گیربکس ۷ سرعته Ddct Vw

POLITECNICO DI TORINO
MASTER OF SCIENCE IN
MECHATRONIC ENGINEERING
Master Degree Thesis
LQI control of dual-clutch transmission

systems in hybrid vehicles with P2
architecture
Supervisors: Candidate:
Prof. CARLO NOVARA ILARIA GIRONE
Ing. EMANUEL CORIGLIANO
JULY 2020
Summary
In this dissertation, realized in collaboration with Centro Ricerche Fiat (CRF), a

Linear Quadratic Integral approach is implemented in order to control a Dry Dual
Clutch Transmission system in hybrid vehicles with P2 architecture, during the
slipping phase. The main objectives of the control are related to an improvement of
the vehicle comfort, drivability and robustness; a better gearbox/clutch maneuver
smoothness and the minimization at the same time of vibrations and abrupt move-
ments. In particular, the analysis is focused on the Dry Dual Clutch Transmission
(DDCT) developed by Fiat Power-train Technologies. The aim of the control is
that to ensure the slipping speed, that is the speed difference between the disks of
the clutch, reaches the zero value in a smooth way, after following a given reference.
A simplified model of the transmission is here implemented with clutch actuator,
dynamic engine model and elasticity of the transmission shafts. Some uncertainties
have to be considered in the control: the delays of the clutch and engine actuators,
the clutch transmissibility, vehicle mass and the drive shaft torsion effects.
Firstly a Linear Quadratic Integral controller for the plant without actuators is
simulated: both the engine and clutch torques are considered as inputs computed
by the controller and the outputs are equal to the states: engine speed, slipping
speed, drive shaft torsional speed and angle.
Then, with the aim of improving the model making it more realistic, the actua-
tors and delays are introduced in the model and some techniques are introduced
together with the LQI control strategy in order to compensate them, obtaining a
more complex state space representation of the plant that in this case, being more
accurate, gives more reliable results.
The impossibility to measure the actuators states, makes necessary the introduction
of a state estimation with the Kalman Filter that results to be the best way to
manage the uncertainties related to the presence of the delays.
In conclusion, the Linear Quadratic Integral approach for the clutch slipping control
turns out to be advantageous in terms of performances that are assessed in terms of
Standard deviation and Root Mean Square Error (RMSE) and also for its flexibility
that makes possible the introduction of other control strategies suitable for the
actuators management.
ii
iii
Acknowledgements
Preme ergo quod coepisti,

et fortasse perduceris aut ad summum
aut eo quod summum nondum esse solus intellegas
Seneca, Epistulae morales ad Lucilium,Liber II
La stesura della tesi di laurea rappresenta la conclusione del percorso universitario:

tante sono state in questi anni le esperienze e le difficoltà vissute, che hanno
certamente consentito una crescita non solo dal punto di vista professionale ed
accademico, ma anche una grandissima maturazione personale.
Ringrazio il Prof. Carlo Novara, relatore della presente tesi e il correlatore Ing.
Emanuel Corigliano per avermi dato l’opportunità di occuparmi di un progetto
dai caratteri innovativi, grazie per la straordinaria disponibilità e supporto che mi
hanno accompagnata negli ultimi mesi. Un ringraziamento speciale ai miei genitori
che nonostante ogni difficoltà mi hanno sempre spronata ad andare avanti con forza
e determinazione, che mi hanno insegnato il valore del sacrificio e hanno sempre
gioito con me al raggiungimento di ogni singolo obiettivo, grazie per avermi sempre
accompagnata, seppur talvolta da lontano, in ciascun istante della mia esperienza
universitaria e non solo.
Un grazie a mia madre, per me esempio di coraggio e grinta, che mi ha insegnato
che la forza d’animo consente di superare ogni avversità anche la più ardua.
Un grazie al mio papà, che ha da sempre fatto in modo che seguissi i miei desideri
ed inclinazioni ed ha sempre guardato con orgoglio ed ammirazione ciascun mio
traguardo.
Un ringraziamento a chi mi ha accompagnata in questi cinque anni, chi c’è sempre
stato e chi è andato via, ma anche le persone che ho incontrato in questo percorso
e con le quali ho potuto avere sempre un confronto costruttivo di condivisione.
Ringrazio gli amici di sempre, per avermi sempre ascoltata, grazie perché mi siete
stati sempre vicini e avete cercato di incoraggiarmi nei momenti più difficili, vi sarò
per sempre grata.
iv
Table of Contents
List of Tables viii
List of Figures ix
1 Introduction 1
1.1 Thesis structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 DCT System in hybrid vehicles with P2 architecture 4

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 C635 Dry Dual Clutch Transmission . . . . . . . . . . . . . . . . . 9
2.2.1 Operation of the DDCT . . . . . . . . . . . . . . . . . . . . 11
2.2.2 Dry Dual Clutch Unit . . . . . . . . . . . . . . . . . . . . . 12
2.2.3 Electro-hydraulic actuation system . . . . . . . . . . . . . . 13
2.2.4 DDCT Control Unit . . . . . . . . . . . . . . . . . . . . . . 14
2.3 Micro slip . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4 State of art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.5 Model configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.5.1 Actuators . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.5.2 Engine model . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.5.3 Load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.6 Hybrid electric vehicles with P2 architecture . . . . . . . . . . . . . 23
3 Model Analytical description 25

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2 State Space representation . . . . . . . . . . . . . . . . . . . . . . . 26
3.3 Complete model with actuators . . . . . . . . . . . . . . . . . . . . 29
4 LQI technique 32
4.1 Linear Quadratic regulator . . . . . . . . . . . . . . . . . . . . . . . 32
4.1.1 Finite-horizon, continuous-time LQR . . . . . . . . . . . . . 32
4.1.2 Infinite-horizon, continuous-time LQR . . . . . . . . . . . . 33
vi
4.1.3 Finite-horizon, discrete-time LQR . . . . . . . . . . . . . . . 33
4.1.4 Infinite-horizon, discrete-time LQR . . . . . . . . . . . . . . 34
4.2 Incorporating integral action in LQR . . . . . . . . . . . . . . . . . 34
4.3 Optimal management of the delays . . . . . . . . . . . . . . . . . . 37
4.3.1 Smith Predictor . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.3.2 Kalman state observer . . . . . . . . . . . . . . . . . . . . . 40
4.4 LQI in Simulink . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.5 Plant without actuators and delays . . . . . . . . . . . . . . . . . . 43
5 Tuning and Simulation results 49

5.1 LQI technique with actuators and delays . . . . . . . . . . . . . . . 49
5.2 Linearization of the actuators and delays . . . . . . . . . . . . . . . 54
5.3 LQI technique with Smith Predictor . . . . . . . . . . . . . . . . . . 58
5.4 State space representation of the actuators and the Plant with LQI
technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
5.5 Kalman State Estimation . . . . . . . . . . . . . . . . . . . . . . . 64
5.6 Comparison between simulation results . . . . . . . . . . . . . . . . 68
5.6.1 Observations . . . . . . . . . . . . . . . . . . . . . . . . . . 69
6 Conclusions 70
6.1 Future works . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
Bibliography 72
vii
List of Tables
2.1 C635 DDCT Main Mechanical characteristics . . . . . . . . . . . . 9
3.1 DDCT main parameters . . . . . . . . . . . . . . . . . . . . . . . . 27
5.1 Indexes computation . . . . . . . . . . . . . . . . . . . . . . . . . . 68

5.2 Convergence interval . . . . . . . . . . . . . . . . . . . . . . . . . . 68
viii
List of Figures
2.1 The basic configurations of HEVs . . . . . . . . . . . . . . . . . . . 4

2.2 MHEV system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.3 MHEV with P2 architecture . . . . . . . . . . . . . . . . . . . . . . 7
2.4 C635 Manual Transmission and DDCT . . . . . . . . . . . . . . . . 8
2.5 Cross section of the C635 DDCT . . . . . . . . . . . . . . . . . . . 10
2.6 Dual Clutch Unit, Clutch Actuators and installation . . . . . . . . . 12
2.7 Hydraulic Power Unit and CAM . . . . . . . . . . . . . . . . . . . . 13
2.8 C635 DDCT complete actuation system . . . . . . . . . . . . . . . 14
2.9 DCT and clutches K1 and K2 . . . . . . . . . . . . . . . . . . . . . 16
2.10 Schematic representation of the DDCT . . . . . . . . . . . . . . . . 16
2.11 Decoupling control implemented in [4] . . . . . . . . . . . . . . . . 17
2.12 MPC control from [5] . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.13 Engine speed and torque control [6] . . . . . . . . . . . . . . . . . . 18
2.14 General scheme presented in [7] . . . . . . . . . . . . . . . . . . . . 19
2.15 Overall model of the DDCT system . . . . . . . . . . . . . . . . . . 20
2.16 Actuators for Cm and Cf . . . . . . . . . . . . . . . . . . . . . . . . 20
2.17 Reference angular velocity . . . . . . . . . . . . . . . . . . . . . . . 21
3.1 General driveline scheme . . . . . . . . . . . . . . . . . . . . . . . . 25

3.2 DDCT driveline scheme . . . . . . . . . . . . . . . . . . . . . . . . 26
3.3 Reduced linear driveline scheme . . . . . . . . . . . . . . . . . . . . 26
3.4 Complete model with actuators . . . . . . . . . . . . . . . . . . . . 30
4.1 General LQI scheme . . . . . . . . . . . . . . . . . . . . . . . . . . 35

4.2 General architecture for the control . . . . . . . . . . . . . . . . . . 37
4.3 Controller without Smith Predictor . . . . . . . . . . . . . . . . . . 39
4.4 System considered for Smith Predictor . . . . . . . . . . . . . . . . 39
4.5 Smith Predictor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.6 General observer scheme . . . . . . . . . . . . . . . . . . . . . . . . 41
4.7 LQI controller scheme . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.8 Plant without actuators . . . . . . . . . . . . . . . . . . . . . . . . 44
ix
4.9 General scheme with controller . . . . . . . . . . . . . . . . . . . . 44
4.10 Delta speed behavior without actuators . . . . . . . . . . . . . . . . 45
4.11 Drive shaft torsion speed and angle . . . . . . . . . . . . . . . . . . 45
4.12 Delta speed behavior without actuators T=10ms . . . . . . . . . . . 46
4.13 Drive shaft torsion speed and angle T=10 ms . . . . . . . . . . . . 46
4.14 Total engine and clutch torque T=10 ms . . . . . . . . . . . . . . . 47
5.1 Plant fed with the actuated inputs . . . . . . . . . . . . . . . . . . 49

