## Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

Thesis No. 564

Driveline Modeling and Principles

for Speed Control and

Gear-Shift Control

Magnus Pettersson

Division of Vehicular Systems

Department of Electrical Engineering

Link¨oping University, S–581 83 Link¨ oping, Sweden

Link¨oping 1996

Driveline Modeling and Principles for Speed Control and

Gear-Shift Control

c 1996 Magnus Pettersson

magnusp@isy.liu.se

Department of Electrical Engineering

Link¨ oping University

S–581 83 Link¨ oping

Sweden

ISBN 91-7871-744-2

ISSN 0280-7971

LiU-TEK-LIC-1996:29

To Anna and Oscar

Abstract i

Abstract

A vehicular driveline consists of engine, clutch, transmission, shafts, and wheels,

which are controlled by a driveline management system.

Experiments and modeling using a heavy truck show that there are signiﬁcant

torsional resonances in the driveline. A linear model with a drive shaft ﬂexibility

is able to suﬃciently explain the measured engine speed and wheel speed.

Engine control for automatic gear shifting is an approach at the leading edge of

technology. A critical step is the controlling of the engine such that the transmission

transfers zero torque, whereafter neutral gear can be engaged. Driveline oscillations

is a limiting factor in this system. A model of the transmission torque is developed

and a state-feedback controller is used to drive this torque to zero. The result is a

possibility to optimize the time needed for a gear shift. Furthermore, neutral gear

can successfully be engaged also when facing load disturbances and initial driveline

oscillations.

Traditionally in diesel trucks, the engine speed is controlled by a system called

RQV. This system has the desired property of a load dependent stationary error,

and the undesired property of vehicle shuﬄe following a change in pedal position.

A model based state-feedback controller is derived that actively reduces wheel

speed oscillations. The performance and driveability is signiﬁcantly improved,

while maintaining the desired load characteristics for RQV control.

In conclusion, the proposed strategies improve performance and driveability in

both speed control and gear-shift control.

ii Abstract

Acknowledgment iii

Acknowledgment

This work has been carried out under the excellent guidance of Professor Lars

Nielsen at Vehicular Systems, Link¨ oping University, Sweden. By inspiring me and

taking time for many discussions he has contributed to this work in many ways.

I am indebted to Lars-Gunnar Hedstr¨ om, Anders Bj¨ornberg, Kjell Gestl¨ ov, and

Bj¨orn Westman at Scania in S¨ odert¨ alje for the help during this work, and for

interesting discussions regarding control and modeling in heavy trucks.

I am also grateful to Simon Edlund, Lars Eriksson, and Mattias Nyberg for read-

ing the manuscript. Thanks for the remarks and suggested improvements. Thanks

also to Tomas Henriksson, my former oﬃce colleague, for our many discussions

regarding research and courses.

I am indebted to Dr Joakim Petersson, Dr Fredrik Gustafsson, Dr Anders

Helmersson, and Dr Tomas McKelvey for help and discussions.

Thanks to Dr Peter Lindskog and Magnus Sundstedt for support on computers

and L

A

T

E

X.

I am very grateful to my parents Birgitta and Nils and my sister Katharina for

their love and support in whatever I do.

Finally, I would like to express my deepest gratitude to my wife Anna and our

son Oscar for their encouragements, patience, and love during this work.

Link¨ oping, April 1996

Magnus Pettersson

iv Acknowledgment

Contents

1 Introduction 1

1.1 Outline and Contributions . . . . . . . . . . . . . . . . . . . . . . . . 2

2 Driveline Modeling 3

2.1 Basic Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2.2 Shaft Flexibilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.2.1 Model 1: Drive Shaft Flexibility . . . . . . . . . . . . . . . . 7

2.2.2 Model 1 Extended with a Flexible Propeller Shaft . . . . . . 10

2.3 Models Including the Clutch . . . . . . . . . . . . . . . . . . . . . . . 11

2.3.1 Model 2: Flexible Clutch and Drive Shafts . . . . . . . . . . 11

2.3.2 Model 3: Nonlinear Clutch and Drive Shaft Flexibility . . . . 12

2.4 Additional Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

3 Field Trials and Modeling 17

3.1 The Truck . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

3.2 Measurement Description . . . . . . . . . . . . . . . . . . . . . . . . 19

3.3 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

3.4 Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

3.4.1 Inﬂuence from the Drive Shaft . . . . . . . . . . . . . . . . . 24

3.4.2 Inﬂuence from the Propeller Shaft . . . . . . . . . . . . . . . 24

3.4.3 Deviations between Engine Speed and Transmission Speed . . 26

3.4.4 Inﬂuence from the Clutch . . . . . . . . . . . . . . . . . . . . 27

3.4.5 Model Validity . . . . . . . . . . . . . . . . . . . . . . . . . . 31

3.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

v

vi Contents

4 Architectural Issues for Driveline Control 35

4.1 State-Space Formulation . . . . . . . . . . . . . . . . . . . . . . . . . 36

4.1.1 Disturbance Description . . . . . . . . . . . . . . . . . . . . . 37

4.1.2 Measurement Description . . . . . . . . . . . . . . . . . . . . 37

4.2 Controller Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . 38

4.3 Some Feedback Properties . . . . . . . . . . . . . . . . . . . . . . . . 39

4.4 Driveline Control with LQG/LTR . . . . . . . . . . . . . . . . . . . 41

4.4.1 Transfer Functions . . . . . . . . . . . . . . . . . . . . . . . . 42

4.4.2 Design Example with a Simple Mass-Spring Model . . . . . . 43

4.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

5 Speed Controller Design and Simulations 49

5.1 RQV Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

5.2 Problem Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

5.2.1 Mathematical Problem Formulation . . . . . . . . . . . . . . 50

5.3 Speed Control with Active Damping and RQV Behavior . . . . . . . 53

5.3.1 Extending with RQV Behavior . . . . . . . . . . . . . . . . . 55

5.4 Inﬂuence from Sensor Location . . . . . . . . . . . . . . . . . . . . . 56

5.4.1 Inﬂuence from Load Disturbances . . . . . . . . . . . . . . . 58

5.4.2 Inﬂuence from Measurement Disturbances . . . . . . . . . . . 60

5.4.3 Load Estimation . . . . . . . . . . . . . . . . . . . . . . . . . 62

5.5 Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

5.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

6 Gear-Shift Controller Design and Simulations 67

6.1 Problem Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

6.2 Transmission Torque . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

6.2.1 Transmission Torque for Model 1 . . . . . . . . . . . . . . . . 68

6.2.2 Transmission Torque for Model 2 . . . . . . . . . . . . . . . . 71

6.2.3 Transmission Torque for Model 3 . . . . . . . . . . . . . . . . 72

6.3 Preliminary Trials . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

6.3.1 Unconstrained Active Damping . . . . . . . . . . . . . . . . . 73

6.3.2 Undamped Gear-Shift Condition . . . . . . . . . . . . . . . . 74

6.4 Gear-Shift Control Criterion . . . . . . . . . . . . . . . . . . . . . . . 76

6.5 Gear-Shift Control Design . . . . . . . . . . . . . . . . . . . . . . . . 77

6.6 Inﬂuence from Sensor Location . . . . . . . . . . . . . . . . . . . . . 80

6.6.1 Inﬂuence from Load Disturbances . . . . . . . . . . . . . . . 80

6.6.2 Inﬂuence from Measurement Disturbances . . . . . . . . . . . 81

6.7 Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

6.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86

7 Conclusions 89

Bibliography 91

Notations 93

1

Introduction

The main parts of a vehicular driveline are engine, clutch, transmission, shafts,

and wheels. Since these parts are elastic, mechanical resonances may occur. The

handling of such resonances is of course basic for driveability, but is also other-

wise becoming increasingly important since it is a linking factor in development of

new driveline management systems. Two systems where driveline oscillations limit

performance is speed control and automatic gear shifting.

Fundamental driveline equations are obtained by using Newton’s second law.

The result is a series of models consisting of rotating inertias, connected with

damped torsional ﬂexibilities. Experiments are performed with a heavy truck with

diﬀerent gears and road slopes. The aim of the modeling and experiments is to ﬁnd

the most important physical eﬀects that contribute to driveline oscillations. Some

open questions are discussed, regarding inﬂuence of sensor dynamics and nonlinear

eﬀects.

The ﬁrst problem is wheel speed oscillations following a change in accelera-

tor pedal position, known as vehicle shuﬄe (Mo, Beaumount, and Powell 1996;

Pettersson and Nielsen 1995). Traditionally in diesel trucks, the fuel metering is

governed by a system called RQV. With RQV, there is no active damping of wheel

speed oscillations resulting in vehicle shuﬄe. Another property is that a load de-

pendent stationary error results from downhill and uphill driving. The thesis treats

model based speed control with active damping of wheel speed oscillations while

maintaining the stationary error characteristic for RQV control.

Engine controlled gear shifting without disengaging the clutch is an approach at

the leading edge of technology (Orehall 1995). The engine is controlled such that

the transmission transfers zero torque, whereafter neutral gear can be engaged.

1

2 Chapter 1 Introduction

The engine speed is then controlled to a speed such that the new gear can be

engaged. A critical part in this scheme is the controlling of the engine such that

the transmission torque is zero. In this state, the vehicle is free rolling, which must

be handled. Driveline oscillations is a limiting factor in optimizing this step. In

this thesis the transmission torque is modeled, and controlled to zero by using state

feedback. With this approach, it is possible to optimize the time needed for a gear

shift, also when facing existing initial driveline oscillations.

A common architectural issue in the two applications described above is the

issue of sensor location. Diﬀerent sensor locations result in diﬀerent control prob-

lems. A comparison is made between using feedback from the engine speed sensor

or the wheel speed sensor, and the inﬂuence in control design is investigated.

1.1 Outline and Contributions

In Chapters 2 and 3, a set of three driveline models is derived. Experiments with a

heavy truck are described together with the modeling conclusions. The contribution

of the chapter is that a linear model with one torsional ﬂexibility and two inertias

is able to ﬁt the measured engine speed and wheel speed within the bandwidth

of interest. Parameter estimation of a model with a nonlinear clutch and sensor

dynamics explains that the diﬀerence between experiments and model occurs when

the clutch transfers zero torque.

Control of resonant systems with simple controllers is, from other technical

ﬁelds, known to have diﬀerent properties with respect to sensor location. These

results are reviewed in Chapter 4. The extension to more advanced control design

methods is a little studied topic. The contribution of the chapter is a demonstration

of the inﬂuence of sensor location in driveline control when using LQG/LTR.

Chapters 5 treats the design and simulation of the speed controller. A key

contribution in this chapter is the formulation of a criterion for the speed con-

trol concept described above with active damping and retained RQV feeling. A

simulation study shows signiﬁcantly improved performance and driveability.

Chapters 6 deals with the design and simulation of the gear-shift controller.

A major contribution in this thesis is a gear shifting strategy, based on a model

describing the transmission torque, and a criterion for a controller that drives

this torque to zero. The design improves the performance also in the case of

load disturbances and initial driveline oscillations. Conclusions are summarized in

Chapter 7.

2

Driveline Modeling

The driveline is a fundamental part of a vehicle and its dynamics has been modeled

in diﬀerent ways depending on the purpose. The frequency range treated in this

work is the regime interesting for control design (Mo, Beaumount, and Powell

1996; Pettersson and Nielsen 1995). Vibrations and noise contribute to a higher

frequency range (Suzuki and Tozawa 1992; Gillespie 1992) which is not treated

here. This chapter deals with building models of a truck driveline. The generalized

Newton’s second law is used together with assumptions about how diﬀerent parts

in the driveline contribute to the model. The aim of these assumptions is to ﬁnd the

most important physical eﬀects, contributing to driveline oscillations. Modeling is

an iterative process in reality. Nevertheless, a set of three models of increasing

complexity is presented. Next chapter will validate the choices.

First, a linear model with ﬂexible drive shafts is derived. Assumptions about

stiﬀ clutch, stiﬀ propeller shaft, viscous friction in transmission and ﬁnal drive,

together with a linear model of the air drag constitute the model. A second linear

model is given by using the assumptions made above, and adding a second ﬂexibility

which is the clutch. Finally, a more complete nonlinear model is derived which

includes a clutch model with a static nonlinearity.

2.1 Basic Equations

A vehicular driveline is depicted in Figure 2.1. It consists of an engine, clutch,

transmission, propeller shaft, ﬁnal drive, drive shafts, and wheels. In this section

fundamental equations for the driveline will be derived. Furthermore, some basic

equations regarding the forces acting on the wheel, are obtained. These equations

3

4 Chapter 2 Driveline Modeling

Engine Clutch

Transmission

Propeller shaft

Final drive

Drive shaft

Wheel

Figure 2.1 A vehicular driveline.

are inﬂuenced by the complete dynamics of the vehicle. This means that eﬀects

from, for instance, vehicle mass and trailer will be described by the equation de-

scribing the wheel. Next, a relation between the inputs and outputs of each part

is obtained, in order to get a complete physical model. Inputs and outputs of each

subsystem are labeled according to Figure 2.2.

Engine: The output torque of the engine is characterized by the driving torque

(M

m

) resulting from the combustion, the internal friction from the engine

(M

fr:m

), and the external load from the clutch (M

c

). The generalized New-

ton’s second law of motion (Meriam and Kraige 1987) gives the following

model

J

m

¨

θ

m

= M

m

−M

fr:m

−M

c

(2.1)

where J

m

is the mass moment of inertia of the engine and θ

m

is the angle of

the ﬂywheel.

Clutch: A friction clutch found in vehicles equipped with a manual transmis-

sion consists of a clutch disk connecting the ﬂywheel of the engine and the

transmission’s input shaft. When the clutch is engaged, no internal friction is

assumed, giving M

c

= M

t

, according to Figure 2.2. The transmitted torque

is a function of the angular diﬀerence (θ

m

− θ

c

) and the angular velocity

diﬀerence (

˙

θ

m

−

˙

θ

c

) over the clutch

M

c

= M

t

= f

c

(θ

m

−θ

c

,

˙

θ

m

−

˙

θ

c

) (2.2)

2.1 Basic Equations 5

M

d

r

w

F

w

θ

f

θ

w

Wheel

M

m

M

fr:m

M

p

θ

t

Clutch

Trans-

M

p

M

f

M

d

θ

t

θ

p

θ

f

Engine

θ

m

M

t

θ

c

M

w

θ

w

M

fr:t

M

fr:f

M

fr:w

M

c mission

Propeller

shaft

Final

drive

Drive

shaft

Figure 2.2 Subsystems of a vehicular driveline with its input and output angle

and torque.

Transmission: A transmission has a set of gears, each with a conversion ratio

i

t

. This gives the following relation between the input and output torque of

the transmission

M

p

= f

t

(M

t

, M

fr:t

, θ

c

−θ

t

i

t

,

˙

θ

c

−

˙

θ

t

i

t

, i

t

) (2.3)

where the internal friction torque of the transmission is labeled M

fr:t

. The

reason for considering the angle diﬀerence θ

c

−θ

t

i

t

in (2.3) is the possibility

of having torsional eﬀects in the transmission.

Propeller shaft: The propeller shaft connects the transmission’s output shaft

with the ﬁnal drive. No friction is assumed (M

p

= M

f

), giving the following

model of the torque input to the ﬁnal drive

M

p

= M

f

= f

p

(θ

t

−θ

p

,

˙

θ

t

−

˙

θ

p

) (2.4)

Final drive: The ﬁnal drive is characterized by a conversion ratio i

f

in the same

way as the transmission. The following relation for the input and output

torque holds

M

d

= f

f

(M

f

, M

fr:f

, θ

p

−θ

f

i

f

,

˙

θ

p

−

˙

θ

f

i

f

, i

f

) (2.5)

6 Chapter 2 Driveline Modeling

F

a

v

F

w F

r

+mg sin(α)

Figure 2.3 Forces acting on a vehicle.

where the internal friction torque of the ﬁnal drive is labeled M

fr:f

.

Drive shafts: The drive shafts connects the wheels to the ﬁnal drive. Here it is

assumed that the wheel speed is the same for the two wheels. Therfore, the

drive shafts are modeled as one shaft. When the vehicle is turning and the

speed diﬀers between the wheels, both drive shafts have to be modeled. No

friction (M

w

= M

d

) gives the model equation

M

w

= M

d

= f

d

(θ

f

−θ

w

,

˙

θ

f

−

˙

θ

w

) (2.6)

Wheel: In Figure 2.3 the forces acting on a vehicle with mass m and speed v is

shown. Newton’s second law in the longitudinal direction gives

F

w

= m˙ v +F

a

+F

r

+mg sin(α) (2.7)

The friction force (F

w

) is described by the sum of the following quantities

(Gillespie 1992).

• F

a

, the air drag, is approximated by

F

a

=

1

2

c

w

A

a

ρ

a

v

2

(2.8)

where c

w

is the drag coeﬃcient, A

a

the maximum vehicle cross section

area, and ρ

a

the air density. However, eﬀects from, for instance, open

or closed windows will make the force diﬃcult to model.

2.2 Shaft Flexibilities 7

• F

r

, the rolling resistance, is approximated by

F

r

= m(c

r1

+c

r2

v) (2.9)

where c

r1

and c

r2

depends on, for instance, tires and tire pressure.

• mg sin(α), the gravitational force, where α is the slope of the road.

The coeﬃcients of air drag and rolling resistance, (2.8) and (2.9), can be iden-

tiﬁed e.g. by a identiﬁcation scheme (Henriksson, Pettersson, and Gustafsson

1993).

The resulting torque due to F

w

is equal to F

w

r

w

, where r

w

is the wheel

radius. Newton’s second law gives

J

w

¨

θ

w

= M

w

−F

w

r

w

−M

fr:w

(2.10)

where J

w

is the mass moment of inertia of the wheel, M

w

is given by (2.6),

and M

fr:w

is the friction torque. Including (2.7) to (2.9) in (2.10) together

with v = r

w

˙

θ

w

gives

(J

w

+mr

2

w

)

¨

θ

w

= M

w

−M

fr:w

−

1

2

c

w

A

a

ρ

a

r

3

w

˙

θ

2

w

(2.11)

−r

w

m(c

r1

+c

r2

r

w

˙

θ

w

) −r

w

mgsin(α)

A complete model for the driveline with the clutch engaged is described by

Equations (2.1) to (2.11). So far the functions f

c

, f

t

, f

p

, f

f

, f

d

, and the friction

torques M

fr:t

, M

fr:f

, and M

fr:w

are unknown. In the following section assumptions

will be made about the unknowns, resulting in a series of driveline models, with

diﬀerent complexities.

2.2 Shaft Flexibilities

In the following two sections, assumptions will be made about the unknowns. First,

a model with one torsional ﬂexibility (the drive shaft) will be considered, and then

a model with two torsional ﬂexibilities (the drive shaft and propeller shaft) will be

considered.

2.2.1 Model 1: Drive Shaft Flexibility

Assumptions about the fundamental equations in Section 2.1 are made in order

to obtain a model with drive shaft ﬂexibility. Labels are according to Figure 2.2.

The clutch and the propeller shafts are assumed to be stiﬀ, and the drive shaft is

described as a damped torsional ﬂexibility. The transmission and the ﬁnal drive

are assumed to multiply the torque with the conversion ratio, without losses.

8 Chapter 2 Driveline Modeling

Clutch: The clutch is assumed to be stiﬀ, which gives the following equations

for the torque and the angle

M

c

= M

t

, θ

m

= θ

c

(2.12)

Transmission: The transmission is described by one rotating inertia J

t

. The

friction torque is assumed to be described by a viscous damping coeﬃcient

b

t

. The model of the transmission, corresponding to (2.3), is

θ

c

= θ

t

i

t

(2.13)

J

t

¨

θ

t

= M

t

i

t

−b

t

˙

θ

t

−M

p

(2.14)

By using (2.12) and (2.13), the model can be rewritten as

J

t

¨

θ

m

= M

c

i

2

t

−b

t

˙

θ

m

−M

p

i

t

(2.15)

Propeller shaft: The propeller shaft is also assumed to be stiﬀ, which gives the

following equations for the torque and the angle

M

p

= M

f

, θ

t

= θ

p

(2.16)

Final drive: In the same way as the transmission, the ﬁnal drive is modeled by

one rotating inertia J

f

. The friction torque is assumed to be described by a

viscous damping coeﬃcient b

f

. The model of the ﬁnal drive, corresponding

to (2.5), is

θ

p

= θ

f

i

f

(2.17)

J

f

¨

θ

f

= M

f

i

f

−b

f

˙

θ

f

−M

d

(2.18)

Equation (2.18) can be rewritten with (2.16) and (2.17) which gives

J

f

¨

θ

t

= M

p

i

2

f

−b

f

˙

θ

t

−M

d

i

f

(2.19)

Reducing (2.19) to engine speed is done by using (2.12) and (2.13) resulting

in

J

f

¨

θ

m

= M

p

i

2

f

i

t

−b

f

˙

θ

m

−M

d

i

f

i

t

(2.20)

By replacing M

p

in (2.20) with M

p

in (2.15), a model for the lumped trans-

mission, propeller shaft, and ﬁnal drive is obtained

(J

t

i

2

f

+J

f

)

¨

θ

m

= M

c

i

2

t

i

2

f

−b

t

˙

θ

m

i

2

f

−b

f

˙

θ

m

−M

d

i

f

i

t

(2.21)

Drive shaft: The drive shaft is modeled as a damped torsional ﬂexibility, having

stiﬀness k, and internal damping c. Hence, (2.6) becomes

M

w

= M

d

= k(θ

f

−θ

w

) +c(

˙

θ

f

−

˙

θ

w

) = k(θ

m

/i

t

i

f

−θ

w

) (2.22)

+ c(

˙

θ

m

/i

t

i

f

−

˙

θ

w

)

2.2 Shaft Flexibilities 9

θ

w

θ

m

J

m

+J

t

/i

2

t

+J

f

/i

2

t

i

2

f J

w

+mr

2

w

k

c

M

m

−M

fr:m

r

w

m(c

r1

+gsin(α))

Figure 2.4 Model 1: Stiﬀ clutch and drive shaft torsional ﬂexibility.

where (2.12), (2.13), (2.16), and (2.17) are used. By replacing M

d

in (2.21)

with (2.22) the equation describing the transmission, propeller shaft, ﬁnal

drive, and drive shaft, becomes

(J

t

i

2

f

+J

f

)

¨

θ

m

= M

c

i

2

t

i

2

f

−b

t

˙

θ

m

i

2

f

−b

f

˙

θ

m

(2.23)

−k(θ

m

−θ

w

i

t

i

f

) −c(

˙

θ

m

−

˙

θ

w

i

t

i

f

)

Wheel: If (2.11) is combined with (2.22), and if the linear part of the air drag

in (2.11) is used, the following equation for the wheel results

(J

w

+mr

2

w

)

¨

θ

w

= k(θ

m

/i

t

i

f

−θ

w

) +c(

˙

θ

m

/i

t

i

f

−

˙

θ

w

) (2.24)

−b

w

˙

θ

w

−c

w

A

a

ρ

a

r

3

w

˙

θ

w

−mc

r2

r

2

w

˙

θ

w

−r

w

m(c

r1

+gsin(α))

where the friction torque is described as viscous damping, with label b

w

.

The complete model is obtained by inserting M

c

from (2.23) into (2.1), together

with (2.24), which gives the following equations. An illustration of the model can

be seen in Figure 2.4.

Deﬁnition 2.1 Resulting equations for Model 1 - drive shaft ﬂexibility.

(J

m

+J

t

/i

2

t

+J

f

/i

2

t

i

2

f

)

¨

θ

m

= M

m

−M

fr:m

−(b

t

/i

2

t

+b

f

/i

2

t

i

2

f

)

˙

θ

m

(2.25)

−k(θ

m

/i

t

i

f

−θ

w

)/i

t

i

f

−c(

˙

θ

m

/i

t

i

f

−

˙

θ

w

)/i

t

i

f

(J

w

+mr

2

w

)

¨

θ

w

= k(θ

m

/i

t

i

f

−θ

w

) +c(

˙

θ

m

/i

t

i

f

−

˙

θ

w

) (2.26)

−(b

w

+c

w

A

a

ρ

a

r

3

w

+mc

r2

r

2

w

)

˙

θ

w

−r

w

m(c

r1

+gsin(α))

Possible states describing Model 1 are

˙

θ

m

,

˙

θ

w

, and θ

m

/i

t

i

f

−θ

w

.

10 Chapter 2 Driveline Modeling

2.2.2 Model 1 Extended with a Flexible Propeller Shaft

It is also possible to consider two torsional ﬂexibilities, the propeller shaft and the

drive shaft. In the derivation of the model, the clutch is assumed stiﬀ, and the

propeller and drive shafts are modeled as damped torsional ﬂexibilities. As in the

derivation of Model 1, the transmission and ﬁnal drive are assumed to multiply the

torque with the conversion ratio, without losses.

