gorithm is based on the echolocation of micro-bats, where micro-bat used echolo-
cation to detect their prey, and avoid obstacles. Additionally BA can be more
effective as it uses frequency tuning and parameters control to influence explo-
ration and exploitation (Yang 2010). Different variants of the algorithm have
been published. Algorithm proposed by Yang (2011b) was applied to solve Multi-
objective functions in design of structure, and results shows that BA is an efficient
optimiser. Fister et al. (2013) have proposed an hybrid BA integrated with Dif-
ferential Evolution (DE) algorithm, where different experiments were realised on
test functions. It was shown that hybrid BA outstandingly improved the results.
Firefly algorithm is based on the flashing patterns and behaviour of tropical
fireflies (Goel & Panchal 2014). This algorithm is mainly a mutation based. Two
24
Background and literature research
iterative loops are used to compare each firefly brightness (objective functions),
and the firefly with the strongest brightness is attracted by the firefly with weakest
brightness. Consequently, the fireflies can be subdivided into various groups, and
each group can swarm around a local mode dominated by the firefly with the
strongest brightness. The drawback of firefly algorithm, in the case of multi-
objective problem is that weighted sum method must be applied to reduce them
into one single scalar, as the only concern is the attractiveness of fireflies. Liu
et al. (2012) have demonstrated the effectiveness of Firefly algorithm applied to
path planning problems. Arora & Singh (2013) have proposed a conceptual study,
by comparing Firefly algorithm, Bat algorithm and Cuckoo Search, where it was
concluded that FA was better in terms of finding optimum solution, as well as
performing local search.
Flower Pollination algorithm is a new type of optimisation developed by Yang
(2012), where the main concept of the algorithm is based on the flower pollination
process of flowering plants. In this algorithm, there are no explicit crossovers.
Additionally, it uses the current best solution among the population to make the
next move. Consequently, the algorithm can possibly be trapped in a local mode.
Flower Pollination algorithm has been extended to multi-objective function in
Yang et al. (2013), where it is applied to solve a disc brake design problem.
The main concept of the algorithm evoked in this thesis is stochastic-based. There
are also deterministic-based algorithms, where they are designed to search for
global best solution (Arora 2011). The basic deterministic algorithms are Steep-
est Descent method, the quasi Newton method, the Newton-Raphson method or
the Levenberg-Marquandt method (Colaço & Dulikravich 2011). These type of
algorithms are mainly used for non-linear minimisation problems. Their func-
tion is based on an iterative process, where after a certain number of iterations,
25
Background and literature research
the objective function converges to its minimum value. A more general form
describing the iterative process can be defined as follows:
Y k+1 = Y k + ξ k E k (2.14)
where Y , ξ, E and k, denote for the vector of design variables, the search step
size, the direction of descent and the number of iterations, respectively. The
following statement summarises some of the deterministic algorithms:
• Steepest Descent (Fletcher & Powell 1963): this method is based on basic
gradient method, where the principle is to focus the search on the opposite
direction of the locally highest variation of the objective function, such that
to locate its minimum value .
• Quasi Newton (Shanno 1970): This is another Newton based, except there
is no need of second derivative. However, it utilises the Hessian based on
the first derivative. This method is computationally faster, but it had a
slower convergence.
26
Background and literature research
27
Background and literature research
vehicle speed characteristic curve, tractive effort and maximum vehicle speed of
each gear ratio (Kasseris & Heywood 2007). In design criteria for gear ratio, the
weight is considered to be one of the most important aspect in terms of cost and
fuel consumption, as well as the bending and wear strength of the gear tooth
(Chen & Usman 2001).
Shamekhi et al. (2014) presented a gear ratio optimisation, where a neural
network was used to obtain a model of the transmission system which was fast to
execute. It was observed that such an approach required a massive set of training
data to obtain an accurate model. The model was subsequently exploited by a ge-
netic algorithm (GA) to optimise the gear ratio parameters. Shariatpanahi et al.
(2004) have combined GA with neural network to optimised gear ratio. Two mod-
els were defined using neural network and used in parallel during optimisation.
The first model was used to calculate the acceleration, maximum vehicle speed,
and gradability, while the second model was applied to calculate fuel consump-
tion and emissions. The reason of using two models based on neural network was
to obtain two simplified accurate models. However the computational time can
be costly by running the two models during optimisation. Casavola et al. (2010)
have proposed gear ratios optimisation, where a fuzzy logic was used to defined a
gear shifting strategy based on engine speed, engine torque, vehicle speed, brake
pedal and throttle position. This simplified model was used to minimise fuel
consumption by optimising gear ratios. Therefore, this model and optimal gear
ratios are only used for benchmark, however they cannot be used for commercial
vehicles as their driveability is poor.
Yokota et al. (1998) proposed a GA optimisation and a nonlinear integer
programming (NIP) problem formulation for gears weigh reduction based on gear
dimension, bending and torsional strength. 44.8% reduction of weigh and 18.5%
reduction in mean radius were achieved. Whilst such results were promising, the
authors required to carry out additional simulation to ensure that the solutions
28
Background and literature research
were realistic. Golabi et al. (2014) have proposed a gear ratio optimisation for
volume and weight reduction, where the optimisation algorithm was implemented
within the MATLAB function Fmincon. The optimisation methods described are
used to improve the existing results or to meet the design requirements. There
is a lack of benchmark comparison in terms of algorithm. In addition, the gear
ratio bandwidth is missing in most cases. The latter is however a requirement
according to the technical standard of SAIC Motor and BorgWarner Design for
Dual Clutch Transmission (see Section 3.4.5, in Chapter 3).
In literature, several approaches have been applied to optimise vehicle power-
train system. Hu et al. (2010) have applied GA to optimise the shift quality of a
DCT, where a trade-off was made between jerk and friction as objective function.
Ye et al. (2004) have applied GA to minimise fuel consumption and emissions
of a four cylinder engine. The main objective was to find the speed-load point
with the minimum Brake Specific Fuel Consumption (BSFC) when using Vari-
able Valve Timing (VVT) and Variable Compression Ratio (VCR) engine. It was
noticed that the improvement of fuel economy and emissions was occurred at low
speed and mid load region.
Having reviewed the gear ratio design and optimisation methods, it can be ob-
served that the most important design criteria are fuel consumption and emissions
as well as gears weight, with driveability mentioned in only a few publications.
In addition, the optimisation algorithms are often used without modifications
and the problem formulation consider the variables to optimise directly. There
is therefore an opportunity to develop more efficient problem formulation as well
as introduce problem specific features in the optimisation algorithms to improve
the coverage and convergence of the algorithms.
29
Background and literature research
Gear shift map design methodologies are complex. The first shift maps are de-
termined using design based-software. Initially the shift points are obtained from
vehicle dynamic characteristic by determining the intersection between two accel-
eration curves of two adjacent gears (Xi et al. 2009), (Liu et al. 2009) and (Kirtane
n.d.). Most of the methods developed have not focused on ensuring that gear shift
map results in good vehicle’s driveability. Such task is traditionally reserved to
calibration engineers. Ngo, Hofman, Steinbuch & Serrarens (2013) have proposed
alternative gear shift map design, which alleviates the issue of driveability, where
the shift points are obtained from statistical data of various driving cycles, also
showing an acceptable driveability by comparing different acceleration profiles.
Gear shifting strategy is a very important aspect of reducing fuel consumption,
as it can be adapted to driver behaviour as well as the vehicle’s environment.
Santiciolli et al. (2015) described a gear shift strategy optimisation over the drive
cycle FTP-72 (Barlow et al. 2009b), using a multi-objective genetic algorithm
(MOGA), where the trade-off was to maximise performance and fuel economy.
The optimisation results demonstrated a good trade-off between performance and
fuel economy using a gear shift strategy. This method is only based on a known
driving cycle, and use the current driving cycle shift position to shift earlier.
Ha & Jeon (2013) described an adaptive gear shifting strategy based on torque
and traffic condition, where the constraints are limited to the maximum and
minimum engine rotation and engine torque. Dovgan et al. (2012) used throttle,
brake and gear management as decision variables to improve performance and
fuel consumption over a known drive cycle, however the optimum solutions were
exhibiting an uncomfortable jerk, which can compromise the vehicle driveability.
Kim et al. (2007a) proposed an integrated transmission control algorithm to assist
the driver power demand via throttle pedal acceleration. This method was based
30
Background and literature research
31
Background and literature research
zones for each gear ratio. Then, Cruise GSP Optimization was used to adjust
the upshift and downshift points around the optimum zones, which in fact reflects
the optimum gear shifting point. This method gave good optimised gear shift
map. However the problem formulation was not detailed.
Various techniques have been applied to optimise gear shift map. Li & Hu
(2010) have proposed a fuzzy neural network to define an optimum gear shifting
decision maker. Kim et al. (2007b) have proposed an optimum decision maker
for gear shift and throttle position using dynamic programming (DP), where fuel
consumption is considered as a cost function to be minimised. Fu & Bortolin
(2012) have proposed a model predictive controller combined with DP to find
an optimum gear shift sequence, considering fuel consumption as cost function.
These method are mainly designed for real time driving, where an indicator can
advise the driver, when to change gear.
Having reviewed the algorithms used to solve the problems considered in this
work, has highlighted research opportunities in applying other evolutionary algo-
rithms such as cuckoo search and variants of multi objective genetic algorithms.
Most optimisation strategies presented do not exploit problem specific features
nor design variable formulation to improve both solution coverage and speed of
convergence. Finally, whilst normalisation procedures are used, there is little
mention of the details of these normalisation procedures. The main objectives
considered or gear shift map optimisation relate to fuel/CO2 and driveability/re-
serve power. There is therefore an opportunity to investigate alternative objective
formulation to help quantify the differences between alternative solutions with
similar objective values for fuel/CO2 and driveability/reserve power. The next
section reviews the software tools used by industry and academia to support the
design, optimisation and calibration of gear shift map and gear ratio.
32
Background and literature research
2.4.6 Software tools used for gear ratio and gear shift map
Simulation tools are extensively used in the automotive industry to model, reduce
the cost and speed up the vehicle systems development. Simulation tools are
used as part of the design process, to predict and understand specific vehicle
systems behaviour and performance, as well as for testing, systems improvement
(or optimisation), validation and calibration before the vehicle can be put into
production.
They are many simulations tools in the automotive area. This review focuses
on software tools associated with gear ratio and gear shift map design and opti-
misation. These includes: AVL Cruise, MATLAB/Simulink, dSPACE, CATIA,
AMESim, GT-Suite, ADAMS, ADVISOR, veDYNA, Modellica, ROMAX.
MATLAB/Simulink is the main environment used in the automotive indus-
try from an electrical and software perspective for vehicle simulation, algorithm
development, optimisations and research (Xi et al. 2009).
For example MATLAB/Simulink was used in General Motor to analyse and
determine transmission gear content required to minimise fuel consumption for
various powertrain system developed (Robinette 2014). These tools can be ap-
plied to a variety of vehicle simulation applications. More specifically, AVL has
developed a simulation package (Le Guen et al. 2011) to optimise gear shift pat-
tern.
dSPACE (Lucente et al. 2007) is mostly used for rapid prototyping, designing
and testing of mechatronic systems. AMESim is used to simulate vehicle dynam-
ics for design and optimisation before integration. More specifically it has been
used in Song et al. (2014) to analyse the effect of gear shift characteristics when
developing an electric oil pump for automatic transmission and in Xiang et al.
(2013) to study the shifting schedule for a speed electric vehicle.
CATIA is well known for optimising and analysing gears. Rajan & Usmansha
33
Background and literature research
(2014) have optimised an automotive transmission gear box for weight reduction
and improve fuel consumption. GT-Suite (Ortiz-Soto et al. 2012) was used to
model combustion and heat transfer of an engine, in order to optimised fuel
consumption over various drive cycles. Lin et al. (2009) have used ROMAX
package to assess gearbox optimisation based on design specification.
Simulations packages have played an integral part in the development of tools
for fuel economy, Argonne National Laboratory had developed various simulation
to study fuel efficiency for electric drive vehicle technologies (Moawad & Rousseau
2014), and gear reduction study to improve energy management strategy (Kim
et al. 2012).
Ford Motor has described a simulation study to design a gear shift map for a
Dual Clutch Transmission (Liu et al. 2009), where Downshift and Upshift point
are initially defined using a vehicle dynamic model. FEV has developed a tool
referred to as ShiftAnalyzer (Kirschstein et al. 2009), for online calibration of
vehicle powertrain. It has been used to optimise and calibrate the shift quality
of the BMW mini Cooper automatic gearbox.
The software tools have a massive advantages, as they can be used to reduce
vehicle development cost, reduce weight, and optimise software applications, re-
sulting in reduced fuel consumption. However software usage had some limitation
as it cannot give accurate and uniform views onto systems based on theories (Broy
2006).
There are various industry standards used for the design and optimisation of
gearbox such as, AGMA standards for gears as well as the ISO6336. ANSI, JIS
and DIN standards are also used for gear teeth.
34
Background and literature research
Intelligence
Considering the analysis above, the use of EA and swarm intelligence is quite
simple to justify the motivation of their applications. Firstly, avoiding the for-
mulation into a specific mathematical framework, as it can be costly and time
consuming and some of the constraints do not need to be reformulated in a pre-
defined mathematical structure. Secondly, reducing CO2 emissions, is the main
objective in this thesis. However, diminishing CO2 emissions can conduct to de-
terioration of different aspects of vehicle good response. Very often, driveability
is the measure delimiting how far the optimisation can produce solutions to re-
duce CO2 emissions. Additionally, different objective functions are considered to
precisely measure qualitatively an obtained gear shift map or gear ratio from the
optimiser.
In order to achieve these goals, the use of EA and swarm intelligence appears
to be a promising alternative to traditional approaches. These algorithms can
consider any objective functions and constraints, regardless of their mathematical
framework.
EA and swarm intelligence have been used in many applications, with multi-
objective functions. GA has been applied across many fields, as per number of
publications (Khajehzadeh et al. 2011). Shariatpanahi et al. (2004) have applied
a GA with neural network to optimise gear ratio, Yin et al. (2013) have used
aggregation weighted sum to combine several objective functions in attempt to
optimise gear shift map. Swarm intelligence have gained enormous attentions,
due to their successful applications, CS has been applied to welded beam design
and disc brake design (Yang & Deb 2013).
In particular, all these methods do not require any gradient evaluation, which
means that the reformulation of objective functions is not necessary any more.
35
Background and literature research
EA and swarm intelligence have also got a drawback. They cannot guarantee the
actual optimality of the solution for a given problem. Several arguments can be
raised to tackle this issue:
• An algorithm having the ability to find a good gear shift map or gear ratio,
by satisfying all constraints, and considerably better than the initial one, is
a massive gain for the industry and environment.
• EA and swarm intelligence can permit to solve the initial problem, by ob-
taining satisfactory results within a reasonable time.
This chapter has given an overview of traditional analytical gear shift map and
gear ratio design. Amongst these, the methods adopted by SAIC motors will be
used in Chapter 3 to create the initial solutions that will be further refined by
the optimisation approaches described in Chapters 4 and 5.
Both deterministic and stochastic methods are applicable to this constrained
multi-objective optimisation problem. This work will develop methods based on
the most appropriate stochastic algorithms, namely genetic algorithms and cukoo
search, as well as deterministic algorithm, namely the interior-point algorithm.
Common to all the optimisation approaches considered is the need to evaluate
the appropriateness of candidate solutions. The validated model required to
evaluatealternative gear shift map and gear ratio is described in chapter 3.
The criteria traditionally used to differentiate between alternative solutions
have been reviewed. The adopted criteria as well as the additional criteria de-
veloped to help quantify the benefits of alternative solutions are formulated in
Chapter 4. These criteria have been combined using both weighted sum and
Pareto approaches, see Chapter 4.
36
Chapter 3
Powertrain modelling
3.1 Introduction
This chapter describes the models and algorithms exploited in subsequent chapter
to optimise gear ratio as well as gear shift map. The chapter starts with a de-
scription of the ROEWE 950 characteristics and the validation of its proprietary
transmission model. It is followed by a technical description of the traditional
methods adopted to design gear shift map and gear ratio based on technical
standard of SAIC Motor and BorgWarner Design for Dual Clutch Transmission.
These methods have been included to provide an understanding of the process
which is to be complemented by the optimisation strategies developed in Chap-
ters 4 and 5.
In order to study and validate the optimisation algorithms in this thesis, a DCT
model has been made available by SAIC Transmission team. This section presents
the model description of ROEWE 950, a conventional style saloon car, equipped
with 2 litre turbocharged 4 cylinder engine featuring variable camshaft timing
37
Powertrain modelling
(VCT) and a start-stop system. The engine is mated to a 6 speed Dual Clutch
Transmission (DCT 360 variant). Within the Transmission Control Unit (TCU)
resides the so-called shift map. This mapping is designed with respect to the
engines operating range so that the transmission is in the right gear at the right
time e.g. when the driver requires maximal torque, depending on the prevailing
conditions, typically the transmission will select a lower gear moving the engine
further into maximum torque producing range (high RPM combined with a wide
open throttle).
The DCT 360 essentially comprises two transmission units that are indepen-
dent of each other. Each transmission unit is constructed in the same way as
a manual gearbox. Allocated to each transmission unit is a multi-plate clutch.
Both multi-plate clutches are of the wet type (operating in oil) and hydraulically
actuated. They are regulated, opened and closed by the mechatronics system
using hydraulic oil depending on the gear to be selected.
1st, 3rd, 5th and reverse gear are selected via multi-plate clutch 1. 2nd,
4th and 6th gears are selected via multi-plate clutch 2. One transmission unit
is always in gear and the other transmission unit has the next gear selected in
preparation but with the clutch still in the open position. Every gear is allocated
a conventional manual gearbox synchronisation and selector element (Kulkarni
et al. 2007). The following section describes the vehicle model including the DCT
and engine model.
The vehicle model with DCT was implemented using the commercial software
MATLAB/Simulink, which is the standard package of modelling uses in industry.
The vehicle model is divided in four subsystems (see Figure 3.1) which are:
38
Powertrain modelling
duce CO2 emissions by using the well-known NEDC drive cycle, approved
among and homologue by the United Nations Economic Community for
Europe (ECE) regulation 101 (United Nations 2006). The vehicle model
is designed to consider as input a drive cycle. By default, the NEDC in
implemented, however the model also include different drive cycle such as
WLTP, ARTEMIS Urban, JAPANESE 10-15 Drive cycle. A more detail of
these drive cycle can be found in Barlow et al. (2009a).
• Engine model: knowing the driver demand (torque and throttle position)
and engine speed, the engine fuel map BSFC is then utilised to estimate
fuel consumption, where the CO2 emission can be derived. Noticed that,
this task is crucial in this thesis, as the optimisation is mainly aimed to
reduce CO2 emissions.
39
Powertrain modelling
Having described the vehicle model in Simulink, a verification of the model must
be realised in order to validate and accepted it as an abstract representation of the
real system. This is achieved by comparing vehicle model against approved data
made available from rolling road data. Measured data were obtained from rolling
road over the NEDC, then converted into excel spreadsheet, which made easier to
extract the data using MATLAB/Simulink environment for comparison studies.
The original data runs over a 1250s period with sample interval at 1ms and
10ms. For practical use, the simulation and experimental data were re-sampled
at the same rate at 10 ms. Figures 3.2 and 3.2 represent a comparison between
simulation and experimental results on rolling road, where both considered initial
shift map (see Appendix B) and the NEDC.
The vehicle speeds are almost comparatively similar, except the driver data
included a maximum error of magnitude ± 2 km/h for set of data. Therefore, both
engine speed show a good compromise in both dynamic and steady state, with
some minor error difference. Regarding the gear shift map, a minor difference can
be noticed on the first set of data at higher speed, which could be due to driver
throttle position and vehicle speed variation.
40
Powertrain modelling
50
0
0 200 400 600 800 1000
Engine Speed [RPM]
1500
1000
6
Theor Gear Sel
5 Exp Gear Sel
4
3
2
1
0 200 400 600 800 1000
Time [Sec]
Figure 3.2: First set of validation data representing comparison between simula-
tion and experimental results from rolling road over the NEDC. Theor V Speed,
Exp V Speed, Theor Eng Speed, Exp Eng Speed, Theor Gear Sel and Exp Gear
Sel denote for vehicle speed, engine speed and gear selection used in simulation
and rolling road, respectively.
Vehicle Speed [km/h]
Theor V Speed
100
Exp V Speed
50
0
0 200 400 600 800 1000
Engine Speed [RPM]
1500
1000
6
5
4 Theor Gear Sel
3 Exp Gear Sel
2
1
0 200 400 600 800 1000
Time [Sec]
Figure 3.3: Second set of validation data representing comparison between simu-
lation and experimental results from rolling road over the NEDC. Theor V Speed,
Exp V Speed, Theor Eng Speed, Exp Eng Speed, Theor Gear Sel and Exp Gear
Sel denote for vehicle speed, engine speed and gear selection used in simulation
and rolling road, respectively.
41
Powertrain modelling
This subsection describes fuel consumption (Mashadi & Crolla 2012), and CO2
emissions (US EPA 1999) calculation use in this thesis.
• Fuel consumption
where Fuel(g/s) denotes the fuel mass. Pe is the engine power expressed by
the product of engine speed and torque that are also used to define the
BSFC from the engine map.
The total fuel map of an entire cycle is defined by summing the individual
fuel masses:
N
X
F uel(g) = F uel(g/s) (i). (3.2)
i=1
where N denotes the number of sampled data over the driving cycles con-
sidered.
F uel(g)
F C(l/100km) = × 100. (3.3)
dcycle × ρf uel
where dcycle and ρf uel denotes for a given driving cycle distance, and the den-
sity of the fuel respectively. The factor 100 is used as the fuel consumption
is defined by l/100km.
• CO2 emissions
42
Powertrain modelling
CO2(g/mol)
CO2 emissions = F uel(g) Cc,c(%) × . (3.4)
C(grams)
where Cc,c(%) , CO2(g/mol) and C(grams) denote for percentage of carbon con-
tent from fuel, molecular weight of CO2 and molecular weight of carbon
respectively.
The longitudinal vehicle dynamic equation defining the equilibrium relation, be-
tween drive forces and resistance forces is applied to determine the vehicle ac-
43
Powertrain modelling
celeration, speed, traction forces, rolling resistance force, and aerodynamic force.
The characteristics of engine torque map alongside with fuel consumption map
(see Figures 3.4 and 3.5) are required to design the shift schedule. The engine
torque map was modelled using the so-called Magic Torque formulae (Mashadi &
Crolla 2012), where the fuel map was created using MATLAB handle function.
350
300
250
Torque [Nm]
200
150
100
50
−50
100
90
80 6000
70 5500
60 5000
50 4500
4000
40 3500
30 3000
2500
20 2000
10 1500
1000
Throttle position [%] Speed [RPM]
44
Powertrain modelling
900
700
600
500
400
300
200
100
90
80 6000
70 5500
60 5000
50 4500
4000
40 3500
30 3000
2500
20 2000
10 1500
1000
Load [%] Speed [RPM]
The driving forces developed by engine must overcome the rolling resistance
(Froll ), aerodynamic drag (FaeroDy ), climbing resistance (FClimb ) and acceleration
resistance (Fac ) as describes in Xi et al. (2009) and Kirtane (n.d.).
where Acc is the acceleration of the vehicle, Mv is the vehicle mass, δn is the
equivalent mass of rotary mass of vehicle, influenced by the inertia of the engine
flywheel (Je ), vehicle wheels (Ji ) and transmission ratio (Ji ), which is expressed
as follows:
JW Ji i2g i2F Je
δn = 1 + 2
. (3.7)
RW Mv
45
Powertrain modelling
Te ig iF ηT
FT F = . (3.8)
RW
where Te is the engine output torque, ig is the gear ratio of transmission, iF is the
final gear ratio, ηT is the transmission efficiency, and RW is the radius of wheel.
The air resistance force is expressed as:
CD AV 2
FaeroDy = . (3.9)
21.15
where Vi and ωe is the vehicle speed at given gear position and engine speed
respectively.
46
Powertrain modelling
hicle acceleration and speed (see Equation (3.11)) can be calculated at different
gear position with different throttle pedal angle. The shift schedule design for
dynamic performance (see Figure 3.6) is to drive at the maximum vehicle accel-
eration. However, the shifting point is selected at the intersection point of two
acceleration curves adjacent gear at the same throttle angle.
−1
0 50 100 150 200 250 300 350
Speed [km/h]
Figure 3.6: Vehicle dynamic performance curves with different gear position and
throttle pedal angle
where g(i)→(i+1) is the gear shift at a same throttle position from g(i) to gear
g(i+1) at the intersection of acceleration curves Acc(i) and Acc(i+1) . If there is no
intersection between acceleration curves Acc(i) and Acc(i+1) , then the maximum
speed of the gear g(i) is considered. The Upshift schedule is obtained by connect-
ing together shift point of the same gear between different throttle positions (see
Figure 3.7).