5.2 Delta speed behavior . . . . . . . . . . . . . . . . . . . . . . . . . . 50
5.3 Drive shaft torsion speed and angle . . . . . . . . . . . . . . . . . . 50
5.4 Total engine and clutch torque . . . . . . . . . . . . . . . . . . . . . 51
5.5 Delta speed behavior with T=10 ms . . . . . . . . . . . . . . . . . . 52
5.6 Drive shaft torsion speed and angle with T=10 ms . . . . . . . . . . 52
5.7 Total engine and clutch torque with T=10 ms . . . . . . . . . . . . 53
5.8 Model plant to be linearized . . . . . . . . . . . . . . . . . . . . . . 54
5.9 Delta speed behavior with linearization . . . . . . . . . . . . . . . . 55
5.10 Drive shaft torsion speed and angle with linearization . . . . . . . . 55
5.11 Total engine and clutch torque with linearization . . . . . . . . . . 56
5.12 Delta speed behavior with linearization T=10 ms . . . . . . . . . . 57
5.13 Drive shaft torsion speed and angle with linearization T=10 ms . . 57
5.14 Total engine and clutch torque with linearization T=10 ms . . . . . 58
5.15 Smith Predictor scheme for clutch actuator . . . . . . . . . . . . . . 58
5.16 Delta speed behavior with LQI and Smith Predictor . . . . . . . . . 59
5.17 Drive shaft torsion speed and angle with LQI and Smith Predictor . 59
5.18 Total engine and clutch torque with LQI and Smith Predictor . . . 60
5.19 Simulink scheme of the complete system . . . . . . . . . . . . . . . 61
5.20 Delta speed behavior with LQI and complete State Space . . . . . 62
5.21 Drive shaft torsion speed and angle with complete State Space . . . 62
5.22 Total engine and clutch torque with LQI and complete State Space 63
5.23 LQI with Kalman filter . . . . . . . . . . . . . . . . . . . . . . . . . 64
5.24 Full model with observer . . . . . . . . . . . . . . . . . . . . . . . . 65
5.25 Plant with actuators . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.26 Delta speed behavior with LQI and Kalman filter . . . . . . . . . . 66
5.27 Drive shaft torsion speed and angle and Kalman filter . . . . . . . . 66
5.28 Total engine and clutch torque with LQI with Kalman filter . . . . 67
5.29 Expected performance . . . . . . . . . . . . . . . . . . . . . . . . . 67
x
Chapter 1
Introduction
The automotive industry is facing many challenges related to different fields, in
particular it is becoming more and more important the improvement of the tech-
nologies related to the reduction of emissions, fuel consumption and driving comfort.
Considering all these factors, that are becoming crucial for the success of a new
product, the companies are giving particular attention to the developement of more
efficient transmission systems.
The environmental needs and legislative pressure for a significant CO2 emissions
reduction in the coming years forces the automotive industry to optimize every
single component of the next generation vehicles. [1]
The transmission,which is the element used for matching the engine’s charac-
teristics with the vehicle drivability requirements, assumes an important role in
this optimization process. High mechanical efficiency and the ability to enable the
operation of the prime mover in its most efficient operating points while delivering
the required power are clearly the main development objectives of any new trans-
mission projects.
The Manual Transmission, despite its intrinsically high overall efficiency which can
also benefit of the related dry clutches technology, cannot guarantee high Real-life
powertrain efficiency due to the impossibility to control it in coordination with the
rest of the powertrain and vehicle.
The step necessary to ensure this possibility involves transmission automation, lead-
ing to the adoption of a Non Manual Transmission (NMT) technology. Premising
that the specific market, brand, mission and user requirements must be taken into
account for the proper transmission choice, the conclusion is that there is no single
Non Manual Transmission technology able to satisfy all vehicle, market, mission
and user requirements:
1
Introduction
• AMTs (Automated Manual Transmissions) are the most cost-effective solu-

tions for transmission automation and are suitable for utilitarian vehicle applications
• DCTs (Dual Clutch Transmissions) represent a perfect balance between pow-

ertrain efficiency and driving comfort and are suitable for medium segment ap-
plications, furthermore thanks to sportiveness’ driving experience that they can
transmit to the driver are now one of the preferred choice for sports cars.
• CVTs (Continuously Variable Transmissions) are particularly suitable for some

specific market and vehicle segments.
• Multi-speed ATs (planetary Automatic Transmissions) are competitive in terms

of comfort/performance for the higher segment vehicles.
In this thesis project, the DCT trasmission system is taken into account, es-
pecially because it allows the engine’s energy to flow directly to the wheels. On
the other hand, the driver comfort also depends on the way in which the clutches
engage: this engagement must be realised as smooth as possible.
Adopting some specific control strategies, it is possible to reduce vibrations and

abrupt movements, managing the transitions between combustion engine and elec-
tric engine and improving the clutch maneuver smoothness.
The clutch engagement problem is here analysed, focusing on the Dual Clutch
Transmission (DCT) system, in the following operational conditions: one DCT
clutch is in slipping phase, the other is opened, while the combustion engine is ON.
The LQI (Linear Quadratic Integral) control is used in order to control the dual
clutch during the slipping phase making possible the convergence to zero of the
speed difference between the clutch disks (delta speed).
2
Introduction
1.1 Thesis structure

The second chapter of this dissertation deals with the explanation of the characteris-
tics of the Dual Clutch transmission system in hybrid vehicles with P2 architecture,
the description of the main components involved and the control targets.
Then, in the following chapter the micro-slip problem is presented, the state
of art and the detailed model configuration, together with the analytical description
of the system, the equations developed and the determined state space representa-
tion.
The fourth chapter is reserved to the description of the LQI control strategy,
the operational conditions and the control targets. Here, also the other techniques
adopted in order to compensate the delays of the actuators are explained, paying
specific attention to the advantages of combining different methods to obtain better
results in terms of outputs.
In the last part of the thesis, the system realized in Simulink environment is
shown, along with the simulation results that are interpeted and commented. A
detailed description is carried out for the different strategies adopted.
In the last chapter the conclusions about the obtained results are exhibited with all
the consideration needed to validate the improvement of the transmission system,
achieving the control targets.
3
Chapter 2
DCT System in hybrid

vehicles with P2
architecture
2.1 Introduction
Nowadays, one of the aims of the automotive industry is a more efficient use of
energy in a vehicle. Hybrid vehicles are possible candidates to reduce both emissions
and consumptions. Hybrid powertrains utilize a conventional ICE together with a
battery, an electric motor, and an electronic controller. HEVs have two possible
configurations: series and parallel hybrids.
Figure 2.1: The basic configurations of HEVs
4
DCT System in hybrid vehicles with P2 architecture
In the series structure, the ICE is exploited to drive an alternator to generate

electricity, which is then sent directly to the electric motor, which powers the wheels
or can be stored in a battery system. The engine is not giving power directly to
the wheels and its efficiency is maximised because it operates within a small range
of speeds. The parallel hybrid has two paths for the power, so that the wheels can
be directly powered by either the Integrated Electric Machine or the electric motor
(or both).
The fuel consumption is reduced and power output does not change because the
electric motor and the battery provide added power to the ICE. In order to save
fuel, the engine automatically stops when the vehicle comes to a stop, and starts
automatically when accelerating again. The battery is recharged using the energy
from the regenerative braking, avoiding the IEC from doing it, and saving more
fuel. A hybrid powertrain does not need to be plugged in and charged up because
it is completely self-sufficient.
There is another, simpler version of the hybrid powertrain that is the so-called "mild
hybrid", which has an electronically controlled integrated starter-generator (ISG),
instead of an electric propulsion motor and is also capable of efficient heavy-duty
towing. [2] The ISG is not able to move the car, but it can help its propulsion and
recover energy through regenerative braking. The mild hybrid cars are normally
operated in the same manner as the full hybrids. One of the differences with respect
to a full hybrid is that a mild hybrid has a much smaller battery than a full hybrid
and cannot run on battery power alone.
A typical mild hybrid vehicle system’s structure is given below:
Figure 2.2: MHEV system
One of the most important advantages of the mild HEV system is that its cost is
lower compared to the full HEV system. For sure, a less expensive hybrid system
is more likely to appeal to a greater number of simple consumers who do not need
more robust and more expensive hybrid systems, especially in the initial phase of
market development. When consumers will better understand the differences in
hybrid technologies, they will be better able to match their need for increased fuel
saving with the most cost-effective system.
5
Presently, the most common mild hybrid topology is the P0 architecture, in which
the electric machine is integrated in the Front End Accessory Drive (FEAD). In
this case, the electric machine is replacing the alternator in terms of usage.
The belt-integrated starter generator (BiSG or e-machine) can be an asynchronous
or synchronous electric machine, and it has two main functions:
1. provide torque to the powertrain in motor mode
2. produce electricity in generator mode
For example, a BiSG can provide 4-6 kW of nominal power and around 14-16 kW
of peak power. The output torque can be around 60 Nm which can reach up to
160 Nm at the crankshaft when amplified by the belt pulley ratio.
In case of a P1 mild hybrid architecture, the e-machines need to be flat, in order
to be positioned between the internal combustion engine and the transmission. For
these applications, in order to have a high torque output, the e-machines need to
have high power density. Therefore permanent magnet synchronous e-machines are
used instead of induction asynchronous e-machines.
Moving the electric motor between the engine and the transmission the architecture
becomes P2. The change from P1 to P2, realized also introducing a clutch that
disconnects the thermal engine, realizes many possibilities for hybrid cars.
It is possible, for example, to improve existing transmissions, for example using
the electric motor group – clutch instead of the torque converter.
Effectively, the electric motor is connected to the input of the transmission and
gives energy instead of the thermal one. The clutch gives the possibility to start
the combustion engine, start from a standstill and travel in electric mode without
pulling it. The P2 system also allows the hybrid car to "boot" with only the electric
motor and thermal engine turned off.
The P3 configuration changes the position of the engine "downstream" of the
gearbox. It minimizes losses, as the electric motor drags only the final part of the
transmission. However, this kind of solution is a bit expensive because it forces to
change the design of a large part of the transmission.
Hybrid cars with the acronym P4 have an electric motor that acts exclusively on
the axis not connected to the thermal engine. In this way it is possible to have a
four-wheel drive without mechanical connection with the thermal engine. Even the
pure electric mode is easily achieved, just disconnecting the "main" transmission
using only the axle connected to the electric motor.
Considering the P2 architecture, it is possible to employ the Dual Clutch Trans-
mission (DCT), in order to solve the problem of the torque interruption during
the automated gear shifting of the Automated Manual Transmissions (AMT) that
penalizes the comfort of the driver.
6
Figure 2.3: MHEV with P2 architecture
The Dual Clutch Transmission is made-up of two independent sub-gearboxes,

one for the even gear sets and the other for the odd gear sets, each one activated by
separate clutches: the engaging and the disengaging ones. The gear shift process
involves the engagement of the on-coming clutch and the release of the off-going
clutch to ensure a shift without traction interruption.
There are two different kinds of DCTs: two wet multi-disc clutches bathed in oil
for cooling (WDCT), or two dry single-disc clutches (DDCT). Wet Dual Clutch
Transmissions are usually used in high torque applications, while the Dry-Dual
Clutch is employed for smaller vehicles where lower torques are needed.
In this thesis project the Dry Dual Clutch Transmission is considered, in particular
the transmission developed by Fiat Power-train Technologies that is part of the
new C365 transmission family. This new transmission family consists of a range
of Manual, All Wheel Drive and DCT transversal, 6-speed transmissions with a
maximum input torque of 350 Nm and output torque of 4200 Nm. The DCT
version of this transmission is the highest rated amongst the double dry dual clutch
transmissions in the market. The C635 transmissions have been designed taking
into account the current trends in vehicle and powertrain technology.
7
Figure 2.4: C635 Manual Transmission and DDCT
The advantage of employing the dry clutch instead of the wet dual clutch is related
to the improvement of the fuel efficiency, that is directly linked to the reduction of
pumping losses of the fluid in the clutch housing. This research takes into account
the developement of a control method in order to track a reference trajectory for the
clutch slipping speed avoiding oscillations, throught LQI Linear Quadratic Integral
control strategy. The adoption of other techniques together with the main one
has been necessary in order to manage the delays of the clutch and motor actuators.
8
2.2 C635 Dry Dual Clutch Transmission