The derivation of Model 1 is repeated here with the diﬀerence that the model

for the propeller shaft (2.16) is replaced by a model of a ﬂexibility with stiﬀness k

p

and internal damping c

p

M

p

= M

f

= k

p

(θ

t

−θ

p

) +c

p

(

˙

θ

t

−

˙

θ

p

) = k

p

(θ

m

/i

t

−θ

p

) +c

p

(

˙

θ

m

/i

t

−

˙

θ

p

) (2.27)

where (2.12) and (2.13) are used in the last equality. This means that there are

two torsional ﬂexibilities, the propeller shaft and the drive shaft. Inserting (2.27)

into (2.15) gives

J

t

¨

θ

m

= M

c

i

2

t

−b

t

˙

θ

m

−

_

k

p

(θ

m

/i

t

−θ

p

) +c

p

(

˙

θ

m

/i

t

−

˙

θ

p

)

_

i

t

(2.28)

By combining this with (2.1) the following diﬀerential equation describing the

lumped engine and transmission results

(J

m

+J

t

/i

2

t

)

¨

θ

m

= M

m

−M

fr:m

−b

t

/i

2

t

˙

θ

m

(2.29)

−

1

i

t

_

k

p

(θ

m

/i

t

−θ

p

) +c

p

(

˙

θ

m

/i

t

−

˙

θ

p

)

_

The ﬁnal drive is described by inserting (2.27) in (2.18), and repeating (2.17)

θ

p

= θ

f

i

f

(2.30)

J

f

¨

θ

f

= i

f

_

k

p

(θ

m

/i

t

−θ

p

) +c

p

(

˙

θ

m

/i

t

−

˙

θ

p

)

_

−b

f

˙

θ

f

−M

d

(2.31)

Including (2.30) in (2.31) gives

J

f

¨

θ

p

= i

2

f

_

k

p

(θ

m

/i

t

−θ

p

) +c

p

(

˙

θ

m

/i

t

−

˙

θ

p

)

_

−b

f

˙

θ

p

−i

f

M

d

(2.32)

The equation for the drive shaft (2.22) is repeated with new labels

M

w

= M

d

= k

d

(θ

f

−θ

w

) +c

d

(

˙

θ

f

−

˙

θ

w

) = k

d

(θ

p

/i

f

−θ

w

) +c

d

(

˙

θ

p

/i

f

−

˙

θ

w

) (2.33)

where (2.30) is used in the last equality.

The equation for the ﬁnal drive (2.32) now becomes

J

f

¨

θ

p

= i

2

f

_

k

p

(θ

m

/i

t

−θ

p

) +c

p

(

˙

θ

m

/i

t

−

˙

θ

p

)

_

−b

f

˙

θ

p

(2.34)

−i

f

_

k

d

(θ

p

/i

f

−θ

w

) +c

d

(

˙

θ

p

/i

f

−

˙

θ

w

)

_

2.3 Models Including the Clutch 11

θ

w

θ

m

J

m

+J

t

/i

2

t

J

w

+mr

2

w

k

d

c

d

M

m

+M

fr:m r

w

m(c

r1

+gsin(α))

k

p

c

p

θ

p

J

f

Figure 2.5 Model with stiﬀ clutch and two torsional ﬂexibilities.

The equation for the wheel is derived by combining (2.11) with (2.33). If the linear

part of the of the air drag in (2.11) is used, the following equation for the wheel

results

(J

w

+mr

2

w

)

¨

θ

w

= k

d

(θ

p

/i

f

−θ

w

) +c

d

(

˙

θ

p

/i

f

−

˙

θ

w

) (2.35)

−b

w

˙

θ

w

−c

w

A

a

ρ

a

r

3

w

˙

θ

w

−mc

r2

r

2

w

˙

θ

w

−r

w

m(c

r1

+gsin(α))

where again the friction torque is assumed to be described by a viscous damping

coeﬃcient b

w

. The complete model with drive shaft and propeller shaft ﬂexibilities

is the following, which can be seen in Figure 2.5.

(J

m

+J

t

/i

2

t

)

¨

θ

m

= M

m

−M

fr:m

−b

t

/i

2

t

˙

θ

m

(2.36)

−

1

i

t

_

k

p

(θ

m

/i

t

−θ

p

) +c

p

(

˙

θ

m

/i

t

−

˙

θ

p

)

_

J

f

¨

θ

p

= i

2

f

_

k

p

(θ

m

/i

t

−θ

p

) +c

p

(

˙

θ

m

/i

t

−

˙

θ

p

)

_

−b

f

˙

θ

p

(2.37)

−i

f

_

k

d

(θ

p

/i

f

−θ

w

) +c

d

(

˙

θ

p

/i

f

−

˙

θ

w

)

_

(J

w

+mr

2

w

)

¨

θ

w

= k

d

(θ

p

/i

f

−θ

w

) +c

d

(

˙

θ

p

/i

f

−

˙

θ

w

) (2.38)

−(b

w

+c

w

A

a

ρ

a

r

3

w

+mc

r2

r

2

w

)

˙

θ

w

−r

w

m(c

r1

+gsin(α))

2.3 Models Including the Clutch

The clutch is so far considered to be stiﬀ and lumped together with the engine

mass moment of inertia. In this section this assumption is relaxed and ﬁrst, the

clutch is modeled as a linear ﬂexibility. Secondly, a nonlinear model of the clutch

is derived.

2.3.1 Model 2: Flexible Clutch and Drive Shafts

A model with a linear clutch and one torsional ﬂexibility (the drive shaft) is derived

by repeating the procedure for Model 1 with the diﬀerence that the model for the

12 Chapter 2 Driveline Modeling

clutch is a ﬂexibility with stiﬀness k

c

and internal damping c

c

M

c

= M

t

= k

c

(θ

m

−θ

c

) +c

c

(

˙

θ

m

−

˙

θ

c

) = k

c

(θ

m

−θ

t

i

t

) +c

c

(

˙

θ

m

−

˙

θ

t

i

t

) (2.39)

where (2.13) is used in the last equality. By inserting this into (2.1) the equation

describing the engine inertia is given by

J

m

¨

θ

m

= M

m

−M

fr:m

−

_

k

c

(θ

m

−θ

t

i

t

) +c

c

(

˙

θ

m

−

˙

θ

t

i

t

)

_

(2.40)

Also by inserting (2.39) into (2.14), the equation describing the transmission is

J

t

¨

θ

t

= i

t

_

k

c

(θ

m

−θ

t

i

t

) +c

c

(

˙

θ

m

−

˙

θ

t

i

t

)

_

−b

t

˙

θ

t

−M

p

(2.41)

M

p

is derived from (2.19) giving

(J

t

+J

f

/i

2

f

)

¨

θ

t

= i

t

_

k

c

(θ

m

−θ

t

i

t

) +c

c

(

˙

θ

m

−

˙

θ

t

i

t

)

_

−(b

t

+b

f

/i

2

f

)

˙

θ

t

−M

d

/i

f

(2.42)

which is the lumped transmission, propeller shaft, and ﬁnal drive inertia.

The drive shaft is modeled according to (2.22) as

M

w

= M

d

= k

d

(θ

f

−θ

w

) +c

d

(

˙

θ

f

−

˙

θ

w

) = k

d

(θ

t

/i

f

−θ

w

) +c

d

(

˙

θ

t

/i

f

−

˙

θ

w

) (2.43)

where (2.16) and (2.17) is used in the last equality.

The complete model is obtained by inserting (2.43) into (2.42) and (2.11), and

using the linear part of the air drag. An illustration of the model can be seen in

Figure 2.6.

Deﬁnition 2.2 Resulting equations for Model 2 - ﬂexible clutch and drive shaft

ﬂexibility.

J

m

¨

θ

m

= M

m

−M

fr:m

−

_

k

c

(θ

m

−θ

t

i

t

) +c

c

(

˙

θ

m

−

˙

θ

t

i

t

)

_

(2.44)

(J

t

+J

f

/i

2

f

)

¨

θ

t

= i

t

_

k

c

(θ

m

−θ

t

i

t

) +c

c

(

˙

θ

m

−

˙

θ

t

i

t

)

_

(2.45)

−(b

t

+b

f

/i

2

f

)

˙

θ

t

−

1

i

f

_

k

d

(θ

t

/i

f

−θ

w

) +c

d

(

˙

θ

t

/i

f

−

˙

θ

w

)

_

(J

w

+mr

2

w

)

¨

θ

w

= k

d

(θ

t

/i

f

−θ

w

) +c

d

(

˙

θ

t

/i

f

−

˙

θ

w

) (2.46)

−(b

w

+c

w

A

a

ρ

a

r

3

w

+c

r2

r

w

)

˙

θ

w

−r

w

m(c

r1

+gsin(α))

2.3.2 Model 3: Nonlinear Clutch and Drive Shaft Flexibility

When studying a clutch in more detail it is seen that the torsional ﬂexibility comes

from an arrangement of smaller springs in series with springs with much higher

stiﬀness. The reason for this arrangement is vibration insulation. When the angle

diﬀerence over the clutch starts from zero and increases, the smaller springs, with

2.3 Models Including the Clutch 13

θ

w

θ

m

J

m

J

w

+mr

2

w

k

d

c

d

M

m

+M

fr:m r

w

m(c

r1

+gsin(α))

k

c

c

c

θ

t

J

t

+J

f

/i

2

f

Figure 2.6 Model 2: Linear clutch and drive shaft torsional ﬂexibility.

Torque

θ

m

−θ

c

θ

c1

θ

c2

−θ

c1

−θ

c2

mechanical stop

k

c2

k

c1

mechanical stop

k

c2

k

c1

Figure 2.7 Nonlinear clutch characteristics.

stiﬀness k

c1

, are being compressed. This ends when they are fully compressed at

θ

c1

radians. If the angle is increased further, the stiﬀer springs, with stiﬀness k

c2

,

are beginning to be compressed. When θ

c2

is reached, the clutch hits a mechanical

stop. This clutch characteristics can be modeled as in Figure 2.7. The resulting

stiﬀness k

c

(θ

m

−θ

c

) of the clutch is given by

k

c

(x) =

_

_

_

k

c1

if |x| ≤ θ

c1

k

c2

if θ

c1

< |x| ≤ θ

c2

∞ otherwise

(2.47)

14 Chapter 2 Driveline Modeling

2α

θ

2

θ

1

k c

Figure 2.8 A shaft with stiﬀness k and internal damping c with a backlash of 2α

rad.

The torque M

kc

(θ

m

−θ

c

) from the clutch nonlinearity is

M

kc

(x) =

_

¸

¸

_

¸

¸

_

k

c1

x if |x| ≤ θ

c1

k

c1

θ

c1

+k

c2

(x −θ

c1

) if θ

c1

< x ≤ θ

c2

−k

c1

θ

c1

+k

c2

(x +θ

c1

) if −θ

c2

< x ≤ −θ

c1

∞ otherwise

(2.48)

The nonlinear model is given by the following equations. The linear part of the air

drag is included, as in the previous models.

Deﬁnition 2.3 Resulting equations for Model 3 - nonlinear clutch and drive shaft

ﬂexibility.

J

m

¨

θ

m

= M

m

−M

fr:m

−M

kc

(θ

m

−θ

t

i

t

) (2.49)

−c

c

(

˙

θ

m

−

˙

θ

t

i

t

)

(J

t

+J

f

/i

2

f

)

¨

θ

t

= i

t

_

M

kc

(θ

m

−θ

t

i

t

) +c

c

(

˙

θ

m

−

˙

θ

t

i

t

)

_

(2.50)

−(b

t

+b

f

/i

2

f

)

˙

θ

t

−

1

i

f

_

k

d

(θ

t

/i

f

−θ

w

) +c

d

(

˙

θ

t

/i

f

−

˙

θ

w

)

_

(J

w

+mr

2

w

)

¨

θ

w

= k

d

(θ

t

/i

f

−θ

w

) +c

d

(

˙

θ

t

/i

f

−

˙

θ

w

) (2.51)

−(b

w

+mc

r2

r

w

+c

w

A

a

ρ

a

r

3

w

)

˙

θ

w

−r

w

m(c

r1

+gsin(α))

where M

kc

(·) is given by (2.48) and c

c

denotes the damping coeﬃcient of the clutch.

2.4 Additional Dynamics

For high speeds, the linear part of the air drag, is not suﬃcient. Then the diﬀer-

ential equation describing the wheel and the vehicle (2.26), (2.46), and (2.51) can

be changed to include the nonlinear model of the air drag, described in (2.8). The

2.4 Additional Dynamics 15

model describing the wheel is

(J

w

+mr

2

w

)

¨

θ

w

= k

d

(θ

t

/i

f

−θ

w

) +c

d

(

˙

θ

t

/i

f

−

˙

θ

w

) (2.52)

−(b

w

+mc

r2

r

w

)

˙

θ

w

−r

w

m(c

r1

+gsin(α))

−

1

2

c

w

A

a

ρ

a

r

3

w

˙

θ

2

w

It is well known that elements like transmissions and drives introduce backlash.

Throughout this thesis the dead zone model will be used (Liversidge 1952). The

torque resulting from a shaft connected to a drive with backlash 2α is

M =

_

_

_

k(θ

1

−θ

2

−α) +c(

˙

θ

1

−

˙

θ

2

) if θ

1

−θ

2

> α

k(θ

1

−θ

2

+α) +c(

˙

θ

1

−

˙

θ

2

) if θ

1

−θ

2

< −α

0 if |θ

1

−θ

2

| < α

(2.53)

where k is the stiﬀness and c is the internal damping of the shaft, according to

Figure 2.8.

16 Chapter 2 Driveline Modeling

3

Field Trials and Modeling

Field trials are performed with a Scania truck. Diﬀerent road slopes and gears are

tested to study driveline resonances. The driving torque, engine speed, transmission

speed, and wheel speed are measured. As mentioned already in Chapter 2, these

measurements are used to build models by extending an initial model structure by

adding the eﬀect that seems to be the major cause for the deviation still left. There

has been some open questions regarding model structure in this study. One such

question is whether diﬀerences in engine speed and transmission speed is due to

clutch dynamics or has other causes. The parameters of the models are estimated.

The result is a series of models that describe the driveline in increasing detail.

3.1 The Truck

Tests were performed with a Scania 144L530 truck (Figure 3.1) on test roads in

S¨odert¨alje, Sweden, September 1995. The 6x2 truck (6 wheels, 2 driven) has a 14

liter V8 diesel engine (Figure 3.2) with maximum power of 530 Hp and maximum

torque of 2300 Nm. The DSC14 engine is connected to a manual range-splitter

transmission (Figure 3.3) via a clutch. The transmission has 14 gears and a hy-

draulic retarder. It is also equipped with the gear shifting system Opticruise (Ore-

hall 1995). A propeller shaft connects the output shaft of the transmission with

the ﬁnal drive. The drive shafts connect the ﬁnal drive to the wheels which has a

radius of r

w

= 0.52 m. The weight of the truck is m = 24 ton and the front area

is A

a

= 9 m

2

. The drag coeﬃcient is equal to c

w

= 0.6.

17

18 Chapter 3 Field Trials and Modeling

Figure 3.1 Scania 6x2 144L530 truck.

Figure 3.2 Scania DSC14 engine.

3.2 Measurement Description 19

Figure 3.3 Scania GRS900R range-splitter transmission with retarder and Opti-

cruise gear changing system.

3.2 Measurement Description

The truck is equipped with three sensors measuring the angle of the ﬂywheel of the

engine (θ

m

), the output shaft from the transmission (θ

t

), and the driving wheel

(θ

w

). The velocity of a rotating part is measured by using an inductive sensor

(Nwagboso 1993), which detects the time when cogs from a rotating cogwheel are

passing. This time sequence is then inverted to get the angle velocity. Hence, the

bandwidth of the measured signal depends on the speed and the number of cogs

the cogwheel is equipped with.

If the cogwheels of the three sensors are compared, the transmission speed

sensor has fewer cogs than the other two sensors, indicating that the bandwidth of

this signal is lower.

By measuring the amount of fuel that is fed to the engine, a measure of the

driving torque (M

m

) is obtained. The friction torque of the engine (M

fr:m

) is also

calculated online from a function given by Scania. From these two signals, the

torque u = M

m

−M

fr:m

acting on the driveline is calculated.

Hence, ﬁve signals are sampled (

˙

θ

m

,

˙

θ

t

,

˙

θ

w

, M

m

, M

fr:m

) with the Scania sam-

pling program “Truck-view”. Sampling is not equidistant in time, and the sample

period range from 0.05 s to 0.11 s (corresponding to sampling frequencies between

9 Hz and 20 Hz). The data has information up to half the sample period, which

means that there is information up to 10 Hz frequency.

The four signals used in the following modeling are calculated from the ﬁve

sampled signals. The four signals are (

˙

θ

m

,

˙

θ

t

,

˙

θ

w

, u = M

m

−M

fr:m

).

In the rest of this thesis, the control signal u = M

m

−M

fr:m

is assumed to be a

20 Chapter 3 Field Trials and Modeling

6.8 7 7.2 7.4 7.6 7.8 8 8.2 8.4

7

7.5

8

8.5

9

9.5

10

Time, [s]

[

r

a

d

/

s

]

Figure 3.4 Example of resampling a signal not equidistant in time (x). The

dotted line is the linear interpolation between the samples and the straight line is

the signal ﬁltered with 6 Hz.

continuous signal. This is reasonably for the frequency range considered for control

design. A motivation for this is that an eight-cylinder engine makes 80 strokes/s

at an engine speed of 1200 rev/min. The dynamics from fuel amount to engine

torque is not considered in this work.

Preprocessing Data

Since the sampling is not equidistant in time, the data sets are resampled. A new

data set is obtained by interpolating the old data using linear interpolation. This

introduces higher frequencies than those in the original data set. Therefore, the

interpolated data is low-pass ﬁltered with a frequency corresponding to half the

sampling frequency in the original data. This means a frequency in the interval 4.5

to 10 Hz. The chosen frequency is 6 Hz. This is done oﬄine and therefore without

phase shifts in the signals. An example of the resampling is seen in Figure 3.4.

Parameter Estimation Software

To estimate the parameters of the linear models derived in Chapter 2 the Sys-

tem Identiﬁcation Toolbox (Ljung 1995) is used. The prediction error estimation

method (PEM) for parameterized state-space representations is used to estimate

the unknown parameters and initial conditions.

In order to estimate the parameters and initial condition of the nonlinear

Model 3, the continuous model is discretized. This is done by using Euler’s method.

3.3 Experiments 21

For a continuous diﬀerential equation, ˙ x = f(x, u), the discrete version is

x

n

= x

n−1

+hf(x

n−1

, u

n−1

) (3.1)

where h is the sampling time. The global truncation error with this method equals

O(h). Therefore it is necessary to keep h small. A too small h can give numerical

problems and it also gives unnecessarily long iteration time. The data is resampled

at a sampling frequency of 1 kHz. Furthermore, the ﬁve diﬀerential equations,

describing Model 3, are scaled to be of the same magnitude.

For a given set of parameters, initial conditions, and control signal sequence u,

the state vector is calculated at each sample. By comparing the model output (y

m

,

y

t

, y

w

) with the measured signals (

˙

θ

m

,

˙

θ

t

,

˙

θ

w

) a cost function can be evaluated.

The cost function used is

∀i

_

(

˙

θ

m

(i) −y

m

(i))

2

+i

2

t

(

˙

θ

t

(i) −y

t

(i))

2

+i

2

t

i

2

f

(

˙

θ

w

(i) −y

w

(i))

2

_

(3.2)

where ∀i means that the sum is to taken over all samples in the estimation data.

The optimal parameters and initial conditions are the ones minimizing (3.2). The

data sets are divided into two parts to be used with the parameter estimation and

validation respectively.

For Model 1 the following states are used in the parameter estimation

x

1

= θ

m

/i

t

i

f

−θ

w

, x

2

=

˙

θ

m

, x

3

=

˙

θ

w

and for Models 2 and 3,

x

1

= θ

m

−θ

t

i

t

, x

2

= θ

t

/i

f

−θ

w

, x

3

=

˙

θ

m

, x

4

=

˙

θ

t

, x

5

=

˙

θ

w

are used. More details about the state-space representation can be found in Chap-

ter 4.

3.3 Experiments

A number of roads at Scania were used for testing. They have diﬀerent known

slopes. The sensor outputs described above were logged, with the friction torque

(M

fr:m

) subtracted from the driving torque (M

m

). Step input experiments were

done by repeatedly pressing and releasing the accelerator, in order to excite drive-

line oscillations.

Trial 1: The test was performed with step inputs on the accelerator with gear 1.

The road was almost ﬂat. The data is seen in Figure 3.5.

Trial 2: The test was performed with step inputs on the accelerator with gear 5.

The road was almost ﬂat. The data is seen in Figure 3.6.

Trial 3: The test was performed with step inputs on the accelerator with gear 5.

The road has 16 % slope. The data is seen in Figure 3.7.

Trial 4: The test was performed with step inputs on the accelerator with gear 8.

The road was almost ﬂat. The data is seen in Figure 3.8.

22 Chapter 3 Field Trials and Modeling

0 10 20 30

−500

0

500

1000

1500

[

N

m

]

0 10 20 30

50

100

150

200

250

300

[

r

a

d

/

s

]

0 10 20 30

5

10

15

20

[

r

a

d

/

s

]

Time, [s]

0 10 20 30

1

2

3

4

5

6

[

r

a

d

/

s

]

Time, [s]

Driving torque u = M

m

−M

fr:m Engine speed

˙

θ

m

Transmission speed

˙

θ

t

Wheel speed

˙

θ

w

Figure 3.5 Torque and angular velocities for a test with gear 1 and ﬂat road.

0 10 20 30

−500

0

500

1000

1500

[

N

m

]

0 10 20 30

100

150

200

250

300

[

r

a

d

/

s

]

0 10 20 30

10

15

20

25

30

35

40

[

r

a

d

/

s

]

Time, [s]

0 10 20 30

2

4

6

8

10

12

[

r

a

d

/

s

]

Time, [s]

Driving torque u = M

m

−M

fr:m Engine speed

˙

θ

m

Transmission speed

˙

θ

t

Wheel speed

˙

θ

w

Figure 3.6 Torque and angular velocities for a test with gear 5 and ﬂat road.

3.3 Experiments 23

0 10 20 30

−500

0

500

1000

1500

2000

[

N

m

]

0 10 20 30

50

100

150

200

250

[

r

a

d

/

s

]

0 10 20 30

10

15

20

25

30

35

[

r

a

d

/

s

]

Time, [s]

0 10 20 30

2

4

6

8

10

[

r

a

d

/

s

]

Time, [s]

Driving torque u = M

m

−M

fr:m Engine speed

˙

θ

m

Transmission speed

˙

θ

t

Wheel speed

˙

θ

w

Figure 3.7 Torque and angular velocities for a test with gear 5 and 16 % slope.

0 10 20 30

−500

0

500

1000

1500

2000

[

N

m

]

0 10 20 30

100

150

200

250

[

r

a

d

/

s

]

0 10 20 30

30

40

50

60

70

[

r

a

d

/

s

]

Time, [s]

0 10 20 30

8

10

12

14

16

18

20

[

r

a

d

/

s

]

Time, [s]

Driving torque u = M

m

−M

fr:m Engine speed

˙

θ

m

Transmission speed

˙

θ

t

Wheel speed

˙

θ

w

Figure 3.8 Torque and angular velocities for a test with gear 8 and ﬂat road.

24 Chapter 3 Field Trials and Modeling

3.4 Models

A number of driveline models were developed in Chapter 2. The choices made in

the modeling are now justiﬁed, by ﬁtting the models to measured data. Besides

the measured states (

˙

θ

m

,

˙

θ

t

,

˙

θ

w

), the load and the states describing the torsion of

the ﬂexibilities are estimated by the models.

The data shown are from Trial 1, where the driveline oscillations are well ex-

cited. Similar results are obtained from the other trials.

3.4.1 Inﬂuence from the Drive Shaft

First, the inﬂuence from the drive shaft is investigated by estimating the parameters

and initial conditions of Model 1. The engine speed and the wheel speed data is

used to estimate the parameters. In Figure 3.9, the results from Trial 1 are shown.

Here, also the transmission speed is plotted together with the model output engine

speed scaled with the conversion ratio in the transmission (i

t

). The plots are typical

examples that show that a major part of the driveline dynamics in the frequency

range up to 6 Hz is captured with a linear mass-spring model with the drive shafts

as the main ﬂexibility.

Result

• The main contribution to driveline dynamics from driving torque to engine

speed and wheel speed is the drive shaft.

• The true angle diﬀerence (x

1

= θ

m

/i

t

i

f

− θ

w

) is unknown, but the value

estimated by the model has physically reasonable values.

• The model output transmission speed (x

2

/i

t

) ﬁts the measured transmission

speed data well, but there are still deviations between model and measure-

ment.

3.4.2 Inﬂuence from the Propeller Shaft

The model equations (2.36) to (2.38) describes Model 1 extended with the propeller

shaft with stiﬀness k

p

and damping c

p

. The three inertias in the model are

J

1

= J

m

+J

t

/i

2

t

J

2

= J

f

(3.3)

J

3

= J

w

+mr

2

w

If the size of the three inertias are compared, the inertia of the ﬁnal drive (J

f

) is

considerably less than J

1

and J

2

in (3.3). Therefore, the model will act as if there

are two damped springs in series. The total stiﬀness of two undamped springs in

series is

k =

k

p

i

2

f

k

d

k

p

i

2

f

+k

d

(3.4)

3.4 Models 25

−0.5

0

0.5

r

a

d

100

150

200

250

300

r

a

d

/

s

5

10

15

r

a

d

/

s

14 16 18 20 22 24 26 28 30

1

2

3

4

5

r

a

d

/

s

Time, [s]

x

1

= θ

m

/i

t

i

f

−θ

w

x

2

=

˙

θ

m

x

2

/i

t

=

˙

θ

t

x

3

=

˙

θ

w

Figure 3.9 Model 1 estimated on data from Trial 1. The top ﬁgure shows the

drive shaft angle diﬀerence, and the bottom ﬁgures show the model outputs (x

2

,

x

3

) in dashed lines, together with the measured data in solid. The plots are typical

examples of that a major part of the dynamics is captured by a linear model with

drive shaft ﬂexibility.

26 Chapter 3 Field Trials and Modeling

whereas the total damping of two dampers in series is

c =

c

p

i

2

f

c

d

c

p

i

2

f

+c

d

(3.5)

The damping and the stiﬀness of the drive shaft in the previous section will thus

typically be underestimated due to the ﬂexibility of the propeller shaft. This eﬀect

will increase with lower conversion ratio in the ﬁnal drive, i

f

. The individual

stiﬀness values obtained from parameter estimation are somewhat lower than the

values obtained from material data.