47
Powertrain modelling
Upshift map
100
90
Upshift 1 to 2
80 Upshift 2 to 3
Upshift 3 to 4
70 Upshift 4 to 5
Throttle position [%]
Upshift 5 to 6
60
50
40
30
20
10
0
0 50 100 150 200 250
Vehicle speed [Km/h]
Shift schedule based traction force is similar to the acceleration curves, how-
ever the Equation (3.8) which is derived from the vehicle longitudinal is used to
establish the shift point (see Figure 3.8).
Upshift map
100
90
Upshift 1 to 2
80 Upshift 2 to 3
Upshift 3 to 4
70 Upshift 4 to 5
Throttle position [%]
Upshift 5 to 6
60
50
40
30
20
10
0
0 50 100 150 200 250
Vehicle speed [Km/h]
Figure 3.8: Upshift gear shift map based on traction force curves
48
Powertrain modelling
Shift schedule based brake specific fuel consumption (BSFC) map is based on the
minimum fuel consumption (see Figure 3.9), where the shift point is following the
same as the acceleration shift point.
Upshift map
100
Upshift 1 to 2
90 Upshift 2 to 3
Upshift 3 to 4
80 Upshift 4 to 5
Upshift 5 to 6
70
Throttle position [%]
60
50
40
30
20
10
0
0 50 100 150
Vehicle speed [Km/h]
Figure 3.11 is to compare different methods to define an initial gear shift map.
It is noticed that the ideal fuel consumption for a gear shift map is converging
toward the left side of the map which corresponds to the most efficient area of the
BSFC map. It might be ideal for fuel consumption to consider a shift map design
based on BSFC map, however the driveability might be worst. The gear shift map
should be design based on a compromise of fuel consumption and performance.
The next of this study will be to design an algorithm capable to select a shift
map in order to satisfy the vehicle driveability as well as fuel consumption and
vehicle performance.
49
Powertrain modelling
Upshiftacc 1 to 2
Upshiftacc 2 to 3
350 400
Upshiftacc 3 to 4
Upshiftacc 4 to 5
300 Upshiftacc 5 to 6 350
Upshifttf 1 to 2
250 Upshifttf 2 to 3
300
Upshifttf 3 to 4
Torque (Nm)
Upshifttf 4 to 5
200
Upshifttf 5 to 6
250
Upshiftbsfc 1 to 2
150 Upshiftbsfc 2 to 3
Upshiftbsfc 3 to 4 200
100 Upshiftbsfc 4 to 5
Upshiftbsfc 5 to 6
Upshiftstd 1 to 2 150
50
Upshiftstd 2 to 3
Upshiftstd 3 to 4
100
1000 1500 2000 2500 3000 3500 4000 4500 5000 5500td 4 to6000
Upshift 5
Speed (rpm) s
Upshiftstd 5 to 6
80 Upshiftacc 4 to 5
Upshiftacc 5 to 6
70
Throttle position [%]
Upshifttf 1 to 2
60 Upshifttf 2 to 3
Upshifttf 3 to 4
50
Upshifttf 4 to 5
40 Upshifttf 5 to 6
Upshiftbsfc 1 to 2
30
Upshiftbsfc 2 to 3
20 Upshiftbsfc 3 to 4
10 Upshiftbsfc 4 to 5
Upshiftbsfc 5 to 6
0
0 50 100 150 200 250
Vehicle speed [Km/h]
Figure 3.11: Upshift gear shift map, based on acceleration, traction force, BSFC
50
Powertrain modelling
This section discussed different criteria use to qualify a Dual Clutch Transmission
gear shift-map. Also, simulation results of different design described in Subsection
3.3.2, 3.3.3 and 3.3.4 are compared with a standard gear shift map. In order
to compare different results against each other, every shift were converted as
a traction point (see Figure 3.12), and the following expression describes the
conversion from gear shift point (in km/h) to traction force:
ig η
Ftstd = Te(ωe ) . (3.13)
Rw
where ig and Rw are the gear ratio and wheel radius respectively. Te is the engine
torque expressed a function of engine speed ωe (see Subsection 2.4, and Subsection
2.3.2, in Chapter 2).
[10% Thr] [20% Thr] [30% Thr] [40% Thr] [50% Thr]
2500 5000 10000
3000 6000
Tractive force [N]
0 0 0 0 0
0 100 200 300 0 100 200 300 0 100 200 300 0 100 200 300 0 100 200 300
Speed [km/h] 1st gear
[60% Thr] [70% Thr] [80% Thr] [90% Thr] 2nd gear
[100% Thr]
12000 3rd gear
4th gear
Tractive force [N]
10000
5th gear
10000 10000 10000 10000
8000 6th gear
Sh. p.(new)
6000
Sh. p.(Std)
4000 5000 5000 5000 5000 Sh. p.(bfs)
2000
0 0 0 0 0
0 100 200 300 0 100 200 300 0 100 200 300 0 100 200 300 0 100 200 300
Speed [km/h]
Figure 3.12: Traction force curves with shift point between two adjacent gears
compare with standard and minimum BSFC shift map
Figure 3.13 represents the minimum fuel consumption curve for each gear
and the intersection between two adjacent curves (black circle). The red circle
represents the standard shift point, while the green is BSFC traction force shift
51
Powertrain modelling
point. It can be noticed that both shift points, traction and BSFC shift point
start almost at the same point for lower throttle position, however at higher
throttle position, traction force are significantly reducing.
[10% Thr] [20% Thr] [30% Thr] [40% Thr] [50% Thr]
700 700 550 600 700
500 600
600 600 500
Bsfc [g/kWh]
450
500
500 500
400 400
400
400 400 350
300
300 300
300 300
250 200 200
0 100 200 300 0 100 200 300 0 100 200 300 0 100 200 300 0 100 200 1st300
gear
Speed [km/h]
2nd gear
[60% Thr] [70% Thr] [80% Thr] [90% Thr] [100% Thr]3rd gear
800 800 800 800 800 4th gear
5th gear
6th gear
Bsfc [g/kWh]
Figure 3.13: BSFC curves with shift point between two adjacent gears compare
with standard and traction BSFC shift map
The Figure 3.12 was also defined in the same manner as Figure 3.13, the
conversion from gear shift point to BSFC values is described as follows:
where BSF Ct is an handle function in MATLAB used to model the BSFC map
as a look up table, where it is a function of engine torque and speed. It can be
noticed at lower throttle position, the BSFC shift point for standard gear are very
low, while the value increased at higher throttle position.
Table 3.1 is to compare standard gear shift map to gear shift map design
based on tractive force and BSFC map. It can be remarked that the minimum
BSFC design gear shift does not guaranty a good CO2 emission over the NEDC.
52
Powertrain modelling
Table 3.1: Zone on BSFC map and CO2 emission over the NEDC
Shif t map CO2 [g/km] Zone1 Zone2 Zone3
Standard 198 0.052896 0.11239 0.013851
T raction 228 1.0929 0.13322 0.01092
Standard 202 0.18389 0.14378 0.011415
Downshift schedule
The Downshift schedule is based on a linear convergence or shift buffer zone (Liu
et al. 2009), (Xi et al. 2009). Lower shift point are defined using the minimum
consumption, meaning from 0% to 30% throttle position. From 30% to 100%, a
linear convergence based on engine RPM is used to define the rest of the gear
shift map. Notice that the range of engine is covered by the throttle position.
Upshift
Up
12
Torque [Nm]
300 Up23
200 Up
34
100 Up45
0
0 Up10 20 30 40 50 60 70 80 90 100
Engine speed [RPM]
56
Throttle position [%]
6000
4000
Engine consumption [g/kWh]
2000
0 10 20 30 40 50 60 70 80 90 100
Throttle position [%]
400
300
200
0 10 20 30 40 50 60 70 80 90 100
Throttle position [%]
53
Powertrain modelling
This section describes different methods for selecting transmission gear ratio.
For a given vehicle and engine specification, gear ratios are designed to satisfy
performance requirement, gradability, fuel economy and acceleration. At first the
low and high gears are defined based on vehicle, engine characteristics and road
condition, after the intermediate gear are calculated.
The starting point to calculate gear ratio is based on longitudinal vehicle dynamic
formulas.
The total resistive acting against the vehicle is expressed as follow:
where:
• Froll : Rolling resistance. The rolling resistance is the resistance force acting
on the rolling wheel
where F rcof , mT otal , g and α denote the rolling resistance coefficient, the
whole vehicle mass, gravitational force and road gradient, respectively.
• FAir : Air resistance. The air resistance is made up of the pressure drag
including induced drag, surface resistance and internal resistance.
1
FAir = ρCw Av 2 . (3.17)
2
where ρ, Cw , A and v denote the air density, drag coefficient, front area and
54
Powertrain modelling
vehicle speed.
The diving resistance also called tractive force developed by the engine power
is described as:
itot
Fdrive = Te ηtot . (3.20)
Rw
where Te , itot , Rw and ηtot denote the engine torque, total gear ratio, wheel
dynamic radius and transmission efficiency, respectively.
The engine power is expressed as follows:
Pe = Te ωe . (3.21)
The equilibrium relation between drive forces and running resistance is nat-
urally obtained from the driving force and total force resistance, using Newton
second law:
55
Powertrain modelling
The main contribution of the powertrain is to offer ratio between engine speed
and road wheel speed enabling the vehicle to move under difficult condition and
reasonably operate in the fuel efficient ranges of the engine performance map.
The following section describes the maximum ratio required iA,max , the smallest
gear ratio and finally the intermediate ratio.
The largest gear ratio (LGR) is the starter gear, which is mainly used for slow
driving and starting up the vehicle. A climbing performance (gradeability) α
greater than 50% is normally required for an unladen passenger car. This is to
ensure that the vehicle can tow a trailer and overcome ramp easily, however the
acceleration and aerodynamic are ignored as the vehicle speed is low. The lower
the weigh of the vehicle, the longer the LGR should be (smaller ratio value). The
higher the weight of the vehicle, the shorter the LGR should be (smaller value).
For low torque engines the LGR must be shorter (higher ratio value). The main
driven equation to determine the largest gear ratio is described as follow:
Rw mvehicle+trailer g F rcof cos(α) + sin(α)
iAmax = . (3.24)
T em axηt
where mvehicle+trailer represents the vehicle mass with a trailer. This is to ensure
that a trailer can be towed and climb a ramp with ease.
The final or the smallest gear ratio will depend on the maximum engine power
delivered to wheels, and the resistive power based on rolling and air resistance.
56
Powertrain modelling
Rw
iA,min = ωe (3.25)
Vmax
where Vmax is the maximum vehicle speed delivered by the maximum engine
power. The theoretical maximum speed is defined at the balance point, which
represents the intersection between the resistive power and the engine maximum
power (see Figure 3.15).
Figure 3.15: Performance power curves, where P Ex1, P Ex2 and P Ex3 are the
excess power of Over-revving (racing car), optimality and Under-revving (pas-
senger car) with their respective maximum vehicle speed Vmax1, Vmax2 and
Vmax3
57
Powertrain modelling
π
Pemax = aωe5 + bωe4 + cωe3 + dωe2 + eωe + f ωe . (3.28)
30
After defining the maximum engine power and speed, the maximum vehicle
speed is naturally derived at the balance point of the maximum tractive force
(FrT ot ) and resistive force. The resistive force is expressed as follows:
1 2
FRR = Fr M g + Cw AρAir V . (3.30)
2
Pmax ηt
FrT ot = . (3.31)
Vmax
By equating both tractive and resistive forces, the following equation is de-
rived:
58
Powertrain modelling
Fr M g + 12 Cw AρAir Vmax
2
Vmax
Pmax = . (3.32)
ηt
1 3
Fr M gVmax + Cw AρAir Vmax − Pmax ηt = 0. (3.33)
2
The smallest gear ratio is influenced by the trade-off between the vehicle fuel
economy and performance. The optimum gear ratio can be either modified for
fuel consumption or for vehicle performance. If the gear ratio iA,min increases, the
engine power curve will move to the left of the optimum engine power, and the
vehicle is over geared (Over-revving), which is good for fuel economy. However
if the gear ratio iA,min decreases, the engine power curve will move to the right
of the optimum engine power, and the vehicle is under geared (Under-revving),
which is good for performance. The following equation expressed over geared and
under geared:
Rw
Vmax = ωe . (3.34)
iA,min × F actgear
where F actgear is a factor to increase or decrease the optimum gear ratio. When
F actgear is less than 1, the vehicle is over geared, however if F actgear greater than
1 the vehicle is under geared.
The intermediate gears also called discrete gears ratio are linked kinematically the
vehicle and engine speed. The gear shifting is realised through the intermediate
gears. The intermediate gears should be large enough to allows the next lower
gears to be engaged when the engine torque is reached, without outreaching
the maximum engine speed. The greater the number of gear ratio, the better
59
Powertrain modelling
the engine can exploits the efficiency of the fuel map, however the gear change
frequency will increase. There are standard methods (Naunheimer et al. 2010),
(Mashadi & Crolla 2012) to determine initial intermediate gears knowing the high
and low gears that will be discussed in the coming section.
The geometric progression method is considered as an ideal case, where the gear
is changed at a uniform speed, which results in an engine working range ωH and
ωL (see Figure 3.21). This method requires the engine to operate within the same
speed range, which is naturally selected for best fuel consumption.
After defining the high and low gears ratio, for example for a 6 speeds gearbox
design, the constant step ratio (Kstep ) for a geometric progression method. This
can be defined as follows:
where ig1 , ig2 , ig3 , ig4 , ig5 and ig6 are the gear ratios for a 6 speeds gearbox.
Also, multiplying the equalities results in:
s
5
ig1
Kstep = . (3.37)
ig6
r
nL
Kstep = N −1
. (3.38)
nH
60
Powertrain modelling
Progression design
It can be noticed that the geometrical method produces smaller speed ranges
(δV ) for lower speed ratio, however it produces larger speed ranges in higher
gear, which leads to define the speed ratio as follows:
δVi+1
= Kstep . (3.40)
δVi
Inversely, the gear ratio step can be expressed throughout tractive force range
as follow:
δFi
= Kstep . (3.41)
δFi+1
The multiplication of the ratio Cstepi together will equate the ratio of first to
the last gears, which can be written as follows:
where iL and iH are the low and high gear ratios. By substituting (3.42) into
(3.43), then simplifying results in:
61
Powertrain modelling
N −1 N −1 −2
Kstep = Cstep 1
× kf1+2+...N
actor . (3.44)
1− N
Cstep1 = Kstep × kf actor
2
, N > 2. (3.45)
By knowing the value of lower and higher gear ratios, other value of gear ratios
can be determined by using the following expression:
The two main methodologies have been described, the following section will
discuss the results.
The criteria described in this section are used to qualify how good are the gear
ratio are designed. The results will compare two different methods with the
standard gear ratio supplied by SMTC. As many parameters are involved in the
gear ratio design, this will focus on one case study just for analysis.
62
Powertrain modelling
40
Geom. design
30 Pro. design
Std. Design
Gradability [%]
20
10
−10
0 50 100 150 200 250 300
Vehicle speed [km/h]
Acceleration-Traction force
Figure 3.17 demonstrates maximum acceleration with a provided set of gear ratios
starting from 1st to 6th, where the Geometric and Progressive ratios are compared
with the standard gear ratios. It can be noticed that the acceleration from gear
1 to 6 reduces while the vehicle speed is increasing, alternatively the reserve
acceleration will naturally decrease. Similarly, Figure 3.18 represents traction
63
Powertrain modelling
Geom. design
6 Pro. design
Std. Design
4
Acceleration [m/s2]
−1
16000
Geom. design
Pro. design
14000 Std. Design
12000
Traction force [N]
10000
8000
6000
4000
2000
0
0 50 100 150 200 250 300
Vehicle speed [km/h]
Figure 3.19 compares standard gear ratio with Geometric and Progressive designs.
It can be noticed that the progressive and standard gear ratios are decreasing
similarly. The ratio step is the division of two adjacent ratios, written as follows:
64
Powertrain modelling
20
Geom. design
10 Prog. design
Standard design
0
1 2 3 4 5 6
Gear [−] Geom. design
20 Pro. design
10 Std design
0
1 2 3 4 5 6
20 Geom. design
R. S [−]
Pro. design
10 Std design
0
1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
R. S change [−]
1 2 3 4
Gear [i(g)/i(g+1)]/[i(g+1)/i(g+2)] [−]
δi ig
= . (3.49)
δgear igi+1
The bandwidth is defined as the division of two adjacent ratio steps and it is
written as follows:
2
δi ig /igi+1
= . (3.50)
δgear igi+1 /igi+2
65
Powertrain modelling
Traction loss
1st gear
15000 2nd gear
10000 3th gear
4th gear
5000 5th gear
0 6th gear
50 100 150 200 250 300
Max traction force
Progression gear ratio
15000
10000
5000
0
50 100 150 200 250 300
Standard gear ratio
15000
10000
5000
0
50 100 150 200 250 300
Vehicle speed [km/h]
The objective of spacing the gear ratio is to minimise the loss of traction force
(i.e intersection between the traction force curve of 1st, 2nd gear and maximum
traction force, see Figure 3.20) because of the discontinuous stepping of gears.
Traction loss is constant for Geometric design, however for progressive design the
traction loss decreases as gear ratio increases.
66
Powertrain modelling
0
0 50 100 150 200 250 300
Engine speed [RPM]
0
0 50 100 150 200 250 300 350
Vehicle speed [km/h]
Fuel consumption
45 Geo. design
Pro. design
40 std. design
Speed for consumption
35
Fuel consumption [1/100km]
30
25
20
15
10
67
Powertrain modelling
BSF C × Pe
FC = . (3.51)
ρf uel × v
where ρf uel denotes fuel density. The fuel consumption criteria is based on the
largest gear ratio (6th) at 120 km/h, and it is defined in the following table:
This section has described the main criteria of gear ratio. The next step will
be to design an algorithm to optimise gear ratio without degrading the vehicle
performance, fuel consumption, emissions and comfort.
This chapter has described the powertrain model of the ROEWE 950 vehicle
with its validation, and traditional design of gear ratio and gear shift map. This
information is then exploited in Chapters 4 and 5 in order to define the problem
formulation and develop algorithms for gear ratio and gear shift map optimisation.
68
Chapter 4
Problem formulation
4.1 Introduction
69
Problem formulation
This section presents the problem formulation for gear shift map optimisation.
The role of the gear shift map is to move the engine towards its most efficient
regions in terms of both emission and performance. However, the designer of
any gear shift map is always constrained by the region that the engine can be
placed on the BSFC map. Limitations arise in standard automated gearbox from
the availability of only a fixed number of discrete gear ratios that are finite in
number. In this thesis a 6 speed SAIC Dual Clutch Transmissions was used to
develop the optimised shift map. The standard shift map is composed of 5 gear
set including Upshift and Downshift.
The optimisation algorithm should find the best solution, however the decision
as to which criteria to use and the relative importance between criteria should lie
with the engineer. This section considers a number of objectives formulation that
can help differentiate alternative solutions. The main objectives are expressed and
grouped in terms of emissions (Ngo, Hofman, Steinbuch & Serrarens 2013) (Yin
et al. 2013), driveability (Le Guen et al. 2011) and durability. The objective
formulation presented in this section have either been adopted from published
work or designed to provide additional means to differentiate between similar
solutions. The formulation was adapted for a minimisation problem, where the
lowest values of the objectives represent the best solutions.
70
Problem formulation
The gear utilisation criterion is novel and relates to the means to achieve a low
CO2 . It is based on the assumption that vehicles consume less fuel and produce
less emissions when they operate on a higher gear. This new criterion aims to
quantify the time spent on each gear, hence identify which gear is contributing to
low CO2 . Accordingly, this objective function was designed to assist the engineer
to quantify the gear usage during the NEDC and identify which gear ratios lead to
lower CO2 emissions. The gear utilisation criterion JP GU is given by the inverse
percentage of gear utilisation over a driving cycle. It is described by (Llamas
et al. 2010):
−1
6
X
JP GU = Gu %αu (4.1)
u=1
where the weighting factors αu (see Table 4.1) are defined in order to favour the
time spent on specific gears, with
Gu × 100
Gu % = PNg =6 (4.2)
u=1 Gu
where Gu % is the percentage of time spent on each gear, and u denotes each gear
ratio.
In this thesis, the aim is to promote the use of higher gears, hence the higher
the gear, the higher the associated weighting factors αu (see Table 4.1).
In addition taking inspiration from Le Guen et al. (2011), a new cost function
was developed to minimise CO2 by moving the engine operating points, expressed
in terms of engine torque, Te , and engine speed, we , towards the left side of the
BSFC map. It is realised by minimising the distance (see Figure 4.1) between
71
Problem formulation
a reference, or anchor, point O(wref , Tref ) on the BSFC map, and the Upshift
points for the throttle positions, tk , of interest. The distance is calculated based
on Upshift of each gear set: g ∈ [1, 2, 3, 4, 5], and throttle positions tk in 10%
increment. The distance dupshif ts is expressed as:
5
X 100
X
dupshif ts = dg,tk (4.3)
g=1 k=0
where dg,tk denotes the distance for each Upshift set, and it is expressed as:
with wref ∈ [780, 2000], Tref ∈ [80, 150], where O(wref , Tref ) and Ug−1,g,tk (we , Te )
represent the position of the anchor point fixed on the left edge of the BSFC map
and the Upshift between the gears (g − 1) and g, respectively.
Figure 4.1: The distance (dtk ) between the reference point O(wref , Tref ), and the
Upshift 1 (Up1 ) and Upshift 2 (Up2 ) at 0%, 40% and 100% throttle positions,
respectively. The grey dotted line represents the engine maximum torque. The
engine speed varies from the minimum stable speed, stalling speed, to the max-
imum engine speed. Z1 represents the zone with the most efficient operating
point, Z2 and Z3 are zones with higher fuel consumption
72
Problem formulation
The second set of objectives aims to improve driveability. The standard Inverse
Reserve Power (IRP) formulation was adopted to characterise the vehicle ability
to accelerate (see Section 2.3.2, in Chapter 2):
A new set of criteria aims to simultaneously optimise CO2 emissions and
driveability, inspired from Le Guen et al. (2011), aims to maximise the percentage
of time spent on the most efficient Engine Operating Point (EOP). It is achieved
by dividing the BSFC map into zones defined based on the range of BSFC values.
Three zones were defined based on cross-correlation study (see Section 6.1.1 in
Chapter 6) between zones and the main objective function, CO2 and IRP. The
zone thresholds were tuned in order to relate zone 1 to CO2 emissions as in
this thesis, the main focus is based on minimising CO2 emission. The ranges of
BSFC values, [g/K m], for zone1 , zone2 and zone3 are [200, 255[, [255, 265[ and
[265 max(BSF C)], respectively. To express this objective for a minimisation
problem, the inverse of the percentage of time the engine spends in each of the
three zones is calculated as follows:
−1
Zone1 100
J =
z1
P 3
k=1 Zonek Wk
−1
Jz2 = P3Zone2 100 (4.5)
k=1 Zonek Wk
−1
Jz3 = P3Zone3 100
Zone W
k=1 k k
where k ∈ [1, 2, 3], and Wk represent the individual weighting for each zones.
Based on the zone definition, Jz1 is the most efficient, followed by Jz2 and Jz3 .
The optimisation algorithm will therefore aim to generate solutions corresponding
to the engine operating most of the time in the most efficient regions, resulting
in the smallest possible values for Jz1 .
73
Problem formulation
The last criteria aims to improve gearbox durability by minimising the number
of gear change in order to prolong gearbox longevity, and as a by-product very
short successive up/down-down/up gear changes. It is expressed as sum of the
absolute of the difference between successive gear changes:
N
X −1
JGchange = ˙
dG(k) (4.6)
k=1
Two Downshift or two Upshift are not allowed to cross over. This is implemented
by calculating the distance between two adjacent Downshift (Dg,g−1,tk , Dg+1,g,tk )
or Upshift (Ug−1,g,tk , Ug,g+1,tk ) and ensure that the distance is positive. It is
expressed as follows:
74
Problem formulation
Ug−1,g,tk − Ug,g+1,tk > 0
(4.7)
Dg,g−1,t − Dg+1,g,t > 0
k k
The engine speed should not be less than the minimum stable speed MinEngspeed ,
nor should it be greater than the maximum allowable speed MaxEngspeed :
Fourth, the shape of the shift map is also controlled, by observing the percentage
of slope, between two adjacent throttle positions (Ug−1,g,tk+1 , Ug−1,g,tk ) of the
same Upshift gear. It is given as follows:
Ug−1,g,tk+1 − Ug−1,g,tk
U pSlopek+1 = 100%. (4.9)
Ug−1,g,tk+1
Fifth and last constraint is to avoid a shift map with rapid gear change, Up/Down
or Down/Up. A conservative value of minimum gear change time was carefully
chosen based on the average gear change time of SAIC DCT, which is 400 ms.
75
Problem formulation
Design variables represent the free variables to be optimised. They are mapped
from the gear shift points that represent the gear shift map. The new map-
ping presented in this thesis is applicable to any optimisation techniques. It
has been designed to enforce the following engineering constraints: i) prevent
crossing between Downshift and Upshift ii) maintain a minimum hysteresis be-
tween Downshift and Upshift to avoid frequent gear changes for small velocity
variations.
The variable mapping expresses the Downshift (V D) and Upshift (V U ), from
a set of independent variables to a set of relative increments (∆V D).