The C635 Dry Dual Clutch Transmission is a six-speed, automated manual gearbox
developed by Fiat Powertrain Technologies (FPT) together with Magneti Marelli
and BorgWarner, and is fabricated by FPT at the Verrone, Italy plant. The trans-
mission adopts a control system developed by Magneti Marelli that incorporates
BorgWarner’s hydraulic actuation module into its own power and transmission
control units.The DDCT is selled in different ways under the trade names TCT,
Twin Clutch Transmission (Alfa Romeo), Euro Twin Clutch Transmission (Fiat
USA), and Dual Dry Clutch Auto Transmission (Dodge).
This transmission is the highest-torque transverse dry dual clutch application and
it is able to receive torque inputs of up to 350 Nm and realizes a maximum output
of 4200 Nm. It weights 81 kg, inclusive of oil and transmission control unit. The
C635 DDCT transmission mechanical characteristics are summarised in Table 1:
Layout 3-shaft plus RG idler

Number of speeds 6-7
Housing Two piece with intermediate support plate
P-D Centre Distance 197mm
Input Torque 350 Nm max.
Output Torque 4200 Nm max.
Gears Ratios in production Torque Limitations
I 3.9 260 Nm
II 2.269 295 Nm
III 1.435-1.522 350 Nm
IV 0.978-1.116 350 Nm
V 0.755-0.915 350 Nm
VI 0.622-0.767 350 Nm
RG 4.00 250 Nm
Final Drive 3.579-3.833-4.118-4.438
Maximum speed ratio in I 17.3
Spread 6.27
Synchronisers I-II-III 3-cone
IV-RG 2-cone
V-VI 1-cone
Weight 81 kg with hydraulic and lubrification oil
Table 2.1: C635 DDCT Main Mechanical characteristics
As shown in Figure 2.5 below, the 3-shaft transmission architecture is contained

in a 2-piece aluminium housing with an intermediate support plate for the shaft
bearings. This solution allows the positioning of the differential group closer to the
engine.
The gear set housing is characterised by reduced upper secondary shaft length, a
feature which is also necessary in order to ensure packaging in the lower segment
vehicles where the longitudinal crash beam imposes serious installation constrains.
The 6th gear is shared with the 4th while an eventual, non-powershift, 7th gear
9
would be shared with the 5th, with either increased or equal overall gear spread
compared to the 6-speed version. The most important feature of this transmission
in terms of packaging characteristics is the adoption of a coaxial pull-rod for the
actuation of the odd-gear clutch (K1), while the even gear clutch (K2) is actu-
ated with a rather conventional hydraulic Concentric Slave Cylinder (CSC). This
pull-rod is connected to a hydraulic piston actuator located on the rear face of the
transmission housing in a manner identical to the one adopted in the past in an
earlier Fiat Powertrain technical demonstrator. In fact, despite the optimised axial
dimension of the dry Dual Clutch Unit (DCU) the eventual addition of a second
actuator mechanism within the clutch housing would prohibit installation of the
transmission in the lower Segment vehicles. Finally, all synchroniser groups share
the same base elements and are identical to those of the Manual version. Logistics
and economical considerations favoured this solution which may not be necessary
in a DCT transmission. The four forks are guided on two rods. A specific form of
the rods allows production of a single part number, sheet metal, fork body welded
to the prong and driven by the hydraulic pistons. The MT and DCT forks differ
only in terms of prong shape.
Figure 2.5: Cross section of the C635 DDCT
10
2.2.1 Operation of the DDCT
The C635 DDCT is an automatically shifted transaxle, electronically controlled

with a dual-clutch system for torque conversion. The most important part of the
transaxle is the main shaft that is made up of two reciprocally coaxial driveshafts.
The transmission of the torque to the main shaft is done employing two dry clutches,
operated by two separate slave cylinders.
A basic concentric slave cylinder, placed under the gearbox casing, operates the
even-number gear clutch. Another slave cylinder, located at the back of the gearbox
casing, is used to operate the odd-number gear clutch. These two slave cylinders are
actuated and the gears are shifted by the hydraulic power unit which is controlled
by a special-purpose transmission control module.
The design of the transaxle is done in order to optimize the gear shift points,
as if driving with a manual gearbox, while delivering an uninterrupted drive torque
to the wheels. Another important benefit is guaranteeing smooth gear shifting
that is normally associated with an automatic transmission. The C635 DDCT
has the benefits of manual shifting without the complexity found in an automatic
transmission.
The shift from one gear to the next one is realized as follows: when acceler-
ating, the engine speed increases before reaching the point where the shift to the
next gear should be done. If the transaxle is manual, the driver depresses the clutch
pedal, moves the shift lever to the next position and finally releases the clutch.
Adopting this dry dual clutch transmission, even if the clutch for the current gear
remains engaged, the gears are moved to the next position by the hydraulic power
unit because the clutch for the next gears is not engaged. Actually, it is a sort of
preselection of the next gear before the clutch is engaged.
At the appropriate moment, the clutch for the current gear disengages and the
clutch for the next gear engages. Everything is done without interrupting torque
delivery to the wheels, considering that the disengagement of current gear clutch
and the engagement of the next gear clutch actually overlap. In order to shift from
an odd-number gear to an even-number one, it is only required that the clutches
are actuated so that one engages and the other one disengages.
The hydraulic power unit represents a key part of the entire transaxle. This
gives hydraulic energy for performing power-unit two functions. In primis, it selects
and engages the gears in the transaxle, and then it actuates the slave cylinders that
engage and disengage the clutches. The hydraulic power unit has an accumulator
11
very similar to the one present in the anti-locking brake system, that maintains
the operating pressure (435 psi). The hydraulic power unit primes itself before the
ignition is turned on, so that, when the front doors of the vehicle are open a slight
pump noise can be heard.
The C635 DDCT engages each gear in a sequence; a 1-2 shift is followed by
a 2-3 shift and so on. In certain conditions, the transaxle is allowed to skip gears,
so that a 6-4 downshift can occur when passing a vehicle on the freeway. This
ability allows a downshift of one, two or three gears, in order to meet the pedal
requests of the driver.
If the vehicle increase its speed when traveling downhill with a gear engaged
and the accelerator released, the clutch closes automatically, after reaching a preset
speed, and provides the engine brake function. The quantity of engine braking
transmitted through the clutch is usually based on the vehicle operating conditions.
2.2.2 Dry Dual Clutch Unit

The analysis of the model, divided into different subsystems, is directly related to
the knowledge of the different parts that make up the DDCT as it is. The torque
transmission is obtained, during the shifting phase, throught the engagement of the
closing clutch and the release of the opening one. The position sensor, integrated
in the rear-hydraulic piston actuator, controls the position of the clutch K1 that
is closed as in manual transmissions. The clutch K2 is in this phase open and
is controlled in force throught hydraulic pressure given by the Concentric Slave
Cylinder (CSC). [3] Both clutches act on a centre plate with characteristics related
to the need of thermal dissipation especially during the critical phases.
Figure 2.6: Dual Clutch Unit, Clutch Actuators and installation
12
2.2.3 Electro-hydraulic actuation system

The C635 DDCT clutches and gear shifting mechanisms are electro-hydraulically
operated through a dedicated hydraulic oil circuit. The reason for this choice was
the good system efficiency and compactness as well as Fiat Powertrain Technologies
long experience with similar systems in its various Automated Manual Transmission
applications. The system is composed by a hydraulic power unit (PU), consisting
of an electrically driven high pressure pump and accumulator, and an Actuation
Module (CAM) in which the control solenoid valves, gear shift actuators and sensors
are included. The maximum operating pressure of the system under normal driving
conditions does not exceed the 20 bar, even if in some particular and rare situations
it can reach 39 bar. The electric oil pump is driven by a Smart Drive Unit (SDU),
controlled through a Pulse Width Modulation (PWM) signal: in this way the pump
speed can be regulated in order to reduce noise or to increase the efficiency. The
pump’s duty cycle in urban driving rarely exceeds 25% and it is a clear proof of
the system’s efficiency when compared to other concepts that need continuous
operation of a mechanically driven hydraulic pump. This makes possible the use of
a brushed motor for the pump drive.
Figure 2.7: Hydraulic Power Unit and CAM
The clutch and gear actuation module (CAM), is made up of 4 distinct double
action pistons actuating the gear engagement forks, one “shifter” spool which
selects the piston to be actuated and 5 solenoid valves with 4 pressure proportional
(PPV) and one flow proportional (QPV). Two of the pressure proportional valves
actuate the gear engagement piston which is selected by the spool valve operated by
the third PPV. The fourth PPV is used for the control of the K2 clutch. The flow
proportional valve is used for the position control of the K1 clutch. The solenoid
valves are directly derived by those currently used in FPT’s AMT systems and,
therefore, adopt well proven technologies and guarantee robustness. The Actuation
Module also has 5 non-contact linear position sensors, one for each shifting piston
13
and one for the shifter spool, together with two speed sensors reading the speed of
the two primary shafts. One pressure sensor is used for the control of the K2 clutch
and one for the system pressure monitoring and control. The K1 clutch position
sensor, as explained before, is integrated in the clutch piston actuator placed at
the rear of the transmission. Figure below represents the hydraulic circuit of the
complete actuation system (CAS).
Figure 2.8: C635 DDCT complete actuation system
2.2.4 DDCT Control Unit

The C635 DDCT control strategies run in a multitasking environment so that the
Main Micro Controller resources are optimally managed. These strategies can be
grouped as:
• Actuator Control: exploits the high performance achievable with electro- hy-
draulic actuators. The main control strategies deal with:
– Engagement Actuators Control: the desired trajectories are realized by com-

manding the two relevant PPVs one against the other.
– Shifter Control: hydraulic power to the required engagement actuator is guar-

anteed by a fast and precise control of the shifter, obtained by commanding the
related PPV to push the shifter piston against the spring in order to reach the
desired position.
– Odd Gears Clutch Control: the normally closed clutch (K1), which is the clutch
of the first and of the reverse gear, is controlled by a position closed loop.
14
– Even Gears Clutch Control: the normally open clutch (K2) is controlled in
force with a pressure feedback signal delivered by one of the CAM sensors.
• Self-Tuning Control: many self-tuning controls are needed in order to com-

pensate for the various parameters’ drift and to guarantee the same high-level
calibrations to all vehicles. The main self-tuning control algorithms concern the
conversion of the requested clutch transmitted torque to K1 position and K2 pres-
sure.
• Launch and Gear Shift strategies: different modes of shift patterns in au-
tomatic and in manual mode are contemplated and are accomplished by specific
control strategies and calibrations on the engine side.
15
2.3 Micro slip