3.4.3 Deviations between Engine Speed and Transmission

Speed

As mentioned above, there is good agreement between model and experiments for

u = M

m

− M

fr:m

,

˙

θ

m

, and

˙

θ

w

, but there is a slight deviation between measured

and estimated transmission speed. This deviation has a character of a phase shift

and some smoothing (signal levels and shapes agree). This indicates that there

is some additional dynamics between engine speed,

˙

θ

m

, and transmission speed,

˙

θ

t

. Two natural candidates are additional mass-spring dynamics in the driveline,

or sensor dynamics. The explanation is that there is a combined eﬀect, with the

major diﬀerence explained by the sensor dynamics. The motivation for this is that

the high stiﬀness of the clutch ﬂexibility (given from material data) can not result

in a diﬀerence of a phase shift form. Neither can backlash in the transmission

explain the diﬀerence, because then the engine and transmission speeds would be

equal when the backlash is at its endpoint.

As mentioned before, the bandwidth of the measured transmission speed is lower

than the measured engine and wheel speeds, due to fewer cogs in the sensor. It is

assumed that the engine speed and wheel speed sensor dynamics are not inﬂuencing

the data for frequencies up to 6 Hz. The speed dependence of the transmission

sensor dynamics is neglected. The following sensor dynamics are assumed, after

some comparison between sensor ﬁlters of diﬀerent order,

f

m

= 1

f

t

=

1

1 +αs

(3.6)

f

w

= 1

where a ﬁrst order ﬁlter with an unknown parameter α models the transmission

sensor. Figure 3.10 shows the conﬁguration with Model 1 and sensor ﬁlter f

m

, f

t

,

and f

w

. The outputs of the ﬁlters are y

m

, y

t

, and y

w

.

Now the parameters, initial condition, and the unknown ﬁlter constant α can

be estimated such that the model output (y

m

, y

t

, y

w

) ﬁts the measured data. The

result of this is seen in Figure 3.11 for Trial 1. The conclusion is that the main

part of the deviation between engine speed and transmission speed is due to sensor

dynamics. In Figure 3.12, an enlarged plot of the transmission speed is seen, with

the model output from Model 1 with and without sensor ﬁltering.

3.4 Models 27

θ

w θ

m

J

m

+J

t

/i

2

t

+J

f

/i

2

t

i

2

f J

w

+mr

2

w

k

c

M

m

−M

fr:m

r

w

m(c

r1

+gsin(α))

f

w

y

w

f

m

y

m

f

t

/i

t

y

t

Figure 3.10 Model 1 with sensor dynamics.

Result

• If Model 1 is equipped with a ﬁrst order sensor ﬁlter for the transmission

speed, all three velocities (

˙

θ

m

,

˙

θ

t

,

˙

θ

w

) are estimated by the model. The

model output ﬁts the data except for a number of time intervals where there

are deviations between model and measured data (see Figure 3.12). However,

these deviations will in the following be related to nonlinearities at low clutch

torques.

3.4.4 Inﬂuence from the Clutch

So far the clutch has been assumed stiﬀ, and the drawback with the models consid-

ered so far is that they are unable to estimate the angle diﬀerence over the clutch

that actually exists. Model 2 and 3 on the other hand estimate a clutch angle

diﬀerence.

Linear Clutch (Model 2)

The parameters and initial conditions of Model 2 are estimated with the sensor

dynamics described above. A problem when estimating the parameters of Model 2

is that the bandwidth of 6 Hz in the data is not enough to estimate the stiﬀness

k

c

in the clutch. Therefore, the value of the stiﬀness given by Scania is used and

ﬁxed, and the rest of the parameters are estimated.

The resulting clutch angle diﬀerence (x

1

= θ

m

− θ

t

i

t

) and the drive shaft an-

gle diﬀerence (x

2

= θ

t

/i

f

− θ

w

) are seen in Figure 3.13. The true values of these

torsions are not known, but the ﬁgure shows that the drive shaft torsion have

realistic values that agree with other experience. However, the clutch angle tor-

sion does not have realistic values, which can be seen when comparing with the

static nonlinearity in Figure 2.7. The model output velocities (

˙

θ

m

,

˙

θ

t

,

˙

θ

w

) show

28 Chapter 3 Field Trials and Modeling

−0.5

0

0.5

r

a

d

100

150

200

250

300

r

a

d

/

s

5

10

15

r

a

d

/

s

14 16 18 20 22 24 26 28 30

1

2

3

4

5

r

a

d

/

s

Time, [s]

x

1

= θ

m

/i

t

i

f

−θ

w

y

m

= x

2

=

˙

θ

m

y

t

= x

2

/i

t

(1 + αs) =

˙

θ

t

/(1 + αs)

y

w

= x

3

=

˙

θ

w

Figure 3.11 Model as in Figure 3.9 but with sensor dynamics included. The

top ﬁgure shows the angle diﬀerence over the drive shaft, and the bottom ﬁgures

show the model outputs (y

m

, y

t

, y

w

) in dashed, together with the measured data

in solid. The main part of the deviation between engine speed and transmission

speed is due to sensor dynamics. See also Figure 3.12.

3.4 Models 29

11 12 13 14 15 16 17 18

100

120

140

160

180

200

220

240

260

280

Time, [s]

r

a

d

/

s

Figure 3.12 Enlargement of part of Figure 3.11. Measured transmission speed

(solid), output from Model 1 without sensor ﬁltering (dashed), and output from

Model 1 with sensor ﬁltering (dash-dotted). The parameters are estimated on data

from Trial 1.

14 16 18 20 22 24 26 28 30

−0.02

−0.01

0

0.01

0.02

0.03

0.04

r

a

d

14 16 18 20 22 24 26 28 30

−0.5

0

0.5

r

a

d

Time, [s]

x

1

= θ

m

−θ

t

i

t

x

2

= θ

t

/i

f

−θ

w

Figure 3.13 Clutch angle diﬀerence (top ﬁgure) and drive shaft angle diﬀerence

(bottom ﬁgure) resulting from parameter estimation of Model 2 with sensor ﬁlter-

ing, on data from Trial 1. The true values of these torsions are not known, but the

plots show that the drive shaft angle has realistic values.

30 Chapter 3 Field Trials and Modeling

14 16 18 20 22 24 26 28 30

0

r

a

d

14 16 18 20 22 24 26 28 30

−0.5

0

0.5

r

a

d

Time, [s]

−θ

c1

θ

c1

x

1

= θ

m

/i

t

−θ

t

x

2

= θ

t

/i

f

−θ

w

Figure 3.14 Clutch angle diﬀerence (top ﬁgure) and drive shaft angle diﬀerence

(bottom ﬁgure) resulting from parameter estimation of Model 3 with sensor ﬁlter-

ing, on data from Trial 1. The true values of these torsions are not known, but the

plots show that they have realistic values.

no improvement compared to those generated by Model 1 with sensor dynamics,

displayed in Figure 3.11.

Result

• The model including a linear clutch does not improve the data ﬁt. The

interpretation of this is that the clutch model does not add information for

frequencies under 6 Hz.

Nonlinear Clutch (Model 3)

When estimating the parameters of Model 3, the clutch static nonlinearity is ﬁxed

with known physical values and the rest of the parameters are estimated, except

for the sensor ﬁlter which is the same as in the previous model estimations.

The resulting clutch angle diﬀerence (x

1

= θ

m

− θ

t

i

t

) and drive shaft angle

diﬀerence (x

2

= θ

t

/i

f

− θ

w

) after minimizing (3.2) are seen in Figure 3.14. The

true values of these torsions are not known as mentioned before. However, the

ﬁgure shows that both angles have realistic values that agree with other experience.

The model output velocities (

˙

θ

m

,

˙

θ

t

,

˙

θ

w

) show no improvement compared to those

generated by Model 1 with sensor dynamics, displayed in Figure 3.11.

In Figure 3.12 it was seen that the model with the sensor ﬁltering ﬁtted the

signal except for a number of time intervals with deviations. The question is if

3.4 Models 31

this is a result of some nonlinearity. Figure 3.15 shows the transmission speed

plotted together with the model output and the clutch angle torsion. It is clear

from Figure 3.15 that the deviation between model and experiments occurs when

the clutch angle passes the area with the low stiﬀness in the static nonlinearity (see

Figure 2.7).

Result

• The model including the nonlinear clutch does not improve the data ﬁt for

frequencies up to 6 Hz.

• The model is able to estimate a clutch angle with realistic values.

• The estimated clutch angle shows that when the clutch passes the area with

low stiﬀness in the nonlinearity, the model deviates from the data. The

reason is unmodeled dynamics at low clutch torques (Bj¨ ornberg, Pettersson,

and Nielsen 1996).

3.4.5 Model Validity

As mentioned before, the data sets are divided into two parts. The parameters are

estimated on the estimation data. The results are then evaluated on the validation

data, and these are the results shown in this chapter.

In the parameter estimation, the unknown load l, which vary between the tri-

als, is estimated. The load can be recalculated to estimate road slope, and the

calculated values agree well with the known values of the road slopes at Scania.

Furthermore, the estimation of the states describing the torsion of the clutch and

the drive shaft shows realistic values. This gives further support to model structure

and parameters.

The assumption about sensor dynamics in the transmission speed inﬂuencing

the experiments, agrees well with the fact that the engine speed sensor and the

wheel speed sensor have considerably higher bandwidth (more cogs) than the trans-

mission speed sensor.

When estimating the parameters of the models investigated, there is a problem

with identifying the viscous friction components b. The sensitivity in the model to

variations in the friction parameters is low, and the same model ﬁt can be obtained

for a range of frictions parameters.

32 Chapter 3 Field Trials and Modeling

11 12 13 14 15 16 17 18

r

a

d

11 12 13 14 15 16 17 18

4

6

8

10

12

14

16

18

r

a

d

Time, [s]

θ

c1

−θ

c1

x

1

= θ

m

/i

t

−θ

t

↑

↑

↑ ↑

↑

y

t

= x

4

/(1 + αs)

Figure 3.15 Clutch angle diﬀerence (top ﬁgure) and measured and estimated

transmission speed (bottom ﬁgure) from estimation of Model 3 with sensor dy-

namics on data from Trial 1. The result is that the miss ﬁt occurs when the clutch

angle passes the area with the low stiﬀness (|θ| < θ

c1

) in the static nonlinearity.

3.5 Summary 33

3.5 Summary

Parameter estimation of the models derived in Chapter 2 shows that a model with

one torsional ﬂexibility and two inertias is able to ﬁt the measured engine speed

and wheel speed. By considering the diﬀerence between the measured transmission

speed and wheel speed it is reasonably to deduce that the main ﬂexibility is the

drive shafts.

In order for the model to ﬁt the data from all three measured velocities, a

ﬁrst order sensor ﬁlter is added to the model, in accordance with properties of

the sensory system. It is shown that all three velocities are ﬁtted. Parameter

estimation of a model with a nonlinear clutch explains that the diﬀerence between

the measured data and the model occurs when the clutch transfers zero torque.

Further supporting facts of the models are that they give values to the non-

measured variables, drive shaft and clutch torsion, that agree with experience from

other sources. Furthermore, the known road slopes are well estimated.

The result is a series of models that describe the driveline in increasing detail

by, in each extension, adding the eﬀect that seems to be the major cause for the

deviation still left.

The result, from a user perspective, is that, within the frequency regime in-

teresting for control design, the mass-spring models with some sensor dynamics

(Model 1 and Model 2) give good agreement with experiments. They are thus suit-

able for control design. The major deviations left are captured by the nonlinear

eﬀects in Model 3 which makes this model suitable for verifying simulation studies

in control design.

34 Chapter 3 Field Trials and Modeling

4

Architectural Issues for Driveline

Control

As seen in the previous chapters, there are signiﬁcant torsional resonances in a

driveline. Active control of these resonances is the topic of the rest of this thesis.

Chapters 5 and 6 treats two diﬀerent problems. Besides formulating the control

problem in this chapter, there is one architectural issue that will be given special

attention. There are diﬀerent possible choices in driveline control between using

diﬀerent sensor locations, e.g. engine speed sensor, transmission speed sensor,

or wheel speed sensor. If the driveline were rigid, the choice would not matter,

since the sensor outputs would diﬀer only by a scaling factor. However, it will be

demonstrated that the presence of torsional ﬂexibilities implies that sensor choice

gives diﬀerent control problems. The diﬀerence can be formulated in control the-

oretic terms e.g. by saying that the poles are the same, but the zeros diﬀer both

in number and values. The issue of sensor location seems to be a little studied

topic (Kubrusly and Malebranche 1985; Ljung 1988), even though its relevance for

control characteristics.

The driveline model equations in Chapter 2 are written in state-space form in

Section 4.1. The formulation of performance output and controller structures used

in the rest of the thesis are given in Section 4.2. Control of resonant systems with

simple controllers is known to have structural properties e.g. with respect to sensor

location (Spong and Vidyasagar 1989), as mentioned before. In Section 4.3, these

diﬀerences are illustrated for driveline models. In Section 4.4, forming the main

contribution of this chapter, an investigation about how these properties transfers

when using more complicated controller structures like LQG/LTR is made. This

part is based on the material in Pettersson and Nielsen (1995).

35

36 Chapter 4 Architectural Issues for Driveline Control

4.1 State-Space Formulation

The input to the open-loop driveline system is u = M

m

− M

fr:m

, the diﬀerence

between the driving torque and the friction torque. Possible physical state variables

in the models of Chapter 2 are torques, angle diﬀerences, and angle velocity of any

inertia. In this work, the angle diﬀerence of each torsional ﬂexibility and the angle

velocity of each inertia is used as states. The state space representation is

˙ x = Ax +Bu +H l (4.1)

where A, B, H, x, and l are deﬁned next for the linear Models 1 and 2 in Chapter 2.

State-space formulation of Model 1:

x

1

= θ

m

/i

t

i

f

−θ

w

x

2

=

˙

θ

m

(4.2)

x

3

=

˙

θ

w

l = r

w

m(c

r1

+gsin(α))

giving

A =

_

_

0 1/i −1

−k/iJ

1

−(b

1

+c/i

2

)/J

1

c/iJ

1

k/J

2

c/iJ

2

−(c +b

2

)/J

2

_

_

, (4.3)

B =

_

_

0

1/J

1

0

_

_

, H =

_

_

0

0

−1/J

2

_

_

(4.4)

where

i = i

t

i

f

J

1

= J

m

+J

t

/i

2

t

+J

f

/i

2

t

i

2

f

J

2

= J

w

+mr

2

w

(4.5)

b

1

= b

t

/i

2

t

+b

f

/i

2

t

i

2

f

b

2

= b

w

+c

w

Aρr

3

w

+mc

r2

r

2

w

according to Deﬁnition 2.1.

State-space formulation of Model 2:

x

1

= θ

m

−θ

t

i

t

x

2

= θ

t

/i

f

−θ

w

x

3

=

˙

θ

m

(4.6)

x

4

=

˙

θ

t

x

5

=

˙

θ

w

4.1 State-Space Formulation 37

A is given by the matrix

_

_

_

_

_

_

0 0 1 −i

t

0

0 0 0 1/i

f

−1

−k

c

/J

1

0 −c

c

/J

1

c

c

i

t

/J

1

0

k

c

i

t

/J

2

−k

d

/i

f

J

2

c

c

i

t

/J

2

−(c

c

i

2

t

+b

2

+c

d

/i

2

f

)/J

2

c

d

/i

f

J

2

0 k

d

/J

3

0 c

d

/i

f

J

3

−(b

3

+c

d

)/J

3

_

_

_

_

_

_

and

B =

_

_

_

_

_

_

0

0

1/J

1

0

0

_

_

_

_

_

_

, H =

_

_

_

_

_

_

0

0

0

0

−1/J

2

_

_

_

_

_

_

(4.7)

where

J

1

= J

m

J

2

= J

t

+J

f

/i

2

f

J

2

= J

w

+mr

2

w

(4.8)

b

2

= b

t

+b

f

/i

2

f

b

3

= b

w

+c

w

Aρr

3

w

+c

r2

r

w

according to Deﬁnition 2.2.

4.1.1 Disturbance Description

The disturbance l can be seen as a slow-varying part resulting from the rolling

resistance and the road slope plus and additive disturbance v. A second state

disturbance n is a disturbance acting on the input of the system. This disturbance

is considered because the ﬁring pulses in the driving torque can be seen as an

additive disturbance acting on the input. The state-space description is

˙ x = Ax +Bu +Bn +H l +Hv (4.9)

with x, A, B, H, and l deﬁned in (4.2) to (4.8).

4.1.2 Measurement Description

For controller synthesis it is of fundamental interest which physical variables of

the process that can be measured. In the case of a vehicular driveline the normal

sensor alternative is an inductive sensor mounted on a cog wheel measuring the

angle, as mentioned before. Sensors that measure torque are expensive, and are

seldom used in a production vehicular applications.

The output of the process is deﬁned as a combination of the states given by the

matrix C in

y = Cx +e (4.10)

38 Chapter 4 Architectural Issues for Driveline Control

F

y

(s)

M

C

e

z

x

u

y

1/s

A

B

˙ x

F

r

(s)

r

D

H

l +v

Figure 4.1 Plant and controllers F

r

and F

y

.

where e is a measurement disturbance.

In this work, only angle velocity sensors are considered, and therefore, the

output of the process is one/some of the state variables deﬁning an angle velocity.

Especially, the following are deﬁned (corresponding to a sensor on

˙

θ

m

and

˙

θ

w

for

Model 1).

C

m

= (0 1 0) (4.11)

C

w

= (0 0 1) (4.12)

4.2 Controller Formulation

The performance output z is the combination of states that has requirements to

behave in a certain way. This combination of the states is described by the matrices

M and D in the following way

z = Mx +Du (4.13)

The control problem can be seen as in Figure 4.1. The unknown controllers F

r

and F

y

are to be designed such that that the performance output (4.13) meets its

requirements (deﬁned later).

In this thesis controllers will be designed as state-feedback controllers exten-

sively except for a few simple examples. The control signal u is a linear function of

the states (if they are all measured) or else the state estimates, which are obtained

from a Kalman ﬁlter,

u = l

0

r −K

c

ˆ x (4.14)

where, r represents the commanded signal with the gain l

0

, and K

c

is the state-

feedback matrix.

4.3 Some Feedback Properties 39

Identifying the matrices F

r

(s) and F

y

(s) in Figure 4.1 gives

F

y

(s) = K

c

(sI −A+BK

c

+K

f

C)

−1

K

f

(4.15)

F

r

(s) = l

0

_

1 −K

c

(sI −A+BK

c

+K

f

C)

−1

B

_

The closed-loop transfer functions from r, v, and e to the control signal u are

given by

G

ru

=

_

I −K

c

(sI −A+BK

c

)

−1

B

_

l

0

r (4.16)

G

vu

= K

c

(sI −A+K

f

C)

−1

N −K

c

(sI −A+BK

c

)

−1

N (4.17)

−K

c

(sI −A+BK

c

)

−1

BK

c

(sI −A+K

f

C)

−1

N

G

eu

= K

c

_

(sI −A+BK

c

)

−1

BK

c

−I

_

(sI −A+K

f

C)

−1

K

f

(4.18)

The transfer functions to the performance output z are given by

G

rz

= (M(sI −A)

−1

B +D)G

ru

(4.19)

G

vz

= M(sI −A+BK

c

)

−1

BK

c

(sI −A+K

f

C)

−1

N (4.20)

+M(sI −A+BK

c

)

−1

N +DG

wu

G

ez

= (M(sI −A)

−1

B +D)G

vu

(4.21)

Two return ratios results, which characterizes the closed-loop behavior at the

plant output and input respectively

GF

y

= C(sI −A)

−1

BF

y

(4.22)

F

y

G = F

y

C(sI −A)

−1

B (4.23)

When only one sensor is used, these return ratios are scalar and thus equal.

LQG/LTR is not directly applicable to driveline control with more than one

sensor as input to the observer. This is because there are unequal number of sensors

and control signals. Therefore, it is important with the type of investigation about

the structural properties made in this chapter, when extending to more sensors.

This is however not considered in this work.

4.3 Some Feedback Properties

The performance output when controlling the driveline to a certain speed is the

velocity of the wheel, deﬁned as

z =

˙

θ

w

= C

w

x (4.24)

When studying the closed-loop control problem with a sensor on

˙

θ

m

or

˙

θ

w

, two

diﬀerent control problems results. In Figure 4.2 a root locus with respect to a

P-controller gain is seen for two gears using velocity sensor

˙

θ

m

and

˙

θ

w

respectively.

The open-loop transfer functions from u to engine speed G

um

has three poles and

40 Chapter 4 Architectural Issues for Driveline Control

−10 −5 0

−6

−4

−2

0

2

4

6

I

m

a

g

−10 −5 0

−6

−4

−2

0

2

4

6

I

m

a

g

−20 −10 0

−15

−10

−5

0

5

10

15

I

m

a

g

Real

−20 −10 0

−15

−10

−5

0

5

10

15

I

m

a

g

Real

Gear 1 and

˙

θ

m

feedback Gear 1 and

˙

θ

w

feedback

Gear 8 and

˙

θ

m

feedback Gear 8 and

˙

θ

w

feedback

Figure 4.2 Root locus with respect to a P controller gain, for gear 1 (top ﬁgures)

and gear 8 (bottom ﬁgures) with sensor on

˙

θ

m

(left ﬁgures), and

˙

θ

w

(right ﬁgures).

The cross represent the open-loop poles, while the rings represents the open-loop

zeros. The system goes unstable when the

˙

θ

w

gain is increased, but is stable for

all

˙

θ

m

gains.

two zeros, as can be seen in Figure 4.2. G

uw

on the other hand has one zero, and

the same poles. Hence, the relative degree of G

um

is one and G

uw

has a relative

degree of two. This means that when

˙

θ

w

feedback is used, and the gain is increased,

two poles must go to inﬁnity which makes the system unstable. When the velocity

sensor

˙

θ

m

is used, the relative degree is one, and the closed-loop system is stable

for all gains. (Remember that

˙

θ

w

is the performance output and thus desirable to

use.)

The same eﬀect can be seen in the step response when the P controller is used.

Figure 4.3 demonstrates the problem with resonances that occurs with increasing

gain for the two cases of feedback. When the engine speed sensor is used, the engine

speed behaves well when the gain is increased, but the resonance in the drive shaft

makes the wheel speed oscillate. When using

˙

θ

w

feedback it is diﬃcult to increase

4.4 Driveline Control with LQG/LTR 41

0

0.5

1

1.5

A

m

p

l

i

t

u

d

e

0

20

40

60

80

A

m

p

l

i

t

u

d

e

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

0

0.5

1

1.5

2

Time, [s]

A

m

p

l

i

t

u

d

e

Wheel speed,

˙

θ

w

feedback

Engine speed,

˙

θ

m

feedback

Wheel speed,

˙

θ

m

feedback

Figure 4.3 Step responses using a P controller with diﬀerent gains on Model 1

with gear 1. With

˙

θ

w

feedback (top ﬁgure), increased rise time results in instability.

With

˙

θ

m

feedback (bottom ﬁgures), increased gain results in a well behaved engine

speed, but an oscillating wheel speed.

the bandwidth, since the poles moves closer to the imaginary axis, giving a resonant

system.

The characteristic results in Figures 4.2 and 4.3 only depend on the relative de-

gree, and are thus parameter independent. However, this observation may depend

on feedback structure, and therefore a more detailed analysis is performed in the

following sections.

4.4 Driveline Control with LQG/LTR

Diﬀerent sensor locations result in diﬀerent control problems with diﬀerent inherent

characteristics, as seen in the previous section. The topic of this section is to show

how this inﬂuences control design when using LQG/LTR. The reason for using

42 Chapter 4 Architectural Issues for Driveline Control

LQG/LTR, in this principle study, is that it oﬀers a control design method resulting

in a controller and observer of the same order as the plant model, and it is also an

easy method for obtaining robust controllers.

4.4.1 Transfer Functions

When comparing the control problem with using

˙

θ

m

or

˙

θ

w

as sensors, open-loop

transfer functions G

um

and G

uw

results. These have the same number of poles but

diﬀerent number of zeros as mentioned before. Two diﬀerent closed-loop systems

results depending on which sensor that is used.

Feedback from

˙

θ

w

A natural feedback conﬁguration is to use the performance output,

˙

θ

w

. Then

among others the following transfer functions results, where (4.16) to (4.21) are

used together with the matrix inversion lemma

G

rz

=

G

uw

F

y

F

r

1 +G

uw

F

y

= T

w

F

r

(4.25)

G

nu

= =

1

1 +G

uw

F

y

= S

w

(4.26)

where n is the input disturbance. The transfer functions S

w

and T

w

are, as usual,

the sensitivity function and the complementary sensitivity function. Also, as usual,

S

w

+T

w

= 1 (4.27)

Feedback from

˙

θ

m

The following transfer functions results if the sensor measures

˙

θ

m

G

rz

=

G

uw

F

y

F

r

1 +G

um

F

y

(4.28)

G

nu

=

1

1 +G

um

F

y

(4.29)

The diﬀerence between the two feedback conﬁgurations is that the return diﬀerence

is 1 +G

uw

F

y

or 1 +G

um

F

y

.

It is desirable to have sensitivity functions that corresponds to y =

˙

θ

m

and

z =

˙

θ

w

. The following transfer functions are deﬁned

S

m

=

1

1 +G

um

F

y

, T

m

=

G

um

F

y

1 +G

um

F

y

(4.30)

These transfer functions corresponds to a conﬁguration where

˙

θ

m

is the output (i.e.

y = z =

˙

θ

m

). Using (4.28) it is natural to deﬁne T

m

by

T

m

=

G

uw

F

y

1 +G

um

F

y

= T

m

G

uw

G

um

(4.31)

4.4 Driveline Control with LQG/LTR 43

The functions S

m

and T

m

describe the design problem when feedback from θ

m

is

used.