Figure 4.2 illustrates the mapping of the gear shift map onto design variables.
Each variable is expressed in terms of a specific throttle position t% and a set of
consecutive gears (g, g − 1), where the subscripts g and (g − 1) denote for even
76
Problem formulation
M
X
Dg,g−1,tM = V Dg,g−1,t0 + ∆V Dg,g−1,tk . (4.10)
k=0
The inverse mapping to obtain the design variables from the points on the
gear shift map is given by:
and
∆V Ug−1,g,tM = Ug−1,g,tM − Vhyst,tM − Dg,g−1,tM . (4.13)
77
Problem formulation
The range and resolution of the design variables were carefully tailored to speed
up the algorithm convergence whilst at the same time ensure that the solutions
produced were suitable. To guide the algorithm towards practical solutions, the
initial population was randomly generated based on a proprietary gear shift map
designed using standard techniques. It was found empirically that the best com-
promise between solution space coverage and generation of practically acceptable
solutions was obtained by restricting the change from the original gear shift map
to ± 27%.
78
Problem formulation
This section describes gear ratio problem formulation. The 6 speed SAIC DCT
gearbox comprises 6 gear ratios. The first gear should allow the vehicle to start
with a trailer under a road grade of 50% gradability.
The 4 intermediates gear ratio (2nd ,3rd , 4th and 5th ) are utilised once the
vehicle has started moving. The last gear ratio (6th ) in this thesis, is mainly
designed for fuel economy and comfort especially for passenger car.
The next subsections describe the intermediate gear ratio problem formulation
(see Section 4.3.1), the last gear formulation (see Section 4.3.2).
The first gear ratio design is not considered in this thesis as it is mainly
dependant on vehicle gradability, and ability to carry a trailer. It is assumed
that it is taken from the proprietary gear shift map.
The aim of intermediate gears is to allow the vehicle to move from high to low
gear or the other way. They also allow to navigate through different zones of
BSFC map and access the most efficient area of fuel consumption.
The performance of a set of gear ratio is mainly assessed by spreading the
gear ratio on the engine BFSC map, which should guide the engineers to observe
where each gear ratio is operating on the BSFC map. Three main objective
functions are considered. The first is fuel consumption which is a by-product
of CO2 emissions. The second is IRP which is inversely proportional to CO2
emissions, which helps the engineers to limit how far the optimiser can minimise
CO2 emissions while still conserving a minimum driveability. The third one is
related to the intermediate gear ratio spacing, it is defined as the bandwidth of
gear step ratio. The following statement lists the objectives function:
79
Problem formulation
• Gear step change bandwidth (Equation (3.50), see Section 3.4.5, in Chap-
ter 3)
After describing the objective functions. The next task is to outline any
engineering constraints on objective functions. They are given in the next section.
Formulate constraints
• The values of a set of gear ratio starting from the first to the last gear, must
be defined in descending order.
• The gear step change bandwidth must be below a fixed constant defined by
the engineer, a more common value should be less than 0.135 (see Equa-
tion (3.50), see Section 3.4.5, Chapter 3).
Design variables
80
Problem formulation
aim to find the optimal design variables values from which the gear ratios will be
reconstructed so that their effect can be simulated within the vehicle model.
G1 − ψa
G2 =
G3 = G1 − ψa − ψb
(4.14)
G4 = G1 − ψa − ψb − ψc
G5 = G1 − ψa − ψb − ψc − ψd
where, ψa , ψb , ψc and ψd denote the design variables that relate to gear ratio
G2 to G5 , G3 to G5 and G4 to G5 , respectively. This is achieved adopting the
formulation described in (4.14) and by constraining the design variable to be
strictly positive.
The final task of the formulation is to bound each design variable, by setting the
lower and upper bounds of the design variables, such as to keep the solution in
the feasible region Rn (Augusto et al. 2012) and to limit the solution space. A set
of gear ratio is defined in descending order, which mean the gear ratio boundaries
must follow the same pattern, and is given as:
where, ψυL and ψυU represent the lower and upper gear ratio design variables. They
have been defined between ± 20% of the original gear ratio of SAIC 6 speed DCT.
81
Problem formulation
The use of the last gear on a passenger vehicle is generally encouraged during
cruising, when the vehicle is expected to favour fuel economy. As described in
Section 3.4.3, in Chapter 3, the three types of design are Under-revving, Optimal
and Over-revving. In this thesis, a saloon passenger car is considered, which leads
the justification of the adoption of Under-revving design.
Objective functions
Similar competing objectives apply to the gear shift map optimisation and to the
gear ratio optimisation. The optimisation of the last gear ratio is mainly dom-
inated by fuel consumption, and the specific objective considered is the vehicle
fuel consumption at 120 km/h (see Section 3.4.5, in Chapter 3). The CO2 emis-
sion reduction should however not be at the total detriment of vehicle comfort
and driveability. The latter is expressed as the vehicle excess power available for
the last gear (see Section 3.4.5, in Chapter 3).
Constraints
The constraints imposed on the last gear ratio, are designed to ensure that the
Over-revving corresponds to high performance (greater excess power and fuel
consumption), unlikely to Under-revving with better fuel consumption (smaller
excess power).
Design variables
The selection of the last gear ratio, considering Under-revving design, is obtained
by increasing the last gear ratio of the optimal design (see Section 3.4.3, in Chap-
ter 3). However there are two factors to be considered, where one factor (iF act,1 ) is
for Under-revving, and a second factor iF act,2 for Over-revving. Both factors must
82
Problem formulation
The boundaries on the factors used to decrease or increase the gear ratio are given
as follows:
where, τ are the two boundaries for the optimiser. The boundaries chosen for
this thesis are given in the following Table 4.2:
This section describes the decision maker to handle multi-objective functions for
gear shift map and gear ratio.
In this thesis, the decision maker for gear shift map is based on two principles.
The first is a weighted sum combining all objective functions into one scalar. The
second is based on Pareto optimal solutions (see Section 2.3.5, in Chapter 2). The
two methods are describe as follows:
83
Problem formulation
The traditional weighted sum method, which combined multiple objective func-
tions into one single scalar (see Section 2.3.5, in Chapter 2) is expressed as follows:
X Job1(i) WGSMi
ObjGSM = (4.17)
JN 1(i)
i ∈ [1, ..., 8]
Job1 ∈ JCO2 , JIRP , JGj , JGch , JDist , Jz1 , Jz2 , Jz3
JN 1 ∈ JCO2 (x0 ), JIRP (x0 ), JGj (x0 ), JGch (x0 ), JDist (x0 ), Jz1 (x0 ), Jz2 (x0 ), Jz3 (x0 )
where WGSMi denotes the weighting associated with individual objectives Job(i)
function. JCO2 , JIRP , JGj , JGch , JDist , Jz1 , Jz2 , Jz3 denote the objective functions
for CO2 , zone 1, IRP, gear change frequency, time spent on each specific gear,
distance, zone 1, zone 2 and zone 3 respectively. JN 1 represents the values for
each objective function corresponding to the current manufacturer gear shit map,
which is used as a reference for subsequent optimisation (see Section 2.3.5, in
Chapter 2). Additionally WGSMi are positive and must satisfy:
i=8
X
WGSMi = 1, WGSMi ∈ (0, 1) (4.18)
i=1
In this thesis, a modified Pareto based on Haas et al. (1998) is applied. It uses ob-
jective weighting Pareto ranking to differentiate between non dominated solutions
given by:
84
Problem formulation
j ∈ [1, 2 and 3]
Job2 ∈ [JCO2 , JIRP , JBwd ]
JN 2 ∈ JCO2 (x0 ), JIRP (x0 ), JBwd (x0 )
where WGRj denotes the weighting associated with individual objectives Job2(i) .
JBwd denotes the gear ratio step bandwidth. JN 2 represents the objective function
with the initial gear ratio design variable. It is used to normalise each objective
function during optimisation process. Also, WGRi must be positive and satisfying:
i=1
X
WGRi = 1, WGRi ∈ (0, 1) (4.21)
i=8
Similar to Equation (4.19), the same pattern is utilised to define the modified
Pareto for gear ratio. It is given as follows:
This chapter has provided the problem formulation for both shift map and gear
ratio optimisations. The problem formulation is the most important stage in
optimisation. An identical approach has been adopted for each problem consid-
85
Problem formulation
ered: objective and constraints formulation, design variable and bounds formu-
lation. Five objective formulations have been proposed in this thesis to supple-
ment widely accepted formulation. These objectives aim to help users identify
the most suitable solutions, typically by observing the performance of engine op-
erating point on the BSFC map, whilst the specific objectives are expressed using
various formulations. Additionally, the overall solution of the problem considered
involves competing objectives.
The key to success is rooted in the handling of objectives which are presented
in Chapter 2. The new design variable formulation is one of the significant con-
tributions of this thesis. It aims to simplify the design process for a gear shift
map, also, it allows the user to specify a range of throttle positions and minimum
hysteresis in order to guide the optimiser more accurately.
Having formulated the optimisation problem, the next chapter describes var-
ious optimisers used to find suitable solutions for either gear shift map, gear
ratio or simultaneously weighing the best combination of gear shift map and gear
ratio.
86
Chapter 5
5.1 Introduction
This Chapter presents the main technical contributions of this thesis. It describes
the algorithms developed to optimise the gear shift map as well as the gear ratio.
The chapter starts with the description of the Multi-Objective Genetic Algorithm
(MOGA) modified to accommodate the proposed problem specific operator and
a repair mechanism. The second nature inspired algorithm is then presented,
namely the Multi-Objective Cuckoo Search (MOCS), which is also modified to
accommodate a new local search applied to optimise gear ratio. It presents the
different behaviour of the nature inspired techniques implemented in this thesis
to justify their selection and their specific features. Finally, it describes the
culmination of the work by merging MOGA, MOCS and constrained optimisation,
implemented within the MATLAB interior-point algorithm, for the combined gear
ratio and gear shift map optimisation.
87
Evolutionary algorithm & swarm intelligence for shift map and gear ratio
optimisation
5.2 Problem specific MOGA
A MOGA was selected to optimise a six speed DCT gear shift map due to its
ability to handle competing and changing objectives (see Section 4.2.1 in Chap-
ter 4). Whilst many objectives have been formulated, the two main objectives are
to lower the CO2 whilst keeping the driveability acceptable. To focus the search
towards lowering CO2 , a modified Pareto ranking (Haas et al. 1998) was adopted
(see Section 4.4.1 , in Chapter 4).
The standard MOGA (Konak et al. 2006) was complemented by a new repair
mechanism and a new local search operator, as shown in Figure 5.1 and 5.2. The
aim of these improvements is to exploit problem specific features to find better
solutions faster. The new repair mechanism was developed to handle minimum
reserve power requirements (see Section 5.2.2, in Chapter 4). The new problem
specific operator referred as gear early shifting (GES) operator was developed to
focus the search towards good regions, in terms of CO2 emissions, of the solution
space (see Section 5.2.1). To prevent the algorithm from becoming trapped in
a local minima and producing gear shift with similar characteristics, the GES
operator is only applied every N generations, where N is a user tunable parameter.
To avoid unrealistic gear shift map to be accepted, each solution is first checked
against the constraints defined in Section 4.2.2. Then the vehicle simulator is run
for each candidate solution. If the simulator does not complete the drive cycle
simulation, the solution is rejected (see Figure 5.2) as it violates the gear shifting
logic and/or is not compliant with the vehicle BSFC map.
88
Evolutionary algorithm & swarm intelligence for shift map and gear ratio
optimisation
Figure 5.1: Multi-Objective Genetic Algorithm with problem specific (GES) op-
erator and repair mechanism to ensure reserve power constraints are met by the
optimised solutions
89
Evolutionary algorithm & swarm intelligence for shift map and gear ratio
optimisation
Figure 5.2: Multi-Objective Genetic Algorithm with GES and repair mechanism
90
Evolutionary algorithm & swarm intelligence for shift map and gear ratio
optimisation
5.2.1 Local search: gear early shifting (GES) operator
The new GES operator aims to reduce CO2 emissions by producing early gear
shift to reach as quickly as possible the most efficient area of the BSFC map. It is
realised by reducing the velocity difference between Upshift and Downshift, where
the speed increment ∆V Ug,g−1,tM are reduced by a ratio βk ∈ [0.25, 0.50, 0.75]
expressed in percentages to form:
∆V U25g,g−1,tM = β25 ∆V Ug,g−1,tM
∆V U50g,g−1,tM = β50 ∆V Ug,g−1,tM (5.1)
∆V U75g,g−1,tM = β75 ∆V Ug,g−1,tM
In this thesis the same set of ratios βk < 1 are applied to each gear set
and each throttle angle. Investigating randomly generated ratio is considered as
further work. Similarly is the investigation of the benefits of using ratios βk > 1
to increase the difference between Upshift and Downshift, thereby increasing the
hysteresis between up and down shift, resulting in making quick gear changes less
likely, but at the cost of higher CO2 .
Parents used by the GES operator are selected based on the following rules.
First, NCO2 candidate solutions are randomly selected among the NbestCO2 indi-
viduals in term of CO2 . Then NIRP candidate solutions are randomly selected
among NbestIRP individuals in term of IRP. Finally NDif f candidate solutions are
selected according to the highest difference in terms of Euclidean distance CO2
and IRP, between successive point on Pareto set. In this work NbestCO2 =NbestIRP
= 10, NCO2 = NIRP = NDif f =1. Figure 5.3 summarises the algorithm with iind
denoting each individual solution, C and P with subscript CO2 , IRP and Diff
denoting the children and parents for the three criteria considered respectively.
91
Evolutionary algorithm & swarm intelligence for shift map and gear ratio
optimisation
1: Randomly select NCO2 amongst NbestCO2 individuals
2: Randomly select NIRP amongst NbestIRP individuals
3: Select NDif f individual
4: for iind = 1 to NCO2 + NIRP + NDif f do
5: Calculate speed difference ∆V Ug,g−1,tM
6: Create 3 new maps per candidate solution to move Upshift towards the
Downshift
7: for k = 1 to 3 do
8: Replace ∆V Ug−1,g,tM in P by βk ∆V Ug,g−1,tM to form C
9: end for
10: end for
The main emphasise of MOGA is to reduce CO2 . This can lead to solutions
that are on the limit or even inappropriate in terms of reserve power, i.e. ability
to accelerate after a gear change. A method called GeneRepair operator was
proposed by Mitchell et al. (2003), and used to correct invalid tours which may
be generated following crossover and mutation in Travelling Salesman Problem.
Similarly, a repair mechanism has been devised to detect conditions when the
reserve power is insufficient at any time during the drive cycle and automatically
adjust the appropriate Upshift to ensure that the minimum requirements in terms
of reserve power are met, see Figure 5.4.
92
Evolutionary algorithm & swarm intelligence for shift map and gear ratio
optimisation
5.2.3 Conclusions on problem specific MOGA
This section has presented a MOGA combined with problem specific operator to
improve solutions quality and rate of convergence. Additionally, a repair mecha-
nism was developed to insure that solutions produced are practically realisable by
enforcing a minimum reserve power constraints. The relative benefits of the pro-
posed modification are evaluated in Chapter 7. The following section describes
Cuckoo Search algorithm develop to optimise gear ratio.
This section, describes Cuckoo Search (CS) algorithm (Yang & Deb 2013) modi-
fied for the context of gear ratio optimisation. In addition to the standard Levy
Flight operator, it includes Bat, Firefly and Flower Pollination. Theses oper-
ators are integrated within the Cuckoo Search to generate new optimised gear
ratio. The operator combination aims to improve the performance of the existing
Cuckoo Search by exploiting the benefits of other operators.
There are three objectives that are minimised. The most important is the CO2
followed by IRP and bandwidth. Note that the bandwidth is also formulated
as a constraint together with the gear ratio step change (see Section 3.4.5 in
Chapter 3).
93
Evolutionary algorithm & swarm intelligence for shift map and gear ratio
optimisation
Figure 5.5: Multi-Objective Cuckoo Search with Levy Flight operator supple-
mented by, Bat, Firefly and Flower Pollination operator for gear ratio optimisa-
tion.
94
Evolutionary algorithm & swarm intelligence for shift map and gear ratio
optimisation
In this research, CS has been developed to solve a Multi-Objective Optimi-
sation Problem (MOOP), and uses a modified Pareto rangking to find a set of
non-dominated solutions based on 4.19. There are four design variables related
to a set of gear ratio to be optimised (see Section 4.3.1 in Chapter 4). The CS
for gear ratio optimisation is illustrated in Figures 5.5 and 5.6 and described in
Figures 5.7, 5.8. The initialisation stage generates xi solutions comprising the
four design variables (ψa , ψb , ψc and ψd ) that convert onto 6 gear ratios (see
Section 4.3.1). Additionally, considering a MOOP, a Cuckoo can lays multiple
eggs (objective functions) in a nest. The main goal is to replace the current value
of the objective functions by new solutions in the nest. At each generation, the
following tasks are accomplished: (i) Generate new gear set solutions by applying
either Levy Fly, Bat, Firefly or Flower Pollination operators. (ii) Reconstruct the
gear set, run the vehicle model, evaluate each solutions and calculate the objec-
tives (JCO2 , JIRP and JBwd ) for each solutions. (iii) Check if any new solutions in
are Pareto optimal, then replace the worse solutions, else choose jn nest randomly
and replace them by in if JBwd is better. (vi) New optimised set of gear ratios
ln are generated by abandoning the worse solutions with the probability of Pa ∈
[0, 1], repeat (ii). (v) Add new solutions created, ln , to the population, where the
modified Pareto cost and fitness of each solutions are calculated. Then classify
the solutions in ascending order, finally keep the n host nest with best gear set for
the next generation. (vi) Repeat until convergence or until a user defined time
limit is reached.
xi = ψai , ψbi , ψci , ψdi (5.2)
95
Evolutionary algorithm & swarm intelligence for shift map and gear ratio
optimisation
Figure 5.7: Multi-Objective Cuckoo with hybrid operators for gear ratio optimi-
sation
Figure 5.8: Multi-Objective Cuckoo with hybrid operators for gear ratio optimi-
sation
96
Evolutionary algorithm & swarm intelligence for shift map and gear ratio
optimisation
5.3.1 Local search component
This subsection describes different operators used within the Cuckoo Search:
Levy Flight
The original Cuckoo search relies on the Levy flight operator to generate new
solutions. It is a random walk, where the step size is based on Levy distribution.
It was adopted in this work, as Levy flight is efficient in terms of exploring large-
scale search space. A new solution xt+1
i is defined as follows:
xt+1
i = xti + θ ⊕ Levy(τ ) (5.3)
where xti is the current solution, and θ0 is a constant, whilst the expression in
the bracket corresponds to the difference between the current and the best so-
lution in the nest respectively. Generating new solutions from Levy flight is not
straightforward, a simple scheme was defined by Yang (2014) can be described as
follows:
u
Levy (τ1 ) = (5.5)
|v|1/τ1
where, u and v are obtained from a normal distribution, and it is given as follows:
with
Γ(1 + λ) sin(πλ/2)
σu = (5.7)
λΓ((1 + λ)/2) 2(λ−1)/2
97
Evolutionary algorithm & swarm intelligence for shift map and gear ratio
optimisation
where Γ is the standard Gamma function.
Bat operator
The Bat operator is used in this search, as it has the particularity to use frequency
tuning by updating a current solution to obtain a new solutions. The main idea is
inspired by echolocation of microbats. A solution is represented by a virtual bat
position and its corresponding velocity. It is given by the following expression:
xt+1
i = xti + vit+1 (5.10)
where, xtcurbest2 denotes for current best location (current optimal gear ratio) which
has been found so far among the nhost virtual bats location. The original Bat
algorithm includes a local search, presented by the loudness, in this research
loudness is not considered as only Bat algorithm operator is utilised. Also, it
is assumed that the pulsation rate is fixed. The interesting particularity of Bat
algorithm, it captures the advantage of many algorithm such as the standard of
PSO where the frequency controls the space area of swarming particles motions,
and Harmony Search(HS) by varying the loudness and pulsation rate. It can be
98
Evolutionary algorithm & swarm intelligence for shift map and gear ratio
optimisation
noticed that the updated solution xi is similar to Arithmetic crossover from GA
(Ladkany 2012).
Firefly
The Firefly algorithm (FA) is based on the flashing pattern of fireflies. The
operator of Firefly is obtained from the original Fireflies Algorithm. A potential
solution of gear ratio is defined as a firefly location. The following assumptions
are made to mimic the behaviour of the algorithm:
• All fireflies are unisex, which mean that any firefly can be attracted by any
other firefly.
2
βf f ly = βf f ly0 e−γf f ly .rf f ly (5.11)
2
xt+1
i = xti + βf f ly0 e−γf f ly .rf f ly,(i,j) .(xti − xtj ) + αt µti (5.12)
where xi is the motion of Firefly i attracted to another, rf2 f ly,(i,j) , is the distance
between any two Fireflies i and j located at xi and xj , it is given by the Euclidean
distance:
99
Evolutionary algorithm & swarm intelligence for shift map and gear ratio
optimisation
v
u υ=d
uX
rf f ly,(i,j) =t (xi,υ − xj,υ )2 (5.13)
υ=a
where, xi,υ is the υth component of the gear ratio design variables, which rep-
resents the spatial coordinate xi . Considering gear ratio design variables, the
coordinates between two Fireflies are described as follows:
q
rf f ly,(i,j) = (xi,a − xj,a )2 + (xi,b − xj,b )2 + (xi,c − xj,c )2 + (xi,d − xj,d )2 (5.14)
Figure 5.9 illustrates the core component of the algorithm, where two iterative
loops are used to compare each firefly (gear ratio) light intensity (given by the
weighted sum of the objective functions, see Section 4.4.2, in Chapter 4), and
move any firefly towards the firefly with the strongest light. Ii and Ij denote the
light intensity of iterative loop 1 and loop 2 respectively. The advantage of Firefly
is to always look forward to move all Firefly toward the current best solution,
however the drawback of the algorithm might limit the exploration of different
research space. The last component is based on Flower Pollination algorithm
which is described in the following section.
100
Evolutionary algorithm & swarm intelligence for shift map and gear ratio
optimisation
Flower pollination
After defining the assumptions, a new gear ratio design is obtained using the
Flower Pollination operator which is given as follows:
xt+1
i = xti + F P (xtj − xtk ) (5.15)
where xtj and xtk denote pollen from flower 1 (randomly selected gear ratio among
the current population) and flower 2 of the same species respectively. This allows
to mimic the flower constancy in a restricted neighbourhood. F P is drawn from
a uniform distribution. FA operator is similar to heuristic crossover (Kaya et al.
2011), it has the advantage of directing the search in a promising direction, also
it has the particularity to relocate the search when the solutions are clustered.
However, it has the drawback of preventing the search space to focus on one
direction.
101
Evolutionary algorithm & swarm intelligence for shift map and gear ratio
optimisation
5.3.2 Conclusion
This section has presented a hybrid MOCS algorithm. The hybridization supple-
mented Levy Flight operator with operators originally developed for Bat, Firefly
and Flower Pollination. These operators are arithmetic and heuristic crossover
operator, and the combine of these operators can improve the algorithm in term
of exploration and exploitation.
The following section describes how MOGA and MOCS are combined together
to simultaneously optimise gear ratio and shift map.
Search
The main concept of this hybrid design combining MOGA, cuckoo and con-
strained optimisation is to exploit the relative strengths as well as solutions pre-
viously obtained when considering each problem independently. CS is employed
to obtain a set of optimised gear ratio. GA is used to generate a set of optimised
gear shift map in favour to low CO2 emission whilst keeping a good driveability.
The following section describes the algorithms combination core.
The approach adopted in this work is illustrated in Figure 5.10, where three
iterative loops are used to combine the algorithms for gear ratio and gear shift
map optimisation. The first iterative loop vary a set of Starter gear ratio including
the initial starter gear ratio, which has been defined manually with respect to
gradeability (see Section 3.4.5, in Chapter 3). A second iterative loop is then
used to select the last gear ratio pre-defined by interior-point algorithm. CS
algorithm is integrated in the second iterative loop, where a set of intermediate
102
Evolutionary algorithm & swarm intelligence for shift map and gear ratio
optimisation
gear ratios are optimised. The third iterative loop is used to select a set of gear
ratio, then applied GA to obtain various optimised shift maps.
103
Evolutionary algorithm & swarm intelligence for shift map and gear ratio
optimisation
algorithm to evolve toward the desired search space.
It is worth pointing out that mutation can take different forms. The generation
of new offspring created by the operator of Levy Flight (5.3), Bat (5.10), Firefly
(5.12) and Flower pollination (5.15) algorithms are mainly mutation based. These
operator use stochastic moves or randomisation method to generation the next
offspring.
Levy Flight, Bat and Flower pollination algorithms operators use current best
solution among the population to make the next moves. In FA operator, there is
no current best solution, however it uses a ranking and selection methods during
the update of offspring based on two iterative loops. It can subdivide into multiple
subgroups, where each subgroup can potentially swarm around a local mode.
In order to assess each operator abilities of reproducing offspring, an exper-
imental test was set up. The experimental set up consists of generating var-
ious offspring based on different crossover operators (direct, intermediate and
extended line recombination) from GA, and operators (Levy Flight, Bat, Firefly
and Flower pollination) from hybrid MOCS algorithm developed in this thesis.