The Dry Dual Clutch is made up of two different clutches K1 and K2: the first
one is connected with the inner primary shaft and the odd gears, the second one is
linked with the outer primary shaft and the even gears. The center plate has two
different contact surfaces where the sliding problem is located.
Figure 2.9: DCT and clutches K1 and K2
The micro-slip problem affects both clutches at different times, so in this work
the design of the controller is referred only to one of the clutches. The clutch is
essentially composed by two rotating friction surfaces that are pressed against each
other by a normal force Fn .
The presence of friction at the contact surface realizes the transmission of the
clutch torque Cf . The effective torque that can be transmitted depends on the
value of Fn .
In the following Figure 2.10 it is possible to see a schematic representation of the
DDCT including the engine torque Cm and the torque given by the reaction of the
driveline Cr , the angular speeds are also represented (ωm ,ωp ).
Figure 2.10: Schematic representation of the DDCT
The three different phases in which the clutch can be in ordinary operating condi-
tions can be the following ones:
16
-Open clutch phase: the two disks are completely disengaged Cf = 0;
-Slipping phase: the difference between the engine speed and the primary shaft
speed is not zero ωd = ωm − ωp ;
-Closed clutch phase: the synchronization of the engine shaft and the primary shaft
is completed and the torque is fully transmitted Cf = Cm .
The micro-slip effect is related to the slipping phase and the objective of the
control design studied in this thesis is to guarantee a smooth slip between the
clutch disks. The control determines a better engagement of the clutch reducing
driveline oscillations and jerks and allowing a partial decoupling from the high
dynamics of the engine to obtain a better driving comfort.
The aim is that to design a feedback controller using the Linear Quadratic Integral
control in order to make the slipping speed ωd to track a given reference signal.
Another control target is that to bring ωsr that is the torsion velocity of the
transmission to zero.
2.4 State of art

The dry dual clutch engagement together with the micro-slip problem have been
already analysed in literature and different solutions have been developed in order
to realize the clutch slipping control and the reduction of torque oscillations.
In [4] a control with two independent PID controllers P1 and P2 is presented: the
engine speed ωe and the slipping speed are tracked independently as shown in the
figure below:
Figure 2.11: Decoupling control implemented in [4]
The system represents a two input and two output decoupled control.
17
Both models, the clutch and the vehicle ones, provide a feedback in terms of
speed in order to guarantee a smooth clutch engagement and a better reference
tracking.
Another control is proposed in [5]: only the clutch torque is here used as a control
variable and the engine torque is considered a non- controllable signal. The engine
torque is subjected to a lot of constraints that are the result of a trade-off between
fuel efficiency, emissions and traction control. These constraints do not allow to
adoperate the engine torque as a control variable.
The throttle pedal position xp determines the output engine torque given by the
ECU, engine control unit, that is used by the vehicle model in order to compute
the angular velocity ωe and x0 .
Figure 2.12: MPC control from [5]
The total engagement time tf , that is computed in function of the throttle pedal
position, is used to calculate the time control horizon tr . The clutch torque is then
computed throught the MPC control strategy solving the optimal control problem
with a suitable cost function.
In [6], in order to prevent undesirable effects about non linearities and saturation,
a torque rate limitation strategy is implemented for the gearshift control. The
controlled clutch is actuated only in one direction: the oncoming clutch normal
force can be only increased or held, while the off-going clutch normal force can be
decreased or held. By measurements of the two clutches engine speeds, the target
engine speed is computed and the engine torque controller is implemented as a
proportional-integrative PI. The engine torque is controlled also by the feedback of
the speed tracking error.
Figure 2.13: Engine speed and torque control [6]
18
In article[7] the author suggests a controller based on the different five AMT
operating conditions: engagement, slipping opening, synchronization, go to slip
and slipping closing. In this case the measurement of engine speed, clutch speed
and the estimation of the transmitted torque give the possibily to realize decoupled
and cascade feedback loops. In this case the control target is that to reach the
engagement phase in a short amount of time avoiding the engine stall.
Figure 2.14: General scheme presented in [7]
2.5 Model configuration

In order to describe the model into details it is necessary to analyze all the elements
connected to the Dry Dual Clutch used by Centro di Ricerche Fiat.
In the Figure below, a general scheme of the model is presented with the following
components:
- the engine that gives the torque Cm ;
- the torque CLoad that includes the effects of the air;
- the reference generator that gives the reference angular speed ωref for the clutch
slipping speed ωd ;
- the actuator of the clutch torque Cf given by the controller;
- DDCT system that includes the clutch and the transmission.
All these elements are described in details and their implementation in Matlab
Simulink environment is shown.
19
Figure 2.15: Overall model of the DDCT system
2.5.1 Actuators
The actuators are an important part of the system because they allow the trans-
mission of the two torques Cm and Cf :
Figure 2.16: Actuators for Cm and Cf
The actuator block AC m (s) contains the transfer function for the engine torque
actuator that is:
14.05
ACm (s) = e−0.02·s · (2.1)
s + 14.15
The block SysClutch(s) represents the behavior of the actuator that realizes the
requested value of the transmitted torque Cf :
25.305(s + 138)(s2 − 161.3s + 161.3s + 3.134 · 104 )

SysClutch(s) = (2.2)
(s2 − 82.23s + 2563)(s2 − 134.1s + 4.27 · 104 )
A delay block is also present (ActuatorDelay) with a delay of 20 ms.

20
2.5.2 Engine model

In this section the model of the engine is presented. This input of the plant is
computed by the LQI controller throught the ωref that has the following behavior:
Figure 2.17: Reference angular velocity
The state-feedback control is of the form
u = −k[x; xi ] (2.3)
where xi is the integrator output and u is given by the two torques Cm and Cf .
2.5.3 Load
The load contribution is given by three different terms:
- Aerodynamic resistance force:
Fa = 0.5ρa Af Ca (va + vv )2 (2.4)
Where:
- ρa is the air density;
- Af is the frontal area of the vehicle;
- Ca is the aerodynamic drag coefficient;
- va is the wind speed;
21
- vv is the vehicle speed.
- Rolling resistance force:

Fr = mv gµr cos(β) (2.5)
With the following terms:
- mv is the vehicle mass;
- g is the gravity acceleration;
- µr is the rolling friction coefficient;
- β is the slope of the road.
- Uphill driving force: resistance force when driving on a non-horizontal surface. It

is given by:
Fg = mv gsin(α) (2.6)
The final expression of the total load torque, considering the wheel radius rw is:
CLoad = (Fa + Fr + Fg )rw (2.7)
22
2.6 Hybrid electric vehicles with P2 architecture

This section is focused on the explication of the implementation of the parallel
hybrid vehicle model.
Accordingly to the analyzes carried out by Yalian Yang researchers, Xiaosong Hua,
Huanxin Pei, Zhiyuan Peng reported in the scientific article "Comparison of power-
split and parallel hybrid powertrain architectures with a single electric machine:
Dynamic programming approach ", among the various parallel architecture, the P2
is advantageous in terms of fuel saving with respect to the P1 architecture.
The parallel hybrid vehicle is characterized by having a thermal engine and an
electric motor constrained to the wheels, the power and torque provided for the
propulsion is the result of an excellent contribution of both.
Unlike other hybrids, it only requires a single motor, a standard transmission and
two clutches. This clutch is used to operate the gas engine and the electric motor
separately without the need for two separate engines. This is where the cost saving
comes in.
With the P2 technology, the cost can be reduced by more than one-thirds. While
P2 hybrid cars have clear advantages, they also face a problem when starting the
engine when running down the road in electric mode. Different car manufacturers
are trying to find ways to solve this problem.
The fuel efficient hybrid cars have surely secured a place for the richer class of
people but its high price has kept the hybrids far from the reach of most new car
buyers. The P2 technology is sure to bring its price down and make the hybrids
affordable enough for widespread use.
Key advantages include:
- The major advantage is its low cost which will open the hybrid market for
all customers.
- The P2 system gives better regenerative braking.
- The P2 technology does not require a torque converter and hence its cost is
reduced further.
- They use high power lithium-ion batteries which are smaller, lighter and have a
low cost.
- Vehicle Load is reduced by using only single motor and other lighter parts.
Some key limitations of P2 Hybrids are related to:

23
- Inability to start the engine while running on electric mode only.
- The driving may not be as smooth with the manual transmission and their
might be more wear and tear because of the complex gear system.
24
Chapter 3
Model Analytical
description
3.1 Introduction
In order to design a Linear Quadratic Integral controller LQI, together with the
other techniques needed in order to compensate the delays and non-linearities
of the system, the computation of the state-space model is needed. The general
driveline scheme, that is a linear model, gives a good representation of the system
that is essentially made up of: the engine, the crankshaft, the clutch, the main
shaft, the gearbox, the secondary shaft, the differential and the wheel shaft.
Figure 3.1: General driveline scheme
From this basic scheme it is possible to derive the specific DDCT driveline scheme
where it is possible to observe the odd gears and the even ones.
In order to derive the linear model of the driveline three assumptions are made:
25
Model Analytical description
Figure 3.2: DDCT driveline scheme
1) The main shaft is perfectly rigid;
2) The wheels motion is pure rolling;
3) The two branches of the driveline are perfectly simmetric.
Starting from these considerations,the following scheme of the driveline can be

built, considering linear conditions:
Figure 3.3: Reduced linear driveline scheme
3.2 State Space representation

The values and meaning of the parameters shown in Figure 3.3 are explained in
the table below:
26
Component Symbol Value

Torsional damping coefficient csa 105 (Nm/(rad/s))
Torsional stiffnes coefficient ksa 6719 (Nm/rad)
Gear ratio (second gear) τ 10.473
Wheel radius rW 0.308 (m)
Wheel inertia JW 3 (kgm2 )
Vehicle mass M 1550 (kg)
2
Vehicle inertia Jv M · rW + JW (kgm2 )
Motor damping coefficient cm 105 (Nm/(rad/s))
Motor inertia Jm 0.28 (kgm2 )
Table 3.1: DDCT main parameters
Imposing the rotational equilibrium for the system in Figure 3.3, the system of
differential equations is obtained:
Jm ω̇m = Cm − cm ωm − Cf (3.1)
1
Jp ω̇p = Cf + (−csa ωsr − ksa θsr ) (3.2)
τ
Jv ω̇r = csa ωsr + ksa θsr − CLoad (3.3)
θ̇sr = ωsr (3.4)

With:
θsr = θsa − θr (3.5)
ωp
ωsr = − ωr (3.6)
τ
Putting together all the equations the final form of the equations is:
1 cm
ω̇m = (Cm − Cf ) − ωm (3.7)
Jm Jm
Cf 1 csa ωp cprim ωp
ω̇p = + (− + csa ωr − ksa θsr ) − (3.8)
Jp τ Jp τ Jp
csa ωp csa ωr ksa θsr CLoad
ω̇r = − + − (3.9)
Jv τ Jv Jv Jv
27
ωp
− ωr
θ̇sr = (3.10)
τ
In order to build throught these last equations the state space representation, the
input vector, the output vector and states have to be defined:
- The input vector is u(t) = [Cm (t), Cf (t), CLoad (t)]T
- The state vector is x(t) = [ωm (t), ωp (t), ωr (t), θsr (t)]T
- The output vector is y(t) = [ωd (t), ωsr (t)]T where ωd (t) = ωm (t) − ωp (t) is
ωp (t)
the slipping velocity and ωsr (t) = − ωr (t) is the torsion speed.
τ
Considering the general state space representation:
ẋ(t) = Ax(t) + Bu(t)
y(t) = Cx(t) + Du(t)
with:
cm
 
− 0 0 0 
 Jm
csa cprim csa ksa 


 0 −( 2
+ ) − 
Jp τ Jp Jp τ Jp τ 
 
A= (3.11)

csa csa ksa 
 0 −
 

 Jv τ Jv Jv 
1
 
 
0 −1 0
τ
1 
1 
J − 0 
 m Jm
1

 
 0 0 
B=
 Jp

 (3.12)
 1
0 0 − 
 

 Jv 
0 0 0
1 −1 0 0
 
C= 1  (3.13)
0 −1 0
τ
28
C D
0 0 0
D= (3.14)
0 0 0
ωp (t)
In order to change the states vector to x(t) = [ωm (t), ωm (t) − ωp (t), −
τ
ωp (t)
ωr (t), θsr (t)]T with ωd (t) = ωm (t) − ωp (t) and ωsr (t) = − ωr (t) the following
τ
transformation matrix Tx is involved:
 
1 0 0 0
1 −1 0 0
 
Tx = 

1 
(3.15)
0 −1 0


 τ 
0 0 0 1
And the matrices become:
AÍ = Tx ATx−1
B Í = Tx B
1 0 0 0
 
0 1 0 0
with C Í =   choosing the outputs equal to the states.