When combining (4.30) and (4.31), the corresponding relation to (4.27) is

S

m

+T

m

G

um

G

uw

= 1 (4.32)

If S

m

is made zero for some frequencies in (4.32), then T

m

will not be equal to

one, as in (4.27). Instead, T

m

= G

uw

/G

um

for these frequency domains.

Limitations on Performance

The relations (4.27) and (4.32) will be the fundamental relations for discussing

design considerations. The impact of the ratio G

uw

/G

um

will be analyzed in the

following sections.

Deﬁnition 4.1 T

m

in (4.31) is the modiﬁed complementary sensitivity function,

and G

w/m

= G

uw

/G

um

is the dynamic output ratio.

4.4.2 Design Example with a Simple Mass-Spring Model

Linear Quadratic Design with Loop Transfer Recovery will be treated in four cases,

being combinations of two sensor locations,

˙

θ

m

or

˙

θ

w

, and two models with the

same structure, but with diﬀerent parameters. Design without pre ﬁlter (F

r

= 1)

is considered.

The section covers a general plant with n inertias connected by k −1 torsional

ﬂexibilities, without damping and load, and with unit conversion ratio. There are

(2n − 1) poles, and the location of the poles are the same for the diﬀerent sensor

locations. The number of zeros depends on which sensor that is used, and when

using

˙

θ

w

there are no zeros. When using feedback from

˙

θ

m

there are (2n−2) zeros.

Thus, the transfer functions G

um

and G

uw

, have the same denominators, and a

relative degree of 1 and (2n −1) respectively.

Structural Properties of Sensor Location

The controller (4.15) has a relative degree of one. The relative degree of G

um

F

y

is

thus 2, and the relative degree of G

uw

F

y

is 2n. When considering design, a good

alternative is to have relative degree one in GF

y

, implying inﬁnite gain margin and

high phase margin.

When using G

um

F

y

, one pole has to be moved to inﬁnity, and when using

G

uw

F

y

, 2n−1 poles have to be moved to inﬁnity, in order for the ratio to resemble

a ﬁrst order system at high frequencies. It could be expected that a higher control

signal is needed for

˙

θ

w

feedback in order to move the poles towards inﬁnity.

When the return ratio behaves like a ﬁrst order system, also the closed-loop

transfer function behaves like one. This conﬂicts with the design goal of having

44 Chapter 4 Architectural Issues for Driveline Control

10

−1

10

0

10

1

10

2

−60

−50

−40

−30

−20

−10

0

10

Frequency (rads^−1)

G

a

in

(

d

B

)

Figure 4.4 G

w/m

for a) (solid) and b) (dashed).

a steep roll-oﬀ rate for the closed-loop system in order to attenuate measurement

noise. Hence, there is a trade-oﬀ when using

˙

θ

w

feedback.

When using

˙

θ

m

feedback, there is no trade-oﬀ, since the relative degree of G

um

is one.

Structure of G

w/m

We have in the previous simple examples seen that the relative degree and the

zeros are important. The dynamic output ratio contains exactly this information

and nothing else.

For low frequencies the dynamic output ratio has gain equal to one,

¸

¸

G

w/m

(0)

¸

¸

= 1

(if the conversion ratio is equal to one). Furthermore, G

w/m

has a relative degree

of 2n − 2 and thus, a high frequency gain roll-oﬀ rate of 20(2n − 2) dB/decade.

Hence, the dynamic output ratio gives the closed-loop transfer function T

m

a high

frequency gain roll-oﬀ rate of q

m

+ 20(2n −2) dB/decade, where q

m

is the roll-oﬀ

rate of G

um

F

y

. When using

˙

θ

w

feedback, T

w

will have the same roll-oﬀ rate as

G

uw

F

y

.

Parametric properties of G

w/m

Typical parametric properties of G

w/m

can be seen in the following example.

Example 4.1 Two diﬀerent plants are considered:

a) J

1

= 0.0974, J

2

= 0.0280, k = 2.80, c = 0, b

1

= 0.0244, b

2

= 0.566, l = 0.

b) J

1

= 0.0974, J

2

= 0.220, k = 5.50, c = 0, b

1

= 1.70, b

2

= 0.660, l = 0.

with labels according to the state-space formulation in Section 4.1. The shape of

G

w/m

can be seen in Figure 4.4.

4.4 Driveline Control with LQG/LTR 45

LQG Designs

Integral action is included by augmenting the state to attenuate step disturbances

in v (Maciejowski 1989). The state-space realization A

a

, B

a

, M

a

, C

wa

, and C

ma

re-

sults. The Kalman-ﬁlter gain, K

f

, is derived using a Riccati equation (Maciejowski

1989)

P

f

A

T

+AP

f

−P

f

C

T

V

−1

CP

f

+BWB

T

= 0 (4.33)

The covariances W and V , of v and e respectively, are adjusted until the return

ratio

C(sI −A)

−1

K

f

, K

f

= P

f

C

T

V

−1

(4.34)

and the closed-loop transfer functions S and T show satisfactory performance.

The Nyquist locus remains outside the unit circle centered at −1. This means that

there is inﬁnite gain margin, and a phase margin of at least 60

◦

. Furthermore, the

relative degree is one, and |S| ≤ 1.

Design for

˙

θ

w

feedback. W is adjusted (and thus F

y

(s)) such that S

w

and T

w

show a satisfactory performance, and that the desired bandwidth is obtained. The

design in Example 4 is shown in Figure 4.5. Note that the roll-oﬀ rate of T

w

is 20

dB/decade.

Design for

˙

θ

m

feedback. W is adjusted (and thus F

y

(s)) such that S

m

and T

m

(and thus

˙

θ

m

) show a satisfactory performance. Depending on the shape of G

w/m

for middle high frequencies, corrections in W must be taken such that T

m

achieves

the desired bandwidth. If there is a resonance peak in G

w/m

, the bandwidth in

T

m

is chosen such that the peak is suppressed. Figure 4.5 shows such an example,

˙

θ

m

feedback in b), where the bandwidth is lower in order to suppress the peak in

G

w/m

. Note also the diﬀerence between S

w

and S

m

.

The parameters of the dynamic output ratio are thus important in the LQG

step of the design.

Loop Transfer Recovery, LTR

The next step in the design process is to include K

c

, and recover the satisfactory re-

turn ratio obtain previously. When using the combined state feedback and Kalman

ﬁlter, the return ratio is GF

y

= C(sI −A)

−1

BK

c

(sI −A+BK

c

+K

f

C)

−1

K

f

. A

simplistic LTR can be obtained by using K

c

= ρC and increasing ρ. As ρ is in-

creased, 2n − 1 poles move towards the open system zeros. The remaining poles

move towards inﬁnity (compare to Section 5.1). If the Riccati equation

A

T

P

c

+P

c

A−P

c

BR

−1

B

T

P

c

+C

T

QC = 0 (4.35)

is solved with Q = ρ, and R = 1, K

c

=

√

ρC is obtained in the limit, and to

guarantee stability, this K

c

is used for recovery.

In Figures 4.6 and 4.7, the recovered closed-loop transfer functions, Nyquist

locus, and control signal are seen.

46 Chapter 4 Architectural Issues for Driveline Control

10

0

10

2

−60

−40

−20

0

G

a

i

n

(

d

B

)

a)

10

0

10

2

−60

−40

−20

0

G

a

i

n

(

d

B

)

a)

10

0

10

2

−60

−40

−20

0

Frequency (rads^−1)

G

a

i

n

(

d

B

)

b)

10

0

10

2

−60

−40

−20

0

Frequency (rads^−1)

G

a

i

n

(

d

B

)

b)

Figure 4.5 Closed-loop transfer functions S (left ﬁgures), and T (right ﬁgures).

Feedback from

˙

θ

w

in solid curves, and feedback from

˙

θ

m

in dashed curves. T

m

is

seen in right ﬁgures in dash-dot curves. W = 15 (θ

w

, a), W = 5 · 10

4

(θ

m

, a),

W = 5 · 10

2

(θ

w

, b), and W = 50 (θ

m

, b).

Recovery for

˙

θ

w

feedback. There is a trade-oﬀ when choosing an appropriate

ρ. A low ρ gives good attenuation of measurement noise and a low control signal,

but in order to have good stability margins, a high ρ must be chosen. This gives

an increased control signal, and a 20 dB/decade roll-oﬀ rate in T

w

for a wider

frequency range.

Recovery for

˙

θ

m

feedback. There is no trade-oﬀ when choosing ρ. It is possible to

achieve good recovery with reasonable stability margins and control signal, together

with a steep roll-oﬀ rate.

The structural properties i.e. the relative degrees are thus dominant in deter-

mining the LTR step of the design.

4.5 Summary

Control and damping of torsional oscillations in vehicular drivelines is an impor-

tant problem. Diﬀerent sensor locations give diﬀerent transfer functions, G

um

or

G

uw

. These functions have the same poles, but have diﬀerent relative degree and

diﬀerent zeros. The dynamic output ratio, G

w/m

, exactly captures these diﬀerences

and nothing else. The problem that the performance output signal is not the same

4.5 Summary 47

10

0

10

2

−60

−40

−20

0

G

a

i

n

(

d

B

)

a)

10

0

10

2

−60

−40

−20

0

G

a

i

n

(

d

B

)

a)

10

0

10

2

−60

−40

−20

0

Frequency (rads^−1)

G

a

i

n

(

d

B

)

b)

10

0

10

2

−60

−40

−20

0

Frequency (rads^−1)

G

a

i

n

(

d

B

)

b)

Figure 4.6 Closed-loop transfer functions S (left ﬁgures), and T (right ﬁgures)

after recovery. Feedback from

˙

θ

w

in solid curves, and feedback from

˙

θ

m

in dashed

curves. T

m

is seen in right ﬁgures in dash-dot curves. For the

˙

θ

m

design ρ = 10

6

(a) and ρ = 10

5

(b) is used, and for the

˙

θ

w

design ρ = 10

4

, 10

8

, and 10

11

is used

in both a) and b).

as the measured output signal is handled by introducing a modiﬁed complementary

sensitivity function, being modiﬁed with G

w/m

. Both structural and parameter

dependent aspects of sensor location have been characterized. In LQG/LTR, pa-

rameter dependent properties dominate in the LQG step of the design, whereas

structural properties, i.e. sensor location, dominate in the LTR step.

48 Chapter 4 Architectural Issues for Driveline Control

−0.2 −0.1 0

−0.2

−0.1

0

0.1

0.2

I

m

a

g

A

x

i

s

a)

10

0

10

5

−50

0

50

G

a

i

n

(

d

B

)

a)

−0.2 −0.1 0

−0.2

−0.1

0

0.1

0.2

Real Axis

I

m

a

g

A

x

i

s

b)

10

0

10

5

−50

0

50

Frequency (rads^−1)

G

a

i

n

(

d

B

)

b)

Figure 4.7 Nyquist plot of return ratio (left ﬁgures) and F

y

/(1 +G

uw

F

y

) (right

ﬁgures). Feedback from

˙

θ

w

in solid curves, and feedback from

˙

θ

m

in dashed curves.

For the

˙

θ

m

design ρ = 10

6

(a) and ρ = 10

5

(b) is used, and for the

˙

θ

w

design

ρ = 10

4

, 10

8

, and 10

11

is used in both a) and b). A dash-dotted circle with radius

one and centered at -1, is also shown in the Nyquist plots.

5

Speed Controller Design and

Simulations

Driveline oscillations may occur in diﬀerent modes of operation. Active damping

in two modes will be treated in this and next chapter. The ﬁrst problem is wheel

speed oscillations following a change in accelerator pedal position, known as vehicle

shuﬄe (Mo, Beaumount, and Powell 1996; Pettersson and Nielsen 1995). Tradi-

tionally in diesel powered trucks, the relation between the accelerator pedal and

the amount of fuel metered by the diesel pump is governed by a system called RQV

control. The RQV control gives a speciﬁc character to the driving feeling e.g. when

going uphill and downhill. This driving character is important to maintain when

extending speed control with active damping. Traditional RQV control is explained

in Section 5.1. Thereafter, the speed control problem keeping RQV characteristics

is formulated in Section 5.2. The sections following study the problem using avail-

able computationally powerful methods like LQG/LTR. Sensor location, inﬂuence

from disturbances, and load estimations are treated.

5.1 RQV Control

RQV control is the traditional diesel engine control scheme steaming from the

mechanical centrifugal governor, used to control the diesel pump (Bosch 1993). In

todays electronically controlled engines, the RQV scheme is still used for controlling

the fuel amount to the engine, since the driver wants the engine to behave as with

the mechanical governor.

RQV control is essentially a P controller with the accelerator as reference value

and a sensor measuring the engine speed. The RQV controller has no information

about the load, and a nonzero load (e.g. going uphill or downhill) gives a stationary

49

50 Chapter 5 Speed Controller Design and Simulations

error. The RQV controller is described by

u = u

0

+K

p

(ri −

˙

θ

m

) (5.1)

where i = i

t

i

f

is the conversion ratio of the driveline, K

p

is the controller gain, and

r is the reference velocity. The constant u

0

is a function of the speed, but not the

load since this is unknown to the RQV controller. RQV control is demonstrated in

the following example.

Example 5.1 Consider the truck modeled in Chapters 2 and 3 traveling at a speed

of 2 rad/s (3.6 km/h) with gear 1 and a total load of 3000 Nm (≈ 2 % road slope).

Let the new desired velocity be r = 2.3 rad/s. Figure 5.1 shows the RQV control

law (5.1) applied to Model 1 with three gains K

p

. In the plots, u

0

is calculated

such that the stationary level is the same for the three gains. (Otherwise there

would be a gain dependent stationary error.)

When the controller gain is increased, the rise time and the overshoot is in-

creasing. Hence, there is a trade-oﬀ between short rise time and little overshoot.

Furthermore, the engine speed behaves well, but the ﬂexibility of the driveline

causes the wheel speed to oscillate, when the gain is increased.

Figure 5.2 shows the transfer functions from load and measurement disturbances

v and e to the performance output, when the RQV controller is used. The resonance

peak in the transfer functions is increasing when the controller gain is increased.

5.2 Problem Formulation

The performance output for the speed controller is the wheel speed, z =

˙

θ

w

, as

deﬁned in Chapter 4. In Figure 5.3, the wheel speed z is seen for Models 1 and

2. Model 2 adds a second resonance peak from the clutch. Furthermore, the high

frequency roll-oﬀ rate is steeper for Model 2 than for Model 1. Note that the

transfer function from the load l to the performance output z is the same for the

two models. This chapter deals with the development of a controller based on

Model 1.

5.2.1 Mathematical Problem Formulation

A ﬁrst possible attempt for speed control is a scheme of applying the engine torque

to the driveline such that the following cost function is minimized

lim

T→∞

_

T

0

(z −r)

2

(5.2)

where r is the reference velocity given by the driver. If a control law is to minimize

the cost function, then (5.2) can be made arbitrarily small if there are no restrictions

on the control signal u, since the plant model is linear. A diesel engine can only

produce torque in a certain range, and therefore, (5.2) is extended such that a large

control signal adds to the cost function.

5.2 Problem Formulation 51

0 1 2 3 4 5 6 7 8

115

120

125

130

135

140

S

p

e

e

d

,

[

r

a

d

/

s

]

0 1 2 3 4 5 6 7 8

1.9

2

2.1

2.2

2.3

2.4

S

p

e

e

d

,

[

r

a

d

/

s

]

0 1 2 3 4 5 6 7 8

0

500

1000

1500

2000

Time, [s]

T

o

r

q

u

e

,

[

N

m

]

Engine speed

˙

θ

m

Wheel speed

˙

θ

w

(performance output)

Control signal u

Figure 5.1 RQV control (5.1) of Model 1. Controller gains K

p

= 8, K

p

= 25, and

K

p

= 85 are shown in solid, dashed and dash-dotted lines respectively. Increased

gain results in a well behaved engine speed, but an oscillating wheel speed.

The stationary point z = r is reached if a control signal u

0

is used. This torque

is a function of the reference value r and the load l. For a given wheel speed

˙

θ

w

and load l the driveline has the following stationary point

x

0

(

˙

θ

w

, l) =

_

_

b

2

/k 1/k

i 0

1 0

_

_

_

˙

θ

w

l

_

= δ

x

˙

θ

w

+δ

l

l (5.3)

u

0

(

˙

θ

w

, l) =

_

(b

1

i

2

+b

2

)/i 1/i

_

_

˙

θ

w

l

_

= λ

x

˙

θ

w

+λ

l

l (5.4)

The stationary point is obtained by solving

Ax +Bu +Hl = 0 (5.5)

for x and u, where A, B, H, and x is given by (4.2) to (4.5).

52 Chapter 5 Speed Controller Design and Simulations

10

−2

10

−1

10

0

10

1

10

2

10

3

−140

−120

−100

−80

G

a

i

n

[

d

B

]

10

−2

10

−1

10

0

10

1

10

2

10

3

−150

−100

−50

Frequency [rad/s]

G

a

i

n

[

d

B

]

Closed-loop transfer function G

vz

Closed-loop transfer function G

ez

Figure 5.2 Closed-loop transfer functions G

vz

and G

ez

when using the RQV

control law (5.1) for the controller gains K

p

= 8 (solid), K

p

= 25 (dashed), and

K

p

= 85 (dash-dotted). The resonance peaks increase with increasing gain.

By using these equations, the cost function can be written such that a control

signal u that deviates from the stationary value u

0

(r, l) adds to the cost function.

The extended cost function is given by

lim

T→∞

_

T

0

(z −r)

2

+η(u −u

0

(r, l))

2

(5.6)

where η is used to control the trade-oﬀ between rise time and control signal ampli-

tude.

The controller that minimizes (5.6) has no stationary error, since the load l is

included and thus compensated for. However, it is desirable that the stationary

error characteristic for the RQV controller is maintained in the speed controller, as

mentioned before. A stationary error comparable with that of the RQV controller

can be achieved by using only a part of the load l in the criterion (5.6), as will

be demonstrated in Section 5.3.1. Furthermore, the following demands should be

considered.

• The control signal can not exceed u

min

= −300 Nm or u

max

= 2300 Nm.

• The inﬂuence from load and measurement disturbances on the performance

output, wheel speed, should be minimized. Load disturbances result from,

for instance, road roughness or impulses from towed trailers.

5.3 Speed Control with Active Damping and RQV Behavior 53

−200

−150

−100

−50

0

G

a

i

n

[

d

B

]

10

−2

10

−1

10

0

10

1

10

2

10

3

−140

−120

−100

−80

−60

Frequency [rad/s]

G

a

i

n

[

d

B

]

G

uz

for Model 1 and 2

G

lz

for Model 1 and 2

Figure 5.3 Transfer functions from control signal u and load l to performance

output z. Model 1 is shown in solid and Model 2 is shown in dashed. The modeled

clutch gives a second resonance peak and a steeper roll-oﬀ rate.

5.3 Speed Control with Active Damping and RQV

Behavior

The problem formulation (5.6) will be treated in two steps. First without RQV

behavior i.e. using the load l, and then extending to RQV behavior. The problem

formulation (5.6) is in this section solved with LQG technique. This is done by lin-

earizing the driveline model and rewriting (5.6) in terms of the linearized variables.

A state-feedback matrix is derived that minimizes (5.6). This is done by solving a

Riccati equation. The derived feedback law is a function of η which is chosen such

that a feasible control signal is used.

The model (4.1)

˙ x = Ax +Bu +Hl (5.7)

is aﬃne since it includes a constant term l. The model is linearized in the neigh-

borhood of a stationary point (x

0

, u

0

). The linear model is

∆˙ x = A∆x +B∆u (5.8)

where

∆x = x −x

0

∆u = u −u

0

(5.9)

x

0

= x

0

(x

30

, l)

54 Chapter 5 Speed Controller Design and Simulations

u

0

= u

0

(x

30

, l)

where the stationary point (x

0

, u

0

) is given by (5.3) and (5.4). Note that the linear

model is the same for all stationary points.

The problem is to devise a feedback-control law that minimizes the cost function

(5.6). The cost function is expressed in terms of ∆x and ∆u by using (5.9)

lim

T→∞

_

T

0

(M(x

0

+ ∆x) −r)

2

+η(u

0

+ ∆u −u

0

(r, l))

2

(5.10)

= lim

T→∞

_

T

0

(M∆x +r

1

)

2

+η(∆u +r

2

)

2

(5.11)

with

r

1

= Mx

0

−r (5.12)

r

2

= u

0

−u

0

(r, l)

In order to minimize (5.10) a Riccati equation is solved. Then the constants r

1

and

r

2

must be expressed in terms of state variables. This can be done by augmenting

the plant model (A, B) with models of the constants r

1

and r

2

. Since these models

will not be controllable, they must be stable in order to solve the Riccati equation

(Maciejowski 1989). Therefore the model ˙ r

1

= ˙ r

2

= 0 cannot be used because the

poles are located on the imaginary axis. Instead the following are used

˙ r

1

= −αr

1

(5.13)

˙ r

2

= −αr

2

(5.14)

which with a low α indicates that r is a slow varying constant.

The augmented model is given by

A

r

=

_

_

_

_

_

_

0 0

A 0 0

0 0

−α 0

0 0 0 0 −α

_

_

_

_

_

_

, (5.15)

B

r

=

_

_

B

0

0

_

_

, x

r

= (∆x

T

r

1

r

2

)

T

(5.16)

By using these equations, the cost function (5.10) can be written in the form

lim

T→∞

_

T

0

x

T

r

Qx

r

+R∆u

2

+ 2x

T

r

N∆u (5.17)

with

Q = (M 1 0)

T

(M 1 0) +η(0 0 0 0 1)

T

(0 0 0 0 1)

N = η(0 0 0 0 1)

T

(5.18)

R = η

5.3 Speed Control with Active Damping and RQV Behavior 55

The cost function (5.10) is minimized by using

∆u = −K

c

∆x (5.19)

with

K

c

= Q

−1

(B

T

r

P

c

+N

T

) (5.20)

where P

c

is the solution to the Riccati equation

A

T

r

P

c

+P

c

A

r

+R −(P

c

B

r

+N)Q

−1

(P

c

B

r

+N)

T

= 0 (5.21)

The control law (5.19) becomes

∆u = −K

c

x

r

= −

_

K

c1

K

c2

K

c3

_

∆x −K

c4

r

1

−K

c5

r

2

(5.22)

By using (5.9) and (5.12) the control law is written

u = K

0

x

30

+K

l

l +K

r

r −

_

K

c1

K

c2

K

c3

_

x (5.23)

with

K

0

=

_

K

c1

K

c2

K

c3

_

δ

x

−K

c4

Mδ

x

+λ

x

−K

c5

λ

x

K

r

= K

c4

+K

c5

λ

x

(5.24)

K

l

=

_

K

c1

K

c2

K

c3

_

δ

l

−K

c4

Mδ

l

+λ

l

where δ

x

, δ

l

, λ

x

, and λ

l

are described in (5.3) and (5.4).

When this control law is applied to Example 5.1 the controller gains becomes

u = 0.230x

30

+ 4470r + 0.125l −

_

7620 0.0347 2.36

_

x (5.25)

where η = 5 · 10

−8

and α = 0.0001 are used. With this controller the phase margin

is guaranteed to be at least 60

◦

and the amplitude margin is inﬁnity (Maciejowski

1989). The result is seen in Figure 5.4.

The rise time of the LQG controller is shorter than for the RQV controller. Also

the overshoot is less when using LQG control. The driving torque is controlled such

that the oscillations in the wheel speed are actively damped. The controlled driving

torque makes the engine speed oscillate, as seen in Figure 5.4.

5.3.1 Extending with RQV Behavior

The RQV controller has no information about the load l, and therefore a stationary

error will be present when the load is diﬀerent from zero. The LQG feedback law

(5.25) is a function of the load, and the stationary error is zero if the load is known.

There is however a demand by the driver that the load should give a stationary

error, and only when using a cruise controller the stationary error should be zero.

The LQG controller can be changed such that a load diﬀerent from zero gives a

stationary error. This is done by using β

l

l instead of the complete load l in (5.23).

The constant β

l

range from β

l

= 0 which means no compensation for the load, to

56 Chapter 5 Speed Controller Design and Simulations

0 1 2 3 4 5 6 7 8

120

130

140

150

S

p

e

e

d

,

[

r

a

d

/

s

]

0 1 2 3 4 5 6 7 8

1.9

2

2.1

2.2

2.3

2.4

S

p

e

e

d

,

[

r

a

d

/

s

]

0 1 2 3 4 5 6 7 8

0

500

1000

1500

2000

Time, [s]

T

o

r

q

u

e

,

[

N

m

]

Engine speed

˙

θ

m

Wheel speed

˙

θ

w

(performance output)

Control signal u

Figure 5.4 Model 1 controlled with the LQG control law (5.25). RQV control

(5.1) with K

p

= 25 is seen in dashed lines. With active damping, the engine speed

oscillates, resulting in a well behaved wheel speed.

β

l

= 1 which means fully compensation of the load and no stationary error. The

compensated LQG control law becomes

u = K

0

x

30

+K

l

β

l

l +K

r

r −

_

K

c1

K

c2

K

c3

_

x (5.26)

In Figure 5.5, the RQV controller with its stationary error (remember the reference

value r = 2.3 rad/s) is compared to the compensated LQG controller (5.26) applied

to Example 5.1 for three values of β

l

. By adjusting β

l

, the speed controller with

active damping is extended with a stationary error comparable with that of the

RQV controller.