This case study considered the gear ratio with four design variables denoted by
ψa , ψb , ψc and ψd . Two parents (P1 and P2) are selected from a preliminary test
and replicated 3000 times in order to be used by various operators to generate
offspring.
Figures 5.11 and 5.12 illustrate the spread of offspring produces by crossover
operators from GA and operators from MOCS. In Figure 5.12, it is clear that
Levy flight only focus on the two parents to generate offspring. However GA
crossover, Bat and Flower pollination algorithm operators generate offspring near
both sides of the parents and in between. Bat algorithm operators seem to cover
a bit more search space than the last operators. It can be noticed that Levy
Flight operators is intensively exploitation. However the operators from GA, Bat
and Flower pollination algorithms are exploitation, but they also consider a small
104
Evolutionary algorithm & swarm intelligence for shift map and gear ratio
optimisation
2 2
10 10
0 0
10 10
1.3 1.4 1.5 1.6 1.7 1.8 0.7 0.8 0.9 1 1.1
ψc [−] ψ [−]
d
Figure 5.11: Semi-log plot results representing the spread of offspring based on
two parents P1 and P2. The offspring are generated using GA crossover operators:
direct, intermediate and extended line recombination, also with hybrid MOCS
operators: Levy Flight, Bat, Firefly and Flower pollination
range of exploration.
In Figure 5.11, it can be seen that Firefly algorithm outperforms all operators
in terms of offspring distribution, as it has the largest range of exploitation dis-
tribution. Overall, Firefly algorithm had a promising potential of exploring the
search space, additionally it can also act as exploitation.
105
Evolutionary algorithm & swarm intelligence for shift map and gear ratio
optimisation
ψa, P1=5.63, P2=5.58 ψc, P1=2.9, P2=2.9
2 2
10 10
0
0 10
10
5.55 5.6 5.65 2.9 2.905 2.91 2.915 2.92 2.925 2.93
ψa [−] GA recint ψc [−]
ψc, P1=1.54, P2=1.56 GA reclin ψd, P1=0.961, P2=0.978
GA recdis
Levy
bat
Firefly
2
10 flower 102
0 0
10 10
1.52 1.53 1.54 1.55 1.56 1.57 1.58 1.59 0.94 0.95 0.96 0.97 0.98 0.99 1 1.01
ψc [−] ψd [−]
Figure 5.12: Semi-log plot results representing the spread of offspring based on
two parents P1 and P2. The offspring are generated using GA crossover operators:
direct, intermediate and extended line recombination, also with hybrid MOCS
operators: Levy Flight, Bat, Firefly and Flower pollination. This figure illustrates
a zooming view around the two parents.
106
Chapter 6
6.1 Introduction
This chapter builds on Chapters 4 and 5. Chapter 4 presented the new formu-
lation proposed to solve both the gear ratio and the gear shift map optimisation
problems. Chapter 5 described the new optimisation framework exploiting the
proposed problem formulations and the problem specific knowledge.
The first section focuses on the objectives formulation and in particular the
parameter selection for the new zone definition within the BSFC map and the
correlation between the zones and the other objectives.
The second section starts by classifying the solutions obtained according to dif-
ferent objectives to relate objective value to engineering requirements and specific
features to differentiate the various solutions. A correlation analysis between CO2
emission and all the other objectives including the alternative zones parametrisa-
tion is performed to identify appropriate zone thresholds. This section concludes
with a proposed method to select objective weightings to express their relative
importance based on a user classification of a sample of solutions.
107
Simulation settings and parameters selection
Figure 6.1 and Figure 6.2 illustrate the correlation between various objective
functions. Figure 6.1 shows the group of solutions selected to be analysed later.
In Figure 6.2, it can be seen that CO2 emission is correlated to z1 and a bit to
Gj% and Dist, however it is inversely correlated to IRP, Gch and z2. There is no
real correlation between CO2 and z3. Figure 6.2 also shows clearly, the evolution
of the group of solution namely Set A, Set B and Set C across various objective
functions. They are in general grouped in the same search area. All red dots
group have the lowest IRP, but the highest CO2 alongside Gj% and z1. All green
dots group are in most cases in the middle area of various objectives, whilst the
magenta dots group are in the lower bottom. Consequently they have the lowest
CO2 emissions and higher IRP alongside Gch and z2.
Both a correlation analysis and visual inspection of the solutions distribu-
tion were used to evaluate the relationship between the eight possible objectives
considered in this work. The subset of the non-dominated solutions was selected
from the solutions obtained using the MOGA described in Section 7.6.5. The two
most important criteria, namely CO2 and IRP were used to select and classify
the solutions into three groups denoted by A, B and C (see Figures 6.1 and 6.2).
Note that the worse solution in terms of CO2 of group C, could potentially be-
long to group B when considering the gear change frequency (Gch), and the time
spent on higher gear (Gj%). Figure 6.1 illustrates the outcome of the correlation
analysis on a small number of sample solutions, compared to the number used
for the correlation study presented in Table 6.1. It can be observed that JCO2
is correlated with Jz1 , JDist and time spent on higher gear (JGj% ). By contrast
JCO2 and JIRP are non-correlated and can therefore not be met simultaneously.
JIRP is correlated with JGch , and to some extend with Jz2 and Jz3 .
108
Simulation settings and parameters selection
1.005
0.995 A1 A3
A2
A4
0.99 B1
B2
B3
JCO2
0.985
B4
0.98 C1
C2
0.975
C3
0.97
C4
0.965
1 1.05 1.1 1.15 1.2 1.25
J
IRP
Figure 6.1: Pareto plot representing competing criteria CO2 emissions versus
Inverse Reserve Power (IRP). Three sets of solutions are defined, where each set
is comprised of four optimised shift maps. The first set is marked from A1 to A4,
the second set is marked from B1 to B4, and finally the third is marked from C1
to C4
109
Simulation settings and parameters selection
1 1 1 1
JCO2
CO2
0.99
J
0.97 0.97 0.97
JZ1 J
GC
0.98 1 1 1
JCO2
JCO2
0.98 0.98 0.98
J
0.97 Set A
Set B 0.97 0.97 0.97
Set C
Figure 6.2: Pareto plot representing competing criteria CO2 emissions versus
Inverse Reserve Power (IRP), Distance (Dist), zone 1 (z1), Gear change frequency
(Gch), zone 2 (z2), time spent on higher (Gj) and zone 3 (z3). Three sets of
solutions are defined, where each set is comprised of four optimised shift maps.
The first set is marked red circles, the second set is marked green circles, and
finally the third set is marked magenta circles
110
Simulation settings and parameters selection
Table 6.1: Cross correlation results between different objective functions indi-
cating high correlation between JCO2 and BSFC map distance Dist objectives
and high non correlation between the group CO2 Dist and the group IRP Gch
confirming the results of Figure 6.2
JCO2 Jdist Jz1 JIRP JGch Jz2 JGj % Jz3
JCO2 100 62 89 -99 -95 -76 93 -2
Jdist 62 100 60 -58 -48 -30 51 -35
Jz1 89 60 100 -89 -80 -77 81 -12
JIRP -99 -58 -89 100 97 81 -97 -6
JGch -95 -48 -80 97 100 80 -99 -19
Jz2 -76 -30 -77 81 80 100 -78 -54
JGj% 93 51 81 -97 -99 -78 100 15
Jz3 -2 -35 -12 -6 -19 -54 15 100
Having identified the relative trade off required to be addressed, the next
section aims to develop objectives that could capture these trade-offs.
The objectives relating to the zones were designed to attempt to find the most
desirable trade off solution by rewarding gear shift map that results in a good
percentage of the engine operating points (EOP) in the most efficient region of
the BSFC map, i.e. zones 1 and 2. Zone 3 reflects higher fuel consumption
characterised by operating the engine at low or very high revolution per minute.
Ideally a higher degree of correlation between some of the zones objective and both
IRP and CO2 would be desirable to identify a criteria able to capture both these
conflicting requirements. Alternatively the determination of the fuel consumption
thresholds should help engineers identify the difference between solutions more
clearly.
To address these objectives, three different settings were investigated empir-
ically. Table 6.2 presents three different settings applied to the zones objective
function Jz1 , Jz2 and Jz3 . Each setting identifies three zones based on three user
111
Simulation settings and parameters selection
selectable fuel consumption thresholds [g/kW]. To identify the most suitable set
of thresholds the following investigations were performed:
The setting on different zones is to guide the optimiser to favour a shift map
with the most desirable EOP (zones 1 and 2). The largest difference between the
EOP visualisation on the BSFC maps (see Figures 6.3 and 6.4) can be observed
for the setting 2. Similarly the highest difference in terms of cost values is for
setting 2 (see Tables 6.3 and 6.4) with setting 2 clearly differentiating solutions,
increasing the number of operating points in the most fuel efficient zone: +3
points for zone 1 and +2 points for zone 2.
Table 6.2: Threshold adopted to differentiate the different zones on the BSFC
map
Setting 1 (g/kW ) Setting 2 (g/kW ) Setting 3 (g/kW )
Zone 1 < 270 < 255 < 252
Zone 2 < 350 < 265 < 268
Zone 3 ≥ 350 ≥ 265 ≥ 268
Figures 6.3 and 6.4 are the EOP of initial shift map and optimised shift map
respectively, with three different settings based on Table 6.2. The setting on
different zones is to guide the optimiser to favour a shift map with the ideal EOP
(zone 1). From Setting 1 to 3, it can be noticed that zones 1 and 2 are gradually
decreasing (similar remark can be made from Table 6.3 and 6.4 on the percentage
of EOP spent on each zones). The most suitable compromise is represented by
112
Simulation settings and parameters selection
Figure 6.3: EOP of initial gear shift map, where the first is based on setting 1, the
second is based on setting 2 and the third is based on setting 3. The red, magenta
and blue circles represent the EOP of zone 1, zone 2 and zone 3, respectively.
Table 6.3: Percentage of EOP in zone 1, zone 2 and zone 3 for the initial shift
map over the NEDC
Setting 1 Setting 2 Setting 3
Zone 1 EOP (%) 33 10 4
Zone 2 EOP (%) 15 22 29
Zone 3 EOP (%) 52 67 67
To evaluate the effect of the zone thresholds selection on the correlation with
the other objectives a correlation analysis was performed. The cross-correlation
results are based on a set of 26 different shift maps obtained from a Pareto
plot of CO2 versus IRP. Figure 6.5 represents the correlation plot for CO2 , IRP
and gear change frequency against the three zones Z1, Z2 and Z3. It can be
observed that solutions are grouped in clusters for all three settings. This can
help categorises the type of solutions produced. The difference between the groups
is more significant in zone 1 for the setting 1 whereas it is more significant in
113
Simulation settings and parameters selection
Figure 6.4: EOP of optimised gear shift map, where the first is based on setting
1, the second is based on setting 2 and the third is based on setting 3. The
red, magenta and blue circles present the EOP of zone 1, zone 2 and zone 3,
respectively.
Table 6.4: Percentage of EOP in zone 1, zone 2 and zone 3 for an optimised shift
map over the NEDC
Setting 1 Setting 2 Setting 3
Zone 1 EOP (%) 34 13 4
Zone 2 EOP (%) 14 20 30
Zone 3 EOP (%) 52 67 66
zones 2 and 3 for the other two settings. Using setting 2, Zone 2 is proportional
to CO2 and inversely proportional to IRP and Gch. The degree of correlation
corresponding negative-correlation between zone 1 with CO2 and IRP respectively
increases in magnitude between setting 1 and setting 2.
The following section will describe the influence of different setting on objec-
tive functions.
This section plots the cross-correlation between the three different settings
(see Table 6.2) on zones (zone 1, zone 2 and zone 3) against CO2 and IRP. The
cross-correlation results are based on a set of 26 different shift maps obtained
114
Simulation settings and parameters selection
Figure 6.5: Objective functions representing CO2 and IRP versus zone 1, zone 2
and zone 3. Marker with blue ’point’, red ’circle’ and magenta ’plus sign’ present
setting 1, setting 2 and setting 3, respectively.
Figure 6.5 represents the correlation plot of CO2 and IRP versus zones. Ma-
jor remarks are based on settings 1 and 3, when most objective functions are
proportional in setting 1, they are inversely proportional in setting 3. Setting 2
represents the most appropriate compromise, therefore considering setting 2, it
can be noticed that zone 2 is proportional to CO2 and inversely proportional to
IRP and Gch, however zone 2 is partially CO2 , IRP and Gch.
Table 6.5 defines the correlation results of the first setting. Zone 1 is 56%
correlated to CO2 , while it is 62% inversely correlated to IRP. Zone 2 is slightly
correlated and inversely correlated to CO2 and IRP.
Table 6.6 defines the correlation results of the second setting. Zone 1 is 86%
115
Simulation settings and parameters selection
Table 6.5: Cross correlation results between different objective functions indicat-
ing high correlation between JCO2 and BSFC map distance Dist objectives and
high non correlation between the group CO2 and Dist and the group IRP and
Gch. This table is based on setting 1.
JCO2 Jdist Jz1 JIRP JGch Jz2 JGj % Jz3
JCO2 100 90 56 -98 -96 -32 51 -56
Jdist 90 100 54 -89 -89 -35 52 -44
Jz1 56 54 100 -62 -43 -91 -32 -16
JIRP -98 -89 -62 100 94 42 -41 42
JGch -96 -89 -43 94 100 21 -64 48
Jz2 -32 -35 -91 42 21 100 53 -22
JGj% 51 52 -32 -41 -64 53 100 -51
Jz3 -56 -44 -16 42 48 -22 -51 100
116
Simulation settings and parameters selection
Table 6.6: Cross correlation results between different objective functions indicat-
ing high correlation between JCO2 and BSFC map distance Dist objectives and
high non correlation between the group CO2 and Dist and the group IRP and
Gch. This table is based on setting 2.
JCO2 Jdist Jz1 JIRP JGch Jz2 JGj % Jz3
JCO2 100 90 86 -98 -96 36 51 -59
Jdis 90 100 84 -89 -89 35 52 -56
Jz1 86 84 100 -80 -82 15 62 -44
JIRP -98 -89 -80 100 94 -45 -41 64
JGch -96 -89 -82 94 100 -22 -64 45
Jz2 36 35 15 -45 -22 100 -52 -95
JGj% 51 52 62 -41 -64 -52 100 30
Jz3 -59 -56 -44 64 45 -95 30 100
Table 6.7: Cross correlation results between different objective functions indicat-
ing high correlation between JCO2 and BSFC map distance Dist objectives and
high non correlation between the group CO2 and Dist and the group IRP and
Gch. This table is based on setting 3.
JCO2 Jdist Jz1 JIRP JGch Jz2 JGj % Jz3
JCO2 100 90 -56 -98 -96 59 51 -58
JDist 90 100 -56 -89 -89 57 52 -56
Jz1 -56 -56 100 62 55 -90 15 87
JIRP -98 -89 62 100 94 -65 -41 64
JGch -96 -89 55 94 100 -47 -64 45
Jz2 59 57 -90 -65 -47 100 -29 -100
JGj% 51 52 15 -41 -64 -29 100 30
Jz3 -58 -56 87 64 45 -100 30 100
117
Simulation settings and parameters selection
To minimise the number of objective functions and reduce the solution space, the
objectives that were correlated, according to cross-correlation of Table 6.6, were
grouped together. The first group comprises JCO2 , JDist and Jz1 and focuses on
emission reduction. The second group comprises JIRP , JGch and Jz3 and focuses
on driveability at the expense of emission. The last group 3 is durability and
currently only considers JGj based on the assumption that increasing the gear
box use through fast and frequent gear changes would reduce its life expectancy.
The groups are implemented through the use of a group weighting combined
with individual objective weightings within each group. The cost being nor-
malised against the costs obtained for the current gear shift map, the group
weightings are normalised such that their sum equates unity.
The selection of the individual objective weightings is a complex problem
in itself. The main objective being to reduce CO2 emissions, the weightings
associated with emissions should therefore be comparatively high. To ensure
that the vehicle is enjoyable to drive care should also be taken to ensure that the
ability of the vehicle to accelerate after a gear change is not reduced excessively.
Based on these general requirements and with the support of the correlation study
presented in the previous section, the objectives were ranked empirically in terms
of preference (see Table 6.8). The final weighting used in the optimisation is
then the ratio between each individual objective weighting and the sum of all the
objective weightings.
A MOGA was then used to further refine the determination of the most ap-
propriate objective weighting. Having obtained a set of previously optimised
solutions, six of these solutions were examined and ranked empirically from the
best to the worse. The MATLAB GA toolbox was then configured to optimise
the weight for each group of objective to replicate the proposed ranking. The
118
Simulation settings and parameters selection
weightings were initially selected from Table 6.8. Due to time constraints, only a
limited amount of simulation could be carried out. It was found that the weight-
ings determined by the GA could only replicate the ranking of 4 out of the 6
solutions selected.
Due to the difficulty in selecting the correct combination of objectives to merge
all the conflicting requirement into a single expression, this work favours the use
of a modified Pareto ranking. It ranks non dominated solutions according to a
weighted sum based on the weightings identified in Table 6.8. Such an approach
enables to overcome erroneous choice of objective weightings whilst concentrating
on the most promising regions within the solution space.
The next stage in the optimisation procedure is to select the actual ideal
solution. Such a choice is challenging as a single car can have many different
gear shift maps. These maps are selected based on the user requirement, sport
or eco driving, as well as the vehicle environment. The next section aims to
identify qualitative features to augment the information given by the objective
weightings or if possible identify the objective values that correspond to the most
appropriate solutions.
In order to minimise the number of objective functions and ease the multi-
objective optimiser to select a solution based on Pareto optimal point, a group of
objective functions were defined based on cross-correlation results. According to
cross-correlation of Table 6.5, the group of objective function is defined as follows:
Objective group 1 is emissions related, where JCO2 is combined with Jz1 and
Jdis as their correlation values are 90% and 86% respectively. Objective group 2 is
119
Simulation settings and parameters selection
driveability related, where JIRP is combined with JGch and Jz3 as their correlation
values are 94% and 64% respectively. The last objective group 3 is durability
related and only considers JGj% .
A weighted sum method is applied to individual objective function to form
one scalar as group. The group formulation is given as follows:
where WGSM1 , WGSM2 and WGSM3 denote the weighted associated with individual
objective JCO2 , Jdis and Jz1 respectively. JCO2 (x0 ), JDist (x0 ) and Jz1 (x0 ) are
the corresponding objective function with the initial solution, as this allows to
normalise the objective function. The solutions selected by the optimiser will
strongly depend on the weighting factors, and these weights must be positive,
and satisfying:
φ=1
X
Wφ = 1, Wφ ∈ (0, 1) (6.2)
L
where Wφ and L denote the weighting ratio and the maximum number of objective
function, respectively. Consequently, each individual weighting ratio of group 1
is given as follows:
WGSMd
WGr1d = Pd=1 (6.3)
d=3 WGSMd
120
Simulation settings and parameters selection
where WGSM4 , WGSM5 and WGSM6 denote the weighted associated with individual
objective JIRP , JGch and Jz3 respectively. JIRP (x0 ), JGch (x0 ) and Jz3 (x0 ) are
the corresponding objective function with the initial solution, as this allows to
normalise the objective function. Each individual weighting ratio follows the
same pattern as defined by Equation (6.3).
After defining three different groups as reduced objective functions, a combine
multi-group objective function is proposed as follows:
Wg
WGlobale = Pe=1 e (6.6)
e=3 Wge
where the subscript e denotes each individual group objective function. The
weighting in percentage is given in Table 6.8.
121
Table 6.9: Objective functions results of selected solution for optimised weight
Solution JCO2 Jdis Jz1 JIRP JGch Jz3 JGj% Jz2 ObjF RankDes
1 0.9780 0.7696 0.8148 1.1490 1.2667 1.0063 0.8287 1.0934 1.0230 1
2 0.9872 0.8071 0.8641 1.0930 1.1333 1.0179 0.8847 1.0194 1.0088 6
3 0.9903 0.9596 0.8932 1.0769 1.1333 1.0051 0.8924 1.0414 1.0132 5
4 0.9951 1.0501 0.8983 1.0331 1.0000 1.0089 0.9785 1.0260 1.0056 4
5 0.9962 0.9104 0.9403 1.0196 1.0000 1.0114 0.9867 0.9950 0.9979 3
6 0.9931 1.0855 0.8969 1.0382 1.0000 1.0205 0.9826 0.9920 1.0071 2
GA was run 10 times, with a crossover rate of 0.7. The weight of each objective
was selected from Table 6.8 initially, then was randomly changed at each run,
therefore the weighted coefficient was always kept higher as it is the main objective
targeted. GA has managed to find a coefficient of each weight, however only 4
solutions out of 6 were classified. The coefficients of the group weights are given
in Table 6.10:
Table 6.10: Optimised weighting coefficient for Group1 , Group2 and Group3
Group1 Group2 Group3
W eight 6400 2300 440
To identify features of the solutions found by optimising the gear shift map from
a reference gear shift, use is made of solutions belonging to the Pareto optimal
set illustrated in Figure 6.1. Three groups (denoted by A, B and C) have been
identified based on normalised objective values for both IRP and CO2 .
Group A represents solutions that are most suited for sport mode, providing
greater ability to accelerate after a gear change (see Figures 6.6, 6.7, 6.8 and 6.9).
Group B (B1, B2, B3 and B4) represents solutions close to the reference gear
shift map, against which all the costs are normalised, with slightly lower CO2
emissions and similar IRP (see Figures 6.10, 6.11, 6.12 and 6.13).
Group C (C1, C2, C3 and C4) represents solutions that significantly improve
CO2 emissions (see Figures 6.14, 6.15, 6.16 and 6.17).
The features exhibited by these solutions are qualitatively assessed against the
shape of the gear shift maps (see Figures 6.6, 6.10 and 6.14), the engine operating
point (EOP) (see Figures 6.7, 6.11 and 6.15), the gear changes against time (see
Figures 6.8, 6.12 and 6.16), and radar plots of their objective function to visually
assess the relative distribution of the objectives (see Figures 6.9, 6.13 and 6.17).
123
Simulation settings and parameters selection
124
Simulation settings and parameters selection
Norm. CO2: 0.996 [−], Norm. IRP: 1.02 [−] Norm. CO2: 0.994 [−], Norm. IRP: 1.03 [−]
100 100
Throttle Position [%]
60 60
A1 A2
40 40
Initial Upshift Initial Upshift
Initial Downshift Initial Downshift
20 20
Optimised Upshift Optimised Upshift
Optimised Downshift Optimised Downshift
0 0
0 100 200 300 0 100 200 300
Vehicle Speed [km/h] Vehicle Speed [km/h]
Norm. CO2: 0.993 [−], Norm. IRP: 1.03 [−] Norm. CO2: 0.993 [−], Norm. IRP: 1.04 [−]
100 100
Throttle Position [%]
80 80
60 60
40 A3 40 A4
Initial Upshift Initial Upshift
Initial Downshift Initial Downshift
20 Optimised Upshift 20 Optimised Upshift
Optimised Downshift Optimised Downshift
0 0
0 100 200 300 0 100 200 300
Vehicle Speed [km/h] Vehicle Speed [km/h]
Figure 6.6: Set of Optimised shift map obtained from Pareto optimal solution
denoted by A1, A2, A3, and A4, respectively (see Figure 6.1)
125
Simulation settings and parameters selection
Torque (Nm)
[150 KW]
Torque (Nm)
Torque (Nm)
[150 KW]
250 300 250 [140 KW] 300
[130 KW]
[120 KW]
200 200 [110 KW]
250 [100 KW] 250
[90 KW]
150 150 [80 KW]
200 [70 KW] 200
[60 KW]
100 100 [50 KW]
[40 KW]
150 [30 KW] 150
50 50 [20 KW]
[10 KW]
100 100
1000 1500 2000 2500 1000 1500 2000 2500
Speed (rpm) Speed (rpm)
Figure 6.7: Set of Optimised shift map obtained from Pareto optimal solution
denoted by A1, A2, A3, and A4, respectively (see Figure 6.1). The NEDC was
used to calculate the engine operating point (EOP). The EOP of standard shift
map is presented with red mark (+), and optimised shift map presented with
blue mark (o).
126
Simulation settings and parameters selection
6 6
Gear change [km/h]
4 4
3 3
2 2
1 1
0 200 400 600 800 1000 1200 0 200 400 600 800 1000 1200
Time (sec) Time (sec)
Initial gear change
Optimised gear change
6 6
Gear change [km/h]
5 5
4 4
3 3
2 2
1 1
0 200 400 600 800 1000 1200 0 200 400 600 800 1000 1200
Time (sec) Time (sec)
Figure 6.8: Gear change position based on 4 optimised shift maps (A1, A2, A3
and A4 (see Figure 6.1)) compared with initial shift map.
127
Simulation settings and parameters selection
0.6
−0.4
−0.6
−0.8
JIRP = 1.00 ... 1.04 JDist = 0.89 ... 1.09
−1
−1 −0.5 0
JZ1 = 0.90 ...0.51.00 1
Figure 6.9: Radar plot of objectives functions (CO2 emissions, IRP, Distances,
Gear change frequency, time spent on higher gear, different zone on BSFC map
defined as Zone 1, Zone 2, Zone 3) based on 4 optimised shift maps (A1, A2, A3
and A4 (see Figure 6.1)) compared with initial shift map.