0 0 1 0

0 0 0 1
3.3 Complete model with actuators

In this section the full implementation of the model is explained where both the
engine and the clutch actuators are considered. The final form of the state space
representation used in order to implement the Kalman observer together with the
LQI controller is here computed.
From the transfer functions of the actuators, built in Simulink, a linearized state
space model is derived.
For the clutch actuator the general state space equations are:
ż(t) = F z(t) + Gv(t)
u(t) = Hz(t)
And for the engine actuator:

29
ġ(t) = Lg(t) + M r(t)
uÍ (t) = N g(t)
Figure 3.4: Complete model with actuators
At this point a new complete state space representation can be built with 9 states,
where the states determined by the actuators transfer functions need to be predicted
in order to control the system.
 
x C D
v
X= z  U=
 
r
g
x1
 
 .. 
     
 
ẋ A B(: ,2)H B(: ,1)N  x n zeros(4,1) zeros(4,1) C D
v

  
 ż  = zeros(4,4) F zeros(4,1) 
 z1  +  G zeros(4,1)
    
r
ġ zeros(1,4) zeros(1,4) L  .. 
 
0 M
 zn 
 
g
Ẋ(t) = ĀX(t) + B̄U (t)
The matrices concerning the state space models for the actuators are computed by
Simulink using the command ”Linmod”. This complete model is used in order to
improve the performance of the LQI control associating it with other techniques in
order to guarantee an optimal management of the actuators effects and the delays
related to them.
30
The starting model taken into consideration is realized without actuators and delays
in order to verify the performance of the controller in the initial conditions. The
following techniques are then associated to the LQI controller to compensate the
delays of the real model:
- LQI technique with and without the delay of the clutch actuator;
- LQI linearization of the entire nonlinear model and actuator delay;
- LQI control with Smith Predictor;
- State space representation of the actuators and the Plant with LQI technique;
- State space representation of the actuators and the Plant with LQI technique and
Kalman observer.
31
Chapter 4
LQI technique
4.1 Linear Quadratic regulator

The optimal control theory is related to operating a dynamic system at minimum
cost. The situation where the system dynamics are represented by a set of linear
differential equations and the cost is described by a quadratic function is referred
to the LQ problem. In this case, one of the main results in the theory is that the
solution is provided by the linear–quadratic regulator (LQR), a feedback controller
whose equations are given below.
The LQR algorithm is essentially an automated way of finding an appropriate
state-feedback controller.
4.1.1 Finite-horizon, continuous-time LQR

The finite horizon, linear quadratic regulator (LQR) for a continuous-time linear
system deined on t ∈ [t0 , t1 ] is given by [8]:
ẋ = Ax + Bu (4.1)
Ú t1
J = xT (t1 )F (t1 )x(t1 ) + (xT Qx + uT Ru + 2xT N u) dt (4.2)
t0
The feedback control law that minimizes the value of the cost is:
u = −Kx with K = R−1 (B T P (t) + N T ) (4.3)

P is then found by solving the continuous time Riccati differential equation:
AT P (t) + P (t)A − (P (t)B + N )R−1 (B T P (t) + N T ) + Q = −Ṗ (t) (4.4)

32
LQI technique
with the boundary condition: P (t1 ) = F (t1 ).
The first order conditions to find Jmin are:
- State equation : ẋ = Ax + Bu
- Co-state equation: −λ̇ = Qx + N u + At λ
- Stationary equation: 0 = Ru + N T x + B T λ
- Boundary conditions: x(t0 ) = x0 and λ(t1 ) = F (t1 )x(t1 )
4.1.2 Infinite-horizon, continuous-time LQR

For a continuous-time linear system described by:
ẋ = Ax + Bu (4.5)
with a cost function defined as:

Ú ∞
J= (xT Qx + uT Ru + 2xT N u) dt (4.6)
0
The feedback control law that minimizes the value of the cost is:
u = −Kx with K = R−1 (B T P + N T ) (4.7)

P is then found by solving the continuous time algebric Riccati equation:
AT P + P A − (P B + N )R−1 (B T P + N T ) + Q = 0 (4.8)
4.1.3 Finite-horizon, discrete-time LQR

For a discrete-time linear system described by:
xk+1 = Axk + Buk (4.9)
with a performance index defined as:

N −1
J = xT N QxN + (xTk Qxk + uTk Ruk + 2xTk N uk )
Ø
(4.10)
k=0
33
LQI technique
the optimal control sequence minimizing the performance index is given by:
uk = −Fk xk with Fk = (R + B T Pk+1 B)−1 (B T PK+1 A + N T ) (4.11)

and Pk is found iteratively backwards in time by the dynamic Riccati equation:
Pk−1 = AT Pk A − (AT Pk B + N )(R + B T Pk B)−1 (B T Pk A + N T ) + Q (4.12)
from terminal condition PN = Q.

Consider that uN is not defined, since x is driven to its final state xN by AxN −1 +
BuN −1 .
4.1.4 Infinite-horizon, discrete-time LQR

For a discrete-time linear system described by:
xk+1 = Axk + Buk (4.13)
with a performance index defined as:

∞
(xTk Qxk + uTk Ruk + 2xTk N uk )
Ø
J= (4.14)
k=0
the optimal control sequence minimizing the performance index is given by:
uk = −F xk with F = (R + B T P B)−1 (B T P A + N T ) (4.15)

and P is the unique positive definite solution to the discrete time algebraic Riccati
equation (DARE):
P = AT P A − (AT P B + N )(R + B T P B)−1 (B T P A + N T ) + Q (4.16)
One of the ways to solve the algebraic Riccati equation is by iterating the dynamic
Riccati equation of the finite-horizon case until it converges.
4.2 Incorporating integral action in LQR

The LQR control law fundamentally gives a multivariable proportional regulator.
An integral action has been incorporated in the LQR controller by augmenting
system with error in the state variable. One new state is added to the space model
of the system. In this way a Linear Quadratic Integral control is obtained with the
following general scheme:
34
LQI technique
Figure 4.1: General LQI scheme
In order to obtain steady state zero tracking error it is possible to introduce, within
the system equations, the information of the discrete time integral of the tracking
error:
q(k + 1) = q(k) + T (r(k) − y(k)) = q(k) + T e(k) (4.17)

And considering the SISO case, by adding to the system state the discrete time
integral of the tracking error an augmented state is introduced:
C D
q(k)
xtot = ∈ Rn+1 q(k) ∈ R (4.18)
x(k)
The system equations become:

C D C DC D C D C D
q(k + 1) 1 −T C q(k + 1) 0 T
xtot (k + 1) = = 1x1 + u(k) + r(k) (4.19)
x(k + 1) 0nx1 A x(k + 1) B 0
C D
è é q(k)
yk = 0k C (4.20)
x(k)
With the new matrices Atot , Btot , Ctot . The augmented system is reachable if the
original one is.
The selected control law is modified as:
è é
u(k) = −Kxtot (k) with K = Ki Ko ∈ Rn+1 (4.21)
C D
è é q(k + 1)
u(k) = − Ki Ko (4.22)
x(k + 1)
The system guarantees zero steady state tracking error when r(k) is constant.
Since the feedback law guarantees asymptotic stability, at steady state xtot (k)
will reach constant value. Since xi (k) is the integral of the tracking error, the
equilibrium should be for e(k) = 0, thus at steady state r(k) = y(k).
Finally the design of a LQ state feedback control law with the integral state is
35
LQI technique
performed as the canonical case, using as system matrices Atot , Btot .

In Matlab the following syntax is used [K, S, e] = lqi(SY S, Q, R, N ) where S is the
solution of the Riccati equation and K is the optimal gain matrix with a state-space
model SYS for the plant and weighting matrices Q, R, N.
The problem data must satisfy:
- The pair Atot , Btot is stabilizable;
- R > 0 and Q − N R−1 N T ≥ 0;
- (Q − N R−1 N T , Atot − Btot R−1 N T ) has no unobservable mode on the imaginary

axis (or unit circle in discrete time).
In this specific case a simplified transmission model has been used, with clutch
actuator, dynamic engine model and elasticity of transmission trees. A MIMO
control architecture project with LQI technique with trajectory feedback (reference
– state variable) is here adopted:
o Controls: engine torque and clutch torque;
o State variables to adjust: difference in engine-primary speed, derivative of

engine-primary speed difference, semi-axis twisting speed, semi-axis twist angle,
actuators’ state variables.
o Project Plant has as equivalent state variables the difference in engine-primary

speed, angle and twisting speed of the semiaxes.
o Control objective: get a great compromise between the pursuit of the engine-
primary delta reference and the vehicle driveability.
o Uncertainties to consider: clutch and engine implementation delays, clutch

transmission, engine implementation, vehicle mass, semi-axis twist elasticity.
o Evaluate the use of a predictor (backing of the predicted variables) for op-
timal management of implementation delays.
o Evaluate a state observer development to estimate unmeasured states (actu-

ators states).
o 10ms control discretization and performance loss assessment between continuous

and discrete control.
36
LQI technique
The general architecture used for the control is the following one:
Figure 4.2: General architecture for the control
4.3 Optimal management of the delays

The main activity of this thesis project, realized in collaboration with CRF (Centro
Ricerche Fiat), consists in developing innovative feedback control algorithms for
hybrid and electric vehicles, related in particular to the control of the dual clutch
transmission during its slipping phase. In this regard, the gear shifting phase is
relevant because the clutch disks, when closing, do not get immediately the same
angular velocity: the main control target is that of making possible a smooth
convergence to zero of the difference between the combustion engine and electric
motor speeds (delta speed). One of the problems related to the realization of an
optimal control is the management of the delays of the actuators that creates some
difficulties in the tuning of the Q and R matrices of the LQI controller. In order to
solve these difficulties, different control strategies are here developed in order to
realize an optimal management of the delays.
The main control strategies developed are the following ones:
-LQI technique with and without the delay of the clutch actuator;
37
LQI technique
-LQI linearization of the entire nonlinear model and actuator delays;
-LQI control with Smith Predictor;
-State space representation of the actuators and the Plant with LQI technique;
-State space representation of the actuators and the Plant with LQI technique and
Kalman observer.
4.3.1 Smith Predictor