5.4 Inﬂuence from Sensor Location

The LQG controller investigated in the previous section uses feedback from all

states (x

1

= θ

m

/i

t

i

f

− θ

w

, x

2

=

˙

θ

m

, and x

3

=

˙

θ

w

). This is not possible if only

5.4 Inﬂuence from Sensor Location 57

0 1 2 3 4 5 6 7 8

1.95

2

2.05

2.1

2.15

2.2

2.25

2.3

2.35

Time, [s]

S

p

e

e

d

,

[

r

a

d

/

s

]

Figure 5.5 Example 5.1 controlled with the RQV controller (5.1) in dashed line,

and the LQG controller with stationary error (5.26) with β

l

= 0, 0.5, 1. The LQG

controller achieves the same stationary as the RQV controller by adjusting β

l

.

one sensor is used, which is the case considered in this work. The sensor either

measures the engine speed

˙

θ

m

or the wheel speed

˙

θ

w

. In this section an observer

is used to estimate the rest of the states. The observer gain is calculated using

LTR technique. Then two diﬀerent observer problems results depending on which

sensor location that is used.

The LQG feedback law (5.23) then becomes

u = K

0

x

30

+K

r

r +K

l

l −

_

K

c1

K

c2

K

c3

_

ˆ x (5.27)

with K

0

, K

r

, and K

l

given by (5.24). The estimated states ˆ x are given by the

Kalman ﬁlter

∆

˙

ˆ x = A∆ˆ x +B∆u +K

f

(∆y −C∆ˆ x) (5.28)

K

f

= P

f

C

T

V

−1

(5.29)

where P

f

is found by solving the Riccati equation

P

f

A

T

+AP

f

−P

f

C

T

V

−1

CP

f

+W = 0 (5.30)

The covariance matrices W and V corresponds to v and e respectively. The output

matrix C is either equal to C

m

(4.11) when measuring the engine speed, or C

w

(4.12) when measuring the wheel speed.

58 Chapter 5 Speed Controller Design and Simulations

Loop-Transfer Recovery (LTR) is used to recover the properties achieved in the

previous design step when all states are measured. This is done by selecting

V = 1

W = ρBB

T

(5.31)

C = C

m

or C

w

ρ = ρ

m

or ρ

w

and solving (5.29) and (5.30) for K

f

.

When using LQG with feedback from all states, the phase margin ϕ is at least

60

◦

and the amplitude margin a is inﬁnity as stated before. This is obtained also

when using the observer by increasing ρ towards inﬁnity. For Example 5.1 the

following values are used

ρ

m

= 5 · 10

5

⇒ ϕ

m

= 60.5

◦

, a

m

= ∞ (5.32)

ρ

w

= 10

14

⇒ ϕ

w

= 59.9

◦

, a

w

= 35.0 (5.33)

where the aim has been to have at least 60

◦

phase margin.

The observer dynamics is cancelled in the transfer function from reference value

to performance output and control signal. Hence, these transfer functions are not

aﬀected by the sensor location. However, the observer dynamics will be included

in the transfer functions from disturbances both to z and u.

5.4.1 Inﬂuence from Load Disturbances

Figure 5.6 shows how the performance output and the control signal are aﬀected

by the load disturbance v. There is a resonance peak in G

vz

when using feedback

from the engine speed sensor, which is not present when feedback from the wheel

speed sensor is used. The reason to this can be seen when studying the transfer

function G

vz

in (4.20). By using the matrix inversion lemma (4.20) is rewritten as

(G

vz

)

cl

=

G

vz

+F

y

(G

uy

G

vz

−G

uz

G

vy

)

1 +G

uy

F

y

(5.34)

where G

ab

means the transfer function from signal a to b, and cl stands for closed

loop. The signal y in (5.34) mean the output of the system, i.e. either

˙

θ

w

or

˙

θ

m

.

The controller F

y

is given by (4.15) as

F

y

(s) = K

c

(sI −A+BK

c

+K

f

C)

−1

K

f

(5.35)

with C either as C

m

for engine speed feedback, or C

w

for wheel speed feedback.

For the speed controller (z =

˙

θ

w

), Equation (5.34) becomes

(G

vz

)

cl

=

G

vw

1 +G

uw

F

y

(5.36)

5.4 Inﬂuence from Sensor Location 59

10

−2

10

−1

10

0

10

1

10

2

10

3

−140

−120

−100

−80

G

a

i

n

[

d

B

]

10

−2

10

−1

10

0

10

1

10

2

10

3

−150

−100

−50

0

Frequency [rad/s]

G

a

i

n

[

d

B

]

Closed-loop transfer function G

vz

Closed-loop transfer function G

vu

Figure 5.6 Closed-loop transfer functions from load disturbance v to performance

output z and control signal u. Feedback from

˙

θ

w

is shown in solid and feedback

from

˙

θ

m

is shown in dashed lines. With

˙

θ

m

feedback the transfer functions have a

resonance peak, resulting from the open-loop zeros.

when the sensor measures the wheel speed, and

(G

vz

)

cl

=

G

vw

+F

y

(G

um

G

vw

−G

uw

G

vm

)

1 +G

um

F

y

(5.37)

when the sensor measures the engine speed. Hence, when using the wheel speed

sensor, the controller is cancelled in the numerator, and when the engine speed

sensor is used, the controller is not cancelled.

The optimal return ratio in the LQG step is

K

c

(sI −A)

−1

B (5.38)

Hence the poles from A is kept, but there are new zeros that are placed such that

the relative degree of (5.38) is one, the phase margin is at least 60

◦

, and the gain

margin is inﬁnite. In the LTR step the return ratio is

F

y

G

uy

= K

c

(sI −A−BK

c

−K

f

C)

−1

K

f

C(sI −A)

−1

B (5.39)

When ρ in (5.31) is increased towards inﬁnity, (5.38) equals (5.39). This means that

the zeros in the open-loop system C(sI − A)

−1

B are cancelled by the controller.

Hence, the open-loop zeros will become poles in the controller F

y

. This means that

60 Chapter 5 Speed Controller Design and Simulations

10

−2

10

−1

10

0

10

1

10

2

10

3

−150

−100

−50

0

G

a

i

n

[

d

B

]

10

−2

10

−1

10

0

10

1

10

2

10

3

0

50

100

Frequency [rad/s]

G

a

i

n

[

d

B

]

Closed-loop transfer function G

ez

Closed-loop transfer function G

eu

Figure 5.7 Closed-loop transfer functions from measurement noise e to perfor-

mance output z and control signal u. Feedback from

˙

θ

w

is shown in solid and

feedback from

˙

θ

m

is shown in dashed. The diﬀerence between the two feedback

principles is described by the dynamic output ratio. The eﬀect increases with lower

gears.

the closed-loop system will have the open-loop zeros as poles when using the engine

speed sensor. This means that the G

vz

will have the poles −0.5187±3.0753i which

causes the resonance peak in Figure 5.6.

5.4.2 Inﬂuence from Measurement Disturbances

The inﬂuence from measurement disturbances e is seen in Figure 5.7. The transfer

functions from measurement noise (4.21) can be rewritten with the matrix inversion

lemma as

(G

ez

)

cl

= −

G

uz

F

y

1 +G

uy

F

y

(5.40)

The complementary sensitivity function is deﬁned for the two sensor alternatives

as

T

w

=

G

uw

F

y

1 +G

uw

F

y

, T

m

=

G

um

F

y

1 +G

um

F

y

(5.41)

Then

(G

ez

)

cl

= −T

w

with

˙

θ

w

feedback (5.42)

5.4 Inﬂuence from Sensor Location 61

10

−2

10

−1

10

0

10

1

10

2

10

3

−100

−80

−60

−40

−20

0

Frequency [rad/s]

G

a

i

n

[

d

B

]

Figure 5.8 The dynamic output ratio G

w/m

for gear 1 (solid), gear 7 (dashed),

and gear 14 (dash-dotted).

(G

ez

)

cl

= −T

m

G

uw

G

um

= T

m

G

w/m

with

˙

θ

m

feedback (5.43)

where the dynamic output ratio G

w/m

was deﬁned in Deﬁnition 4.1. For Model 1

the dynamic output ratio is

G

w/m

=

cs +k

i(J

2

s

2

+ (c +b

2

)s +k)

(5.44)

where the state-space description in Chapter 4 is used. Especially for low frequen-

cies, G

w/m

(0) = 1/i = 1/i

t

i

f

. The dynamic output ratio can be seen in Figure 5.8

for three gears.

When ρ in (5.31) is increased towards inﬁnity, (5.38) equals (5.39). Then (5.42)

and (5.43) gives

(G

ez

)

cl,m

= (G

ez

)

cl,w

G

w/m

(5.45)

where cl, m and cl, w means closed loop with feedback from

˙

θ

m

and

˙

θ

w

respectively.

The frequency range in which the T

m

= T

w

is valid depends on how large ρ in

(5.31) is made. Figure 5.9 shows the sensitivity functions

S

w

=

1

1 +G

uw

F

y

, S

m

=

1

1 +G

um

F

y

(5.46)

and the complementary sensitivity functions T

w

and T

m

(5.41) for the two cases of

feedback. It is seen that T

m

= T

w

is valid up to about 16 Hz. The roll-oﬀ rate at

62 Chapter 5 Speed Controller Design and Simulations

10

−2

10

−1

10

0

10

1

10

2

10

3

−70

−60

−50

−40

−30

−20

−10

0

10

Frequency [rad/s]

G

a

i

n

[

d

B

]

Figure 5.9 Sensitivity function S and complementary sensitivity function T.

The dash-dotted lines correspond to the case with all states known. When only

one velocity is measured, the solid lines correspond to

˙

θ

w

feedback, and the dashed

lines correspond to

˙

θ

m

feedback.

higher frequencies diﬀer between the two feedback principles. This is due to that

the open-loop transfer functions G

uw

and G

um

have a diﬀerent relative degree.

G

uw

has a relative degree of two, and G

um

has a relative degree of one. Therefore,

T

w

has a steeper roll-oﬀ rate than T

m

.

Hence, the diﬀerence in G

ez

depending on sensor location is described by the

dynamic output ratio G

w/m

. The diﬀerence in low frequency level is equal to the

conversion ratio of the driveline. Therefore, this eﬀect increases with lower gears.

5.4.3 Load Estimation

The feedback law with unknown load is

u = K

0

x

30

+K

r

r +K

l

ˆ

l −

_

K

c1

K

c2

K

c3

_

ˆ x (5.47)

where

ˆ

l is the estimated load. In order to estimate the load, the model used in the

Kalman ﬁlter is augmented with a model of the load. The load is hard to model

correctly since it is a function of road slope. However it can be treated as a slow

varying constant. The augmented model is

x

4

= l, with ˙ x

4

= 0 (5.48)

5.5 Simulations 63

Controller

Design based on Model 1

Control law (5.27)

Observer (5.28)

z (

˙

θ

w

)

y (

˙

θ

w

or

˙

θ

m

)

r

Vehicle

Model 3: (2.49) to (2.51)

Figure 5.10 Simulation conﬁguration. As a step for demonstrating feasibility for

real implementation, Model 3 is simulated with the controller based on Model 1.

This gives

˙

ˆ x = A

l

ˆ x

l

+B

l

u +K

f

(y −C

l

ˆ x

l

) (5.49)

with

ˆ x

l

=

_

ˆ x

ˆ

l

_

, (5.50)

A

l

=

_

_

_

_

0

A 0

−1/J

2

0 0 0 0

_

_

_

_

, (5.51)

B

l

=

_

B

0

_

, C

l

=

_

C 0

_

(5.52)

The feedback law is

u = K

0

x

30

+K

r

r −

_

K

c1

K

c2

K

c3

−K

l

_

ˆ x

l

(5.53)

5.5 Simulations

An important step in demonstrating feasibility for real implementation is that

a controller behaves well when simulated on a more complicated vehicle model

than it was designed for. Here, the control law based on the reduced driveline

model is simulated with a more complete nonlinear model, derived in Chapter 2.

The purpose is also to study eﬀects from diﬀerent sensor locations as discussed in

Section 5.4. The simulation situation is seen in Figure 5.10.

The nonlinear Model 3, given by (2.49) to (2.51), is used as vehicle model. The

steady-state level for Model 3 is calculated by solving the model equations for the

equilibrium point when the load and speed are known.

The controller used is based on Model 1, as seen in the previous sections. The

wheel speed or the engine speed is input to the observer (5.28), and the control law

(5.27) with β

l

= 0 generates the control signal.

64 Chapter 5 Speed Controller Design and Simulations

The simulation case presented here is the same as in Example 5.1, i.e. a velocity

step response. The stationary point is given by

˙

θ

w

= 2, l = 3000 ⇒ x

0

=

_

0.0482 119 2.00

_

, u

0

= 109 (5.54)

where (5.3) and (5.4) are used, and the desired new speed is

˙

θ

w

= 2.3 rad/s. At

steady state, the clutch transfers the torque u

0

= 109 Nm. This means that the

clutch angle is in the area with higher stiﬀness (θ

c1

< θ

c

≤ θ

c2

) in the clutch

nonlinearity, seen in Figure 2.7. This is a typical driving situation when speed

control is used. However, at low clutch torques (θ

c

< θ

c1

) the clutch nonlinearity

can produce limit cycle oscillations (Bj¨ ornberg, Pettersson, and Nielsen 1996). This

situation occurs when the truck is traveling downhill with a load of the same size

as the friction in the driveline, resulting in a low clutch torque. This is however not

treated here. At t = 6 s, a load impulse disturbance is simulated. The disturbance

is generated as a square pulse with 0.1 s width and 1200 Nm height.

In order to simulate the nonlinear model, the diﬀerential equations (2.49) to

(2.51) are scaled such that the ﬁve diﬀerential equations (one for each state) have

about the same magnitude. The model is simulated using Runge Kutta (45)

(Simulink 1993) with a low step size to catch the eﬀect of the nonlinearity.

Figures 5.11 to 5.13 show the result of the simulation. These should be com-

pared with the same control law applied to Model 1 in Figure 5.4. From these

plots it is demonstrated that the performance does not critically depend on the

0 1 2 3 4 5 6 7 8 9 10

1.95

2

2.05

2.1

2.15

2.2

2.25

2.3

2.35

Time, [s]

S

p

e

e

d

,

[

r

a

d

/

s

]

Figure 5.11 Wheel speed when controlling Model 3 with the LQG control law

(5.27) derived from Model 1. The solid line corresponds to

˙

θ

w

feedback and feed-

back from

˙

θ

m

is seen in dashed line. At t = 6 s, an impulse disturbance v acts on

the load. The design still works when simulated with extra clutch dynamics.

5.5 Simulations 65

0 1 2 3 4 5 6 7 8 9 10

−400

−200

0

200

400

600

800

1000

1200

1400

1600

Time, [s]

T

o

r

q

u

e

,

[

N

m

]

Figure 5.12 Control signal corresponding to Figure 5.11. There is no diﬀerence

between the two sensor alternatives in the step response at t = 1 s. However,

the load impulse (at t = 6 s) generates a control signal that damps the impulse

disturbance when feedback from the wheel speed sensor is used, but not with engine

speed feedback.

0 1 2 3 4 5 6 7 8 9 10

Time, [s]

A

n

g

l

e

d

i

f

f

e

r

e

n

c

e

,

[

r

a

d

]

θ

c1

Figure 5.13 Clutch angle diﬀerence corresponding to Figure 5.11. The inﬂuence

from the clutch nonlinearity can be neglected, because the area with low stiﬀness

(θ

c

< θ

c1

) is never entered.

66 Chapter 5 Speed Controller Design and Simulations

simpliﬁed model structure. The design still works if the extra dynamics are added.

Further evidence supporting this is seen in Figure 5.13. The area with low stiﬀness

in the clutch nonlinearity (θ

c

< θ

c1

) is never entered. The load impulse distur-

bance is better attenuated with feedback from the wheel speed sensor, which is a

veriﬁcation of the behavior that was discussed in Section 5.4.

5.6 Summary

Speed control with active damping and RQV behavior has been proposed in this

chapter. RQV control is the traditional way speed control is performed in diesel

engines. RQV control gives a certain driving character with a load dependent

stationary error when going uphill or downhill. With RQV, there is no active

damping of wheel speed oscillations, resulting in vehicle shuﬄe. An increased

controller gain results in more wheel speed oscillations while the engine speed

behaves well.

A major contribution in this chapter is a formulation of a criterion for speed

control with active damping of wheel speed oscillations and a stationary error

giving RQV behavior. To solve the criterion, a linear driveline model with drive

shaft ﬂexibility, and parameters estimated from experiments are used. Simulations

show that the performance of the design, based on the simpliﬁed model, works well

for a more complicated model, with a nonlinear clutch characteristics.

An investigation of the inﬂuence from diﬀerent sensor locations on the control

design shows that when using LQG/LTR the open-loop zeros are cancelled by

the controller. With engine speed feedback this is critical, because the open-loop

transfer function has a resonant zero couple. It is shown that this zero couple

becomes poles of the transfer functions from load disturbances to wheel speed.

This results in undamped load disturbances when engine speed feedback is used.

When feedback from the wheel speed sensor is used, no resonant open-loop poles

are cancelled. Load disturbances are thus better attenuated with this feedback

conﬁguration.

Measurement disturbances are better attenuated when the engine speed sensor

is used, than when using the wheel speed sensor. This eﬀect increases with lower

gears. Two diﬀerent closed-loop transfer functions result, depending on feedback

conﬁguration. The diﬀerence between these two is described by the dynamic output

ratio.

In conclusion, even though there are sensor choices, the use of active damping

signiﬁcantly improves the behavior for both sensor cases. Further, the formulation

is natural, it allows eﬃcient solution, and there is a simple tuning of the amount

of RQV feeling.

6

Gear-Shift Controller Design and

Simulations

Traditionally a gear shift is performed by disengaging the clutch, engaging neutral

gear, shifting to a new gear, and engaging the clutch again. In todays traﬃc it is

desired to have an automatic gear shifting system, where the complete shift action

is controlled by a microprocessor. If the automatic gear shifting system is to work

with a clutch and a manually shifted transmission, one of the following strategies

can be taken.

• The gear shift is performed with a microprocessor controlling the clutch and

the shift event.

• The gear shift is performed without using the clutch (Orehall 1995). In this

case the engine is controlled such that the torque in the transmission is zero,

whereafter neutral gear is engaged. The engine speed is then controlled to the

propeller shaft speed (scaled with the new conversion ratio). Following that,

the new gear is engaged, and then the speed controller controls the driveline

to the speed demanded by the driver.

When using the second approach, neutral gear is engaged when the transmission

transfers zero torque. It is clear that driveline oscillations is an important perfor-

mance limiting factor if they are not damped out. This is because the system has

to wait until satisfactory gear shift conditions are reached, and thus seriously in-

creasing the total time needed for a gear shift. One reason this is not acceptable

is that, since there is no torque, the vehicle is free rolling which may be serious

with heavy loads and large road slopes. The observation that the vehicle is free

rolling, i.e. changing its velocity, when obtaining gear shift conditions shows that

the desired control goal is not a stationary point, which has to be handled.

67

68 Chapter 6 Gear-Shift Controller Design and Simulations

This chapter is devoted to study a new idea that the transmission torque can

be estimated and controlled to zero while having active damping. The problem

formulation is further discussed in Section 6.1. A model of the transmission is

developed and the torque transmitted in the transmission is modeled as a function

of the states and the control signal in Section 6.2. Some ﬁrst primitive attempts

are then discussed in Section 6.3.

A key result in this chapter is, in light of the simplistic attempts in Section 6.3,

the formulation of the gear-shift control criterion in Section 6.4, and its treatment

in Section 6.5. Inﬂuence from sensor location and simulations are presented in the

sections following.

6.1 Problem Formulation

The gear-shift controller is the controller that drives the transmission torque to

zero, while damping oscillations. If a gear shift is commanded when the driveline

is oscillating, the gear-shift controller should still drive the transmission torque to

zero.

The control signal is restricted to be in the interval between u

min

= −300

Nm and u

max

= 2300 Nm. The time it takes for a gear shift should be possible

to optimize. The inﬂuence from load and measurement disturbances should be

minimized.

6.2 Transmission Torque

The performance output z for the gear-shift controller is the torque transmitted

between the cog wheels in the transmission. A more detailed study of the trans-

mission is depicted in Figure 6.1. Here, the input shaft is connected to bearings

with a viscous friction component b

t1

. A cog wheel is mounted at the end of the

input shaft which is connected to a cog wheel mounted on the output shaft. The

conversion ratio between these are i

t

, as mentioned in Chapter 2. The output shaft

is also connected to bearings with the viscous friction component b

t2

.

Two equations describe the inputs and outputs of the transmission

J

t1

¨

θ

c

= M

t

−b

t1

˙

θ

c

−z (6.1)

J

t2

¨

θ

t

= i

t

z −b

t2

˙

θ

t

−M

p

(6.2)

6.2.1 Transmission Torque for Model 1

By using (2.1)

J

m

¨

θ

m

= M

m

−M

fr:m

−M

c

(6.3)

together with (2.12)

M

c

= M

t

, θ

m

= θ

c

(6.4)

6.2 Transmission Torque 69

M

t

Transmission

θ

c

θ

t

b

t2

b

t1

J

t1

J

t2

Input shaft Output shaft

M

p

Figure 6.1 Transmission with two cogwheels with conversion ratio i

t

. The cog-

wheels are connected to the input and output shaft respectively.

equation (6.1) is expressed in terms of engine speed

(J

m

+J

t1

)

¨

θ

m

= M

m

−M

fr:m

−b

t1

˙

θ

m

−z (6.5)

To describe the performance output in terms of state variables,

¨

θ

m

(which is not a

state variable) is replaced with (2.25)

(J

m

+J

t

/i

2

t

+J

f

/i

2

t

i

2

f

)

¨

θ

m

= M

m

−M

fr:m

−(b

t

/i

2

t

+b

f

/i

2

t

i

2

f

)

˙

θ

m

(6.6)

−k(θ

m

/i

t

i

f

−θ

w

)/i

t

i

f

−c(

˙

θ

m

/i

t

i

f

−

˙

θ

w

)/i

t

i

f

which together with u = M

m

−M

fr:m

gives

u −b

t1

˙

θ

m

−z =

J

m

+J

t1

J

m

+J

t

/i

2

t

+J

f

/i

2

t

i

2

f

_

M

m

−M

fr:m

−(b

t

/i

2

t

+b

f

/i

2

t

i

2

f

)

˙

θ

m

−k(θ

m

/i

t

i

f

−θ

w

)/i

t

i

f

−c(

˙

θ

m

/i

t

i

f

−

˙

θ

w

)/i

t

i

f

_

(6.7)

From this it is possible to express the performance output as a function of the

control signal u and the state variables x, according to the state-space description

(4.2) to (4.5).

Deﬁnition 6.1 The performance output for Model 1 is

z = Mx +Du with

70 Chapter 6 Gear-Shift Controller Design and Simulations

M

T

=

_

_

_

(J

m

+J

t1

)k

J

1

i

J

m

+J

t1

J

1

(b

1

+c/i

2

) −b

t1

−

(J

m

+J

t1

)c

J

1

i

_

_

_ (6.8)

D = 1 −

J

m

+J

t1

J

1

where the labels from (4.5) are used.

The unknown parameters in (6.8) are J

m

+J

t1

and b

t1

. The other parameters

are estimated in Chapter 3.

One way of estimating these unknowns would be to decouple Model 1 into two

models, corresponding to neutral gear. Then a model including the engine, clutch,

and the input shaft of the transmission results, in which the performance output z

is equal to zero. Trials with neutral gear would then give a possibility to estimate

the unknowns.

In the derivation of Model 1 in Chapter 2 the performance output z is elimi-

nated. If z is eliminated in (6.1) and (6.2) and (6.4) is used, the equation for the

transmission is

(J

t1

i

2

t

+J

t2

)

¨

θ

m

= i

2

t

M

c

−i

t

M

p

−(b

t1

i

2

t

+b

t2

)

˙

θ

m

(6.9)

By comparing this with the equation describing the transmission in Chapter 2,

(2.15)

J

t

¨

θ

m

= i

2

t

M

c

−b

t

˙

θ

m

−i

t

M

p

(6.10)

the following equations relating the parameters are obtained

J

t

= i

2

t

J

t1

+J

t2

(6.11)

b

t

= i

2

t

b

t1

+b

t2

(6.12)

For the rest of this chapter the following assumption about the parameters in

the transmission is used.

Assumption 6.1 J

t1

= J

t2

and b

t1

= b

t2

.