128
Simulation settings and parameters selection
Norm. CO2: 0.991 [−], Norm. IRP: 1.07 [−] Norm. CO2: 0.989 [−], Norm. IRP: 1.08 [−]
100 100
60 60
40 B1 40 B2
Initial Upshift Initial Upshift
Initial Downshift Initial Downshift
20 20
Optimised Upshift Optimised Upshift
Optimised Downshift Optimised Downshift
0 0
0 100 200 300 0 100 200 300
Vehicle Speed [km/h] Vehicle Speed [km/h]
Norm. CO2: 0.987 [−], Norm. IRP: 1.09 [−] Norm. CO2: 0.986 [−], Norm. IRP: 1.1 [−]
100 100
Throttle Position [%]
60 60
40 B3 40 B4
Initial Upshift Initial Upshift
Initial Downshift Initial Downshift
20 20 Optimised Upshift
Optimised Upshift
Optimised Downshift Optimised Downshift
0 0
0 100 200 300 0 100 200 300
Vehicle Speed [km/h] Vehicle Speed [km/h]
Figure 6.10: Set of Optimised shift map obtained from Pareto optimal solution
denoted by B1, B2, B3, and B4, respectively (see Figure 6.1)
Figure6.10, 6.11 and 6.13 illustrate the plot of the second optimal set (B1,
B2, B3 and B4 (see Figure 6.1)) shift maps, engine operating point (EOP) and
radar plot of their objective function, respectively.
129
Simulation settings and parameters selection
Torque (Nm)
[150 KW]
250 300 250 [140 KW] 300
[130 KW]
[120 KW]
200 200 [110 KW]
250 [100 KW] 250
[90 KW]
150 150 [80 KW]
200 [70 KW] 200
[60 KW]
100 100 [50 KW]
[40 KW]
150 [30 KW] 150
50 50 [20 KW]
[10 KW]
100Standard shift map
EOP 100
1000 1500 2000 2500 1000 1500 2000 2500
EOP Optimised shift map
Speed (rpm) Speed (rpm)
Constant power
EOP of optimised Sh. Map EOP of optimised Sh. Map
350 400 350 400
[150 KW]
Torque (Nm)
Figure 6.11: Set of Optimised shift map obtained from Pareto optimal solution
denoted by B1, B2, B3, and B4 respectively (see Figure 6.1). The NEDC was
used to calculate the engine operating point (EOP). The EOP of standard shift
map is presented with red marked (+), and optimised shift map presented with
blue marked (o).
130
Simulation settings and parameters selection
6 6
Gear change [km/h]
4 4
3 3
2 2
1 1
0 200 400 600 800 1000 1200 0 200 400 600 800 1000 1200
Time (sec) Initial gear change
Time (sec)
Optimised gear change
6 6
Gear change [km/h]
5 5
4 4
3 3
2 2
1 1
0 200 400 600 800 1000 1200 0 200 400 600 800 1000 1200
Time (sec) Time (sec)
Figure 6.12: Gear change position based on 4 optimised shift maps (B1, B2, B3
and B4 (see Figure 6.1)) compared with initial shift map.
131
Simulation settings and parameters selection
−0.2
−0.4
−0.6
−0.8
JDist = 0.70 ... 1.07
JIRP = 1.00 ... 1.10
−1
−1 −0.5 0 0.5 1
JZ1 = 0.83 ... 1.00
Figure 6.13: Radar plot of objectives functions (CO2 emissions, IRP, Distances,
Gear change frequency, time spent on higher gear, different zone on BSFC map
defined as Zone 1, Zone 2, Zone 3) based on 4 optimised shift maps (B1, B2, B3
and B4 (see Figure 6.1)) compared with initial shift map.
132
Simulation settings and parameters selection
Norm. CO2: 0.981 [−], Norm. IRP: 1.11 [−] Norm. CO2: 0.978 [−], Norm. IRP: 1.13 [−]
100 100
60 60
40 C1 40 C2
Initial Upshift Initial Upshift
Initial Downshift Initial Downshift
20 Optimised Upshift 20
Optimised Upshift
Optimised Downshift Optimised Downshift
0 0
0 100 200 300 0 100 200 300
Vehicle Speed [km/h] Vehicle Speed [km/h]
Norm. CO2: 0.975 [−], Norm. IRP: 1.17 [−] Norm. CO2: 0.965 [−], Norm. IRP: 1.22 [−]
100 100
Throttle Position [%]
60 60
40 C3 40 C4
Initial Upshift Initial Upshift
Initial Downshift Initial Downshift
20 Optimised Upshift 20 Optimised Upshift
Optimised Downshift Optimised Downshift
0 0
0 100 200 300 0 100 200 300
Vehicle Speed [km/h] Vehicle Speed [km/h]
Figure 6.14: Set of Optimised shift map obtained from Pareto optimal solution
denoted by C1, C2, C3, and C4 respectively (see Figure 6.1)
133
Simulation settings and parameters selection
Torque (Nm)
[150 KW]
250 300 250 [140 KW] 300
[130 KW]
[120 KW]
200 200 [110 KW]
250 [100 KW] 250
[90 KW]
150 150 [80 KW]
200 [70 KW] 200
[60 KW]
100 100 [50 KW]
[40 KW]
150 [30 KW] 150
50 50 [20 KW]
[10 KW]
100 Standard shift map
EOP 100
1000 1500 2000 2500 1000 1500 2000 2500
Speed (rpm) EOP Optimised shift map Speed (rpm)
Constant power
EOP of optimised Sh. Map EOP of optimised Sh. Map
350 400 350 400
Torque (Nm)
[150 KW]
250 300 250 [140 KW] 300
[130 KW]
[120 KW]
200 200 [110 KW]
250 [100 KW] 250
[90 KW]
150 150 [80 KW]
200 [70 KW] 200
[60 KW]
100 100 [50 KW]
[40 KW]
150 [30 KW] 150
50 50 [20 KW]
[10 KW]
100 100
1000 1500 2000 2500 1000 1500 2000 2500
Speed (rpm) Speed (rpm)
Figure 6.15: Set of Optimised shift map obtained from Pareto optimal solution
denoted by C1, C2, C3, and C4 respectively (see Figure 6.1). The NEDC was
used to calculate the engine operating point (EOP). The EOP of standard shift
map is presented with red mark (+), and optimised shift map presented with
blue mark (o).
134
Simulation settings and parameters selection
6 6
Gear change [km/h]
4 4
3 3
2 2
1 1
0 200 400 600 800 1000 1200 0 200 400 600 800 1000 1200
Time (sec) Initial gear change
Time (sec)
Optimised gear change
6 6
Gear change [km/h]
5 5
4 4
3 3
2 2
1 1
0 200 400 600 800 1000 1200 0 200 400 600 800 1000 1200
Time (sec) Time (sec)
Figure 6.16: Gear change position based on four optimised shift maps (C1, C2,
C3 and C4 (see Figure 6.1)) compared with initial shift map.
135
Simulation settings and parameters selection
0.6
Init. Shift map
Opt. Shift map 1
0.4 Opt. Shift map 2
Opt. Shift map 3
0.2 Opt. Shift map 4
JGC = 1.00 ... 1.40
0 JCO2 = 0.97 ... 1.00
−0.2
−0.4
−0.6
−0.8
JIRP = 1.00 ... 1.22 JDist = 0.67 ... 1.01
−1
−1 −0.5 0 0.5 1
JZ1 = 0.80 ... 1.00
Figure 6.17: Radar plot of objectives functions (CO2 emissions, IRP, Distances,
Gear change frequency, time spent on higher gear, different zone on BSFC map
defined as Zone 1, Zone 2, Zone 3) based on four optimised shift maps (C1, C2,
C3 and C4 (see Figure 6.1)) compared with initial shift map.
136
Simulation settings and parameters selection
This chapter has identified existing trade-offs in terms of the objectives imple-
mented in this thesis, to evaluate the appropriateness of the optimised solutions
produced. A cross correlation study highlighted the degree of correlation between
the different objectives and to identify the tuning parameters for the new criteria
proposed in this work. The outcome of the correlation study was the division
of the objectives into three groups focusing on emissions, driveability and dura-
bility, respectively. These tuning parameters for the zones 1-3 were selected to
emphasise the difference between solutions that promote an efficient distribution
of the engine operating point on the BSFC map and non optimised solutions. z1
promotes better fuel consumption and is correlated with CO2 , Gj% and Dist.
The second group, which is not correlated with the first group includes IRP, Gch
and z2. To characterise the differences between different solutions, three groups
of optimal solutions were selected based on the trade-off between CO2 and IRP.
It was demonstrated that all optimised gear shift maps were shifted to the left in
comparison to the original shift map. This resulted in an early shifting and led
to reduction in CO2 . The group with the lowest CO2 emissions have very rapid
successive changes for their gear sets 2 and 3, especially at low throttle position.
This study has enabled identification of features on the BSFC map as well as
gear shift maps to help select solutions which are optimal but also likely to be
acceptable in terms of driveability.
137
Chapter 7
Algorithm performances
7.1 Introduction
Chapter 6 identified the type of solutions that can be obtained from focusing on
different objectives as well as features of desirable solutions. Having investigated
the formulation of the objectives in the previous chapters, this chapter focuses on
the algorithms exploiting such objective formulations. It aims to demonstrate,
through simulation studies, the benefits of problem specific features as well as
generic algorithm modifications applicable to other optimisation problems. The
generic algorithms modifications include the new hybrid Multi-Objective Cuckoo
Search (MOCS) which combines Levy Flight function with Firefly, Bat and Flower
Pollination operators. The problem specific developments include the gear shift
map repair mechanism, the new gear early shifting (GES) operator and the overall
optimisation framework for the combined gear shift and gear ratio optimisation.
Each contribution is evaluated independently and then in combination against
two benchmarks algorithms: the Interior-Point Optimisation and MOGA from
MATLAB toolbox. Finally the benefit of the combined MOGA and MOCS to
concurrently optimise gear ratio and gear shift map is compared to independent
gear ration and gear shift map optimisation.
138
Algorithm performances
• the MOGAOp (Mp2 ) which is based on Mp1 but modified to include various
operators from the Cuckoo Search Algorithm (see Section 5.3, in Chapter 5)
The Non Pareto based optimisation (weighted sum) algorithms include Mp1 ,
Mp2 as well as:
139
Algorithm performances
The initial comparison is problem independent and therefore does not include
the repair mechanism and GES. Each set up is summarised in a table giving the
ideal objective function values achieved at the last generation for each setting.
The rate of convergence is illustrated by recording the most suitable objective
function at each generation, where the number of generation of each algorithm is
set to 30 with a population of 60 individuals. The maximum functions evaluated
for interior-point algorithm is set to 2000. Every algorithm used the same initial
condition. Pareto based optimisation algorithms use identical initial population
whilst weighted sum approaches use a valid solutions of average quality.
mance
This section describes the criteria used in this chapter to evaluate the benefit of
the problem specific objectives formulation as well as the proposed algorithms.
The objective formulation is evaluated by performing a correlation analysis with
existing objectives and observing from a qualitative perspective (e.g. shape of
the gear shift map, type of solutions produced) and a quantitative perspective
(CO2 emissions, time spent on higher gears and ability of the vehicle to accelerate
after a gear change, expressed as inverse reserve power) the differences between
the optimised solutions. The effectiveness of the algorithms is evaluated using:
140
Algorithm performances
v
u n
u 1 X 2
Ssprd =t d − di (7.1)
n − 1 i=1
where di =minj ( f1i (~x) − f1j (~x) + f2i (~x) − f2j (~x) ), i,j=1,...n, d is the mean
of all di , and n is the number of Pareto optimal set. If Sspread is equal to
zero, it means that all members of the Pareto optimal set are equidistantly
spaced.
One of the major contributions of this work is the solution repair mechanism. It
has been designed such that it can be applied to an existing gear shift map or
a new solution generated by the optimisation algorithm. It is a convenient tool
to rescue solutions, with significant CO2 saving potential, that would otherwise
have been rejected.
To illustrate the benefit of the approach, the vehicle powertrain behaviour
is simulated for a previously optimised gear shift with low CO2 emission. The
gear position, the engine operating point and the reserve power are then plotted
against time at which the reserve power is below the user defined limit set at 3.2
kW Ngo, Colin Navarrete, Hofman, Steinbuch & Serrarens (2013). The corre-
sponding reserve power and throttle position are identified by green circles and
triangles on Figures 7.1, 7.2 and 7.3, respectively.
141
Algorithm performances
Plotting the engine operating point on top of the gear shift map, see Figure 7.1,
clearly indicates that the lower limit constraints are only infringed for a small
region in the gear shift map. Applying the repair mechanism according to the
method described in Section 5.2.2, removes the issue associated with the Upshift
to 2nd gear by increasing the velocity at which the Upshift occurs for low throttle
angles. The ability of the repair mechanism to adjust only a few points in the gear
shift map is believed to be very valuable. The same principle could be applied
to other criteria and constraints that should be met. The effect of the repaired
gear shift map on the vehicle performance assessed against the NEDC was then
re-simulated.
100
90
80
Throttle Position [%]
70
60
50
Initial map
Optimised map
40 Gear 1
Gear 2
30 Gear 3
Gear 4
Gear 5
20 Gear 6
Gear under limit
10
0
0 50 100 150 200 250
Vehicle Speed [km/h]
Figure 7.1: Optimised gear shift with reserve power under 3.2 kW
142
Algorithm performances
Reserve power
10 Reserve power limit 100
Optimised gear change[−] Reserve power under limit
Vehicle speed
Gear 1
Gear 2
Gear 3
Reserve power[kW]
Gear 4
Gear change[−]
Gear 5
Gear 6
Gear under limit
5 50
0 0
0 20 40 60 80 100 120 140 160 180 200
Time (Sec)
Figure 7.2: Optimised gear shift results over the NEDC with reserve power under
3.2 kW for one period of urban driving cycle
Reserve power
10 100
Optimised gear change[−] Reserve power limit
Reserve power under limit
Vehicle speed
Gear 1
Gear 2
Gear 3
Reserve power[kW]
Gear 4
Gear change[−]
Gear 5
Gear 6
Gear under limit
5 50
0 0
800 850 900 950 1000 1050 1100 1150 1200
Time (Sec)
Figure 7.3: Optimised gear shift results over the NEDC with reserve power under
3.2 kW for extra-urban driving cycle
143
Algorithm performances
It is clear that, following the application of the gear shift repair mechanism,
all the occurrences of reserve power below the limit have been removed. The
repaired gear shift map illustrated in Figure 7.4 is only modified for the 2nd gear
Upshift. This small modification is able to remove all the occurrences of the
reserve power that were under the limit over the whole NEDC, see Figure 7.5
and 7.6. Table 7.1 illustrates the benefit and consequences of applying the repair
mechanism to the gear shift map (GSM) illustrated in Figure 7.1 to produce the
repaired gear shift map, GSMr , as illustrated in Figure 7.4.
The application of the repair mechanism results in a slight increase in CO2
emissions. This increase is unavoidable and results in the necessity to meet the
minimum requirement in terms of IRP. Note that similar approaches could be used
to design gear shift map for sport mode, the only difference being an increase in
the minimum reserve power.
100
90
80
Throttle Position [%]
70
60
50
40 Initial map
Optimised map
30 Gear 1
Gear 2
Gear 3
20
Gear 4
Gear 5
10 Gear 6
0
0 50 100 150 200 250
Vehicle Speed [km/h]
Figure 7.4: Optimised gear shift map repaired to enforce minimum reserve power
above 3.2 kW
144
Algorithm performances
10 100
Optimised gear change[−] Reserve power
Reserve power limit
Vehicle speed
Gear 1
Gear 2
Gear 3
Reserve power[kW]
Gear 4
Gear change[−]
Gear 5
Gear 6
5 50
0 0
0 20 40 60 80 100 120 140 160 180 200
Time (Sec)
Figure 7.5: Simulation with a repaired gear shift map showing that the minimum
reserve power is met at all time over the whole NEDC. This illustration shows
one period of urban driving cycle
Table 7.1: Optimised shift map results. GSM is the optimised shift map with
reserve power under the limit. GSMR is the optimised shift map after applying
repair mechanism
Solution JCO2 JDist Jz1 JIRP JGch Jz3 JGj % Jz2
GSM 0.9773 0.7944 0.8076 1.1541 1.2667 1.0060 0.8267 1.1005
GSMR 0.9800 0.8031 0.8173 1.1425 1.2667 0.9947 0.8285 1.1348
145
Algorithm performances
Reserve power
10 Reserve power limit 100
Optimised gear change[−]
Vehicle speed
Gear 1
Gear 2
Gear 3
Gear 4
Reserve power[kW]
Gear 5
Gear change[−]
Gear 6
5 50
0 0
800 850 900 950 1000 1050 1100 1150 1200
Time (Sec)
Figure 7.6: Simulation with a repaired gear shift map showing that the minimum
reserve power is met at all time over the whole NEDC. This illustration shows
one period of extra-urban driving cycle
146
Algorithm performances
Another notable contribution in this thesis is the problem specific GES operator
(see Section 5.2.1, in Chapter 5) designed to generate gear shift map with reduced
CO2 emissions. Table 7.2 confirms that the application of the GES operator gives
rise to three solutions with improved CO2 emissions, at the cost, however, of worse
IRP.
Figure 7.7: Illustration of GES. The full red, blue and green lines denote the
three solutions produced by the GES
Table 7.2: Optimised shift map results. GSMGESinit is the optimised shift map
with . GSMGES25%, 50% and 75% being the optimised shift map after applying the
repair mechanism
Solution JCO2 JDist J z1 JIRP JGch Jz3 JGj % J z2
GSMGESinit 0.9964 1.0137 0.9697 1.0198 1.0000 1.0089 0.9907 0.9879
GSMGES25% 0.9927 0.7558 0.9150 1.0421 1.0000 1.0198 0.9769 0.9842
GSMGES50% 0.9938 0.8344 0.9283 1.0349 1.0000 1.0169 0.9813 0.9854
GSMGES75% 0.9949 0.9216 0.9469 1.0273 1.0000 1.0153 0.9860 0.9805
The benefit of applying GES has improved the current solution generated by
the optimiser by up to 0.37%.
147
Algorithm performances
This section describes the comparison between the algorithms developed in this
thesis and benchmarks from the MATLAB toolbox. Following the evaluation of
the proposed modifications to generic MOGA and MOCS algorithms, the overall
scheme is evaluated in combination with problem specific features developed in
this work, namely the repair mechanism and GES.
The overall outcome of the simulation studies is presented in Table 7.3. Ta-
bles D.1, . . . , D.21 focusing on each feature evaluation are presented in Appendix
D. Each table contains the leading objective value found for each objective con-
sidered. The objectives used within the optimisation algorithms were normalized
against the initial gear shift map provided by the manufacturer. JCO2 is presented
without normalisation in the tables of results to clearly identify the relative merits
of the algorithms investigated.
Table 7.3 identifies the most suitable algorithm, which abbreviations are sum-
marised in Table 7.5, and the settings defined in Section 7.2 and identified using
the colour coding shown in Table 7.4, for both Pareto and weighted sum objective
formulations.
Considering the Pareto case, no repair with GES (green) gives rises to the
leading results in terms of CO2 . This is expected as the application of repair
increases the JCO2 . Mp1 and Mp2 are the most suitable algorithms for JCO2 , JDist ,
JIRP and JGj% when Pareto ranking is used. Note however that GES may lead
to premature convergence and prevent exploration.
Jz3 reaches a minimum value that is similar, irrespective of each algorithm
used. This means that the objective Jz3 cannot be used to differentiate between
alternative solutions. This behaviour can be explained by the drive cycle used to
evaluate alternative solutions, where the vehicle is not operating at high speed
for significant periods of time.
148
Algorithm performances
The ideal algorithm for the weighted sum (WSum ) approach was Mp7 for most
objectives, with the exception of JGch and Jz1 , where Mp6 was better. Mp7 offered
the most suitable performance in terms of JObj , the latter being used to deter-
mine the most suitable solution. Compared to the Pareto based optimisation
approach, it can be seen that the selection of the objective weightings is critical
to the achievable performance against individual objectives. In general Pareto
based optimisation approaches are better at enabling individual objectives to be
minimised.
The performance results are provided in more detail in Appendix D. The ideal
algorithms with different settings are defined in the following Table 7.3:
149
Table 7.3: Most suitable algorithm for each objective function
JCO2 JDist Jz1 JIRP JGch J z3 JGj % J z2 JObj
Set 1 (Par) Mp1 (2) Mp2 (1) Mp4 (2) Mp1 /Mp2 (1) Mp1 (1) All(1) Mp1 (1) Mp2 (1) Mp2 /Mp5 (2)
Set 2 (Par) Mp2 (1) Mp1 (2) Mp2 /Mp4 (3) Mp1 /Mp1 (2) All Mp1 (2) All(1) Mp1 (3) Mp3 (2) All(4)
Set 3 (Par) Mp1 (1) Mp1 (1) Mp4 (1) Mp1 (3) Mp1 (3) All(1) Mp4 (2) Mp1 (3) Mp1 /Mp2 /Mp5 (1)
Set 1 (WSum ) Mp7 (5) Mp7 (3) Mp7 (4) Mp7 (5) Mp6 /Mp6 (4) Mp7 /Mp7 (2) Mp7 (6) Mp6 /Mp6 (4) Mp7 (3)
Set 2 (WSum ) Mp7 (3) Mp7 (4) Mp7 (6) Mp7 (6) Mp6 /Mp6 (4) Mp7 (2) Mp7 (4) Mp6 /Mp6 (4) Mp7 (5)
Set 3 (WSum ) Mp7 (4) Mp7 (5) Mp7 (5) Mp7 (4) Mp6 /Mp6 (4) Mp7 (2) Mp7 (5) Mp6 /Mp6 (4) Mp7 (6)
Table 7.4: Colour coding for various setting in algorithms performance assessment
150
Algorithm performances
1
Applicable only with Pareto based optimisation algorithms.
Algorithm performances
The impact of repair mechanism has limited in some condition the optimiser to
reduce further down JCO2 , as it is designed to improve the driveability, however
will increase JCO2 as illustrates in Table D.10, in Appendix D.
151
Algorithm performances
0.99
0.98
MOGA(set3) [−]
0.96 MOGAOp(set1) [−]
MOGAOp(set2) [−]
MOGAOp(set3) [−]
0.95
0.94
0.93
0.92
0 5 10 15 20 25 30
Generation
Figure 7.8: Convergence of MOGA and MOGAOp based on Pareto under different
objective combinations denoted set 1, set 2, and set 3
Noticed that the application of GES was only implemented MOGA and MOGAOp .
Additionally, as it was described that GES improve the value of JCO2 , however
care need to be taken by applying GES, as it can result to a premature conver-
gence (Pandey et al. 2014). GES was applied in average at every 3 generations,
also at each generation in order to highlight the impact of GES. Noticed that
GESAg (Mp4 ), GESAg/Op (Mp5 ) are referred to as a more aggressive application of
GES, meaning that GES is applied at each generation on MOGA and MOGAOp
respectively.
152
Algorithm performances
Tables D.14, Table D.15 and Table D.16 demonstrate the results obtained by ap-
plying GES. In general the average of JCO2 has significantly reduce in comparison
to non-application of GES, see Table D.17.
MOGA(set1)[−]
0.98
MOGA(set2)[−]
MOGA(set3)[−]
MOGAAg(set1)[−]
0.96
Normilised objective
MOGAAg(set2)[−]
MOGA (set3)[−]
Ag
MOGAOp(set1)[−]
0.94
MOGAOp(set2) [−]
MOGAOp(set3)[−]
0.92 MOGAOp/Ag(set1)[−]
MOGAOp/Ag(set2) [−]
MOGAOp/Ag(set3)[−]
0.9
0.88
0 5 10 15 20 25 30
Generation
Figure 7.9 illustrates the impact of applying GES on the rate of convergence.
As expected, it has contributed a higher, faster rate of convergence of fitness
values in compared to Figure 7.10 and Figure 7.8. Additionally the application
at each generation has also accelerated the rate of convergence.
153
Algorithm performances
The weighted sum method is the simplest technique in optimisation when deal-
ing with more than one objective functions (see Section 4.4.1, in Chapter 4),
therefore it is difficult to reflect the user desire when considering the multiple
154
Algorithm performances
objective function. In this thesis, Pareto outperformed and weighted sum perfor-
mances were compared under 3 settings (see Section 7.2). It can be noticed that
Pareto outperformed the weighted sum in terms of rate of convergence as well as
minimum CO2 emissions (see Figures 7.8, 7.9, 7.10 and 7.11).
Tables D.1, D.2 and D.3 are the results of Pareto-based optimisation for set 1, set
2 and set 3 respectively. Noticed that JCO2 used in the algorithm was normalized
against the initial gear shift map, it is presented without normalisation in the
tables of results. This was adopted to assess more easily the potential for CO2
saving. The MOGAOp gave rise to the ideal JCO2 for set 1 (see Table D.1) but
the worse for set 2 (see Table D.2) whilst the MOGA gave the leading JIRP for
both set 1 and set 2.