The field of interest is the control of systems with delays. Feedback control has
several familar advantages - one important one is the decreased sensitivity of a
closed-loop negative feedback system to variations in its parameters.
The Smith predictor is a model-based controller that is effective for processes with
long dead time. It has an inner loop with a main controller that can be simply
designed without the dead time. The effects of load disturbance and modeling
error are corrected through an outer loop. The Smith predictor can also be used
for processes with significant non-minimum phase dynamics and for high order
systems that exhibit apparent dead time.
Smith’s predictor is a controller for dynamic systems with pure delay time dominant,
i.e. in systems where delay far outweighs the time constant (T >> τ ). Delay is a
time quantity described as the time between the effect occurs and that effect is
taken into account to change the system. It is present in almost all real systems.
In most analyses, a frequent error is the non-consideration of this amount, however
in cases where this is presented with an irrelevant value compared to the time
constant it is possible to overlook it, approximating it to a system without delays.
A very high delay, on the other, creates oscillating phenomena creating problems
of stability.
Smith predictor, as mentioned earlier behaves like a controller, is his f.d.t. is
composed of the f.d.t. of the plant Gp(s), the delay component (e−sT ) and the
f.d.t. of a controller designed for the system in feedback without delay Gcr(s). The
latter controller can be designed according to the classic rules (u -Kx) or a PID,
etc. [9]
The transfer function is the following one:
Gcr(s)
Gc(s) = (4.23)
1 + Gcr(s)Gp(s)[1 − e−sT ]
38
LQI technique
The design of a Gc(s) controller for a system whose plant is Gp(s) and there is a
pure T seconds delay in the feedback loop, as shown in the following figure is such
complicated:
Figure 4.3: Controller without Smith Predictor
On the other hand, a less direct technique is that of Smith’s predictor, which goes
to consider a second system (as shown in the following figure) where the delay is
placed outside the control ring allowing the design of a second Gcr(s) controller.
This one can be designed as a PID regulator, through a feedback of type -Kx, etc.
Figure 4.4: System considered for Smith Predictor
However, the two systems are not fully equivalent, so the design of Gcr(s) is not
the same as designing Gc(s). This problem is solved by equaling the two transfer
functions, which is to derive the value of Gc(s) from Gcr(s).
Gc(s)Gp(s)e−sT
f.d.t.1 = (4.24)
1 + Gc(s)Gp(s)e−sT
Gcr(s)Gp(s)e−sT
f.d.t.2 = (4.25)
1 + Gcr(s)Gp(s)
f.d.t.1 = f.d.t.2
Gc(s)Gp(s)e−sT Gcr(s)Gp(s)e−sT
= (4.26)
1 + Gc(s)Gp(s)e−sT 1 + Gcr(s)Gp(s)
39
LQI technique
Gc(s) Gcr(s)
−sT
= (4.27)
1 + Gc(s)Gp(s)e 1 + Gcr(s)Gp(s)
Gcr(s)(1 + Gc(s)Gp(s)e−sT )
Gc(s) = (4.28)
1 + Gcr(s)Gp(s)
Gcr(s)(1 + Gc(s)Gp(s)e−sT ) Gcr(s)

Gc(s)(1 − )= (4.29)
1 + Gcr(s)Gp(s) 1 + Gcr(s)Gp(s)
1 + Gcr(s)Gp(s) − Gcr(s)Gp(s)e−sT Gcr(s)

Gc(s)( )= (4.30)
1 + Gcr(s)Gp(s) 1 + Gcr(s)Gp(s)
Gc(s)(1 + Gcr(S)Gp(s)[1 − e−sT ]) = Gcr(s)

Gcr(s)
Gc(s) = (4.31)
1 + Gcr(s)Gp(s)[1 − e−sT ]
Figure 4.5: Smith Predictor
4.3.2 Kalman state observer

In the theoretical analysis so far, the state is always been supposed to be measured.
In some cases this could not be possible and a state observer (SO) is needed in
order to provide an estimate x̂(k) of the system states x(k). The estimation is
generally based on the input/output measurements of the system to be controlled.
The general structure of an observer is reported below:
40
LQI technique
Figure 4.6: General observer scheme
In statistics and control theory, Kalman filtering, also known as linear quadratic
estimation (LQE), is an algorithm that uses a series of measurements observed
over time, containing statistical noise and other inaccuracies, and produces esti-
mates of unknown variables that tend to be more accurate than those based on a
single measurement alone, by estimating a joint proba- bility distribution over the
variables for each time frame.
In order to design a Kalman filter, the following hypotheses are introduced: [10]
- the initial state x0 is a Gaussian variable such that:

E[x0 ] = 0
E[x0 xT0 ] = X0
X0 ≥ 0
where E[ ] indicates the expected value.
- the process disturbance w(k) and the measurement error v(k) are white noises
with zero mean value and known variance which are uncorrelated with each other:
E[w(k)] = 0, E[v(k)] = 0
E[w(k)w(k)T ] = wQ̃, E[v(k)v(k)T ] = v R̃
E[v(k)w(k)T ] = E[x0 v(k)T ] = E[x0 w(k)T ] = 0
The covariance matrix of the error is defined as:

P̃ (k|k − 1) = E[ee (k|k − 1)ee (k|k − 1)T ] and P̃ (0| − 1) = P̃0 (4.32)
41
LQI technique
The gain observer L is determined by minimizing a quadratic expression of the

error covariance:
minL (ψ T P̃ (k + 1|k)ψ) (4.33)
where ψ is a generic vector.
The gain L derives from the solution to the previous minimization problem:
L = AP̃ (k|k − 1)C T + [C P̃ (k|k − 1)C T + R̃]−1 (4.34)
The expression P̃ (k|k−1) comes from the solution of the Difference Riccati Equation
(DRE):
P̃ (k + 1|k) = AP̃ (k|k − 1)AT + Q̃ − AP̃ C T [C P̃ (k|k − 1)C T + R̃]−1 C P̃ AT (4.35)
Q̃ and R̃ are weight matrices that can me conveniently made.

The matrices Q̃ and R̃ are related to the information regarding the noise. The R̃
diagonal elements are given by the variance of the output errors. The definition
of Q̃ is more complicated because the process disturbances are often unknown.
Therefore, more frequently the matrices Q̃ and R̃ are considered as two project
parameters.
4.4 LQI in Simulink

Matlab provides Simulink blocks for simulations. In this way it is possible to adjust
the behaviour of the controller by varying its weights and constraints at run time.
In this thesis the controller is designed as follows:
Figure 4.7: LQI controller scheme
The controller computes an optimal state-feedback control law for the tracking
loop shown above.
Different blocks can be recognized:
42
LQI technique
- The reference input that is a a reference signal for the delta speed ωd = ωm − ωp ;
ωp (t)
- The vector of the states that in general is: x(t) = [ωm (t), ωm (t) − ωp (t), −
τ
ωr (t), θsr (t)]T even if in the full model the states of the actuators are added;
- The two inputs of the model u1 and u2 that are the engine torque and the
clutch torque;
- The gain K given by the controller implementation throught the Matlab command
lqi.
4.5 Plant without actuators and delays

In a first approach the LQI technique is taken into account without considering the
non-linearities introduced by the transfer fuctions of the actuators. The simplified
ωp (t)
plant model is considered with four states x(t) = [ωm (t), ωm (t) − ωp (t), −
τ
ωr (t), θsr (t)]T .
Tuning LQ regulators implies choosing the weight matriches Q and R involving
some kind of trial and error procedure.
These matrices are chosen as diagonal matrices so that for a system with n states
and p controls, n+p parameters have to be defined.
The diagonal values qjj and rjj of Q and R respectively are chosen according to
the relative importance of each state and control variable, bearing in mind that
qjj ≥ 0 and rjj > 0.
In principle, a first choice of the diagonal elements of Q and R is performed so that
all the state and input components values appear in the cost function with almost
the same order of magnitude.
After this first iteration, values are modified to impose the desired performance.
Anyway the important aspect is the relative value of a single weighting coefficient
with respect to the others.
The scheme of the plant and the general scheme including the controller are shown
below together with the main results concerning the convergence of the delta speed
ωd and the drive shaft torsion speed and angle ωsr and θsr .
The control can be discretized at T=10 ms in order to verify if it implies a loss of
performance, that in this case remains more or less the same.
43
LQI technique
Figure 4.8: Plant without actuators
Figure 4.9: General scheme with controller
44
LQI technique
500
400
Slipping speed [RPM]

300
200
100
-100
2 2.5 3 3.5 4 4.5 5
Time (s)
Figure 4.10: Delta speed behavior without actuators
2
sr
1.5 sr
Drive shaft torsion speed [RPM]
0.5
-0.5
-1
-1.5
-2
2 2.5 3 3.5 4 4.5 5
Time (s)
Figure 4.11: Drive shaft torsion speed and angle
45
LQI technique
500
400

300
200
100
-100
2.5 3 3.5 4 4.5 5
Time (s)
Figure 4.12: Delta speed behavior without actuators T=10ms
2
sr
1.5 sr
0.5
-0.5
-1
-1.5
-2
2.5 3 3.5 4 4.5 5
Time (s)
Figure 4.13: Drive shaft torsion speed and angle T=10 ms
46
LQI technique
200
Combustion Engine+Electric motor
Clutch
150
100
50
Torque [Nm]
-50
-100
-150
-200
2.5 3 3.5 4 4.5 5
Time (s)
Figure 4.14: Total engine and clutch torque T=10 ms
The controller performance is very good, the control targets are reached, but the
model is different from the reality because all the effects of the actuators and their
delays are here neglected.
In the following section more realistic models will be described together with the
control strategies adopted in order to compensate the delays.
In order to evaluate the goodness of a method, the convergence speed of the pri-
mary output is not the only parameter to take into account: the amplitude of the
oscillations is an important parameter too. In particular, the time interval between
the starting control time (ts = 2s) and the time tf when the slipping speed ωd
reaches the null value, is considered. With respect to the oscillations, the standard
deviation σ is considered [11]: it is a measure of the variation/dispersion of a set of
data with respect to the mean value. A low standard deviation indicates that the
data points tend to be close to the mean of the set, while a high standard deviation
indicates that the data points are spread out over a wider range of values. The
standard deviation formula is the following:
ñq
N
i=1 (xi − µ)
σ= (4.36)
N
47
LQI technique
Another index used in order to evaluate the performance of a certain control

strategy is the Root Mean Square Error.
The root-mean-square deviation (RMSD) [12] or root-mean-square error (RMSE) is
a frequently used index used to measure the differences between values that can be
population values or samples, predicted by a model and the values observed. The
root mean sqaure error represents ÍÍ the square root of the second sample moment of
the differences between predicted values and observed values or the quadratic mean
of these differencesÍÍ . These quantities are called errors (or prediction errors) when
computed out-of-sample and are considered residuals when the calculations are
performed over the data sample that was used for estimatio. The RMSE is used to
evaluate the magnitudes of the errors in predictions for various times into a single
measure of predictive power. This is a measure of accuracy and scale-dependent
used to compare errors of different models for a particular dataset and not between
datasets. This index is always non-negative, and a null value would indicate a
perfect fit to the data. Usually, a lower value is better than a higher one. It is
not valid to do comparisons across different types of data because the measure is
dependent on the scale of the numbers used. RMS deviation is ÍÍ the square root of
the average of squared errorsÍÍ . The consequence of each error on RMSD depends
on the size of the squared error so that larger errors have a very large effect on
RMSD. It is important to remind that the Root Mean Square error is sensitive to
outliers.
48
Chapter 5
Tuning and Simulation

results
5.1 LQI technique with actuators and delays

In order to analyse a more realistic model, the engine and clutch actuators are in-
troduced in before the plant with their respective delays. In this case the actuators
transfer functions are not included in the plant model but the model itself is fed
with the two inputs Cm and Cf after the actuation:
Figure 5.1: Plant fed with the actuated inputs
49
Tuning and Simulation results
500
400