Then (6.11) and (6.12) gives

J

t1

=

J

t

1 +i

2

t

(6.13)

b

t1

=

b

t

1 +i

2

t

(6.14)

In Chapter 3, the estimated combinations of parameters from Model 1 are

J

1

= J

m

+J

t

/i

2

t

+J

f

/i

2

t

i

2

f

(6.15)

b

1

= b

t

/i

2

t

+b

f

/i

2

t

i

2

f

(6.16)

6.2 Transmission Torque 71

From (6.13) and (6.15) J

m

+J

t1

can be derived

J

m

+J

t1

= J

m

+

J

t

1 +i

2

t

= J

m

+

i

2

t

1 +i

2

t

(J

1

−J

m

−J

f

/i

2

t

i

2

f

)

= J

m

1

1 +i

2

t

+J

1

i

2

t

1 +i

2

t

−J

f

1

i

2

f

(1 +i

2

t

)

(6.17)

A combination of (6.14) and (6.16) gives b

t1

b

t1

=

b

t

1 +i

2

t

=

i

2

t

1 +i

2

t

(b

1

−b

f

/i

2

t

i

2

f

) (6.18)

For low gears (i

t

large), and since J

f

and b

f

are considerably less than J

1

and b

1

,

the following assumptions are used

J

m

+J

t1

≈ J

1

i

2

t

1 +i

2

t

(6.19)

b

t1

≈ b

1

i

2

t

1 +i

2

t

(6.20)

6.2.2 Transmission Torque for Model 2

The performance output expressed for Model 2 is given by replacing M

t

in (6.1)

by equation (2.39)

M

c

= M

t

= k

c

(θ

m

−θ

t

i

t

) +c

c

(

˙

θ

m

−

˙

θ

t

i

t

) (6.21)

Then the performance output is

z = k

c

(θ

m

−θ

t

i

t

) +c

c

(

˙

θ

m

−

˙

θ

t

i

t

) −b

t1

i

t

˙

θ

t

−J

t1

i

t

¨

θ

t

(6.22)

This is expressed in terms of state variables by using (2.45)

(J

t

+J

f

/i

2

f

)

¨

θ

t

= i

t

_

k

c

(θ

m

−θ

t

i

t

) +c

c

(

˙

θ

m

−

˙

θ

t

i

t

)

_

(6.23)

−(b

t

+b

f

/i

2

f

)

˙

θ

t

−

1

i

f

_

k

d

(θ

t

/i

f

−θ

w

) +c

d

(

˙

θ

t

/i

f

−

˙

θ

w

)

_

(6.24)

leading to

Deﬁnition 6.2 The performance output for Model 2 is

z = Mx with (6.25)

M

T

=

_

_

_

_

_

_

_

_

k

c

(1 −

J

t1

i

2

t

J

2

)

J

t1

i

t

k

d

J

2

i

f

c

c

(1 −

J

t1

i

2

t

J

2

)

J

t1

i

2

t

J

2

(i

2

t

c

c

+b

2

+c

d

/i

2

f

) −c

c

i

t

−b

t1

i

t

−

J

t1

i

t

c

d

J

2

i

f

_

_

_

_

_

_

_

_

72 Chapter 6 Gear-Shift Controller Design and Simulations

with states and labels according to to the state-space description (4.6) to (4.8).

In Chapter 3, the following combinations of parameters from Model 2 are esti-

mated.

J

2

= J

t

+J

f

/i

2

f

(6.26)

b

2

= b

t

+b

f

/i

2

f

(6.27)

From (6.13) , (6.14) , (6.26), and (6.27), J

t1

and b

t1

can be derived as

J

t1

=

i

2

t

1 +i

2

t

(J

2

−J

f

/i

2

f

) (6.28)

b

t1

=

i

2

t

1 +i

2

t

(b

2

−b

f

/i

2

f

) (6.29)

which are approximated to

J

t1

≈

i

2

t

1 +i

2

t

J

2

(6.30)

b

t1

≈

i

2

t

1 +i

2

t

b

2

(6.31)

since J

f

and b

f

are considerably less than J

1

and b

1

.

6.2.3 Transmission Torque for Model 3

The performance output for Model 3 is derived in the same way as for Model 2,

with the diﬀerence that (6.21) is replaced by

M

c

= M

t

= M

kc

(θ

m

−θ

t

i

t

) +c

c

(

˙

θ

m

−

˙

θ

t

i

t

) (6.32)

where M

kc

is the torque transmitted by the clutch nonlinearity, given by (2.48).

Then the performance output is deﬁned as

Deﬁnition 6.3 The performance output for Model 3 is

z = (M

kc

,

˙

θ

t

/i

f

−

˙

θ

w

,

˙

θ

m

,

˙

θ

t

,

˙

θ

w

)

_

_

_

_

_

_

_

_

1 −

J

t1

i

2

t

J

2

J

t1

i

t

k

d

J

2

i

f

c

c

(1 −

J

t1

i

2

t

J

2

)

J

t1

i

2

t

J

2

(i

2

t

c

c

+b

2

+c

d

/i

2

f

) −c

c

i

t

−b

t1

i

t

−

J

t1

i

t

c

d

J

2

i

f

_

_

_

_

_

_

_

_

(6.33)

The parameters not estimated in the deﬁnition above are approximated in the

same way as for the performance output for Model 2.

6.3 Preliminary Trials 73

0 5 10 15

−400

−300

−200

−100

0

100

200

300

400

500

Time, [s]

T

o

r

q

u

e

,

[

N

m

]

Figure 6.2 Transmission torque z from parameter estimation of Model 1 and

Model 2 on data from Trial 1. The solid line corresponds to Model 1 and the

dashed line corresponds to Model 2.

Comparison

In Figure 6.2 the performance output (6.8) and (6.25) during Trial 1 are shown

from the parameter estimation of the linear Models 1 and 2. Figure 6.3 shows

the performance output in the frequency domain. The low frequency level diﬀers

between the two models, and the main reason to this is the diﬃculties to estimate

the viscous damping coeﬃcients described in Chapter 3. The diﬀerence at higher

frequencies is a result from the clutch which gives a second resonance peak for

Model 2. Furthermore, the roll-oﬀ rate of Model 2 is steeper than for Model 1.

6.3 Preliminary Trials

Two preliminary trials will be performed in this section, to study gear-shift control.

6.3.1 Unconstrained Active Damping

A ﬁrst attempt is to study the performance output, z = Mx + Du, with M and

D given by (6.8). A control law can be derived since z includes the control signal

and D is scalar. If u is chosen as

u = −D

−1

Mx (6.34)

z = 0 is guaranteed.

74 Chapter 6 Gear-Shift Controller Design and Simulations

−100

−50

0

G

a

i

n

[

d

B

]

10

−2

10

−1

10

0

10

1

10

2

10

3

−100

−80

−60

−40

−20

Frequency [rad/s]

G

a

i

n

[

d

B

]

G

uz

for Model 1 and 2

G

lz

for Model 1 and 2

Figure 6.3 Transfer functions from control signal u and load l to transmission

torque z. Model 1 is shown in solid and Model 2 is shown in dashed. The modeled

clutch adds a second resonance peak and a steeper roll-oﬀ rate.

Example 6.1 Consider the truck modeled in Chapters 2 and 3 traveling at a speed

of 3 rad/s (5.4 km/h) with gear 1 and a total load of 3000 Nm (≈ 2 % road slope).

The stationary point is obtained by using (5.3) and (5.4).

x

30

= 3, l = 3000 ⇒ x

0

=

_

0.0511 178 3.00

_

, u

0

= 138 (6.35)

In Figure 6.4 the resulting transmission torque z, control signal u, engine, and

wheel speed is seen when the control signal is chosen as in (6.34). Unconstrained

active damping is achieved which obtains z = 0 instantaneously. The wheel speed

decreases linearly, while the engine speed is oscillating.

Unconstrained active damping (6.34) generates a control signal that is impossi-

ble for the engine to generate. To deal with this situation, it would be desirable to

use an control law which also considers that the control signal must be in a certain

interval.

It can be noted that despite z = 0 is achieved this is not an stationary point,

since the speed is decreasing. This means that the vehicle is free rolling which can

be critical if lasting too long.

6.3.2 Undamped Gear-Shift Condition

The previous approach is not realizable because of the unrealistic control signal. A

second attempt is to explicitly handle the expected vehicle behavior (free rolling)

6.3 Preliminary Trials 75

0 0.5 1 1.5 2

−10

0

10

20

30

40

50

60

T

o

r

q

u

e

,

[

N

m

]

0 0.5 1 1.5 2

−15000

−10000

−5000

0

T

o

r

q

u

e

,

[

N

m

]

0 0.5 1 1.5 2

145

150

155

160

165

170

175

180

[

r

a

d

/

s

]

Time, [s]

0 0.5 1 1.5 2

2.5

2.6

2.7

2.8

2.9

3

3.1

Time, [s]

[

r

a

d

/

s

]

Transmission torque z Engine torque u

Engine speed

˙

θ

m

Wheel speed

˙

θ

w

Figure 6.4 Unconstrained active damping of Model 1. At t = 1 s, the control law

(6.34) drives the transmission torque to zero. The oscillations in the transmission

torque are damped with an unrealizable control signal. The wheel speed decreases

linearly.

when the transmission torque, z, is zero, but without using active damping. This

control law is thus derived by considering a stiﬀ driveline, and solving for z = 0.

By using the labels according to Chapter 4, the diﬀerential equation, describing

the stiﬀ driveline is

(J

1

i +J

2

/i)

¨

θ

w

= u −(b

1

i +b

2

/i)

˙

θ

w

−l/i (6.36)

This equation is developed by using Model 1 in (2.25) and (2.26), and eliminating

the torque transmitted by the drive shaft, k(θ

m

/i −θ

w

) + c(

˙

θ

m

/i −

˙

θ

w

). Then by

using

˙

θ

m

=

˙

θ

w

i, (6.36) results.

76 Chapter 6 Gear-Shift Controller Design and Simulations

Equation (6.5) expressed in terms of wheel speed is

z = u −b

t1

i

˙

θ

w

−(J

m

+J

t1

)i

¨

θ

w

(6.37)

Combining (6.36) and (6.37) gives the performance output for the stiﬀ driveline.

z = (1 −

(J

m

+J

t1

)i

2

J

1

i

2

+J

2

)u−(b

t1

i −

(J

m

+J

t1

)i

J

1

i

2

+J

2

(b

1

i

2

+b

2

))

˙

θ

w

+

(J

m

+J

t1

)i

J

1

i

2

+J

2

l (6.38)

The control signal to force z = 0 is given by solving (6.38) for u while z = 0.

u

shift

(

˙

θ

w

, l) = µ

x

˙

θ

w

+µ

l

l with

µ

x

= (b

t1

i −

(J

m

+J

t1

)i

J

1

i

2

+J

2

(b

1

i

2

+b

2

))(1 −

(J

m

+J

t1

)i

2

J

1

i

2

+J

2

)

−1

(6.39)

µ

l

= −

(J

m

+J

t1

)i

J

1

i

2

+J

2

(1 −

(J

m

+J

t1

)i

2

J

1

i

2

+J

2

)

−1

Figure 6.5 shows Example 6.1 applied to Model 1 controlled with the undamped

gear-shift condition (6.39). This control law achieves z = 0 with a realizable control

signal, but the oscillations introduced are not damped. Therefore, the time needed

to fulﬁll the gear-shift condition is not optimized. The performance of this approach

is worse if the driveline is oscillating at the time for the gear shift.

6.4 Gear-Shift Control Criterion

Neither of the two approaches in the previous section solve the problem satisfactory.

In this section a new idea for gear-shift control is formulated. The transmission

torque is estimated and controlled to zero with active damping. The idea is formu-

lated as a cost criterion which uses a combination of the two previous approaches.

The criterion is formulated such that active damping is obtained with a control law

whose deviation from the undamped gear-shift condition (6.39) adds to the cost

function. Let the cost function be

lim

T→∞

_

T

0

z

2

+η(u −u

shift

(

˙

θ

w

, l))

2

(6.40)

= lim

T→∞

_

T

0

(Mx +Du)

2

+η(u −µ

x

˙

θ

w

−µ

l

l)

2

The controller that minimizes this cost function damps oscillation (since the ﬁrst

parenthesis is minimized), and at the same time, prevents the control signal from

having large deviations from the undamped gear-shift condition u

shift

. The trade-

oﬀ is controlled by η.

If the driveline is stiﬀ, there is no diﬀerence between the two parenthesis in

(6.40). Furthermore, the point at which the cost function is zero is no stationary

point, since the speed of the vehicle will decrease despite z = 0 and u = u

shift

.

6.5 Gear-Shift Control Design 77

0 2 4 6 8

−40

−20

0

20

40

60

T

o

r

q

u

e

,

[

N

m

]

0 2 4 6 8

−150

−100

−50

0

50

100

150

200

T

o

r

q

u

e

,

[

N

m

]

0 2 4 6 8

0

50

100

150

200

Time, [s]

[

r

a

d

/

s

]

0 2 4 6 8

0

0.5

1

1.5

2

2.5

3

3.5

Time, [s]

[

r

a

d

/

s

]

Transmission torque z Engine torque u

Engine speed

˙

θ

m

Wheel speed

˙

θ

w

Figure 6.5 Model 1 controlled with the undamped gear-shift condition (6.39). At

t = 1 s, a gear shift is commanded. The speed dependent realizable control signal

drives the transmission torque to zero. Undamped oscillations in the transmission

torque increase the time needed to fulﬁll the gear-shift condition.

6.5 Gear-Shift Control Design

The new idea for gear-shift control is in this section given eﬃcient treatment by

solving (6.40) for a control law by using LQG technique, using available software.

This is done by linearizing the driveline model and rewriting (6.40) in terms of the

linearized variables. A state-feedback matrix is derived that minimizes (6.40), by

solving a Riccati equation. The derived feedback law is a function of η which is

chosen such that a feasible control signal is used.

The linearized driveline model is given by (5.8) and (5.9) in Section 5.3. The

78 Chapter 6 Gear-Shift Controller Design and Simulations

cost function is expressed in terms of ∆x and ∆u by using (5.9)

lim

T→∞

_

T

0

(M∆x +D∆u +Mx

0

+Du

0

)

2

+ η(∆u −µ

x

∆x

3

+u

0

−µ

x

x

30

−µ

l

l)

2

= lim

T→∞

_

T

0

(M∆x +D∆u +r

1

)

2

+η(∆u −µ

x

∆x

3

+r

2

) (6.41)

with

r

1

= Mx

0

+Du

0

(6.42)

r

2

= u

0

−µ

x

x

30

−µ

l

l

The constants r

1

and r

2

are expressed as state variables, by augmenting the plant

model (A, B) with models of the constants r

1

and r

2

. This was done in (5.13) to

(5.16).

By using these equations, the cost function (6.41) can be written in the form

lim

T→∞

_

T

0

x

T

r

Qx

r

+R∆u

2

+ 2x

T

r

N∆u (6.43)

with

Q = (M 1 0)

T

(M 1 0) +η(0 0 −µ

x

0 1)

T

(0 0 −µ

x

0 1)

N = (M 1 0)

T

D +η(0 0 −µ

x

0 1)

T

(6.44)

R = D

2

+η

The cost function (6.43) is minimized by the state-feedback gain

K

c

= Q

−1

(B

T

r

P

c

+N

T

) (6.45)

where P

c

is the solution to the Riccati equation (5.21). The resulting control law

is

∆u = −K

c

x

r

= −

_

K

c1

K

c2

K

c3

_

∆x −K

c4

r

1

−K

c5

r

2

(6.46)

using (6.42) gives

u = K

0

x

30

+K

l

l −

_

K

c1

K

c2

K

c3

_

x (6.47)

with

K

0

=

_

λ

x

δ

x

µ

x

_

Γ (6.48)

K

l

=

_

λ

l

δ

l

µ

l

_

Γ

where Γ is

Γ =

_

_

1 −K

c4

D −K

c5

_

K

c1

K

c2

K

c3

_

−K

c4

M

K

c5

_

_

(6.49)

6.5 Gear-Shift Control Design 79

0 1 2 3 4

−10

0

10

20

30

40

50

60

T

o

r

q

u

e

,

[

N

m

]

0 1 2 3 4

−300

−200

−100

0

100

200

T

o

r

q

u

e

,

[

N

m

]

0 1 2 3 4

100

120

140

160

180

200

Time, [s]

[

r

a

d

/

s

]

0 1 2 3 4

1.5

2

2.5

3

3.5

Time, [s]

[

r

a

d

/

s

]

Transmission torque z Engine torque u

Engine speed

˙

θ

m

Wheel speed

˙

θ

w

Figure 6.6 Model 1 controlled with the LQG-control law (6.50), solving the gear-

shift criterion (6.40). At t = 1 s, a gear shift is commanded. A realizable control

signal is used such that the transmission torque is driven to zero, while oscillations

are actively damped.

with λ, δ, and µ given by (5.3), (5.4), and (6.39).

When this control law is applied to Example 6.1 the controller gains becomes

u = 2.37 · 10

−4

x

30

−0.0327l −

_

4.2123 0.0207 −1.2521

_

x (6.50)

where η = 0.03 and α = 0.0001 are used. With this controller the phase margin

is guaranteed to be at least 60

◦

and the amplitude margin is inﬁnity (Maciejowski

1989). The result is seen in Figure 6.6. By solving the gear-shift criterion (6.40),

active damping is obtained with a realizable control signal. The control law is a

function of η which is chosen such that the control signal is feasible.

80 Chapter 6 Gear-Shift Controller Design and Simulations

6.6 Inﬂuence from Sensor Location

The LQG controller investigated in the previous section uses feedback from all

states (x

1

= θ

m

/i

t

i

f

− θ

w

, x

2

=

˙

θ

m

, and x

3

=

˙

θ

w

). This is not possible if only

one sensor is used, which is the case considered in this work. The sensor either

measures the engine speed

˙

θ

m

or the wheel speed

˙

θ

w

. In this section an observer is

used to estimate the rest of the states. The observer gain is calculated using LTR

technique. Two diﬀerent observer problems results depending on which sensor

location that is used. The unknown load can be estimated as in Section 5.4.3.

The LQG feedback law (6.47) becomes

u = K

0

x

30

+K

l

l −

_

K

c1

K

c2

K

c3

_

ˆ x (6.51)

with K

0

and K

l

given by (6.48). The estimated state ˆ x is given by the Kalman

ﬁlter

∆

˙

ˆ x = A∆ˆ x +B∆u +K

f

(∆y −C∆ˆ x) (6.52)

K

f

= P

f

C

T

V

−1

(6.53)

where P

f

is found by solving the Riccati equation (5.30).

When using the LQG with feedback from all states, the phase margin ϕ is at

least 60

◦

, and the amplitude margin a is inﬁnity, as stated before. This is obtained

also when using the observer by increasing ρ towards inﬁnity. For Example 6.1 the

following values are used

ρ

m

= 10

4

⇒ ϕ

m

= 77.3

◦

, a

m

= 2.82 (6.54)

ρ

w

= 10

11

⇒ ϕ

w

= 74.3

◦

, a

w

= 2.84 (6.55)

where the aim has been to have at least 60

◦

phase margin.

The observer dynamics is cancelled in the transfer function from reference value

to performance output and control signal. Hence, these transfer functions are not

aﬀected by the sensor location. However, the dynamics will be included in the

transfer functions from disturbances to both z and u.

6.6.1 Inﬂuence from Load Disturbances

Figure 6.7 shows how the performance output and the control signal are aﬀected by

the load disturbance v. In Section 5.4 it was shown that for the speed controller,

the resonant open-loop zeros become poles of the closed-loop system when feedback

from the engine speed sensor is used. The same equations are valid for the gear-

shift controller except the diﬀerence that the D matrix in (6.8) is not equal to

zero, as for the speed controller. Hence, also the transfer function DG

vu

should be

added to (5.34). The closed-loop transfer function G

vu

is given by

(G

vu

)

cl

= −

F

y

G

vy

1 +F

y

G

uy

(6.56)

6.6 Inﬂuence from Sensor Location 81

10

−2

10

−1

10

0

10

1

10

2

10

3

−100

−80

−60

−40

−20

G

a

i

n

[

d

B

]

10

−2

10

−1

10

0

10

1

10

2

10

3

−200

−150

−100

−50

0

Frequency [rad/s]

G

a

i

n

[

d

B

]

Closed-loop transfer function G

vz

Closed-loop transfer function G

vu

Figure 6.7 Closed-loop transfer functions from load disturbance v to performance

output z and control signal u. Feedback from

˙

θ

w

is shown in solid and feedback

from

˙

θ

m

is shown in dashed lines. With

˙

θ

m

feedback the transfer functions have a

resonance peak, resulting from the open-loop zeros.

Thus, the closed-loop transfer function from v to u also has the controller F

y

in the

numerator. Hence, the closed-loop transfer function from v to z has the open-loop

zeros as poles. For

˙

θ

m

feedback, this means that a resonance peak is present in the

transfer functions from v to performance output z and control signal u.

6.6.2 Inﬂuence from Measurement Disturbances

The inﬂuence from measurement disturbances e are seen in Figure 6.8. According

to (5.40) the closed-loop transfer function from e to z is

(G

ez

)

cl

= −

G

uz

F

y

1 +G

uy

F

y

(6.57)

Then

(G

ez

)

cl

= −T

w

G

uz

G

uw

with

˙

θ

w

feedback (6.58)

(G

ez

)

cl

= −T

m

G

uz

G

um

with

˙

θ

m

feedback (6.59)

with T

w

and T

m

from (5.41).

82 Chapter 6 Gear-Shift Controller Design and Simulations

10

−2

10

−1

10

0

10

1

10

2

10

3

−100

−50

0

50

100

G

a

i

n

[

d

B

]

10

−2

10

−1

10

0

10

1

10

2

10

3

−50

0

50

100

Frequency [rad/s]

G

a

i

n

[

d

B

]

Closed-loop transfer function G

ez

Closed-loop transfer function G

eu

Figure 6.8 Closed-loop transfer functions from measurement noise e to perfor-

mance output z and control signal u. Feedback from

˙

θ

w

is shown in solid and

feedback from

˙

θ

m

is shown in dashed. The diﬀerence between the two feedback

principles are described by the dynamic output ratio. The eﬀect increases with

lower gears.

When ρ in (5.31) is increased towards inﬁnity, T

m

= T

w

as was discussed in

Section 5.4. Then (6.58) and (6.59) gives

(G

ez

)

cl,m

= (G

ez

)

cl,w

G

w/m

(6.60)

where cl, m and cl, w means closed loop with feedback from

˙

θ

m

and

˙

θ

w

respectively.

The dynamic output ratio G

w/m

was deﬁned in Deﬁnition 4.1, and is given by

(5.44).

The frequency range in which the T

m

= T

w

is valid depends on how large ρ

in (5.31) is made, as discussed in Section 5.4. Figure 6.9 shows the sensitivity

functions (5.46) and the complementary sensitivity functions T

w

and T

m

(5.41) for

the two cases of feedback. It is seen that T

m

= T

w

is valid up to about 2 Hz.

The roll-oﬀ rate at higher frequencies diﬀer between the two feedback principles.

This is due to that the open-loop transfer functions G

uw

and G

um

have a diﬀerent

relative degree, as discussed in Section 5.4. T

w

has a steeper roll-oﬀ rate than T

m

,

because that G

uw

has a relative degree of two, and G

um

has a relative degree of

one.

Hence, the diﬀerence in G

ez

depending on sensor location is described by the

dynamic output ratio G

w/m

. The diﬀerence in low frequency level is equal to the

conversion ratio of the driveline. Therefore, this eﬀect increases with lower gears.

6.7 Simulations 83

10

−2

10

−1

10

0

10

1

10

2

10

3

−120

−100

−80

−60

−40

−20

0

20

Frequency [rad/s]

G

a

i

n

[

d

B

]

Figure 6.9 Sensitivity function S and complementary sensitivity function T.

The dash-dotted lines correspond to the case with all states known. When only

one velocity is measured, the solid lines correspond to

˙

θ

w

feedback, and the dashed

lines correspond to

˙

θ

m

feedback.

6.7 Simulations

As in the case of the speed controller, in Section 5.5, the feasibility of the gear-shift

controller is studied by simulation on a more complicated vehicle model than it was

designed for. The control design is simulated with the nonlinear Model 3, according

to Figure 6.10. The eﬀects from sensor placement are also studied in accordance

with the discussion made in Section 6.6.

Model 3 is given by Equations (2.49) to (2.51). The steady-state level for

Model 3 is calculated by solving the model equations for the equilibrium point

when the load and speed are known. In Assumption 6.1, the relationship between

the model parameters in the transmission is given. By these, the equation for the

transmission torque is calculated using (6.33).

The controller used is based on Model 1, as seen in the previous sections. The

wheel speed or the engine speed is input to the observer (6.52), and the control law

(6.51) generates the control signal.

Three simulations are performed with the same parameters, given by Exam-

ple 6.1, (i.e. wheel speed

˙

θ

w

= 3 rad/s, and load l = 3000 Nm). In the simulations,

a gear shift is commanded at t = 2 s. The ﬁrst simulation is without disturbances.

In the second simulation, the driveline is oscillating prior to the gear shift. The

oscillations are a result of a sinusoid disturbance acting on the control signal. The

84 Chapter 6 Gear-Shift Controller Design and Simulations

Controller

Design based on Model 1

Control law (6.51)

Observer (6.52)

z (6.33)

y (

˙

θ

w

or

˙

θ

m

)

r

Vehicle

Model 3: (2.49) to (2.51)

Assumption 6.1

Figure 6.10 Simulation conﬁguration. As a step for demonstrating feasibility for

real implementation, Model 3 is simulated with the controller based on Model 1.

third gear shift is simulated with a load impulse at t = 3 s. The disturbance is

generated as a square pulse with 0.1 s width and 1200 Nm height.

In order to simulate the nonlinear model, the diﬀerential equations (2.49) to

(2.51) are scaled such that the ﬁve diﬀerential equations (one for each state) have

about the same magnitude. The model is simulated using Runge Kutta (45)

(Simulink 1993) with a low step size to catch the eﬀect of the nonlinearity.

In Figure 6.11 the simulation without disturbances is shown. This plot should

be compared with Figure 6.6, where the design is tested on Model 1. The result is

that the performance does not critically depend on the simpliﬁed model structure.

The design still works if the extra nonlinear clutch dynamics is added. In the

simulation, there are diﬀerent results depending on which sensor that is used. The

model errors between Model 1 and Model 3 are handled better when using the wheel

speed sensor. However, neither of the sensor alternatives reaches z = 0. This is

due to the low frequency model errors discussed in Section 6.2. In Figure 6.12

the simulation with driveline oscillations prior to the gear shift is shown. The

result is that the performance of the controller is not aﬀected by the oscillations.

Figure 6.13 shows the simulation with load disturbance. The disturbance is better

damped when using feedback from the wheel speed sensor, than from the engine

speed sensor, which is a veriﬁcation of the discussion in Section 6.6.