Figure 7.10 represents the rate of convergence of different optimisers, and the
correlation between various objective functions respectively. In Figure 7.10, it
can be noticed that the convergence of the algorithm depends on the objective
formulation, i.e. set is not improving, in terms of the most suitable solution,
for many generations, whilst sets 1 and 3 lead to a more regular convergence.
MOGAOp is only able to outperform the benchmark MOGA for set 2.
The small population size and use of modified Pareto ranking leads to gaps
between solutions on the Pareto set. This initial comparison is not able to demon-
strate that it is beneficial to include Cuckoo Search operators within the GA.
However, due to time constraints only a few experiments could be run, making
the result not statistically significant.
155
Algorithm performances
0.99
0.98
MOGA(set1) [−]
0.97 MOGA(set2) [−]
Normilised bjective
MOGA(set3) [−]
MOGAOp(set1) [−]
0.96
MOGAOp(set2) [−]
MOGAOp(set3) [−]
0.95
0.94
0.93
0.92
0 5 10 15 20 25 30
Generation
Figure 7.10: Convergence of MOGA and MOGAOp based on Pareto under differ-
ent objective combinations denoted set 1, set 2, and set 3
156
Algorithm performances
GES
Tables D.18, D.19 and D.20 demonstrate the application of repair mechanism and
GES considering set 1, set 2 and set 3 conditions. The application of GES (see
Table D.21) in this case, improves the results of convergence rate.
0.98
MOGA(set1)[−]
MOGA(set2)[−]
0.96
Normilised objective
MOGA(set3)[−]
MOGAOp(set1)[−]
MOGAOp(set2) [−]
0.94
MOGAOp(set3)[−]
0.92
0.9
0.88
0 5 10 15 20 25 30
Generation
Figure 7.11: Convergence of MOGA and MOGAOp based on Pareto under differ-
ent objective combinations denoted set 1, set 2, and set 3
This section has described the performance of MOGA, and MOGAOp under
different settings with the application of repair mechanism and GES. Additionally,
a comparison was made against optimiser benchmark from MATLAB toolbox,
MOGAOp considering Pareto, interior-point algorithm and GA considering Non-
Pareto. It was highlighted that repair mechanism has improved in average JIRP ,
157
Algorithm performances
158
Algorithm performances
7.7 MOCS
This section describes the MOCS algorithm combined with Levy Flight, Bat,
Firefly and Flower Pollination operators. The goal of this section is to verify
the performance of each operator for optimising intermediate gear ratio. The
comparative assessment of each operator is mainly based on the speed of con-
vergence, which measures the quality of difference solutions. An experiment was
realised with a termination condition of 30 generations to assess the performance
of difference operators as follows: (i) the operator were individually run. (ii) the
operators were randomly selected from generation to generation. (iii) the oper-
ators were selected in an ascending order starting from Levy Flight (1), Firefly
(2), Bat (3) and Flower pollination (4). Three objective functions were consid-
ered in this case, namely: CO2 emissions, IRP and gear ratio bandwidth (see
Section 3.4.5, in Chapter 3). The number of available host nest was fixed to
15. Two sets of weights were considered for this study. Table 7.6 describes the
weighted combination for use with the modified Pareto ranking (see Section 4.4.2,
in Chapter 4). The Weighting set 1, favours the bandwidth, and Weighting set 2
favours CO2 emissions.
Table 7.6: Weighting coefficient for gear ratio optimisation. WGR1 , WGR2 and
WGR3 denote the weighting coefficient of CO2 , IRP and bandwidth respectively
WGR1 WGR2 WGR3
Weighting set 1 100 15 300
Weighting set 2 300 15 100
The weighting selection of set 1 is justified by the fact that CO2 emission is
prioritised.
The results for each benchmark settings, are described in Tables 7.7 and 7.8.
159
Algorithm performances
Table 7.7: Comparative performance of MOCS integrated with Levy Flight, Bat,
Firefly, Flower pollination operators separately
Levy F light F iref ly Bat F lower pollination
f unctions evaluated 1152 648 1063 1070
Ideal solution 0.51 0.497 0.5029 0.496
−7 −8
M ean value of solution 1.25e 0 6.77e 4.76−5
Std of solutions 2.44e−7 0 1.13e−7 4.58−5
Spread 2.39e−7 0 1.11e−7 4.48−5
Table 7.8: Comparative performance of MOCS integrated with Levy Flight, Bat,
Firefly, Flower pollination operators. The first test is selecting different operators
randomly. The second test is selecting gradually different operators
Random selection Order selection
f unctions evaluated 1084 1063
Ideal solution 0.499 0.5006
−4
M ean value of solution 5.63 7.31−7
Std of solutions 1.034−3 9.54−7
Spread 1.01e−3 9.33e−7
160
Algorithm performances
Normilised CO2
1.017 1.017 1.017
Figure 7.12: Trade-off between CO2 against IRP, bandwidth and overall weighted
combination cost based on random selection of operators: Levy Flight, Firefly,
Bat and Flower pollination. The blue, red and green circles denote the solutions
trade-off after 10, 20 and 30 generations, respectively.
161
Algorithm performances
0.93
1.016
0.92
Normilised CO2
1.015
Normilised IRP
1.014 0.91
1.013 0.9
1.012
0.89
1.011
0.88
1.01
0.87
0 10 20 30 0 10 20 30
Generation Generation
1 1
Normilised all cost
0.9
Normilised Bdw
0.9
0.8
0.7 0.8
0.6 0.7
0.5
0.6
0.4
0.5
0 10 20 30 0 10 20 30
Generation Generation
162
Algorithm performances
Figure 7.12 and Figure 7.13 illustrate the intermediate gear ratio optimisation
performance considering randomly selected operators. Figure 7.12 demonstrates
the evolution of various objective trade-off against CO2 emissions. It can be
noticed that the CO2 value is increasing while the IRP, bandwidth and overall
cost are decreasing. Figure 7.13 presents the trade-off between CO2 against IRP,
bandwidth and overall weighted combination cost, where the Pareto optimal sets
are shown by the blue, red and green circles after 10, 20 and 30 generations
respectively.
Similarly, it can be seen from Table 7.7 and Table 7.8, that only random
selection managed to maintain a better diversity at the last generation. Thus, it
can be concluded that a bigger number of generation is not necessary to optimise
the intermediated gear ratio considering the first weighting combination.
The second set of weightings puts more emphasis on the gear ratio bandwidth.
Table 7.9: Comparative performance of MOCS integrated with Levy Flight, Bat,
Firefly, Flower pollination operators separately
Levy F light F iref ly Bat F lower pollination
f unctions evaluated 513 147 444 402
Ideal solution 0.822 0.846 0.822 0.822
d 0.0091 0.0109 0.0101 0.0112
Std 0.029 0.0297 0.032 0.0363
Spread 0.028 0.0285 0.0307 0.0349
163
Algorithm performances
Table 7.10: Comparative performance of MOCS integrated with Levy Flight, Bat,
Firefly, Flower pollination operators. The first test is selecting different operators
randomly. The second test is selecting gradually different operators
Random selection Order selection
f unctions evaluated 423 344
Ideal solution 0.846 0.823
d 0.0107 0.0108
Std 0.0355 0.0369
Spread 0.034 0.0355
2
1.017
Normilised CO2
Normilised IRP
1.016 1.5
1.015 1
1.014 0.5
1.013
0
1.012
−0.5
0 10 20 30 0 10 20 30
Generation Generation
Normilised all cost
0.7
Normilised Bdw
0.84
0.65
0.83
0.6
0.55 0.82
0.5
0.81
0.45
0 10 20 30 0 10 20 30
Generation Generation
Figure 7.14 and Figure 7.15 illustrate the intermediate gear ratio optimisation
performance considering Firefly of operators. In most case the IRP stays constant,
and bandwidth naturally decreasing. However, the overall cost increases from
generation to generation. According to Figure 7.15, the spread based on the last
generation is well distributed in comparison to most cases noticed in Section 7.7.1.
164
Algorithm performances
Normilised CO2
Normilised CO2
1.02 1.02 1.02
Figure 7.15: Trade-off with Firefly between CO2 against IRP, bandwidth and
overall weighted combination cost. The blue, red and green circles denote for
solutions trade-off after 10, 20 and 30 generations respectively
tion
The realisation of combined gear ratio and gear shift map is based on the algo-
rithm described in Figure 5.10 (see Section 5.4.1, in Chapter 5). MOGA, MOCS
and interior-point algorithm were combined to obtain a set of gear ratio and op-
timised gear shift map, as illustrated in Figure 7.16. The set of gear shift and
gear ratio combinations contains 15 nests. Each nest is represented by a set of 6
speed gear ratios (G1 ,G2 ,G3 ,G4 ,G5 and G6 ). Each nest is then defined with its
initial shift map, and 40 optimised shift maps.
For this study, 10 generations were considered for MOCS, and 10 generations
for MOGA. Notice that MOCS was setup with random selection of operator at
165
Algorithm performances
Noticed that within a combined set, not all 15 nest are found to be valid due
to constraints. In fact, combSet1 has 15 valid nests, combSet2 has 9 nests valid
nests and combSet3 has also 9 valid nests.
Table 7.11: Performance of MOCS integrated with Levy Flight, Bat, Firefly,
Flower pollination operators. A random selection of operators is considered in
this case. combSet1 , combSet2 , combSet3 denote the sets of combination 1, 2 and
3 respectively
combSet1 combSet2 combSet3
f unctions evaluated 180 436 485
Ideal solution 0.7392 0.6994 0.968
d value of solution 0.0101 0.0023 0.0061
Std of solutions 0.0199 0.0046 0.0119
Spread 0.019 0.0044 0.011
Table 7.11 describes the performance of the intermediate gear ratio optimisa-
tion based on the combined gear ratio and gear shift map optimisation. It can
be seen that the number of function evaluation varies significantly between the
first and the last two set of combinations. This can be justified by the complexity
of this particular gear set, as the first and last gear ratios are different, meaning
166
Algorithm performances
that constraints become more stringent, which can result in a higher function
evaluations.
The leading overall objective function is higher compared to the first two.
Table 7.12: Ideal CO2 emissions for combSet1 , combSet2 and combSet3 with their
corresponding Jdis , Jz1 , JIRP , JGch , Jz3 , JGj % , Jz2 and ObjF
JCO2 Jdis Jz1 JIRP JGch Jz3 JGj % Jz2 ObjF
combSet1 186.18 0.892 0.985 1.47 1.43 1.0 0.934 1.05 0.957
combSet2 186.28 0.9095 0.994 1.49 1.43 1.0 0.69 1.017 0.957
combSet3 186.29 0.814 1.006 1.51 1.43 1.0 0.666 0.983 0.957
100
90
80
70
Throttle Position [%]
60
50
40
30
Initial Upshift
Initial Downshift
20
Optimised Upshift
Optimised Downshift
10
0
0 50 100 150
Vehicle Speed [km/h]
Figure 7.17: Ideal optimised gear shift map in terms of CO2 based on combSet1
Figure 7.17 represents the optimised gear shift maps in terms of CO2 from
combSet1 . It only shows, focuses on the most effective area, between 0 to 70%
of throttle position and 0 to 130 km/h. The similarities between these three
optimised shift maps combSet1 , combSet2 and combSet3 are their very low emissions,
167
Algorithm performances
Reserve power[kW]
60
100 100
40
50 50
20
Initial map
Optimised map
0 0 0
0 20 40 60 80 100 120 0 500 1000 1500
Vehicle Speed [km/h] Time (Sec)
ng [−]
4 Opt gear 2
3 3 Opt gear 3
2 Opt gear 4
2 Opt gear 5
1
Opt gear 6
1 0
0 200 400 600 800 1000 1200 0 10 20 30 40 60 70 80 90 110 130
Time [s] Vehicle Speed [km/h]
Figure 7.18: Gear shift map results based on combSet1 of the most suitable results.
The plots are: optimised gear shift map, reserve power, gear selection and speed
range each gear ratio compared to the original shift map
and as expected the IRP has accordingly increased significantly. The advantage
of this combined method is rapid convergence, with good results obtained after
10 generations. However the drawback of this method is the rapid degradation
of the IRP as it can be seen in Table 7.12, which can consequently affect the
driveability.
Figure 7.18 and Figure 7.19 represent the optimised gear shift map results for
combSet1 , with its corresponding reserve power, gear selection and speed range
on each gear ratio in compared to the standard gear shift map (see Section F.1,
in Appendix F for combSet2 and combSet3 ). As expected, the gear 5th usage has
increased, which benefit the CO2 emissions. Therefore, the reserve power has
massively increased whilst lowering the CO2 emissions.
168
Algorithm performances
70
60
Throttle Position [%]
50
40
30
Initial map
20 Optimised map
Gear 1
Gear 2
10 Gear 3
Gear 4
Gear 5
Gear 6
0
0 20 40 60 80 100 120
Vehicle Speed [km/h]
Figure 7.19: Gear shift map results based on combSet1 of the most suitable results.
The plot presents the vehicle speed of the NEDC reflected on the shift map.
Table 7.13: Ideal CO2 emissions for combSet1 , combSet2 and combSet3 with their
corresponding Jdis , Jz1 , JIRP , JGch , Jz3 , JGj % , Jz2 and ObjF
JCO2 Jdis Jz1 JIRP JGch Jz3 JGj % Jz2 ObjF
combSet1 195.9 0.938 1.28 1.08 1.0 1.0 0.296 0.602 0.989
combSet2 193.3 0.901 1.27 1.15 1.13 1.0 0.287 0.61 0.979
combSet3 193.1 0.941 1.27 1.16 1.13 1.0 0.291 0.608 0.978
169
Algorithm performances
70 70 70
60 60 60
Throttle Position [%]
40 40 40
30 30 30
20 Initial Upshift 20 20
Initial Downshift
Optimised Upshift
Optimised Downshift
10 10 10
0 0 0
0 50 100 0 50 100 0 50 100
Vehicle Speed [km/h] Vehicle Speed [km/h] Vehicle Speed [km/h]
Figure 7.20 represents three optimised gear shift maps selected at early gen-
eration in order to be compared to the last generation described by Figure 7.17.
Table 7.13 represents their objective functions. In terms of CO2 emissions, the
selected solutions from an early generation are similar to the results obtained in
Section 7.6. However the leading solutions in terms of CO2 obtained in the last
generation have significantly lower CO2 values e.g. 186 g/km. The IRP is by
contrast significantly increased, expressing a degradation of the driveability as
CO2 decreases (see Table 7.12 and 7.13). Note that the minimum driveability
constraints are still met by these solutions, however the overall shape of gear
shift map resulting from the combined gear ratio and gear shift map optimisation
is not regular. A comparison was also made between the average and ideal CO2
emissions in terms of shapes of gear shift map. It was found that in the absence of
170
Algorithm performances
Reserve power[kW]
60
100 100
40
Initial map 50 50
20
Optimised map
0 0 0
0 20 40 60 80 100 120 0 500 1000 1500
Vehicle Speed [km/h] Time (Sec)
Gear change frequency: 0.96667 [−] Operating regions of different gear positions
6 7
Init Selected Gear [−]
Opt Selected Gear [−] 6
5
5 Init gears
Gear [−]
4 Opt gear 1
n [−]
4
Opt gear 2
g
3 3 Opt gear 3
2 Opt gear 4
2 Opt gear 5
1
Opt gear 6
1 0
0 200 400 600 800 1000 1200 0 10 30 40 50 60 80 90 110 140
Time [s] Vehicle Speed [km/h]
Figure 7.21: Gear shift map results based on combSet1 at the earlier generation.
The plots are: optimised gear shift map, reserve power, gear selection and speed
range for each gear ratio to the original shift map
stricter constraints on the shape of the gear shift, the optimiser tended to produce
irregular gear shift maps that may not be optimal from a practical perspective
(see F, Section F.2). The selection of the shape constraints has been identified as
an area of further work which requires important inputs from engineers as well as
use of a wide variety of driving cycles to ensure that the engine operating points
cover the whole gear shift map.
The operating points of combined gear ratio and gear shift map are not similar
to the original shift map. This can be explained by the modification of gear ratio.
The engine speed vary from 780 RPM to up to 4000 RPM, in compare to the
original gear ratio, which vary from 780 RPM to 2400 RPM. It can be noticed
that having a wide range of engine speeds and different set of gear ratios benefit
the CO2 emission.
Figure 7.21 and Figure 7.22 represent the optimised gear shift map results for
combSet1 , with its corresponding reserve power, gear selection and speed range on
each gear ratio in compared to the standard gear shift map (see Section F.2, in
171
Algorithm performances
70
60
50
Throttle Position [%]
40
30
Initial map
20
Optimised map
Gear 1
Gear 2
10 Gear 3
Gear 4
Gear 5
Gear 6
0
0 20 40 60 80 100 120
Vehicle Speed [km/h]
Figure 7.22: Gear shift map results based on combSet1 at the earlier generation.
The plot presents the vehicle speed of the NEDC reflected on the shift map.
Appendix F for combSet2 and combSet3 ). Same remarks can be made as observed
in Figure 7.18 and Figure 7.19, except the gear selections are mainly on 4th .
By comparing various objective functions, as seen in Table 7.13 (only opti-
mised gear shift map, see Section 6.1.1, in Chapter 6). It can be noticed that
most objective functions are much higher.
This chapter has reported the simulation results for gear shift map and gear ratio.
It has illustrated the features of the solutions obtained for individual as well as
combined gear shift map and gear ratio optimisation. The chapter started with
a demonstration of the working principle of the new problem specific operator
172
Algorithm performances
and the new repair mechanism. The simulation studies confirmed the ability of
the GES operator to improve, in terms of CO2 existing optimised shift map. The
repair mechanism was shown to be able to correct shift maps which exhibits a
reserve power below a user defined threshold. Considering the usage of weighted
sum to combine a multi-objective function into single objective function. It can be
noticed that the final objective function has the same value (see Table 7.12), which
makes it difficult to select one without observing various objective functions.
A benchmark study was carried out by comparing the MOGA developed in
this thesis against MOGA from MATLAB toolbox considering Pareto ranking
and weighted sum. The same conditions were applied in each case, considering
integration of repair mechanism and problem specific into MOGA. The ideal com-
promised set of solutions was Pareto ranking, as it has demonstrated a better di-
versity among various objective functions. Pareto based optimisation approaches
were also able to find solutions with lower CO2 . The evaluation, through simula-
tion, of the new hybrid MOCS to optimised intermediate gear ratio demonstrated
that optimising intermediate gear ratio was beneficial to bandwidth but not nec-
essarily to CO2 . The evaluation of the MOGA combined with MOCS to optimise
jointly gear shift map and gear ratio demonstrated that such an approach could
identify solutions that could significantly reduce CO2 emissions compared to op-
timising gear ratio and gear shift map independently.
173
Chapter 8
8.1 Introduction
This chapter details the validation of the shift map developed in this thesis, by
testing an optimised gear shift map on a rolling road. This mapping is designed
with respect to the engine operating range so that the transmission is in the
correct gear at all times e.g. when the driver requires maximal torque. Depending
on the prevailing conditions, typically the transmission will select a lower gear
moving the engine further into maximum torque producing range (high RPM
combined with a wide open throttle). Until recently, these shift maps have been
created using a template for the type of engine and vehicle intended and then
calibrated by trial and error until adequate performance is achieved according to
the ride and drive assessment of a test driver.
A new approach to shift map design is proposed in this thesis, such that the
procedure is systematic and automated according to criteria that are selected
in an objective function that guides the final definition of the shift map. In
Chapter 4, the problem formulation has been described detailing the objective
174
Optimised gear shift map experimental results
function (see Section 4.2.1). Many variables may be considered, obviously chief
among them is CO2 , but also driveability considerations, such as power on de-
mand (acceleration curves), gear shift frequency and other important parameters.
In particular a MOGA (see Chapter 5 and Chapter 7) was applied to solve this
multi-variable control problem which balances fuel consumption (CO2 emissions)
and driveability. The emissions testing were carried out in SAIC Anting (China)
test facilities. The first task was the execution of a series of vehicle tests on a
rolling road in controlled conditions in order to establish a baseline average CO2
figure for the particular vehicle. The next testing phase incorporated the same
vehicle but this time with the CO2 reducing (optimised) shift maps.
In this chapter, a rolling road test result is described and compared with
the initial shift map. This chapter is composed as follow: Section 8.2 describes
the rolling road parameters, Section 8.4 explains the test results obtained from
the initial shift map and optimised shift map. Additionally variability of the
simulation model using driver speed input with various optimised shift maps are
considered. Finally Section 8.5 concludes this chapter with a discussion.
This section defines the rolling road test procedure, important parameter settings
before testing the shift map and an explanation of the test results.
175
Optimised gear shift map experimental results
176
Optimised gear shift map experimental results
The testing of emissions and fuel consumption of ROEWE 950 took place on a
rolling road (dynamometer) laboratory at the SAIC Anting test and development
plant (Shanghai, China). Before the emissions test, vehicles are preconditioned
and soaked for at least 6 hours at a test temperature of 20-30C. Emissions are
then measured while vehicle is driven according to the New European Driving
Cycle (NEDC) speed profile. The entire NEDC consists of four repeated ECE-15
driving cycles of 195s duration each and one extra-urban driving cycle (EUDC)
of 400s duration.
Figure 8.1: The New European Driving cycle (NEDC), based one urban part,
composed of four repeated ECE-15 driving cycles, and one extra-urban driving
cycle
177
Optimised gear shift map experimental results
be adjusted within ± 5 sec, if not, then the driving resistance of the rolling road
must be adjusted and a repeat coastdown test performed in order to validate the
experimental data (a more detailed coastdown test description can be found in
Appendix C).
Every test on the rolling road is accompanied by a bag test documentation de-
scribing the useful information regarding the particular vehicle used for fuel and
emissions, test conditions, fuel type, and the respective emissions test results (see
Appendix G). The bag test documentation is composed of four parts as follows:
• Test conditions:
The test conditions section specifies the atmospheric pressure (101.8 kPa),
the ambient temperature (24.3 ◦ C, must be maintained), relative humidity
(39%).
178
Optimised gear shift map experimental results
exhaust tailpipe, the results of the emissions are then calculated and printed
out.
These documents record all the information of the test conditions with fuel and
emissions samples obtained from the rolling road. This information will be used
to further assess various optimised shift maps in terms of emissions performance.
This sub-section hypothesises possible sources of error, regarding the test phase.
It is described as follows:
• The vehicle used for the emissions test was a prototype vehicle, however
each test was carried out with the latest Engine Control Unit (ECU) and
Transmission Control Unit (TCU) calibration. Additionally, the vehicle was
first tested and agreed as valid for rolling road test by calibration engineers,
moreover the tyres were also checked.
• It was advised to retain the same driver for all rolling road tests for this
research, because of the availability of certified drivers for rolling road test-
ing. However, it was not possible to retain the same driver. While different
driver behaviour can affect the final results, it is encouraging to know that
all drivers follow the same strict training within SMTC regarding emissions
test.
• During the test phases, the vehicle encountered some issues with the start
and battery charging system. It was advised to change the vehicle with
another prototype vehicle with identical ECU and TCU calibration.
179
Optimised gear shift map experimental results
• Translation during the test phases. A weekly meeting was held to commu-
nicate with SMTC China calibration engineers to discuss testing and any
related issues with the rolling road and vehicle. English/Chinese translator
was present at the meeting. Additionally, communication with an engineer
who does not speak English was assisted by the translator.
rolling road
This section presents the ideal shift map selected for the rolling road. Table 8.1
illustrates the objective functions of selected gear shift map.
Table 8.1: Best CO2 emissions for GSMSet1 , GSMSet2 and GSMSet3 with their
corresponding Jdis , Jz1 , JIRP , JGch , Jz3 , JGj % and Jz2
JCO2 Jdis Jz1 JIRP JGch Jz3 JGj % Jz2
GSMSet1 193.3 0.82 0.812 1.15 1.27 1.01 0.827 1.09
GSMSet2 190.2 0.86 0.803 1.24 1.4 1.01 0.758 1.09
GSMSet3 190.8 0.673 0.811 1.22 1.4 0.999 0.76 1.12
70
60
50
Throttle Position [%]
40
30
20
10
Initial map
Optimised map
0
0 20 40 60 80 100 120
Vehicle Speed [km/h]
180
Optimised gear shift map experimental results
70
60
40
30
20
10
Initial map
Optimised map
0
0 20 40 60 80 100 120
Vehicle Speed [km/h]
Figure 8.2, Figure 8.3 and Figure 8.4 represent satisfactory optimised gear
shift map obtained.
181
Optimised gear shift map experimental results
70
60
40
30
20
10
Initial map
Optimised map
0
0 20 40 60 80 100 120
Vehicle Speed [km/h]
6
Init Selected Gear [−]
5.5
Opt Selected Gear [−]
4.5
4
Gear [−]
3.5
2.5
1.5
1
0 200 400 600 800 1000 1200
Time [s]
Figure 8.5: Best selected optimised gear shift map 1, gear position
Figure 8.5, Figure 8.6 and Figure 8.7 represent the satisfactory optimised
gear shift map and gear position. The gear change frequency has significantly
increased for Figure 8.6 and Figure 8.7 as they represent the ideal CO2 emissions.