300
200
100
-100
2.5 3 3.5 4 4.5 5
Time (s)
Figure 5.2: Delta speed behavior
2
sr
1.5 sr
0.5
-0.5
-1
-1.5
-2
2.5 3 3.5 4 4.5 5
Time (s)
Figure 5.3: Drive shaft torsion speed and angle
50
200
Clutch
150
100
50
Torque [Nm]
-50
-100
-150
-200
2.5 3 3.5 4 4.5 5
Time (s)
Figure 5.4: Total engine and clutch torque
As it is possible to observe in the figures above, the performance of the controller is

deteriorated because of the introduction of the actuators with the respective delays.
A new tuning of the weight matrices Q and R is here necessary.
Many processes involve dead times, also referred to as transport delays or time
lags. Controlling such processes is challenging because delays cause linear phase
shifts that limit the control bandwidth and affect closed-loop stability.
The Q and R weight matrices used to obtain the result are the following ones:
 
0 0 0 0 0
0 0.5 0 0 0 
 
 
0 0 0 0 0 
Q= 
0 0 0 0 0 
 
0 0 0 0 30
C D
0.2 0
R=
0 0.2
The results are also derived in discrete time with T=10 ms.
51
500
400

300
200
100
-100
2.5 3 3.5 4 4.5 5
Time (s)
Figure 5.5: Delta speed behavior with T=10 ms
2
sr
1.5 sr
0.5
-0.5
-1
-1.5
-2
2.5 3 3.5 4 4.5 5
Time (s)
Figure 5.6: Drive shaft torsion speed and angle with T=10 ms
52
200
Clutch
150
100
50
Torque [Nm]
-50
-100
-150
-200
2.5 3 3.5 4 4.5 5
Time (s)
Figure 5.7: Total engine and clutch torque with T=10 ms
As it is possible to observe from the graphs above, the discretization of the control
at T= 10 ms, deteriorates the performance, also because of the introduction of the
actuators and the delays in the model.
In order to solve the problem, other techniques are introduced together with the
Linear Quadratic Integral control, the actuators are involved into the plant model
and predictors for their states are then introduced to observe them.
53
5.2 Linearization of the actuators and delays

The effects of the delays of the engine and clutch actuators in the control are
minimized linerizing the entire model plant:
Figure 5.8: Model plant to be linearized
The linearization is computed throught the Matlab command linmod that computes
a linear state-space model by linearizing each block in a model individually.
This command obtains linear models from systems of ordinary differential equations
described as Simulink models. Inputs and outputs are denoted in Simulink block
diagrams using Inport and Outport blocks.
The default algorithm uses preprogrammed analytic block Jacobians for most blocks
which should result in more accurate linearization than numerical perturbation of
block inputs and states.
The purpose of the linmod function is to obtain a linearized model of a nonlinear
system. Linear approximations are much more desirable to work with since it
is possible to predict the behavior of linear systems. Nonlinear systems exhibit
behavior which can be radically different for different inputs, making them difficult
to control. This is accomplished by approximating the behavior of a nonlinear
system in a limited range of operation. This range of operation may be considered
the operating point of the system being linearized. Dealing with nonlinearity in a
system is not an easy task. The difficulty in linearizing a nonlinear system increases
with the amount of nonlinearity. In other words, systems described by higher order
polynomials are more difficult to linearize than systems which can be represented
by lower order polynomials.
The limit of this kind of control is that the state space representation derived
contains also the states of the actuators that are not measured: in order to solve
this problem other techniques are introduced to observe them.
The performance in terms of delta speed behavior, drive shaft torsion speed and
engine and clutch torque is shown below:
54
500
400

300
200
100
-100
2 2.5 3 3.5 4 4.5 5
Time (s)
Figure 5.9: Delta speed behavior with linearization
2
sr
1.5 sr
0.5
-0.5
-1
-1.5
-2
2 2.5 3 3.5 4 4.5 5
Time (s)
Figure 5.10: Drive shaft torsion speed and angle with linearization
55
200
Clutch
150
100
50
Torque [Nm]
-50
-100
-150
-200
2 2.5 3 3.5 4 4.5 5
Time (s)
Figure 5.11: Total engine and clutch torque with linearization
As it is possible to observe from the figures above, the approach of delta speed to
zero is not abrupt and undesired oscillations are avoided.
After ωd reaches the zero value, there are not oscillations of the torsion speed and
angle.
The control is also performed discretizing at T=10 ms.
The discretization seems not to cause any loss of performance of the control, even if
an accurate comparison will be shown later, calculating the values of the standard
deviation and Root Mean Square Error together with the convergence time ∆t.
The performance of this technique is very good except for the fact that the control
of the states of the actuators that cannot be measured is here done throught a trial
and error procedure.
After the linearization,an attempt for reducing the computational complexity of
the model is performed, obtaining results full of oscillations for the torsion speed
and a long time of convergence for ωd , so the technique is not taken into account
in this dissertation.
56
500
400

300
200
100
-100
2 2.5 3 3.5 4 4.5 5
Time (s)
Figure 5.12: Delta speed behavior with linearization T=10 ms
2
sr
1.5 sr
0.5
-0.5
-1
-1.5
-2
2 2.5 3 3.5 4 4.5 5
Time (s)
Figure 5.13: Drive shaft torsion speed and angle with linearization T=10 ms
57
200
Clutch
150
100
50
Torque [Nm]
-50
-100
-150
-200
2 2.5 3 3.5 4 4.5 5
Time (s)
Figure 5.14: Total engine and clutch torque with linearization T=10 ms
5.3 LQI technique with Smith Predictor

In order to manage the delay of the clutch actuator, that causes some problems
in the tuning of the Q and R matrices, its transfer function is considered as a
secondary plant to be controlled using the Smith Predictor. The Simulink Scheme
is shown below:
Figure 5.15: Smith Predictor scheme for clutch actuator
Where CS is the controller, Gp is the actuator transfer function without the delay,
P S is the plant of the actuator with delay and Dp is the delay itself. The controller
adopted here is a PI controller.
58
500
Slipping speed [RPM] 400
300
200
100
-100
2 2.5 3 3.5 4 4.5 5
Time (s)
Figure 5.16: Delta speed behavior with LQI and Smith Predictor
2
sr
1.5 sr
0.5
-0.5
-1
-1.5
-2
2 2.5 3 3.5 4 4.5 5
Time (s)
Figure 5.17: Drive shaft torsion speed and angle with LQI and Smith Predictor
59
200 Combustion Engine+Electric motor

Clutch
150
Torque [Nm] 100
50
-50
-100
-150
-200
2 2.5 3 3.5 4 4.5 5
Time (s)
Figure 5.18: Total engine and clutch torque with LQI and Smith Predictor
As it is possible to observe the results are good in terms of delta speed convergence,
even if there are some oscillations in the torsion speed. The correct estimation of
the performance will be done later throught the computation of the RMSE and
standard deviation but in general, the results obtained seem to be worse than those
obtained before, throught the linearization of the entire model.
Even when ωd reaches the null value, some oscillations are present and this behavior
represents a loss of performance of the LQI controller.
60
5.4 State space representation of the actuators

and the Plant with LQI technique
As explained in chapter 3, section 3.3, the final shape of the state space represen-
tation is here derived, where the outputs of the actuators become the inputs of a
new augmented plant with 9 states described by the matrices Ā and B̄.
0 0 0 0 −0.0013 0.0001 −0.0004 −0.0026 0.0014

 
0.0001 −0.0001 0.1117 7.1500 −0.0327 0.0030 −0.0091 −0.0675 0.0014 
 
 0 0 −0.012 −0.7696 0.0034 −0.0003 0.0009 0.0069 0
 

 
 0 0 0.0001 0 0 0 0 0 0 
 
4
Ā = 10  0

0 0 0 −0.0216 −0.022 −0.0118 −0.0052 0 


(5.1)
 0 0 0 0 0.0256 0 0 0 0 
 
 0 0 0 0 0 0.0128 0 0 0 
 
 0 0 0 0 0 0 0.0064 0 0
 

0 0 0 0 0 0 0 0 −0.0014
0 0
 
0 0
 
0 0
 
 
0 0
 
B̄ = 

8 0 (5.2)
 
0 0
 
0 0
 
0 0
 
0 4
The Simulink scheme of the complete model is shown below:
Figure 5.19: Simulink scheme of the complete system
61
500
400
300
200
100
-100
2 2.5 3 3.5 4 4.5 5
Time (s)
Figure 5.20: Delta speed behavior with LQI and complete State Space
2
sr
1.5 sr
0.5
-0.5
-1
-1.5
-2
2 2.5 3 3.5 4 4.5 5
Time (s)
Figure 5.21: Drive shaft torsion speed and angle with complete State Space
62

Clutch
150
100
Torque [Nm]
50
-50
-100
-150
-200
2 2.5 3 3.5 4 4.5 5
Time (s)
Figure 5.22: Total engine and clutch torque with LQI and complete State Space
The computation of a state space representation of the model including the transfer
functions of the actuators and delays makes possible the development of a more
robust and accurate control of the system that takes into account all the uncertain-
ties. As it is possible to observe from the figures shown above, the performance is
very good, the oscillations of the drive shaft torsion speed and angle are reduced
to the minimum and the delays of the actuators are managed avoiding problems in
the control.
The final way in order to manage the states of the actuators that cannot be mea-
sured is that to associate to this kind of control with this state space representation
the Kalman Filter for the three subsystem of the model that are the plant itself
and the two actuators.
This will be the last step in the research of an optimal usage of the LQI control
strategy for the Dry Dual Clutch Transmission avoiding oscillations and a smooth
convergence of the value of ωd that is the most important control target.
63
5.5 Kalman State Estimation

The observation of the states of the actuators is then implemented in Simulink
realizing a Kalman filter design for each single plant of the complete system.
The Kalman filter is designed for the clutch and engine actuators: the plant contains
their outputs as inputs.
In this way the performance of the control is improved and also the tuning of the
Q and R matrices becomes simpler. The weights for the actuators states are tuned
inside the Q and R matrices of the Kalman Filter.
The Simulink scheme is shown below:
Figure 5.23: LQI with Kalman filter
The results of this kind of control are very good in terms of convergence of ωd even
if the interval is longer than the previous model but also in terms of the oscillations
given by the torsion angle that here remains close to the null value.
It is possible to observe that the controller, computing properly the clutch and the
engine torque, is able to ensure a smooth tracking of the reference clutch slipping
speed in order to improve driver comfort and drivability.
The expected performance is reached in the best way adopting the LQI control
strategy together with the Kalman filter.
64
Figure 5.24: Full model with observer
Figure 5.25: Plant with actuators
65
500
400

300
200
100
-100
2 2.5 3 3.5 4 4.5 5
Time (s)
Figure 5.26: Delta speed behavior with LQI and Kalman filter
2
sr
1.5 sr
0.5
-0.5
-1
-1.5
-2
2 2.5 3 3.5 4 4.5 5
Time (s)
Figure 5.27: Drive shaft torsion speed and angle and Kalman filter
66