6.7 Simulations 85

0 2 4 6 8

−10

0

10

20

30

40

50

60

70

Time, [s]

T

o

r

q

u

e

,

[

N

m

]

0 2 4 6 8

−300

−250

−200

−150

−100

−50

0

50

100

150

Time, [s]

T

o

r

q

u

e

,

[

N

m

]

Transmission torque z Control signal u

Figure 6.11 Simulation of Model 3 with observer and control law based on

Model 1. Feedback from the wheel speed sensor is seen in solid, and from the

engine speed sensor is seen in dashed. The design still work when simulated with

extra clutch dynamics.

0 2 4 6 8

−10

0

10

20

30

40

50

60

70

80

90

Time, [s]

T

o

r

q

u

e

,

[

N

m

]

0 2 4 6 8

−300

−250

−200

−150

−100

−50

0

50

100

150

200

Time, [s]

T

o

r

q

u

e

,

[

N

m

]

Transmission torque z Control signal u

Figure 6.12 Simulation of Model 3 with observer and control law based on

Model 1. Feedback from the wheel speed sensor is seen in solid, and from the

engine speed sensor is seen in dashed. The conclusion is that the control law works

well despite initial driveline oscillations.

86 Chapter 6 Gear-Shift Controller Design and Simulations

0 2 4 6 8

−20

−10

0

10

20

30

40

50

60

70

80

Time, [s]

T

o

r

q

u

e

,

[

N

m

]

0 2 4 6 8

−300

−250

−200

−150

−100

−50

0

50

100

150

200

Time, [s]

T

o

r

q

u

e

,

[

N

m

]

Transmission torque z Control signal u

Figure 6.13 Simulation of Model 3 with observer and control law based on

Model 1. An impulse disturbance is acting on the load at t = 3 s. Feedback from

the wheel speed sensor is seen in solid, and from the engine speed sensor is seen

in dashed. The conclusion is that the load disturbance is better attenuated when

using feedback from the wheel speed sensor.

6.8 Summary

Driveline oscillations is a limiting factor in gear shifting with engine control. Based

on a model of the transmission torque, a criterion for a gear-shift controller is

obtained, that actively damps driveline oscillations. The proposed solution handles

the fact that the gear-shift condition is not a stationary point.

When using a driveline model with drive shaft ﬂexibility, it is possible to solve

the criterion for a control law that minimizes the cost function. The control law is

derived with LQG/LTR technique. Simulations show that the performance of the

design, based on the simpliﬁed model, works well for a more complicated model

with a nonlinear clutch characteristics. However, there can be problems with a low

frequency level that gives a stationary error. This diﬀerence in level is a result of

the diﬃculty to estimate the driveline friction parameters.

An investigation of the inﬂuence, from diﬀerent sensor locations, on the control

design results in the same conclusion as in Chapter 5. When using LQG/LTR the

open-loop zeros are cancelled by the controller. This results in undamped load

disturbances when engine speed feedback is used. Therefore, load disturbances are

better attenuated with feedback from the wheel speed sensor.

6.8 Summary 87

Measurement disturbances are better attenuated when the engine speed sensor

is used, than when using the wheel speed sensor. This eﬀect increases with lower

gears. Two diﬀerent closed-loop transfer functions result, depending on feedback

conﬁguration. The diﬀerence between these two is described by the dynamic output

ratio.

In conclusion, actively damped transmission-torque control works well also in

the case of existing initial oscillations. Furthermore, disturbances occuring during

the control action are actively damped, and thus reducing the time needed for a

gear shift.

88 Chapter 6 Gear-Shift Controller Design and Simulations

7

Conclusions

The driveline is a fundamental component in a vehicle, and there is currently

a strong trend in improving performance by adding functionalities in driveline

management systems.

The major contribution of this thesis is a novel gear shifting strategy based

on modeling of the transmission torque, and design of a criterion for a controller

that drives this torque to zero. This controller is to be used with a new automatic

gear shifting system, utilizing engine controlled gear shifting without using the

clutch. The proposed solution oﬀers a possibility to optimize the time needed for

a gear shift, which is important since the vehicle is free rolling when in gear-shift

condition. Furthermore, neutral gear can successfully be engaged also when facing

critical load disturbances and initial driveline oscillations.

A second important contribution is the extension of the traditionally used RQV

controller. A criterion for a controller that actively damps wheel speed oscillations

with a stationary error characteristic for the RQV controller, is obtained. With

this controller the performance and driveability is improved since vehicle shuﬄe is

reduced. Furthermore, the formulation is natural, it allows eﬃcient solution, and

there is a simple tuning of the amount of RQV feeling.

A basis for these results is the modeling conclusions drawn from experiments

and modeling using a heavy truck. A key contribution is the observation that

a linear model with stiﬀ clutch and drive shaft ﬂexibility is able to explain the

measured engine speed and wheel speed. Extra clutch dynamics is not able to

explain more of the experiments for low frequencies. Therefore, the linear model

is concluded to be a basis for control design, which is veriﬁed by simulations on a

model with a nonlinear clutch characteristics.

89

90 Chapter 7 Conclusions

Another important observation from the experiments is the explanation of the

diﬀerence between the measured engine speed and transmission speed. The major

part of the diﬀerence is explained by a simple sensor model. Parameter estimation

of a nonlinear model shows that the deviations still left occur when the clutch

transfers zero torque.

A common architectural issue in driveline control is the issue of sensor location.

Diﬀerent sensors give the same open-loop poles, but diﬀerent zeros. An investiga-

tion of the inﬂuence from diﬀerent sensor locations on the control design shows that

when using LQG/LTR, load disturbances are better damped with feedback from

the wheel speed, due to well damped open-loop zeros. Measurement disturbances

are better attenuated when the engine speed sensor is used, than when using the

wheel speed sensor. The diﬀerence is explained by the dynamic output ratio, and

increases with lower gears.

There are thus issues to be considered in sensor choice, but the overall conclusion

is that the proposed strategies improve performance and driveability in both speed

control and gear-shift control.

Bibliography

Bj¨ornberg, A., M. Pettersson, and L. Nielsen (1996). Nonlinear driveline oscilla-

tions at low clutch torques in heavy trucks. To be presented at Reglerm¨otet ’96 in

Lule˚a, Sweden.

Bosch (1993). Automotive Handbook. Stuttgart, Germany: Robert Bosch GmbH.

Gillespie, T. D. (1992). Fundamentals of Vehicle Dynamics. SAE International.

Henriksson, T., M. Pettersson, and F. Gustafsson (1993). An investigation of the

longitudinal dynamics of a car, especially air drag and rolling resistance. Tech-

nical Report LiTH-ISY-R-1506, Department of Electrical Engineering, Link¨ oping

University.

Kubrusly, C. and H. Malebranche (1985). Sensors and controllers location in

distributed systems - a survey. Automatica 21, 117–128.

Liversidge, J. H. (1952). Backlash and Resilience within Closed Loop of Automatic

Control Systems. Academic Press.

Ljung, L. (1988). Control Theory 1984-1986. Automatica 24, 573–583.

Ljung, L. (1995). System Identiﬁcation Toolbox. User’s Guide. MathWorks, Inc.

Maciejowski, J. M. (1989). Multivariable Feedback Design. Addison-Wesley.

Meriam, J. L. and L. G. Kraige (1987). Dynamics, Volume 2 of Engineering

Mechanics. John Wiley & Sons.

91

92 Bibliography

Mo, C. Y., A. J. Beaumount, and N. N. Powell (1996). Active control of drive-

ability. SAE Paper 960046.

Nwagboso, C. O. (1993). Automotive Sensory Systems. Chapman & Hall.

Orehall, L. (1995). Scania opticruise: Mechanical gearchanging with engine con-

trol. Truck and Commercial Vehicle International ’95.

Pettersson, M. and L. Nielsen (1995). Sensor placement for driveline control.

Preprint of the IFAC-Workshop on Advances in Automotive Control, Ascona,

Switzerland.

Simulink (1993). User’s guide. MathWorks, Inc.

Spong, M. W. and M. Vidyasagar (1989). Robot Dynamics and Control. John

Wiley & Sons.

Suzuki, K. and Y. Tozawa (1992). Inﬂuence of powertrain torsional rigidity on

NVH of 6x4 trucks. SAE Paper 922482.

Notations

Variables

r Radius, reference signal

u Control signal

z Performance output

x State vector

y Sensor output

v State disturbance, velocity

e Measurement disturbance

n Input disturbance

l Load

θ Angle

α Road slope

Symbols

J Mass moment of inertia

i Conversion ratio

k Torsional stiﬀness

93

94 Notations

c Torsional damping

b Viscous friction component

m Vehicle mass

c

r1

, c

r2

Coeﬃcients of rolling resistance

c

w

Air drag coeﬃcient

ρ

a

Air density

A

a

Vehicle cross section area

F

a

Air resistance force

F

r

Rolling resistance force

M Torque, performance output state matrix

A State-space matrix

B Input state matrix

C Output state matrix

H Load state matrix

D Performance output control signal matrix

G Transfer function

G

w/m

Dynamic output ratio

S Sensitivity function

T Complementary sensitivity function

K

c

State-feedback matrix

K

f

Observer gain

ϕ Phase margin

a Amplitude margin

Subscripts

m Engine

c Clutch

t Tranmission

p Propeller shaft

f Final drive

d Drive shafts

w Wheel

fr Friction

0 Stationary value

t1 Transmission input

t2 Transmission output

Driveline Modeling and Principles for Speed Control and Gear-Shift Control c 1996 Magnus Pettersson magnusp@isy.liu.se Department of Electrical Engineering Link¨ping University o S–581 83 Link¨ping o Sweden

ISBN 91-7871-744-2 ISSN 0280-7971 LiU-TEK-LIC-1996:29

To Anna and Oscar

.

. neutral gear can successfully be engaged also when facing load disturbances and initial driveline oscillations. which are controlled by a driveline management system. clutch. Furthermore. The performance and driveability is signiﬁcantly improved. Engine control for automatic gear shifting is an approach at the leading edge of technology. A linear model with a drive shaft ﬂexibility is able to suﬃciently explain the measured engine speed and wheel speed. A critical step is the controlling of the engine such that the transmission transfers zero torque. Experiments and modeling using a heavy truck show that there are signiﬁcant torsional resonances in the driveline. while maintaining the desired load characteristics for RQV control. shafts. In conclusion. whereafter neutral gear can be engaged. and wheels. Driveline oscillations is a limiting factor in this system. Traditionally in diesel trucks. the proposed strategies improve performance and driveability in both speed control and gear-shift control. A model based state-feedback controller is derived that actively reduces wheel speed oscillations. A model of the transmission torque is developed and a state-feedback controller is used to drive this torque to zero. transmission. and the undesired property of vehicle shuﬄe following a change in pedal position. This system has the desired property of a load dependent stationary error. the engine speed is controlled by a system called RQV. The result is a possibility to optimize the time needed for a gear shift.Abstract i Abstract A vehicular driveline consists of engine.

ii Abstract .

I am also grateful to Simon Edlund. I am indebted to Lars-Gunnar Hedstr¨m. and o o o Bj¨rn Westman at Scania in S¨dert¨lje for the help during this work. Finally. I am indebted to Dr Joakim Petersson. Link¨ping University. patience. Kjell Gestl¨v. Link¨ping. April 1996 o Magnus Pettersson . Dr Fredrik Gustafsson. and for o o a interesting discussions regarding control and modeling in heavy trucks. Sweden. I would like to express my deepest gratitude to my wife Anna and our son Oscar for their encouragements. By inspiring me and o taking time for many discussions he has contributed to this work in many ways. and Mattias Nyberg for reading the manuscript. Dr Anders Helmersson. for our many discussions regarding research and courses. Anders Bj¨rnberg. I am very grateful to my parents Birgitta and Nils and my sister Katharina for their love and support in whatever I do. Lars Eriksson. and love during this work. Thanks for the remarks and suggested improvements. Thanks to Dr Peter Lindskog and Magnus Sundstedt for support on computers A and L TEX. Thanks also to Tomas Henriksson. my former oﬃce colleague. and Dr Tomas McKelvey for help and discussions.Acknowledgment iii Acknowledgment This work has been carried out under the excellent guidance of Professor Lars Nielsen at Vehicular Systems.

iv Acknowledgment .

. . . . . . . . . . . . . . 1 2 3 3 7 7 10 11 11 12 14 17 17 19 21 24 24 24 26 27 31 33 . . . .2.1 Outline and Contributions .2 Model 3: Nonlinear Clutch and Drive Shaft Flexibility 2. . .1 Model 2: Flexible Clutch and Drive Shafts . . . . .2 Shaft Flexibilities . . . . . . .4. . . . . . . . . .3. . . 3.4 Additional Dynamics . 2. . . . . . . . . . . . . . . . . . .4. . . . . . . . . . . . . . . . . . . . .1 The Truck . . . . . . .2. . . . . . . . . . . . . . . 2. . . . . .2 Inﬂuence from the Propeller Shaft . . . . . . .3 Models Including the Clutch . . . . . . . 2. . . . . . . . 3. . . . . . . . . . . . . . 3.2 Model 1 Extended with a Flexible Propeller Shaft . . . . . . . . .5 Model Validity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3. 3. . . . . . .3 Experiments . v . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3. . . . . . . 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4. 3 Field Trials and Modeling 3. . . . . . . . . . . . . . 3. . . .4 Inﬂuence from the Clutch . .Contents 1 Introduction 1. . . . . . . . . . . . .3. . . .2 Measurement Description . . . . . . . . . . . .5 Summary .4 Models . . . . . . . . . . . .4. .4. . . . . . . . . . . . 3. . 2. . . . . . . . . . . . . . . . . . . . . . 2. . 3. . . . . . . .3 Deviations between Engine Speed and Transmission Speed . . . . .1 Inﬂuence from the Drive Shaft . .1 Model 1: Drive Shaft Flexibility . 2 Driveline Modeling 2.1 Basic Equations .

. . . . . . . . . . . . . .4. . . . . . . . . . . . . . .1 Extending with RQV Behavior . . . . . . . . . . . . .2 Transmission Torque for Model 2 . . . . . .1 Transmission Torque for Model 1 . 35 36 37 37 38 39 41 42 43 46 49 49 50 50 53 55 56 58 60 62 63 66 67 68 68 68 71 72 73 73 74 76 77 80 80 81 83 86 89 91 93 5 Speed Controller Design and Simulations 5. . . . . . . . . . . . . . . . .1 RQV Control . . . . . .1. . . . . . . . . 4. . . . . . .3. . . . . . . . . . . 5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.7 Simulations . .2 Problem Formulation . . . . 6. . . . . .4 Inﬂuence from Sensor Location . . .1 Transfer Functions . 6 Gear-Shift Controller Design and Simulations 6. . .4. . . . . . . . . . . . . . . . .2. .3 Load Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Inﬂuence from Load Disturbances . . . . . . .3 Speed Control with Active Damping and RQV Behavior 5. . . . . . . . 5. 4. . . . .3 Transmission Torque for Model 3 . . . . . . . . . . . . . . . . . . . . .2 Measurement Description . . .2. .2 Controller Formulation . . . . . . . . .1 State-Space Formulation . . . . . 6. . . . . . . . . . . . . . . .4 Driveline Control with LQG/LTR . . . . .6 Inﬂuence from Sensor Location . . . . . . . . . . . . .4. . . . . .2 Inﬂuence from Measurement Disturbances 6.5 Summary . . . . . .6. . . 6.1 Mathematical Problem Formulation . . . . . . . . . . . . . . . .2 Inﬂuence from Measurement Disturbances . . . 5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Transmission Torque . . . . . . . . . . . . . 6. . . . . . . . 6. . .2 Undamped Gear-Shift Condition . . . . . . . . . . 5. . . . 6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Inﬂuence from Load Disturbances . . . . . . . . . . . . . . . . . . . . . . . . . . 4. . . . . 5. . 7 Conclusions Bibliography Notations .3 Preliminary Trials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6. . . . . . . . . . 4. . .3 Some Feedback Properties . . . . . . . 6. . . . . . . . . . 5.6 Summary . . . . . . . . . . . . .5 Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4. . . . . . . . . . . . . . . . . . . . 5.2 Design Example with a Simple Mass-Spring 4. . . . . . .4. 6. . .1 Disturbance Description . . . . . . . . . . . . . . . . . . . . . . . 6. . . . . . . . . . . . .2. . . .4. . . . . . 5. . . . . . . . . . . . . . 6. 6. . . . . . .5 Gear-Shift Control Design . . . . . . . . . . .vi Contents 4 Architectural Issues for Driveline Control 4. . 4. . . . . . . .1. . . . . . . . . . . . . . . . . . . . . . . Model . . .1 Unconstrained Active Damping . . . . 5. .1 Problem Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . .4 Gear-Shift Control Criterion . . . . . . . . . .3. . . . . . . . .6. . . . . . .3. 6. . . . . . . . .2. .8 Summary .

The engine is controlled such that the transmission transfers zero torque. there is no active damping of wheel speed oscillations resulting in vehicle shuﬄe. Some open questions are discussed. The result is a series of models consisting of rotating inertias. Another property is that a load dependent stationary error results from downhill and uphill driving. the fuel metering is governed by a system called RQV. mechanical resonances may occur. Traditionally in diesel trucks. The ﬁrst problem is wheel speed oscillations following a change in accelerator pedal position. Engine controlled gear shifting without disengaging the clutch is an approach at the leading edge of technology (Orehall 1995). and wheels. Pettersson and Nielsen 1995). The handling of such resonances is of course basic for driveability. regarding inﬂuence of sensor dynamics and nonlinear eﬀects. Since these parts are elastic. Experiments are performed with a heavy truck with diﬀerent gears and road slopes. Beaumount. transmission. whereafter neutral gear can be engaged. The aim of the modeling and experiments is to ﬁnd the most important physical eﬀects that contribute to driveline oscillations. The thesis treats model based speed control with active damping of wheel speed oscillations while maintaining the stationary error characteristic for RQV control.1 Introduction The main parts of a vehicular driveline are engine. connected with damped torsional ﬂexibilities. Fundamental driveline equations are obtained by using Newton’s second law. and Powell 1996. clutch. known as vehicle shuﬄe (Mo. With RQV. shafts. 1 . Two systems where driveline oscillations limit performance is speed control and automatic gear shifting. but is also otherwise becoming increasingly important since it is a linking factor in development of new driveline management systems.

A key contribution in this chapter is the formulation of a criterion for the speed control concept described above with active damping and retained RQV feeling. Experiments with a heavy truck are described together with the modeling conclusions. The contribution of the chapter is a demonstration of the inﬂuence of sensor location in driveline control when using LQG/LTR. Control of resonant systems with simple controllers is. In this state. Driveline oscillations is a limiting factor in optimizing this step. These results are reviewed in Chapter 4. and a criterion for a controller that drives this torque to zero. A critical part in this scheme is the controlling of the engine such that the transmission torque is zero. The extension to more advanced control design methods is a little studied topic. Chapters 5 treats the design and simulation of the speed controller. . known to have diﬀerent properties with respect to sensor location. it is possible to optimize the time needed for a gear shift. Conclusions are summarized in Chapter 7. a set of three driveline models is derived.1 Outline and Contributions In Chapters 2 and 3. from other technical ﬁelds. With this approach. 1. and the inﬂuence in control design is investigated. Chapters 6 deals with the design and simulation of the gear-shift controller. which must be handled. The design improves the performance also in the case of load disturbances and initial driveline oscillations. A comparison is made between using feedback from the engine speed sensor or the wheel speed sensor. In this thesis the transmission torque is modeled. Diﬀerent sensor locations result in diﬀerent control problems. the vehicle is free rolling. The contribution of the chapter is that a linear model with one torsional ﬂexibility and two inertias is able to ﬁt the measured engine speed and wheel speed within the bandwidth of interest. A simulation study shows signiﬁcantly improved performance and driveability. A major contribution in this thesis is a gear shifting strategy. A common architectural issue in the two applications described above is the issue of sensor location. and controlled to zero by using state feedback. based on a model describing the transmission torque. Parameter estimation of a model with a nonlinear clutch and sensor dynamics explains that the diﬀerence between experiments and model occurs when the clutch transfers zero torque.2 Chapter 1 Introduction The engine speed is then controlled to a speed such that the new gear can be engaged. also when facing existing initial driveline oscillations.

drive shafts. In this section fundamental equations for the driveline will be derived. Next chapter will validate the choices. First.1 Basic Equations A vehicular driveline is depicted in Figure 2. Nevertheless. ﬁnal drive. Furthermore. a set of three models of increasing complexity is presented. together with a linear model of the air drag constitute the model. It consists of an engine. transmission. A second linear model is given by using the assumptions made above. The frequency range treated in this work is the regime interesting for control design (Mo. The generalized Newton’s second law is used together with assumptions about how diﬀerent parts in the driveline contribute to the model. This chapter deals with building models of a truck driveline. Assumptions about stiﬀ clutch.1. Vibrations and noise contribute to a higher frequency range (Suzuki and Tozawa 1992. Gillespie 1992) which is not treated here. and adding a second ﬂexibility which is the clutch. Finally. contributing to driveline oscillations. some basic equations regarding the forces acting on the wheel. are obtained. a more complete nonlinear model is derived which includes a clutch model with a static nonlinearity. propeller shaft. and wheels. These equations 3 . The aim of these assumptions is to ﬁnd the most important physical eﬀects. viscous friction in transmission and ﬁnal drive. Pettersson and Nielsen 1995). a linear model with ﬂexible drive shafts is derived. Modeling is an iterative process in reality. Beaumount.2 Driveline Modeling The driveline is a fundamental part of a vehicle and its dynamics has been modeled in diﬀerent ways depending on the purpose. and Powell 1996. stiﬀ propeller shaft. clutch. 2.

4 Chapter 2 Driveline Modeling Wheel Engine Clutch Drive shaft Final drive Transmission Propeller shaft Figure 2. no internal friction is assumed. in order to get a complete physical model. for instance. Next. vehicle mass and trailer will be described by the equation describing the wheel. a relation between the inputs and outputs of each part is obtained. When the clutch is engaged.1) Jm θm = Mm − Mf r:m − Mc where Jm is the mass moment of inertia of the engine and θm is the angle of the ﬂywheel.2. and the external load from the clutch (Mc ).2) . The transmitted torque is a function of the angular diﬀerence (θm − θc ) and the angular velocity ˙ ˙ diﬀerence (θm − θc ) over the clutch ˙ ˙ Mc = Mt = fc (θm − θc . The generalized Newton’s second law of motion (Meriam and Kraige 1987) gives the following model ¨ (2. This means that eﬀects from. the internal friction from the engine (Mf r:m ). θm − θc ) (2. giving Mc = Mt . Engine: The output torque of the engine is characterized by the driving torque (Mm ) resulting from the combustion. Inputs and outputs of each subsystem are labeled according to Figure 2. are inﬂuenced by the complete dynamics of the vehicle.1 A vehicular driveline. according to Figure 2. Clutch: A friction clutch found in vehicles equipped with a manual transmission consists of a clutch disk connecting the ﬂywheel of the engine and the transmission’s input shaft.2.

if ) (2. θp − θf if . This gives the following relation between the input and output torque of the transmission ˙ ˙ Mp = ft (Mt . θc − θt it . Mf r:t .1 Basic Equations 5 Mm θm Engine Mc Clutch θc Mt Transmission Mf r:t θt Mp Mf r:m θt Mp Propeller shaft θp Mf Final drive Mf r:f θf Md θf Md Drive shaft θw Mw Wheel θw rw Fw Mf r:w Figure 2.2. each with a conversion ratio it . Mf r:f .3) is the possibility of having torsional eﬀects in the transmission. giving the following model of the torque input to the ﬁnal drive ˙ ˙ Mp = Mf = fp (θt − θp . The reason for considering the angle diﬀerence θc − θt it in (2.4) Final drive: The ﬁnal drive is characterized by a conversion ratio if in the same way as the transmission. θp − θf if .3) where the internal friction torque of the transmission is labeled Mf r:t . θc − θt it . θt − θp ) (2. Propeller shaft: The propeller shaft connects the transmission’s output shaft with the ﬁnal drive.2 Subsystems of a vehicular driveline with its input and output angle and torque. Transmission: A transmission has a set of gears. No friction is assumed (Mp = Mf ). it ) (2. The following relation for the input and output torque holds ˙ ˙ Md = ff (Mf .5) .

the drive shafts are modeled as one shaft. Therfore. • Fa . Aa the maximum vehicle cross section area. Here it is assumed that the wheel speed is the same for the two wheels.8) where cw is the drag coeﬃcient. the air drag. However.6 Chapter 2 Driveline Modeling v Fa Fw Fr + mg sin(α) Figure 2. for instance.3 the forces acting on a vehicle with mass m and speed v is shown. is approximated by Fa = 1 cw Aa ρa v 2 2 (2. and ρa the air density.7) The friction force (Fw ) is described by the sum of the following quantities (Gillespie 1992). where the internal friction torque of the ﬁnal drive is labeled Mf r:f . open or closed windows will make the force diﬃcult to model. Drive shafts: The drive shafts connects the wheels to the ﬁnal drive. No friction (Mw = Md ) gives the model equation ˙ ˙ Mw = Md = fd (θf − θw . Newton’s second law in the longitudinal direction gives ˙ Fw = mv + Fa + Fr + mg sin(α) (2. When the vehicle is turning and the speed diﬀers between the wheels. θf − θw ) (2. . both drive shafts have to be modeled. eﬀects from.6) Wheel: In Figure 2.3 Forces acting on a vehicle.