182
Optimised gear shift map experimental results
6
Init Selected Gear [−]
5.5
Opt Selected Gear [−]
4.5
4
Gear [−]
3.5
2.5
1.5
1
0 200 400 600 800 1000 1200
Time [s]
Figure 8.6: Best selected optimised gear shift map 2, gear position
6
Init Selected Gear [−]
5.5
Opt Selected Gear [−]
4.5
4
Gear [−]
3.5
2.5
1.5
1
0 200 400 600 800 1000 1200
Time [s]
Figure 8.7: Best selected optimised gear shift map 3, gear position
183
Optimised gear shift map experimental results
Reserve power[kW]
50
0 0
0 200 400 600 800 1000 1200
Time (Sec)
Figure 8.8: Best selected optimised gear shift map 1, reserve power
100
Vehicle speed [km/h]
Reserve power[kW]
50
0 0
0 200 400 600 800 1000 1200
Time (Sec)
Figure 8.9: Ideal selected optimised gear shift map 2, reserve power
Figure 8.8, Figure 8.9 and Figure 8.10 represent the ideal optimised gear
shift map and reserve power. As expected the reserve power of Figure 8.9 and
Figure 8.10 are the highest (see Table 8.1).
184
Optimised gear shift map experimental results
100
Reserve power[kW]
50
0 0
0 200 400 600 800 1000 1200
Time (Sec)
Figure 8.10: Ideal selected optimised gear shift map 3, reserve power
This section describes fuel consumption and emissions testing performed for this
research. The first section defines the initial shift map test on the rolling road
to determine the benchmark CO2 emissions. Optimise shift map results are then
presented and compared with the initial shift map results. Finally, the emissions
saving is also reported and concluded.
8.4.1 Benchmark
In this sub-section the test of the initial shift map on the rolling road is de-
scribed before testing of the optimised shift map. It was agreed to first establish
a benchmark CO2 figure on the rolling road, which will be used to compare the
optimised shift map emissions performance. The test conditions for the Bench-
mark are based on the description given in Section 8.2. Additionally a bag test
example documentation can be found in Appendix G. It would have been ideal to
repeat the test at least 10 times to establish the benchmark CO2 figure, however
due to limited resources, it was only possible to perform the benchmark test 3
times. Table 8.2 describes the results from three tests with the initial shift map
185
Optimised gear shift map experimental results
Table 8.2: Initial shift map fuel economy (FE) and CO2 emissions on a rolling
road over the NEDC. Three test results are presented in this table generated
under the same conditions
Test 1 Test 2 Test 3
CO2 (g/km) 196.39 201.69 204.88
F.E (L/100km) 8.42 8.44 8.58
Table 8.3: Initial shift map fuel consumption and emissions on a rolling road over
the NEDC. This table shows the average and standard (Std) deviation values for
fuel and emissions
Average value Std
CO2 (g/km) 200.98 4.28
F.E (L/100km) 8.4800 0.0872
Table 8.2 shows an overview of fuel consumption and emissions from rolling
road test based on the initial shift map. Table 8.3 describes the average value
and standard deviation of CO2 emissions and FE. The average CO2 listed in
this table is 200.98 g/km, however the official figure of ROEWE 950 is 197.7
g/km. The differences can be attributed the fact that the research is based on a
prototype vehicle which was still in development phases. The overall objective of
this benchmark was to obtain a CO2 figure in order to compare and quantify the
benefit of optimised shift map.
186
Optimised gear shift map experimental results
150
100
50
0
0 500 1000 1500 2000 2500
Time [Sec]
6
Gear shift position [−]
Gear shift
0
0 500 1000 1500 2000 2500
Time [Sec]
Figure 8.11: Driveability test of Optimised shift map before loading on it in ECU
for rolling road test
Every shift map was tested by a calibration engineer before the final test on
the rolling road for FE and CO2 emissions assessment. Figure 8.11 illustrates an
example of an optimised shift map assessment in order to verify its driveability.
The calibration engineer then filed a report on the shift change at various throttle
positions.
Table 8.4: Initial shift map fuel consumption and emissions on a rolling road over
the NEDC. Three various test results are presented in this table under the same
conditions
Test 1 Test 2 Test 3 Test 4
CO2 (g/km) 195.97 198.38 193.42 195.93
F.E (L/100km) 8.2 8.29 8.27 8.42
The percentage change in fuel economy and emissions relative to the optimised
shift map on the rolling road are listed in Table 8.6. The average change in CO2
for the optimised shift map reported in this table is -2.5%. The average change in
FE is -2.2%. Clearly the optimised shift map has significantly improved the CO2
emissions and fuel consumption. The drawback of these tests, is the difficulty of
maintaining consistency between various tests, as the standard deviation is 2.02
187
Optimised gear shift map experimental results
Table 8.5: Initial shift map fuel consumption and emissions on a rolling road over
the NEDC. This table shows the average and standard (Std) deviation values for
fuel and emissions
Average value Std
CO2 (g/km) 195.9250 2.0252
F.E (L/100km) 8.2950 0.0918
Table 8.6: Average change fuel economy and emissions for optimised shift map
in compared to initial shift map on rolling road
Pollutant & fuel Percent change (%)
CO2 -2.5
F.E -2.2
for CO2 emissions, this can be explained by the driver input and the prototype
vehicle. This is why several tests are required to confirm the results, however
considering the average results based on four tests, the method developed in this
thesis shows a significant and promising contribution to reducing CO2 emissions.
350 400
EOP Standard shift map
EOP Optimised shift map
Constant power
300 350
250
300
Torque (Nm)
200
250
150
200
100
150
50
100
800 1000 1200 1400 1600 1800 2000 2200
Speed (rpm)
Figure 8.12: Engine operating point of standard shift map and optimised shift
map 1 from rolling road
Figures 8.12 and 8.13 illustrate two optimised shift maps in comparison to the
initial shift map. Notice that the initial shift map engine speed and torque were
188
Optimised gear shift map experimental results
350 400
EOP Standard shift map
EOP Optimised shift map
Constant power
300 350
250
300
Torque (Nm)
200
250
150
200
100
150
50
100
800 1000 1200 1400 1600 1800 2000 2200
Speed (rpm)
Figure 8.13: Engine operating of standard shift map and optimised shift map 2
from rolling road
not sampled at the same rate, as engine was sampled at 100ms from TCU data
acquisition, and torque was sampled at 10ms from ECU data acquisition. Con-
sequently both signals are re-sampled for comparison study. It can be observed
that both optimised shift maps have an improved spread of operating points than
the initial shift map.
Figures 8.14 and 8.15 demonstrate the engine characteristic of the optimised
shift map in terms of maximum torque, actual torque, reserve torque, engine
speed and reserve power. Notice that the maximum engine power was estimated
using the engine maximum torque data, which has been modelled as a function
of engine speed at full throttle position. The figures are similar, but by taking
a look at reserve torque and maximum power, the result in Figure 8.14 seems
to show more reserve torque than the second in Figure 8.15, at the beginning of
the drive cycle. However, later on during the drive cycle, it tends to converge.
The most significant factor, in terms of fuel consumption can be attributed to
the driver input.
Table 8.7 describes the time spent in each gear with optimised gear shift map
189
Optimised gear shift map experimental results
Torque [Nm]
300
200
100
Max torque
0
0 200 400 600 torque
Actual 800 1000
Engine speed [RPM]
Reserve torque
1500
1000
0 200 400 600 800 1000
Reserve power [Kw]
800
600 Reserve power
400
200
0
0 200 400 600 800 1000
Time [Sec]
Figure 8.14: This figure illustrates the first optimised shift map, engine maximum
torque (estimate), actual torque and reserve torque. Engine speed and estimated
reserve power
Torque [Nm]
300
200
100
Max torque
0 Actual torque
0 200 400 600 torque
Reserve 800 1000
Engine speed [RPM]
1500
1000
0 200 400 600 800 1000
Reserve power [Kw]
800
600 Reserve power
400
200
0
0 200 400 600 800 1000
Time [Sec]
Figure 8.15: This figure illustrates the second optimised shift map, engine max-
imum torque (estimate), actual torque and reserve torque. Engine speed and
estimated reserve power
1 (b) and shift map 2 (c), compared to the initial shift map (a). It is obvious
that spending time in higher gears improves fuel economy and CO2 emissions, as
it can be seen that percentage of gear ratio for fifth gear has increased for both
optimised shift maps (+10.87% and +10.8%).
190
Optimised gear shift map experimental results
Table 8.7: This table illustrates time spent on different gear ratio between opti-
mised gear shift map 1 (b), shift map (c) in comparison to initial gear shift map
(a). Additionally with the increase or decrease time spent of each gear ratio for
optimised shift map
Gear 1 Gear 2 Gear 3 Gear 4 Gear 5 Gear 6
a (%) 33.3 11 20.2 14.3 2.3 18.6
b (%) 32.2 (-1.1) 10.9 (-0.1) 20 (-0.2) 4.8 (-9.5) 13.17 (+10.87) 18.8 (+0.2)
c (%) 32 (-1.3) 11 (0) 20 (-0.2) 4.3 (-10) 13.11 (+10.8) 18.9 (+0.3)
6
Veicle speed [Km/h]
Gear [−]
100
Vehicle speed[Km/h]
Gear [−]
50
0 0
0 200 400 600 800 1000 1200
Time (Sec)
Figure 8.16: Rolling road vehicle speed and gear shift position under initial shift
map
Figures 8.16, 8.17 and 8.18 illustrate the initial shift map, and two optimised
shift maps gear shift position, as mentioned above, the two optimised shift maps
are shifting earlier and maximising the time in fifth gear.
191
Optimised gear shift map experimental results
6
Veicle speed [Km/h]
100
Gear [−]
Vehicle speed[Km/h]
4
Gear [−]
50
2
0 0
0 200 400 600 800 1000 1200
Time (Sec)
Figure 8.17: Rolling road vehicle speed and gear shift position under optimised
shift map 1
6
Veicle speed [Km/h]
100
Gear [−]
Vehicle speed[Km/h]
Gear [−]
50
0 0
0 200 400 600 800 1000 1200
Time (Sec)
Figure 8.18: Rolling road vehicle speed and gear shift position under optimised
shift map 1
This chapter has described the rolling road test of optimised gear shift maps,
which has repeatedly demonstrated CO2 savings, that are predicted in simulation.
Additionally, CO2 emissions were reduced without no perceptible reduction in
192
Optimised gear shift map experimental results
• Optimised gear shift map are then loaded and test drive assessment is con-
ducted on road.
8.5.2 Improvement
193
Chapter 9
This chapter summarises the main contributions from this research on gear shift
map and gear ratio optimisation before presenting the conclusions arising from
this research and opportunities for further work.
The Conclusion starts with the motivation for the work prior to a description
of the novelties and contributions classified in order of importance. This section
finished with a quote from the industrial collaborators describing the commercial
significance of the work.
The further work section presents opportunities to apply some of the work
and opportunities to further develop the methods and software tools developed
in this work.
9.1 Conclusion
194
Discussion, conclusion and further work
emissions (in particular CO2 ). The most significant CO2 emissions saving at the
point of use arises from the adoption of hybrid technologies. However, signif-
icant savings can also be achieved via hardware design and software solutions
applied to automated manual transmissions. This work has demonstrated that
the development of multi objective, nature inspired, optimisation frameworks to
optimise both gear shift map and gear ratio could lead to significant CO2 and fuel
consumption savings whilst maintaining vehicle driveability. The culmination of
the work on gear shift map optimisation was the experimental validation of the
work through rolling road tests performed by the vehicle manufacturer showing
a significant 2.5% CO2 saving compared to the standard vehicle gear shift map.
A further CO2 saving of up to 5.8% was predicted using simulation studies by
combining gear ratio and gear shift map optimisation. Note that these significant
savings were obtained for gear shift maps that meet the minimum requirement
for reserve power, but they also resulted in worsening of the overall reserve power
and were characterised by unusual trajectories for the up and down gear shifts.
There are two types of novelties in this thesis: (i) problem specific formulation
and methodologies and (ii) improvements of generic optimisation algorithms. The
following statements describe them:
• The most significant problem specific contribution in this thesis is the repair
mechanism, which can be applied to any gear shift map (see Section 5.2.2,
in Chapter 5). The application of the repair mechanism leads to a slight
increase in CO2 emissions (see Table 7.1, in Chapter 7). This increase is
unavoidable due to meeting the minimum reserve power requirement set
by the designer (see Section 7.4, in Chapter 7). The simulation results
in Table D.10, in Appendix D have demonstrated the effect of the repair
mechanism.
195
Discussion, conclusion and further work
both gear ratio and gear shift map multi-objective optimisation. These for-
mulations have enabled the efficient and effective development of the overall
optimisation strategies developed in this thesis. The gear shift map prob-
lem formulation enabled the user to specify the range of throttle position to
consider (see Section 4.2.3, Chapter 4). It was designed to enforce the fol-
lowing engineering constraints: i) prevent crossing between Downshift and
Upshift ii) maintain a minimum hysteresis between Downshift and Upshift
to avoid frequent gear changes for small velocity variations. The number of
design variables to optimise was reduced to 90 by reducing the number of
control points in the gear shift map using a sensitivity analysis and prac-
tical implementation constraints. This resulted in the optimised throttle
angle being separated by 10◦ whilst intermediate throttle angle positions
were reconstructed using linear interpolation.
• The intermediates gear ratios were formulated such that the optimiser can
focus more efficiently on gear ratio spacing and maintain a continuously
decreasing ratio from gear G2 to G5 (see Section 4.3.1, Chapter 4).
• The third problem specific contribution is the problem specific GES opera-
tor (see Section 5.2, in Chapter 5). It has the ability to improve the existing
optimised gear shift map by reducing CO2 emissions by up to 0.37%. GES
decreases the difference between Upshift and Downshift, thereby decreasing
the hysteresis between Up and Downshift (with respect to a minimum hys-
teresis), resulting in making quick gear changes more likely (see Section 7.5,
in Chapter 7). Such rapid changes are however acceptable with the use of
the proposed SAIC Dual Clutch Transmission gearbox. The GES effect
was illustrated in Table 7.3, where the application of GES finished with
the lowest CO2 emissions. Additionally the application at each generations
have also accelerated the rate of convergence as demonstrated by comparing
196
Discussion, conclusion and further work
197
Discussion, conclusion and further work
• Key to any optimisation problem is the ability to ask the appropriate ques-
tion to the optimiser. These questions are expressed in terms of objectives
to be optimised or minimised in the case of this thesis. Three new objective
formulations were proposed in this thesis to investigate if the conflicting ob-
jectives of achieving low CO2 and thereby fuel consumption as well as good
driveability expressed in terms of reverse power could be achieved simulta-
neously (see Section 4.2.1, in Chapter 4). These objectives focused on the
percentage of time the engine was operating in the specific regions of the
BSFC map. Zones 1 and 2 correspond to the two most efficient zone, whilst
zones 3 reflects higher fuel consumption characterised by operating the en-
gine at low or very high revolution per minute. It was noticed that zone
3 was proportional to IRP as it represents the highest fuel consumption,
therefore zone 3 can reach a minimum level (0.88) during the optimisation
process.
198
Discussion, conclusion and further work
for the different criteria developed in order to determine the most suitable
trade-off solution. Guiding the algorithm towards the trade-off is achieved
by allocating weighting coefficient as defined in Equation (4.1) (see Sec-
tion 4.2.1, in Chapter 4). To enable the designer a greater flexibility, gear
weightings were introduced to identify the specific gears to be targeted.
This objective, together with the number of gear changes were found to
be particularly useful to identify patterns of gear selection over the NEDC
drive cycle for different solutions. These patterns relate to different features
in the gear shift map and are strongly correlated to the CO2 emissions.
• Taking inspiration from Le Guen et al. (2011), a new cost function (Dist)
was developed to minimise CO2 by moving the engine operating points,
expressed in terms of engine torque, Te , and engine speed, we , towards
the left side of the BSFC map (see Section 4.2.1, in Chapter 4). It is
realised by minimising the distance between a reference, or anchor point
O(wref , Tref ) on the BSFC map, and the Upshift points for the throttle
positions, tk of interest. The problem specific objective aim is to help users
identify the most suitable solutions, by observing the performance of the
engine operating point on the BSFC map, whilst the specific objectives are
expressed using different formulations. It was found that minimising the
distances between the reference point (O(wref , Tref )), and Upshift points
led to a reduction of CO2 emissions.
199
Discussion, conclusion and further work
such applications (see Section 5.2, in Chapter 5). It uses objective weighted
Pareto ranking to differentiate between non dominated solutions. It has the
advantage of focusing the search towards low CO2 regions without neglect-
ing the other objectives. It has shown to be able to find solutions with lower
costs than the standard weighted sum approaches. The combination of a
standard MOGA with the modified Pareto objective formulation and the
new repair mechanism together with the new local search operator, namely
GES. The aim of these modifications were to exploit problem specific fea-
tures to find improved solutions rapidly. It was shown that GES improved
significantly the current solution generated by the optimiser. The applica-
tion of GES was limited to one at every three generations as it can lead to
a premature convergence, as demonstrated in Figure 7.9.
• This method has resulted in solutions which have been selected for testing
first in real life situations to ensure that they offered suitable driveability
and then on rolling road. The rolling road tests confirmed that the signifi-
cant savings found on simulation. On average a saving of 2.5% in terms of
CO2 emissions was achieved.
200
Discussion, conclusion and further work
very well received by SAIC motors. The following quote was written by the line
manager Chris Woolley:
” The excellent work conducted during Adama Fofana’s time at SMTC UK
proved both advantageous and invaluable. His development of control algorithms
used for gear shift schedule optimisation, dynamic performance and improved fuel
economy proved very successful, these were presented throughout to the Global
business and have aided future development in this area. In addition to this
the models generated during Adama’s time with us are stored in a model library
designed and developed my Adama, these models are used today to support product
development through simulation.”
The mathematical framework defined to express optimised gear shift maps into
design variables may still be improved. An optimisation study should be carried
out further to select an appropriate throttle range over various driving cycles.
The hysteresis of each throttle position was fixed in order to reduce the number
of design variables. However it may be beneficial to investigate throttle specific
hysteresis constraints and refined the selection of the hysteresis values adopted
in this thesis.
The powertrain model used in this thesis was only validated against the New
European Drive Cycle (NEDC). To enable the use of the model to evaluate so-
lutions for other drive cycles, the model should be validated against the World
harmonised Light vehicles Test Procedures (WLTP) as well as the road testing
carried out by the manufacturer to evaluate the vehicle behaviour from a quali-
tative perspective. Having validated the simulation model for a range of driving
conditions, the optimisation should therefore evaluate the performance of the
candidate solutions against the NEDC drive cycle as well as WLTP and the road
201
Discussion, conclusion and further work
202
Discussion, conclusion and further work
203
Index
204
References
205
Discussion, conclusion and further work
206
Discussion, conclusion and further work
Fofana, A., Haas, O., Ersanilli, V., Burnham, K., Mahtani, J., Woolley, C. &
Vithanage, K. (2016), ‘Multi-Objective Genetic Algorithm for automatic trans-
mission gear shift map optimisation’, 14-th IFAC Symposium on Control in
Transportation Systems, May 18-20, 2016, Istanbul Technical University, Tak-
sim, Istanbul, Turkey .
Forsgren, A., Gill, P. E. & Wright, M. H. (2002), ‘Interior methods for nonlinear
optimization’, SIAM review 44(4), 525–597.
Fu, J. & Bortolin, G. (2012), ‘Gear Shift Optimization for Off-road Construction
Vehicles’, Procedia - Social and Behavioral Sciences 54(14), 989–998.
Galvagno, E., Velardocchia, M. & Vigliani, a. (2011), ‘Dynamic and kine-
matic model of a dual clutch transmission’, Mechanism and Machine Theory
46(6), 794–805.
Gill, P. E., Murray, W., Saunders, M. a. & Wright, M. H. (1981), ‘Aspects of
mathematical modelling related to optimization’, Applied Mathematical Mod-
elling 5(2), 71–83.
Goel, S. & Panchal, V. (2014), ‘Performance evaluation of a new modi-
fied firefly algorithm’, Reliability, Infocom Technologies and Optimization
(ICRITO)(Trends and Future Directions), 2014 3rd International Conference
on pp. 1–6.
Golabi, S., Fesharaki, J. J. & Yazdipoor, M. (2014), ‘Gear train optimization
based on minimum volume/weight design’, Mechanism and Machine Theory
73, 197–217.
Goldberg, D. E. & Deb, K. (1991), ‘A comparative analysis of selection schemes
used in genetic algorithms’, Foundations of genetic algorithms 1, 69–93.
Goldberg, D. E. & Lingle, R. (1985), Alleles, loci, and the traveling salesman
problem, in ‘Proceedings of the first international conference on genetic al-
gorithms and their applications’, Lawrence Erlbaum Associates, Publishers,
pp. 154–159.
Gopal Dhal, K., Quraishi, I. & Das, S. (2015), ‘A Chaotic Lévy flight Approach
in Bat and Firefly Algorithm for Gray level image Enhancement’, International
Journal of Image, Graphics and Signal Processing 7(7), 69–76.
Ha, S.-H. & Jeon, H.-T. (2013), ‘Development of intelligent gear-shifting map
based on radial basis function neural networks’, International Journal of Fuzzy
Logic and Intelligent Systems 13(2), 116–123.
Haas, O. C., Burnham, K. J. & Mills, J. a. (1998), ‘Optimization of beam orien-
tation in radiotherapy using planar geometry.’, Physics in medicine and biology
43(8), 2179–2193.
Halevy, I., Kava, Z. & Seeman, T. (2006), ‘Normalization and Other Topics in
Multi Objective Optimization’, Optimization 2, 89–101.
207
Discussion, conclusion and further work
Henrique, C., Amendola, F., Augusto, M., Alves, L., Politécnica, E., Paulo, D. S.
& Paulo, S. (2006), ‘Gear shift strategies analysis of the automatic transmission
in comparison with the double clutch transmission comparison with the double
clutch transmission’, SAE Technical Paper .
Holland, J. H. (1992), Adaptation in natural and artificial systems: an introduc-
tory analysis with applications to biology, control, and artificial intelligence,
MIT press.
Hu, H., Wang, X. & Shao, Y. (2010), ‘Optimization of the shift quality of dual
clutch transmission using genetic algorithm’, Proceedings - 2010 6th Interna-
tional Conference on Natural Computation, ICNC 2010 8(Icnc), 4152–4156.
208
Discussion, conclusion and further work
Kita, H., Ono, I. & Kobayashi, S. (1999), ‘Multi-parental extension of the uni-
modal normal distribution crossover for real-coded genetic algorithms’, Proceed-
ings of the 1999 Congress on Evolutionary Computation (CEC’99) pp. 1581–
1587.
Konak, A., Coit, D. W. & Smith, A. E. (2006), ‘Multi-objective optimization
using genetic algorithms: A tutorial’, Reliability Engineering & System Safety
91(9), 992–1007.
Kulkarni, M., Shim, T. & Zhang, Y. (2007), ‘Shift dynamics and control of dual-
clutch transmissions’, Mechanism and Machine Theory 42(2), 168–182.
Ladkany, G. S. (2012), ‘A Genetic Algorithm with Weighted Average Normally-
Distributed Arithmetic Crossover and Twinkling’, Applied Mathematics
03(30), 1220–1235.
Le Guen, D., Weck, T., Balihe, A. & Verbeke, B. (2011), ‘Definition of Gearshift
Pattern: Innovative Optimization Procedures Using System Simulation’, SAE
Int. J. Fuels Lubr. 4(1), 412–431.
Li, G. & Hu, J. (2010), ‘Modeling and analysis of shift schedule for automatic
transmission vehicle based on fuzzy neural network’, 2010 8th World Congress
on Intelligent Control and Automation pp. 4839–4844.
Lin, Y.-s., Shea, K., Johnson, A., Coultate, J. & Pears, J. (2009), A method and
software tool for automated gearbox synthesis, in ‘ASME 2009 International
Design Engineering Technical Conferences and Computers and Information in
Engineering Conference’, American Society of Mechanical Engineers, pp. 111–
121.
Liu, C., Gao, Z. & Zhao, W. (2012), ‘A New Path Planning Method Based on
Firefly Algorithm’, 2012 Fifth International Joint Conference on Computa-
tional Sciences and Optimization pp. 775–778.
Liu, Y., Qin, D., Jiang, H., Liu, C., Zhang, Y. & Lei, Z. (2009), ‘Shift sched-
ule optimization for dual clutch transmissions’, 5th IEEE Vehicle Power and
Propulsion Conference, VPPC ’09 1, 1071–1078.
Llamas, X., Eriksson, L. & Sundstr, C. (2010), ‘Fuel Efficient Speed Profiles for
Finite Time Gear Shift with Multi-Phase Optimization’, Vehicular Systems .
Long, Q. (2014), ‘A constraint handling technique for constrained multi-objective
genetic algorithm’, Swarm and Evolutionary Computation 15, 66–79.
Lourakis, M. I. (2005), ‘A brief description of the levenberg-marquardt algorithm
implemented by levmar’, Foundation of Research and Technology 4, 1–6.