Clutch
200
150
100
Torque [Nm]
50
-50
-100
-150
-200
2 2.5 3 3.5 4 4.5 5
Time (s)
Figure 5.28: Total engine and clutch torque with LQI with Kalman filter
Figure 5.29: Expected performance
67
5.6 Comparison between simulation results

In this section comparisons between the obtained results are shown, with the aim
of understanding which is the model and the control strategy that ensures best
performances.
With respect to the oscillations, the standard deviation σ is considered: it is a
measure of the variation/dispersion of a set of data with respect to the mean value.
In particular, the time interval between the starting control time (ts = 2s) and the
time tf when the slipping speed ωd reaches the null value, is considered.
With respect to the oscillations, the standard deviation is considered: it is a measure
of the variation/dispersion of a set of data with respect to the mean value.
The slipping speed analysis is done in terms of RMSE and standard deviation in
order to have a correct evaluation of the dispersion of the data.
The values are computed throught the commands ÍÍ rmsÍÍ and ÍÍ stdÍÍ given by Matlab
and all the quantities are considered after the starting of the control so at ts =2s.
Model σωsr σωd RMSE ωd

LQI without actuators 0.36 125.05 137.03
LQI with actuators and delays 0.28 144.00 163.30
LQI with linearization 0.58 180.35 247.40
LQI with Smith Predictor 1.81 140.35 158.17
LQI State Space Representation of the plant and actuators 0.08 121.43 134.20
LQI with Kalman Filter 0.28 122.4 135.00
Table 5.1: Indices computation
Model ∆t
LQI without actuators 0.72
LQI with actuators and delays 1.59
LQI with linearization 0.90
LQI with Smith Predictor 1.38
LQI State Space Representation of the plant and actuators 0.70
LQI with Kalman Filter 0.85
Table 5.2: Convergence interval
68
5.6.1 Observations
The convergence time is a very important parameter in the evaluation of the
performance of a certain control strategy. If ωd reaches the null value before, a
faster tracking of the slipping speed reference signal is guaranteed.
As it is possible to observe, the LQI full model with actuators and delays has the
lowest convergence interval and also the other indices demonstrate that it represents
the best control strategy developed in this thesis even if it is not so realistic because
this model does not consider the fact that the states of the actuators cannot be
measured.
The adoption of the Smith Predictor significantly reduces the convergence in-
terval but causes a lot of oscillations as it is possible to see in the high values of the
standard deviation of ωsr and also from the indices calculated for the slipping speed.
The state space representation of the entire model with the actuators increases a
lot the performances also with a trial and error regulation of the weight matrices
for the states of the actuators. The full model without observers guarantees a
good trade off between time convergence and oscillations even if it is not possible
to measure the states of the actuators included in the control.This leads to the
adoption of the Kalman Filter to correctly estimate them.
In this way, the performance deteriorates a little bit, but the model is coher-
ent with the unmeasurability of the states of the actuators.
69
Chapter 6
Conclusions
This thesis has dealt with the control of the dry dual clutch during its slipping
phase. In particular, the main purpose of the project has been to design a controller
that, computing properly the clutch and the engine torque, is able to ensure a
smooth tracking of the reference clutch slipping speed in order to improve driver
comfort and drivability.
The Linear Quadratic Integral control technique has been used: it has been
described from a theoretical point of view, then the advantages of using this
technique have been explained; in particular the main reason is represented by
the fact that this method is more flexible with respect to more traditional techniques.
In this regard, different configurations have been used in order to exploit the
possible advantages caused by different control strategies. As a starting point, a
model whitout the clutch and engine actuators and delays has been considered.
Then these parts have been introduced in order to represent the real model with its
uncertanties : the tuning of the matrices has become more difficult and oscillations
have been introduced in the behavior of the ωd and ωsr .
In addition to the basic model, the engine torque actuator and the clutch torque
one have been inserted in Simulink and subsequently some ways to compensate
them have been experimented.
For all the investigated configurations simulation results corresponding to different

controller parameters settings have been presented. The best set of parameters
used for simulations have been chosen after performing an extensive tuning proce-
dure: the full LQI model with actuators and delays ensures the most satisfactory
performances.
70
Conclusions
In particular, it is shown that the LQI with actuators and delays has the worst
performance in terms of convergence interval even if the drive shaft torson speed
and angle are comparable with the control without actuators.
In changing the configurations of the model, the kind of control turns out to be
very flexible since it can be used through a general approach also introducing other
compensations for the delays.
In the end some important indices have been calculate in order to realize an
analysis of the performance: the main purpose was to show that there is always a
trade-off between the oscillations and the time convergence. By observing simula-
tion results it is clear that the configuration with the state space representation of
the plant and the two actuators with Kalman Fileter ensures very good results in
terms of slipping speed reference tracking and oscillations reduction considering
also the fact that this is the most realistic configuration.
The discretization of the control to T=10 ms strategy does not bring signifi-
cant modifications in terms of performances: the most important elements in this
regard are instead represented by a correct tuning of the weight matrices Q and R.
6.1 Future works

Starting from the results presented in this thesis project, further studies can be
conducted and different applications can be explored.
In particular it is possible to:
• evaluate the performance in terms of vehicle acceleration and drivability;
• compute the feedforward contribution of the engine torque with the inver-
sion on the model and using the the generated references as inputs of the inverse
model;
• use the driver torque (that is function of the accelerator pedal) as feedfor-
ward contribution for the clutch torque command.
It is also possible to validate the control experimenting it on vehicle.
71
Bibliography
[1] F. Cimmino C. Vafidis. «FPT’s High Torque Density Dual Clutch Trans-
mission(HTDDDCT)». In: 8th CTI Symposium Innovative Automotive Trans-
missions (Berlin,2009) (cit. on p. 1).
[2] Ioan-Adrian Viorel Loránd Szabò Lars Lowenstein Cristian Stet. «Integrated
starter-generators for automotive applications». In: Volume 45, Number 3
(2004) (cit. on p. 5).
[3] A. Bordoni G. Gatti. «Development of Fiat powertrain new C725 7-Speed
dual dry clutch transmission for FWD passenger vehicles». In: (Shangai,2013)
(cit. on p. 12).
[4] T. Jin P. Li G. Zhu. «Optimal decoupled control for dry clutch engagement».
In: American Control Conference (ACC), pages 6740-6745 (2013) (cit. on
p. 17).
[5] P. J. Dolcini C. Canudas-de Wit H. Béchart. «Dry clutch control for automo-
tive applications». In: Springer Science and Business Media (2010) (cit. on
p. 18).
[6] L. Glielmo L. Iannelli V. Vacca F. Vasca. «Gearshift control for automated
manual transmissions». In: IEEE/ASME Transactions on, 11(1):17–26 (2006)
(cit. on p. 18).
[7] Jinsung Kim Kwanghyun Cho Seibum B. Choi. «Gear Shift Control of Dual
Clutch Transmissions with a Torque Rate Limitation Trajectory». In: Ameri-
can Control Conference on O’Farrell Street, San Francisco, CA, USA (June
29 - July 01, 2011) (cit. on p. 19).
[8] Massimo Canale. «Slides of the course». In: Digital control systems and
architectures (2019) (cit. on p. 32).
[9] Angelo Salatino. «Notes on Smith Predictor». In: Metodi di controllo per i
sistemi di elaborazione e comunicazione (2012) (cit. on p. 38).
[10] Carlo Novara. «Slides of the course». In: Nonlinear control and aerospace
applications (2018-2019) (cit. on p. 41).
72
BIBLIOGRAPHY
[11] Bland J.M. Altman D.G. «Statistics notes: measurement error». In: BMJ
(1996) (cit. on p. 47).
[12] Hyndman Rob J. Koehler Anne B. «Another look at measures of forecast
accuracy». In: International Journal of Forecasting (2006) (cit. on p. 48).
73

گیربکس ۷ سرعته Ddct Vw

Uploaded by

Copyright:

Available Formats

گیربکس ۷ سرعته Ddct Vw

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

گیربکس ۷ سرعته Ddct Vw

Uploaded by

Copyright:

Available Formats

POLITECNICO DI TORINO

Master Degree Thesis

LQI control of dual-clutch transmission

Prof. CARLO NOVARA ILARIA GIRONE

Ing. EMANUEL CORIGLIANO

In this dissertation, realized in collaboration with Centro Ricerche Fiat (CRF), a

Preme ergo quod coepisti,

La stesura della tesi di laurea rappresenta la conclusione del percorso universitario:

List of Tables viii

2 DCT System in hybrid vehicles with P2 architecture 4

3 Model Analytical description 25

5 Tuning and Simulation results 49

2.1 C635 DDCT Main Mechanical characteristics . . . . . . . . . . . . 9

3.1 DDCT main parameters . . . . . . . . . . . . . . . . . . . . . . . . 27

5.1 Indexes computation . . . . . . . . . . . . . . . . . . . . . . . . . . 68

2.1 The basic configurations of HEVs . . . . . . . . . . . . . . . . . . . 4

3.1 General driveline scheme . . . . . . . . . . . . . . . . . . . . . . . . 25

4.1 General LQI scheme . . . . . . . . . . . . . . . . . . . . . . . . . . 35

5.1 Plant fed with the actuated inputs . . . . . . . . . . . . . . . . . . 49

• AMTs (Automated Manual Transmissions) are the most cost-effective solu-

• DCTs (Dual Clutch Transmissions) represent a perfect balance between pow-

• CVTs (Continuously Variable Transmissions) are particularly suitable for some

• Multi-speed ATs (planetary Automatic Transmissions) are competitive in terms

Adopting some specific control strategies, it is possible to reduce vibrations and

1.1 Thesis structure

DCT System in hybrid

Figure 2.1: The basic configurations of HEVs

In the series structure, the ICE is exploited to drive an alternator to generate

Figure 2.2: MHEV system

1. provide torque to the powertrain in motor mode

2. produce electricity in generator mode

Figure 2.3: MHEV with P2 architecture

The Dual Clutch Transmission is made-up of two independent sub-gearboxes,

Figure 2.4: C635 Manual Transmission and DDCT

2.2 C635 Dry Dual Clutch Transmission

Layout 3-shaft plus RG idler

Table 2.1: C635 DDCT Main Mechanical characteristics

As shown in Figure 2.5 below, the 3-shaft transmission architecture is contained

Figure 2.5: Cross section of the C635 DDCT

2.2.1 Operation of the DDCT

The C635 DDCT is an automatically shifted transaxle, electronically controlled

2.2.2 Dry Dual Clutch Unit

Figure 2.6: Dual Clutch Unit, Clutch Actuators and installation

2.2.3 Electro-hydraulic actuation system

Figure 2.7: Hydraulic Power Unit and CAM

Figure 2.8: C635 DDCT complete actuation system

2.2.4 DDCT Control Unit

– Engagement Actuators Control: the desired trajectories are realized by com-

– Shifter Control: hydraulic power to the required engagement actuator is guar-

• Self-Tuning Control: many self-tuning controls are needed in order to com-

2.3 Micro slip

Figure 2.9: DCT and clutches K1 and K2

Figure 2.10: Schematic representation of the DDCT

-Open clutch phase: the two disks are completely disengaged Cf = 0;

2.4 State of art

Figure 2.11: Decoupling control implemented in [4]

Figure 2.12: MPC control from [5]

Figure 2.13: Engine speed and torque control [6]

Figure 2.14: General scheme presented in [7]

2.5 Model configuration