Lucente, G., Montanari, M. & Rossi, C. (2007), ‘Modelling of an automated
manual transmission system’, IFAC Proceedings Volumes (IFAC-PapersOnline)
7(PART 1), 958–963.
Mahlia, T., Tohno, S. & Tezuka, T. (2012), ‘A review on fuel economy test proce-
dure for automobiles: Implementation possibilities in Malaysia and lessons for
other countries’, Renewable and Sustainable Energy Reviews 16(6), 4029–4046.
209
Discussion, conclusion and further work
Mammetti, M., Gallegos, D., Freixas, A. & Muñoz, J. (2013), ‘The influence of
rolling resistance on fuel consumption in heavy-duty vehicles’, SAE Technical
Paper .
Mani, A. & Patvardhan, C. (2009), ‘A novel hybrid constraint handling technique
for evolutionary optimization’, 2009 IEEE Congress on Evolutionary Compu-
tation pp. 2577–2583.
Martens, D., Baesens, B. & Fawcett, T. (2011), ‘Editorial survey: Swarm intelli-
gence for data mining’, Machine Learning 82(1), 1–42.
Mashadi, B. & Crolla, D. (2012), Wiley.
Mavrotas, G. (2009), ‘Effective implementation of the -constraint method in
Multi-Objective Mathematical Programming problems’, Applied Mathematics
and Computation 213(2), 455–465.
Mitchell, G. G., ODonoghue, D., Barnes, D. & McCarville, M. (2003),
‘GeneRepair-a repair operator for genetic algorithms’, Proc. Genetic and Evolu-
tionary Computation Conference (GECCO) Late Breaking Papers (1999), 235–
239.
Moawad, A. & Rousseau, A. (2014), ‘Impact of electric drive vehicle technologies
on fuel efficiency to support 2017-2025 cafe regulations’, SAE International
Journal of Alternative Powertrains 3(2014-01-1084), 163–175.
Nanda, S. J. & Panda, G. (2014), ‘A survey on nature inspired metaheuristic
algorithms for partitional clustering’, Swarm and Evolutionary computation
16, 1–18.
Naunheimer, H., Bertsche, B., Ryborz, J. & Novak, W. (2010), Automotive trans-
missions: fundamentals, selection, design and application.
Newman, K. A. & Dekraker, P. (2016), Modeling the effects of transmission gear
count, ratio progression, and final drive ratio on fuel economy and performance
using alpha, Technical report.
Ngo, V. D., Colin Navarrete, J. a., Hofman, T., Steinbuch, M. & Serrarens,
a. (2013), ‘Optimal gear shift strategies for fuel economy and driveability’,
Proceedings of the Institution of Mechanical Engineers, Part D: Journal of
Automobile Engineering 227(10), 1398–1413.
Ngo, V. D., Hofman, T., Steinbuch, M. & Serrarens, a. (2013), ‘Gear shift map
design methodology for automotive transmissions’, Proceedings of the Insti-
tution of Mechanical Engineers, Part D: Journal of Automobile Engineering
228(1), 50–72.
Odu, G. O. & Charles-Owaba, O. E. (2013), ‘Review of Multi-criteria Opti-
mization Methods Theory and Applications’, IOSR Journal of Engineering
3(10), 01–14.
Ortiz-Soto, E., Assanis, D. & Babajimopoulos, a. (2012), ‘A comprehensive engine
to drive-cycle modelling framework for the fuel economy assessment of advanced
engine and combustion technologies’, International Journal of Engine Research
13(3), 287–304.
210
Discussion, conclusion and further work
211
Discussion, conclusion and further work
212
Discussion, conclusion and further work
213
Discussion, conclusion and further work
Yao, C., Feng, K. & Hubacek, K. (2015), ‘Driving forces of CO2 emissions in the
G20 countries: An index decomposition analysis from 1971 to 2010’, Ecological
Informatics 26, 93–100.
Yasin, T. P. (1978), ‘The analytical basis of automobile coastdown testing’, SAE
Technical Paper .
Ye, Z. Y. Z., Washko, F. W. F. & Lai, M.-C. L. M.-C. (2004), ‘Genetic algorithm
optimization of fuel economy for PFI engine with VVT-VCR’, Proceedings of
the 2004 IEEE International Conference on Control Applications, 2004. 1, 364–
369.
Yin, X., Wang, W., Chen, X. & Lu, H. (2013), ‘Multi-Performance Optimization
of the Shift Schedule for Stepped Automatic Transmissions’, SAE International
(2013-01-0488).
Yokota, T., Taguchi, T. & Gen, M. (1998), ‘A solution method for optimal weight
design problem of the gear using genetic algorithms’, Computers & industrial
engineering 35(3), 523–526.
Zhou, A., Qu, B.-Y., Li, H., Zhao, S.-Z., Suganthan, P. N., Zhang, Q., H., Z.,
Shi-Zheng, Suganthan, P. N. & Zhang, Q. (2011), ‘Multiobjective evolution-
ary algorithms: A survey of the state of the art’, Swarm and Evolutionary
Computation 1(1), 32–49.
214
Appendix A
where the NLE 2.0L is a gasoline type, naturally aspirated, 4 cylinder engine.
215
Vehicle, Engine & Transmissions Specification
216
Appendix B
100
90
80
70
Throttle position [%]
60
50
Upshift 1 to 2
Upshift 2 to 3
40 Upshift 3 to 4
Upshift 4 to 5
Upshift 5 to 6
30 Dwshift 2 to 1
Dwshift 3 to 2
Dwshift 4 to 3
20 Dwshift 5 to 4
Dwshift 6 to 5
10
0
0 50 100 150 200 250 300
Vehicle speed [Km/h]
Figure B.1: This figure represents the initial calibration shift map of SAIC 6 speed
DCT. The Upshift 1-2, Upshift 2-3, Upshift 3-4, Upshift 4-5 and Upshift 5-6 are
represented by solid black, blue, red, magenta and green lines, respectively. The
Downshift 2-1, Downshift 3-2, Downshift 4-3, Downshift 5-4 and Downshift 6-5
are described by dotted black, blue, red, magenta and green lines, respectively.
217
Initial calibration shift map
Table B.1: Numerical values (Km/h) of initial calibration Upshift (Up) map of
SAIC 6 speed DCT
T hrottle position % U p 1 − 2 U p 2 − 3 U p 3 − 4 U p 4 − 5 U p 5 − 6
0 14 29 43 56 68
10 14 29 43 56 68
20 14 29 43 56 68
30 14 29 43 56 70
40 18 33 51 69 90
50 22 41 64 86 113
60 27 49 77 104 136
70 32 59 94 126 175
80 38 69 110 148 255
90 44 80 127 171 255
100 49 91 143 193 255
Table B.2: Numerical values (Km/h) of initial calibration Downshift (Dw) map
of SAIC 6 speed DCT
T hrottle position % Dw 1 − 2 Dw 2 − 3 Dw 3 − 4 Dw 4 − 5 Dw 5 − 6
0 5 25 39 49 63
10 5 25 39 49 63
20 5 25 39 49 63
30 5 25 39 49 63
40 5 25 39 52 68
50 5 25 42 57 78
60 5 27 46 67 95
70 5 32 54 80 115
80 5 38 64 95 140
90 20 43 78 115 170
100 35 72 110 150 245
218
Appendix C
Figure C.1: Vehicle ROEWE 950 coast down data. These data represents vehicle
speed, time and distance recorded during vehicle deceleration from 130 km/h to
5 km/h
219
Coastdown test data
220
Coastdown test data
Figure C.3: The actual coastdown on the chassis dynamometer must within ± 5
second against the experimental data obtained from the proven ground
Figure C.4: Target coastdown plot test from chassis dynamometer against exper-
imental test from proven ground
221
Appendix D
Table D.1: Most suitable solution upon Pareto optimal set, for each objective
function based on set 1, with no repair and no GES
JCO2 JDist Jz1 JIRP JGch Jz3 JGj % Jz2 JObj Ssprd d Std
Mp1 193.7 0.92 0.84 0.69 0.56 0.87 0.68 1.07 0.93 0.021 0.0043 0.022
Mp2 191.4 0.86 0.82 0.74 0.70 0.87 0.68 1.07 0.94 0.016 0.0031 0.017
Mp3 193.9 0.84 0.84 0.96 1.0 0.88 0.64 0.99 0.97 0.014 0.0038 0.015
Table D.2: Most suitable solution upon Pareto optimal set, for each objective
function based on set 2, with no repair and no GES
JCO2 JDist Jz1 JIRP JGch Jz3 JGj % Jz2 JObj Ssprd d Std
Mp1 191.3 0.89 0.82 0.74 0.70 0.87 0.57 1.06 0.98 0.016 0.0035 0.016
Mp2 195.5 0.80 0.85 0.79 0.83 0.88 0.64 0.98 0.98 0.022 0.0043 0.022
Mp3 190.9 0.87 0.83 0.95 1.0 0.87 0.64 0.96 0.98 0.017 0.0051 0.017
Table D.3: Most suitable solution upon Pareto optimal set, for each objective
function based on set 3, with no repair and no GES
JCO2 JDist Jz1 JIRP JGch Jz3 JGj % Jz2 JObj Ssprd d Std
Mp1 191.2 0.82 0.80 0.93 1.07 0.88 0.61 1.11 0.94 0.029 0.0067 0.029
Mp2 191.4 0.85 0.79 0.90 0.97 0.88 0.68 1.14 0.95 0.025 0.0065 0.026
Mp3 191.0 0.85 0.81 0.94 1.0 0.87 0.64 1.01 0.98 0.024 0.0073 0.025
222
Performance tables of optimisers
Table D.4: Most suitable solution upon Non-Pareto for each objective function
based on set 1
JCO2 JDist Jz1 JIRP JGch Jz3 JGj % Jz2 JObj
Mp6 196.8 0.9196 0.8888 1.0365 1.0 1.0038 0.9789 1.048 0.977
Mp7 196.48 0.8471 0.8272 0.9833 1.1 0.8843 0.6809 1.96 0.944
Table D.5: Most suitable solution upon Non-Pareto for each objective function
based on set 2
JCO2 JDist Jz1 JIRP JGch Jz3 JGj % Jz2 JObj
Mp6 196.8 0.9215 0.8888 1.03 1.0 1.0038 0.9789 1.048 1.0
Mp7 196.1 0.9406 0.858 0.9974 1.23 0.8843 0.6436 1.89 0.9834
Table D.6: Most suitable solution upon Non-Pareto for each objective function
based on set 3
JCO2 JDist Jz1 JIRP JGch Jz3 JGj % Jz2 JObj
Mp6 196.8 0.9215 0.88 1.0365 1.0 1.0038 0.9789 1.048 1.0
Mp7 196.1 0.9429 0.8586 0.9978 1.23 0.8843 0.6433 1.893 0.9833
223
Performance tables of optimisers
Table D.7: Most suitable solution upon Pareto optimal set, for each objective
function based on set 1, with repair mechanism and no GES
JCO2 JDist Jz1 JIRP JGch Jz3 JGj % Jz2 JObj Ssprd d Std
Mp1 192.5 0.88 0.82 0.69 0.57 0.87 0.60 1.09 0.92 0.012 0.0033 0.012
Mp2 193.5 0.86 0.82 0.74 0.70 0.87 0.68 1.07 0.94 0.021 0.0054 0.021
Mp3 194.4 0.85 0.84 0.97 1.00 0.87 0.64 0.94 0.96 0.021 0.0056 0.021
Table D.8: Most suitable solution upon Pareto optimal set, for each objective
function based on set 2, with repair mechanism and no GES
JCO2 JDist Jz1 JIRP JGch Jz3 JGj % Jz2 JObj Ssprd d Std
Mp1 191.4 0.89 0.84 0.75 0.7 0.87 0.52 1.09 0.97 0.022 0.0038 0.022
Mp2 191.8 0.91 0.84 0.89 0.97 0.87 0.60 1.06 0.97 0.015 0.0027 0.015
Mp3 191.9 0.88 0.84 0.95 1.1 0.87 0.57 1.06 0.97 0.019 0.0045 0.019
Table D.9: Most suitable solution upon Pareto optimal set, for each objective
function based on set 3, with repair mechanism and no GES
JCO2 JDist Jz1 JIRP JGch Jz3 JGj % Jz2 JObj Ssprd d Std
Mp1 191.2 0.79 0.82 0.81 0.73 0.99 0.68 1.05 0.93 0.025 0.0044 0.025
Mp2 192.1 0.80 0.82 0.79 0.83 0.87 0.65 1.09 0.94 0.024 0.0079 0.024
Mp3 193.4 0.86 0.83 0.95 1.0 0.87 0.57 1.02 0.97 0.028 0.01 0.029
224
Performance tables of optimisers
GES
Table D.11: Best solution upon Non-Pareto for each objective function based on
set 1
JCO2 JDist Jz1 JIRP JGch Jz3 JGj % J z2 JObj
Mp6 196.8 0.9196 0.8888 1.0365 1.0 1.0038 0.9789 1.0483 0.97
Mp7 197.5 0.8486 0.8866 0.9758 1.1 0.8843 0.7138 1.83 0.95
Table D.12: Most suitable solution upon Non-Pareto for each objective function
based on set 2
JCO2 JDist Jz1 JIRP JGch Jz3 JGj % J z2 JObj
Mp6 196.8 0.9215 0.8888 1.03 1.0 1.0038 0.978 1.048 1.0
Mp7 192.8 0.889 0.8586 1.06 1.36 0.8902 0.5720 1.81 0.975
225
Performance tables of optimisers
Table D.13: Most suitable solution upon Non-Pareto for each objective function
based on set 3
JCO2 JDist Jz1 JIRP JGch Jz3 JGj % J z2 JObj
Mp6 196.8 0.9215 0.8888 1.036 1.00 1.0038 0.978 1.048 1.00
Mp7 197.57 0.9705 0.8545 0.9361 1.1 0.875 0.6513 2.0332 0.9828
Table D.14: Best solution upon Pareto optimal set, for each objective function
based on set 1
JCO2 JDist Jz1 JIRP JGch Jz3 JGj % Jz2 JObj Ssprd d Std
Mp1 193.3 0.56 0.79 0.74 0.7 0.87 0.72 1.03 0.92 0.011 0.0033 0.012
Mp4 191.6 0.58 0.76 0.74 0.7 0.87 0.61 1.04 0.91 0.0036 0.001 0.0037
Mp2 192.6 0.54 0.78 0.69 0.57 0.87 0.65 1.01 0.89 0.0097 0.003 0.0098
Mp5 192.6 0.54 0.80 0.80 0.83 0.88 0.72 0.97 0.89 0.016 0.003 0.016
Table D.15: Most suitable solution upon Pareto optimal set, for each objective
function based on set 2
JCO2 JDist Jz1 JIRP JGch Jz3 JGj % Jz2 JObj Ssprd d Std
Mp1 192.8 0.58 0.81 0.74 0.7 0.87 0.64 1.06 0.98 0.022 0.004 0.022
Mp4 190.7 0.59 0.78 0.92 0.97 0.88 0.56 1.05 0.97 0.016 0.003 0.016
Mp2 190.6 0.63 0.78 0.81 0.97 0.88 0.60 1.09 0.97 0.012 0.0027 0.012
Mp5 190.7 0.71 0.83 0.89 0.97 0.88 0.57 1.02 0.97 0.015 0.0024 0.015
226
Performance tables of optimisers
Table D.16: Most suitable solution upon Pareto optimal set, for each objective
function based on set 3
JCO2 JDist Jz1 JIRP JGch Jz3 JGj % J z2 JObj Ssprd d Std
Mp1 190.6 0.57 0.79 0.79 0.83 0.88 0.58 0.98 0.89 0.028 0.008 0.029
Mp4 191.8 0.61 0.75 0.82 0.97 0.88 0.54 1.89 0.89 0.019 0.0043 0.019
Mp2 190.7 0.54 0.76 0.89 0.97 0.88 0.58 1.02 0.89 0.023 0.0088 0.023
Mp5 190.6 0.54 0.78 0.87 0.97 0.89 0.58 1.03 0.88 0.025 0.007 0.026
Table D.18: Most suitable solution upon Pareto optimal set, for each objective
function based on set 1, with repair mechanism and GES
JCO2 JDist Jz1 JIRP JGch Jz3 JGj % Jz2 JObj Ssprd d Std
Mp1 191.32 0.52 0.81 0.74 0.7 0.87 0.68 1.06 0.92 0.017 0.0046 0.017
Mp2 193.9 0.47 0.84 0.78 0.83 0.87 0.68 1.06 0.9 0.013 0.0027 0.013
227
Performance tables of optimisers
Table D.19: Most suitable solution upon Pareto optimal set, for each objective
function based on set 2, with repair mechanism and GES
JCO2 JDist Jz1 JIRP JGch Jz3 JGj % Jz2 JObj Ssprd d Std
Mp1 190.9 0.65 0.84 0.85 0.87 0.87 0.58 1.05 0.97 0.01 0.003 0.01
Mp2 192.2 0.83 0.84 0.79 0.83 0.87 0.60 1.06 0.97 0.016 0.0026 0.016
Table D.20: Most suitable solution upon Pareto optimal set, for each objective
function based on set 3, with repair mechanism and GES
JCO2 JDist Jz1 JIRP JGch Jz3 JGj % Jz2 JObj Ssprd d Std
Mp1 190.9 0.47 0.78 0.78 0.83 0.87 0.58 1.06 0.88 0.024 0.01 0.025
Mp2 190.7 0.49 0.78 0.79 0.83 0.87 0.57 1.06 0.88 0.031 0.0098 0.03
228
Appendix E
This appendix present the ideal shift map with the minimum CO2 emissions and
their corresponding objective functions.
229
Leading solution for CO2 emissions for Pareto and weighting sum
100
90
80
70
Throttle Position [%]
60
50
40
30
20 Initial map
Optimised map
10
0
0 50 100 150 200 250
Vehicle Speed [km/h]
Figure E.1: Leading Pareto shift map for minimum CO2 emissions under set 1
condition
100
90
80
70
Throttle Position [%]
60
50
40
30
20
10
Initial map
Optimised map
0
0 50 100 150 200 250
Vehicle Speed [km/h]
Figure E.2: Leading Pareto shift map for minimum CO2 emissions under set 2
condition
230
Leading solution for CO2 emissions for Pareto and weighting sum
100
90
80
70
Throttle Position [%]
60
50
40
30
20
10
Initial map
Optimised map
0
0 50 100 150 200 250
Vehicle Speed [km/h]
Figure E.3: Leading Pareto shift map for minimum CO2 emissions under set 3
condition
231
Leading solution for CO2 emissions for Pareto and weighting sum
100
90
80
70
Throttle Position [%]
60
50
40
30
20
10
Initial map
Optimised map
0
0 50 100 150 200 250
Vehicle Speed [km/h]
Figure E.4: Leading weighting sum shift map for minimum CO2 emissions under
set 1 condition
100
90
80
70
Throttle Position [%]
60
50
40
30
20
10
Initial map
Optimised map
0
0 50 100 150 200 250
Vehicle Speed [km/h]
Figure E.5: Leading weighting sum shift map for minimum CO2 emissions under
set 2 condition
232
Leading solution for CO2 emissions for Pareto and weighting sum
100
90
80
70
Throttle Position [%]
60
50
40
30
20
10
Initial map
Optimised map
0
0 50 100 150 200 250
Vehicle Speed [km/h]
Figure E.6: Leading weighting sum shift map for minimum CO2 emissions under
set 3 condition
233
Appendix F
solution
Reserve power[kW]
60
100 100
40
50 50
20
Initial map
Optimised map
0 0 0
0 20 40 60 80 100 120 0 500 1000 1500
Vehicle Speed [km/h] Time (Sec)
4
ng [−]
4 Opt gear 2
3 3 Opt gear 3
2 Opt gear 4
2 Opt gear 5
1
Opt gear 6
1 0
0 200 400 600 800 1000 1200 0 10 20 30 40 60 70 80 90 110 130
Time [s] Vehicle Speed [km/h]
Figure F.1: Gear shift map, engine reserve power, gear selection and operating
speed range for each gear ratio for combSet1
234
Combined gear ratio & shift map optimisation results
Reserve power[kW]
60
100 100
40
50 50
20
Initial map
Optimised map
0 0 0
0 20 40 60 80 100 120 0 500 1000 1500
Vehicle Speed [km/h] Time (Sec)
Gear change frequency: 1.4 [−] Operating regions of different gear positions
6 Init Selected Gear [−] 7
Opt Selected Gear [−] 6
5
5 Init gears
Gear [−]
4 Opt gear 1
n [−]
4
Opt gear 2
g
3 3
Opt gear 3
2 Opt gear 4
2 Opt gear 5
1
Opt gear 6
1 0
0 200 400 600 800 1000 1200 0 10 20 30 40 50 60 70 80 90 100110120130140150
Time [s] Vehicle Speed [km/h]
Figure F.2: Gear shift map, engine reserve power, gear selection and operating
speed range for each gear ratio for combSet2
Reserve power[kW]
60
100 100
40
50 50
20
Initial map
Optimised map
0 0 0
0 20 40 60 80 100 120 0 500 1000 1500
Vehicle Speed [km/h] Time (Sec)
Gear change frequency: 1.4 [−] Operating regions of different gear positions
6 Init Selected Gear [−] 7
Opt Selected Gear [−] 6
5
5 Init gears
Gear [−]
4 Opt gear 1
n [−]
4
Opt gear 2
g
3 3
Opt gear 3
2 Opt gear 4
2 Opt gear 5
1
Opt gear 6
1 0
0 200 400 600 800 1000 1200 0 10 20 30 40 50 60 70 80 90 100110120130140150
Time [s] Vehicle Speed [km/h]
Figure F.3: Gear shift map, engine reserve power, gear selection and operating
speed range for each gear ratio for combSet2
235
Combined gear ratio & shift map optimisation results
map solution
Table F.1: CO2 emissions for combSet1 , combSet2 , combSet3 and combSet4 with
their corresponding Jdis , Jz1 , JIRP , JGch , Jz3 , JGj % , Jz2 and ObjF
JCO2 Jdis Jz1 JIRP JGch Jz3 JGj % Jz2 ObjF
combSet1 195.9 0.937 1.0 1.08 1.0 1.0 0.999 1.00 0.994
combSet2 194.2 0.918 0.78 1.17 1.03 1.0 0.673 1.67 0.986
combSet3 193.3 0.9 0.99 1.15 1.13 1.0 0.938 1.014 0.983
combSet4 193.1 0.94 0.993 1.16 1.13 1.0 0.943 1.011 0.982
Reserve power[kW]
60
100 100
40
Initial map 50 50
20
Optimised map
0 0 0
0 20 40 60 80 100 120 0 500 1000 1500
Vehicle Speed [km/h] Time (Sec)
Gear change frequency: 0.96667 [−] Operating regions of different gear positions
6 7
Init Selected Gear [−]
Opt Selected Gear [−] 6
5
5 Init gears
Gear [−]
4 Opt gear 1
n [−]
4
Opt gear 2
g
3 3 Opt gear 3
2 Opt gear 4
2 Opt gear 5
1
Opt gear 6
1 0
0 200 400 600 800 1000 1200 0 10 30 40 50 60 80 90 110 140
Time [s] Vehicle Speed [km/h]
Figure F.4: Gear shift map, engine reserve power, gear selection and operating
speed range for each gear ratio for combSet1 , combSet2 and combSet3
236
Combined gear ratio & shift map optimisation results
Reserve power[kW]
80
100 100
60
40
50 50
20 Initial map
Optimised map
0 0 0
0 50 100 150 200 250 300 0 500 1000 1500
Vehicle Speed [km/h] Time (Sec)
Gear change frequency: 1.1 [−] Operating regions of different gear positions
6 7
Init Selected Gear [−]
Opt Selected Gear [−] 6
5
5 Init gears
Gear [−]
4 Opt gear 1
n [−]
4
Opt gear 2
g
3 3
Opt gear 3
2 Opt gear 4
2 Opt gear 5
1
Opt gear 6
1 0
0 200 400 600 800 1000 1200 0 10 20 30 40 50 60 70 80 90 100110120130140150
Time [s] Vehicle Speed [km/h]
Figure F.5: Gear shift map, engine reserve power, gear selection and operating
speed range for each gear ratio for combSet2
Reserve power[kW]
80
100 100
60
40
50 50
20 Initial map
Optimised map
0 0 0
0 50 100 150 200 250 300 0 500 1000 1500
Vehicle Speed [km/h] Time (Sec)
Gear change frequency: 1.1 [−] Operating regions of different gear positions
6 7
Init Selected Gear [−]
Opt Selected Gear [−] 6
5
5 Init gears
Gear [−]
4 Opt gear 1
n [−]
4
Opt gear 2
g
3 3
Opt gear 3
2 Opt gear 4
2 Opt gear 5
1
Opt gear 6
1 0
0 200 400 600 800 1000 1200 0 10 20 30 40 50 60 70 80 90 100110120130140150
Time [s] Vehicle Speed [km/h]
Figure F.6: Gear shift map, engine reserve power, gear selection and operating
speed range for each gear ratio for combSet3
237
Appendix G
238