• mg sin(α). So far the functions fc . fd .2.11) A complete model for the driveline with the clutch engaged is described by Equations (2.7) to (2.1 are made in order to obtain a model with drive shaft ﬂexibility. ft . and Gustafsson 1993). ff . In the following section assumptions will be made about the unknowns. The clutch and the propeller shafts are assumed to be stiﬀ.g. Pettersson. by a identiﬁcation scheme (Henriksson.10) where Jw is the mass moment of inertia of the wheel. The coeﬃcients of air drag and rolling resistance. where rw is the wheel radius. without losses. and Mf r:w are unknown.8) and (2.9) where cr1 and cr2 depends on.2. can be identiﬁed e.11). resulting in a series of driveline models. and then a model with two torsional ﬂexibilities (the drive shaft and propeller shaft) will be considered. . and the drive shaft is described as a damped torsional ﬂexibility. Mw is given by (2. Including (2. and the friction torques Mf r:t . fp .9) in (2. The resulting torque due to Fw is equal to Fw rw .2 Shaft Flexibilities In the following two sections. where α is the slope of the road.1 Model 1: Drive Shaft Flexibility Assumptions about the fundamental equations in Section 2. for instance. the rolling resistance. the gravitational force. Mf r:f . Newton’s second law gives ¨ Jw θw = Mw − Fw rw − Mf r:w (2. First. is approximated by Fr = m(cr1 + cr2 v) (2.10) together ˙ with v = rw θw gives 2 ¨ (Jw + mrw )θw = 1 3 ˙2 Mw − Mf r:w − cw Aa ρa rw θw 2 ˙ −rw m(cr1 + cr2 rw θw ) − rw mgsin(α) (2. a model with one torsional ﬂexibility (the drive shaft) will be considered.2 Shaft Flexibilities 7 • Fr .9). and Mf r:w is the friction torque. (2. 2. The transmission and the ﬁnal drive are assumed to multiply the torque with the conversion ratio. Labels are according to Figure 2.2. with diﬀerent complexities.1) to (2. 2. assumptions will be made about the unknowns. tires and tire pressure.6).

having stiﬀness k. the model can be rewritten as ¨ ˙ Jt θm = Mc i2 − bt θm − Mp it t (2.18) can be rewritten with (2.21) Drive shaft: The drive shaft is modeled as a damped torsional ﬂexibility.14) By using (2. the ﬁnal drive is modeled by one rotating inertia Jf .17) which gives ¨ ˙ Jf θt = Mp i2 − bf θt − Md if f (2.19) Reducing (2. and internal damping c.5).20) with Mp in (2.22) . which gives the following equations for the torque and the angle M p = Mf .16) Final drive: In the same way as the transmission.13).17) (2.16) and (2. and ﬁnal drive is obtained ¨ ˙ ˙ (Jt i2 + Jf )θm = Mc i2 i2 − bt θm i2 − bf θm − Md if it f t f f (2. The friction torque is assumed to be described by a viscous damping coeﬃcient bt . corresponding to (2. corresponding to (2.18) Equation (2. The model of the ﬁnal drive. The model of the transmission.13) (2.3). θm = θ c (2. a model for the lumped transmission.19) to engine speed is done by using (2. Hence.12) and (2.13) resulting in ¨ ˙ (2.6) becomes Mw = M d ˙ ˙ k(θf − θw ) + c(θf − θw ) = k(θm /it if − θw ) ˙ ˙ + c(θm /it if − θw ) = (2. which gives the following equations for the torque and the angle Mc = Mt . is θc ¨ Jt θt = θt it ˙ = Mt it − bt θt − Mp (2.15) Propeller shaft: The propeller shaft is also assumed to be stiﬀ.12) and (2.12) Transmission: The transmission is described by one rotating inertia Jt .20) Jf θm = Mp i2 it − bf θm − Md if it f By replacing Mp in (2. is θp ¨ Jf θ f = = θf if ˙ Mf if − bf θf − Md (2. propeller shaft.15). θt = θ p (2.8 Chapter 2 Driveline Modeling Clutch: The clutch is assumed to be stiﬀ. (2. The friction torque is assumed to be described by a viscous damping coeﬃcient bf .

21) with (2.drive shaft ﬂexibility.22) the equation describing the transmission.16). and drive shaft.1).22). . An illustration of the model can be seen in Figure 2. The complete model is obtained by inserting Mc from (2. (2. and θm /it if − θw . propeller shaft.24) (Jw + mrw )θw = k(θm /it if − θw ) + c(θm /it if − θw ) ˙w − cw Aa ρa r3 θw − mcr2 r2 θw − rw m (cr1 + gsin(α)) ˙ ˙ −bw θ w w where the friction torque is described as viscous damping. the following equation for the wheel results 2 ¨ ˙ ˙ (2.23) Wheel: If (2.13).23) into (2.1 Resulting equations for Model 1 .24). which gives the following equations. with label bw .12).2. (2. where (2.4 Model 1: Stiﬀ clutch and drive shaft torsional ﬂexibility. ¨ (Jm + Jt /i2 + Jf /i2 i2 )θm t t f = ˙ Mm − Mf r:m − (bt /i2 + bf /i2 i2 )θm (2. and (2.26) −rw m (cr1 + gsin(α)) ˙ ˙ Possible states describing Model 1 are θm .2 Shaft Flexibilities 9 θm k Mm − Mf r:m c Jm + Jt /i2 + Jf /i2 i2 t t f θw rw m (cr1 + gsin(α)) 2 Jw + mrw Figure 2.4. ﬁnal drive.17) are used. together with (2.25) t t f −k(θm /it if − θw )/it if ˙ ˙ −c(θm /it if − θw )/it if ˙ ˙ k(θm /it if − θw ) + c(θm /it if − θw ) 3 2 ˙ −(bw + cw Aa ρa r + mcr2 r )θw w w 2 ¨ (Jw + mrw )θw = (2.11) is combined with (2. By replacing Md in (2. θw . becomes ¨ (Jt i2 + Jf )θm f = ˙ ˙ Mc i2 i2 − bt θm i2 − bf θm t f f ˙ ˙ −k(θm − θw it if ) − c(θm − θw it if ) (2. and if the linear part of the air drag in (2. Deﬁnition 2.11) is used.

1) the following diﬀerential equation describing the lumped engine and transmission results ¨ (Jm + Jt /i2 )θm t = ˙ Mm − Mf r:m − bt /i2 θm t 1 ˙ ˙ − kp (θm /it − θp ) + cp (θm /it − θp ) it (2.34) (2.18).30) in (2. This means that there are two torsional ﬂexibilities.31) Including (2.30) is used in the last equality. and repeating (2.33) where (2.2. the clutch is assumed stiﬀ.16) is replaced by a model of a ﬂexibility with stiﬀness kp and internal damping cp ˙ ˙ ˙ ˙ Mp = Mf = kp (θt − θp ) + cp (θt − θp ) = kp (θm /it − θp ) + cp (θm /it − θp ) (2.15) gives ˙ ˙ ¨ ˙ Jt θm = Mc i2 − bt θm − kp (θm /it − θp ) + cp (θm /it − θp ) it t (2.27) into (2.10 Chapter 2 Driveline Modeling 2.31) gives ˙ ˙ ¨ ˙ Jf θp = i2 kp (θm /it − θp ) + cp (θm /it − θp ) − bf θp − if Md f The equation for the drive shaft (2. As in the derivation of Model 1. without losses. the transmission and ﬁnal drive are assumed to multiply the torque with the conversion ratio.30) (2. and the propeller and drive shafts are modeled as damped torsional ﬂexibilities. Inserting (2.28) By combining this with (2.27) in (2. In the derivation of the model.2 Model 1 Extended with a Flexible Propeller Shaft It is also possible to consider two torsional ﬂexibilities.12) and (2.27) where (2. The derivation of Model 1 is repeated here with the diﬀerence that the model for the propeller shaft (2. the propeller shaft and the drive shaft.17) θp ¨ Jf θ f = = θf if ˙ ˙ ˙ if kp (θm /it − θp ) + cp (θm /it − θp ) − bf θf − Md (2.13) are used in the last equality.29) The ﬁnal drive is described by inserting (2. the propeller shaft and the drive shaft. The equation for the ﬁnal drive (2.22) is repeated with new labels ˙ ˙ ˙ ˙ Mw = Md = kd (θf − θw ) + cd (θf − θw ) = kd (θp /if − θw ) + cd (θp /if − θw ) (2.32) now becomes ¨ Jf θp = ˙ ˙ ˙ i2 kp (θm /it − θp ) + cp (θm /it − θp ) − bf θp f ˙ ˙ −if kd (θp /if − θw ) + cd (θp /if − θw ) (2.32) .

the following equation for the wheel results 2 ¨ ˙ ˙ (Jw + mrw )θw = kd (θp /if − θw ) + cd (θp /if − θw ) 3 ˙ 2 ˙ ˙ −bw θw − cw Aa ρa rw θw − mcr2 rw θw − rw m (cr1 + gsin(α)) (2.5.38) 3 2 ˙ −(bw + cw Aa ρa r + mcr2 r )θw − rw m (cr1 + gsin(α)) w w 2. The equation for the wheel is derived by combining (2.11) with (2.3 Models Including the Clutch The clutch is so far considered to be stiﬀ and lumped together with the engine mass moment of inertia. In this section this assumption is relaxed and ﬁrst. If the linear part of the of the air drag in (2. The complete model with drive shaft and propeller shaft ﬂexibilities is the following.3.37) = ˙ ˙ kd (θp /if − θw ) + cd (θp /if − θw ) (2.35) where again the friction torque is assumed to be described by a viscous damping coeﬃcient bw .11) is used. 2. the clutch is modeled as a linear ﬂexibility.1 Model 2: Flexible Clutch and Drive Shafts A model with a linear clutch and one torsional ﬂexibility (the drive shaft) is derived by repeating the procedure for Model 1 with the diﬀerence that the model for the .5 Model with stiﬀ clutch and two torsional ﬂexibilities.2. which can be seen in Figure 2. ¨ (Jm + Jt /i2 )θm t ˙ Mm − Mf r:m − bt /i2 θm t 1 ˙ ˙ − kp (θm /it − θp ) + cp (θm /it − θp ) it ˙ ˙ ˙ = i2 kp (θm /it − θp ) + cp (θm /it − θp ) − bf θp f = ˙ ˙ −if kd (θp /if − θw ) + cd (θp /if − θw ) 2 ¨ (Jw + mrw )θw (2.33). a nonlinear model of the clutch is derived.3 Models Including the Clutch 11 θm kp Mm + Mf r:m cp Jm + Jt /i2 t θp kd θw rw m (cr1 + gsin(α)) cd Jf 2 Jw + mrw Figure 2.36) ¨ J f θp (2. Secondly.

44) (2.22) as ˙ ˙ ˙ ˙ Mw = Md = kd (θf − θw ) + cd (θf − θw ) = kd (θt /if − θw ) + cd (θt /if − θw ) (2.19) giving ¨ ˙ ˙ ˙ (Jt + Jf /i2 )θt = it kc (θm − θt it ) + cc (θm − θt it ) − (bt + bf /i2 )θt − Md /if (2. The drive shaft is modeled according to (2.3.39) where (2. ¨ Jm θ m ¨ (Jt + Jf /i2 )θt f = = ˙ ˙ Mm − Mf r:m − kc (θm − θt it ) + cc (θm − θt it ) ˙ ˙ it kc (θm − θt it ) + cc (θm − θt it ) 1 ˙ −(bt + bf /i2 )θt − f if 2 ¨ (Jw + mrw )θw (2.42) and (2.1) the equation describing the engine inertia is given by ˙ ˙ ¨ Jm θm = Mm − Mf r:m − kc (θm − θt it ) + cc (θm − θt it ) (2. When the angle diﬀerence over the clutch starts from zero and increases.12 Chapter 2 Driveline Modeling clutch is a ﬂexibility with stiﬀness kc and internal damping cc ˙ ˙ ˙ ˙ Mc = Mt = kc (θm − θc ) + cc (θm − θc ) = kc (θm − θt it ) + cc (θm − θt it ) (2. propeller shaft.43) where (2.14).45) ˙ ˙ kd (θt /if − θw ) + cd (θt /if − θw ) = ˙ ˙ kd (θt /if − θw ) + cd (θt /if − θw ) (2.16) and (2. and ﬁnal drive inertia.40) Also by inserting (2. with .ﬂexible clutch and drive shaft ﬂexibility.13) is used in the last equality. The reason for this arrangement is vibration insulation.17) is used in the last equality.43) into (2. and using the linear part of the air drag.46) 3 ˙ −(bw + cw Aa ρa r + cr2 rw )θw − rw m (cr1 + gsin(α)) w 2.41) (2.11). the smaller springs. the equation describing the transmission is ˙ ˙ ¨ ˙ Jt θt = it kc (θm − θt it ) + cc (θm − θt it ) − bt θt − Mp Mp is derived from (2.39) into (2. Deﬁnition 2.2 Resulting equations for Model 2 .42) f f which is the lumped transmission. By inserting this into (2. The complete model is obtained by inserting (2. An illustration of the model can be seen in Figure 2.6.2 Model 3: Nonlinear Clutch and Drive Shaft Flexibility When studying a clutch in more detail it is seen that the torsional ﬂexibility comes from an arrangement of smaller springs in series with springs with much higher stiﬀness.

the stiﬀer springs. are being compressed.3 Models Including the Clutch 13 θm kc Mm + Mf r:m cc Jm θt kd θw rw m (cr1 + gsin(α)) cd Jt + Jf /i2 f 2 Jw + mrw Figure 2. Torque mechanical stop kc2 −θc2 −θc1 kc1 kc2 kc1 θc1 θc2 θm − θc mechanical stop Figure 2. the clutch hits a mechanical stop.7 Nonlinear clutch characteristics. This ends when they are fully compressed at θc1 radians.47) .7. are beginning to be compressed.6 Model 2: Linear clutch and drive shaft torsional ﬂexibility.2. If the angle is increased further. This clutch characteristics can be modeled as in Figure 2. stiﬀness kc1 . When θc2 is reached. with stiﬀness kc2 . The resulting stiﬀness kc (θm − θc ) of the clutch is given by kc1 kc2 kc (x) = ∞ if |x| ≤ θc1 if θc1 < |x| ≤ θc2 otherwise (2.

48) and cc denotes the damping coeﬃcient of the clutch.8 A shaft with stiﬀness k and internal damping c with a backlash of 2α rad.8).3 Resulting equations for Model 3 .51) 3 ˙ −(bw + mcr2 rw + cw Aa ρa r )θw − rw m (cr1 + gsin(α)) w where Mkc (·) is given by (2.51) can be changed to include the nonlinear model of the air drag. described in (2.26).50) ˙ ˙ kd (θt /if − θw ) + cd (θt /if − θw ) = ˙ ˙ kd (θt /if − θw ) + cd (θt /if − θw ) (2.46). The linear part of the air drag is included. ¨ J m θm ¨ (Jt + Jf /i2 )θt f = Mm − Mf r:m − Mkc (θm − θt it ) ˙ ˙ −cc (θm − θt it ) ˙ ˙ it Mkc (θm − θt it ) + cc (θm − θt it ) 1 ˙ −(bt + bf /i2 )θt − f if 2 ¨ (Jw + mrw )θw (2. Then the diﬀerential equation describing the wheel and the vehicle (2. and (2. The torque Mkc (θm − θc ) from the clutch nonlinearity is if |x| ≤ θc1 kc1 x kc1 θc1 + kc2 (x − θc1 ) if θc1 < x ≤ θc2 Mkc (x) = −kc1 θc1 + kc2 (x + θc1 ) if −θc2 < x ≤ −θc1 ∞ otherwise (2.14 Chapter 2 Driveline Modeling 2α k θ1 c θ2 Figure 2.nonlinear clutch and drive shaft ﬂexibility. the linear part of the air drag.4 Additional Dynamics For high speeds.49) = (2. as in the previous models. 2. The . is not suﬃcient.48) The nonlinear model is given by the following equations. Deﬁnition 2. (2.

The torque resulting from a shaft connected to a drive with backlash 2α is ˙ ˙ k(θ1 − θ2 − α) + c(θ1 − θ2 ) if θ1 − θ2 > α ˙1 − θ2 ) if θ1 − θ2 < −α ˙ (2.2.8. .52) It is well known that elements like transmissions and drives introduce backlash. according to Figure 2. Throughout this thesis the dead zone model will be used (Liversidge 1952).53) M= k(θ1 − θ2 + α) + c(θ 0 if |θ1 − θ2 | < α where k is the stiﬀness and c is the internal damping of the shaft.4 Additional Dynamics 15 model describing the wheel is 2 ¨ (Jw + mrw )θw = ˙ ˙ kd (θt /if − θw ) + cd (θt /if − θw ) ˙ −(bw + mcr2 rw )θw − rw m (cr1 + gsin(α)) 1 3 ˙2 − cw Aa ρa rw θw 2 (2.

16 Chapter 2 Driveline Modeling .

and wheel speed are measured. 2 driven) has a 14 o a liter V8 diesel engine (Figure 3. As mentioned already in Chapter 2. Sweden. The transmission has 14 gears and a hydraulic retarder. The parameters of the models are estimated. The result is a series of models that describe the driveline in increasing detail. One such question is whether diﬀerences in engine speed and transmission speed is due to clutch dynamics or has other causes. 17 .52 m. The DSC14 engine is connected to a manual range-splitter transmission (Figure 3.3 Field Trials and Modeling Field trials are performed with a Scania truck. The driving torque. There has been some open questions regarding model structure in this study. engine speed. September 1995. these measurements are used to build models by extending an initial model structure by adding the eﬀect that seems to be the major cause for the deviation still left. The 6x2 truck (6 wheels. transmission speed.3) via a clutch. The drive shafts connect the ﬁnal drive to the wheels which has a radius of rw = 0.1 The Truck Tests were performed with a Scania 144L530 truck (Figure 3. 3. The weight of the truck is m = 24 ton and the front area is Aa = 9 m2 .6. Diﬀerent road slopes and gears are tested to study driveline resonances. A propeller shaft connects the output shaft of the transmission with the ﬁnal drive.2) with maximum power of 530 Hp and maximum torque of 2300 Nm. The drag coeﬃcient is equal to cw = 0.1) on test roads in S¨dert¨lje. It is also equipped with the gear shifting system Opticruise (Orehall 1995).

18 Chapter 3 Field Trials and Modeling Figure 3.1 Scania 6x2 144L530 truck.2 Scania DSC14 engine. . Figure 3.

θw . the output shaft from the transmission (θt ). which means that there is information up to 10 Hz frequency. which detects the time when cogs from a rotating cogwheel are passing. the transmission speed sensor has fewer cogs than the other two sensors.2 Measurement Description 19 Figure 3. In the rest of this thesis. indicating that the bandwidth of this signal is lower. This time sequence is then inverted to get the angle velocity.2 Measurement Description The truck is equipped with three sensors measuring the angle of the ﬂywheel of the engine (θm ). θt . If the cogwheels of the three sensors are compared. 3. the bandwidth of the measured signal depends on the speed and the number of cogs the cogwheel is equipped with. θw . The four signals are (θm . The four signals used in the following modeling are calculated from the ﬁve ˙ ˙ ˙ sampled signals. θt . Sampling is not equidistant in time. From these two signals. Mm .3. ﬁve signals are sampled (θm . Mf r:m ) with the Scania sampling program “Truck-view”. The velocity of a rotating part is measured by using an inductive sensor (Nwagboso 1993). ˙ ˙ ˙ Hence. u = Mm − Mf r:m ). the control signal u = Mm − Mf r:m is assumed to be a . a measure of the driving torque (Mm ) is obtained. By measuring the amount of fuel that is fed to the engine.3 Scania GRS900R range-splitter transmission with retarder and Opticruise gear changing system. and the driving wheel (θw ).05 s to 0. and the sample period range from 0. the torque u = Mm − Mf r:m acting on the driveline is calculated.11 s (corresponding to sampling frequencies between 9 Hz and 20 Hz). The friction torque of the engine (Mf r:m ) is also calculated online from a function given by Scania. Hence. The data has information up to half the sample period.

the data sets are resampled. Preprocessing Data Since the sampling is not equidistant in time. the interpolated data is low-pass ﬁltered with a frequency corresponding to half the sampling frequency in the original data. The chosen frequency is 6 Hz.8 Time. The prediction error estimation method (PEM) for parameterized state-space representations is used to estimate the unknown parameters and initial conditions. .5 7 6.4 Example of resampling a signal not equidistant in time (x). [s] 8 8. An example of the resampling is seen in Figure 3.20 Chapter 3 Field Trials and Modeling 10 9.5 to 10 Hz. A motivation for this is that an eight-cylinder engine makes 80 strokes/s at an engine speed of 1200 rev/min.6 7. Therefore. This means a frequency in the interval 4.4.5 9 [rad/s] 8.4 Figure 3.2 8. This is done by using Euler’s method.2 7. This is done oﬄine and therefore without phase shifts in the signals. A new data set is obtained by interpolating the old data using linear interpolation. continuous signal.4 7. This introduces higher frequencies than those in the original data set.8 7 7. the continuous model is discretized. The dynamics from fuel amount to engine torque is not considered in this work. This is reasonably for the frequency range considered for control design. The dotted line is the linear interpolation between the samples and the straight line is the signal ﬁltered with 6 Hz. In order to estimate the parameters and initial condition of the nonlinear Model 3.5 8 7. Parameter Estimation Software To estimate the parameters of the linear models derived in Chapter 2 the System Identiﬁcation Toolbox (Ljung 1995) is used.

The road was almost ﬂat. The road was almost ﬂat. initial conditions. x4 = θt . The road was almost ﬂat. yw ) with the measured signals (θm . For Model 1 the following states are used in the parameter estimation ˙ ˙ x1 = θm /it if − θw . Furthermore. in order to excite driveline oscillations. They have diﬀerent known slopes. 3. Trial 1: The test was performed with step inputs on the accelerator with gear 1. The optimal parameters and initial conditions are the ones minimizing (3. Trial 2: The test was performed with step inputs on the accelerator with gear 5. The sensor outputs described above were logged. More details about the state-space representation can be found in Chapter 4. Trial 4: The test was performed with step inputs on the accelerator with gear 8. x2 = θm . and control signal sequence u. x2 = θt /if − θw . The global truncation error with this method equals O(h). The cost function used is ˙ ˙ ˙ (θm (i) − ym (i))2 + i2 (θt (i) − yt (i))2 + i2 i2 (θw (i) − yw (i))2 t t f ∀i (3.3 Experiments A number of roads at Scania were used for testing. ˙ ˙ ˙ x1 = θm − θt it . x = f (x.2) where ∀i means that the sum is to taken over all samples in the estimation data. the state vector is calculated at each sample. un−1 ) (3. x3 = θm . The data is seen in Figure 3. . the discrete version is ˙ xn = xn−1 + hf (xn−1 .7.8.3 Experiments 21 For a continuous diﬀerential equation. are scaled to be of the same magnitude. A too small h can give numerical problems and it also gives unnecessarily long iteration time.1) where h is the sampling time. θw ) a cost function can be evaluated. The data is seen in Figure 3. with the friction torque (Mf r:m ) subtracted from the driving torque (Mm ). Trial 3: The test was performed with step inputs on the accelerator with gear 5. The data is seen in Figure 3. u). By comparing the model output (ym . The data sets are divided into two parts to be used with the parameter estimation and validation respectively. ˙ ˙ ˙ yt . The data is seen in Figure 3. The data is resampled at a sampling frequency of 1 kHz.3. x5 = θw are used. For a given set of parameters.5. describing Model 3. The road has 16 % slope.6. θt . Therefore it is necessary to keep h small. Step input experiments were done by repeatedly pressing and releasing the accelerator.2). x3 = θw and for Models 2 and 3. the ﬁve diﬀerential equations.

.6 Torque and angular velocities for a test with gear 5 and ﬂat road. [s] 20 30 Figure 3.22 Chapter 3 Field Trials and Modeling 1500 1000 500 0 Driving torque u = Mm − Mf r:m 300 250 [rad/s] 200 150 100 ˙ Engine speed θm [Nm] −500 0 10 20 30 50 0 10 20 30 20 ˙ Transmission speed θt 6 5 ˙ Wheel speed θw 15 [rad/s] [rad/s] 10 5 0 10 Time. [s] 20 30 Figure 3. [s] 20 30 2 0 10 Time.5 Torque and angular velocities for a test with gear 1 and ﬂat road. 1500 1000 500 0 Driving torque u = Mm − Mf r:m 300 250 [rad/s] 200 150 100 0 ˙ Engine speed θm [Nm] −500 0 10 20 30 10 20 30 40 35 30 [rad/s] 25 20 15 10 0 ˙ Transmission speed θt 12 10 [rad/s] 8 6 4 ˙ Wheel speed θw 10 Time. [s] 20 30 4 3 2 1 0 10 Time.

[s] 20 30 Figure 3. [s] 20 30 10 Time. 2000 1500 Driving torque u = Mm − Mf r:m 250 ˙ Engine speed θm 500 0 −500 0 10 20 30 [rad/s] 150 100 0 [Nm] 1000 200 10 20 30 70 60 ˙ Transmission speed θt 20 18 16 ˙ Wheel speed θw [rad/s] 50 40 [rad/s] 10 Time. [s] 20 30 Figure 3. [s] 20 30 14 12 10 30 0 8 0 10 Time.8 Torque and angular velocities for a test with gear 8 and ﬂat road.3.3 Experiments 23 2000 1500 Driving torque u = Mm − Mf r:m 250 200 [rad/s] 150 100 50 0 ˙ Engine speed θm [Nm] 1000 500 0 −500 0 10 20 30 10 20 30 35 30 [rad/s] ˙ Transmission speed θt 10 8 [rad/s] 6 4 2 0 ˙ Wheel speed θw 25 20 15 10 0 10 Time.7 Torque and angular velocities for a test with gear 5 and 16 % slope. .

the inertia of the ﬁnal drive (Jf ) is considerably less than J1 and J2 in (3. Here. Result • The main contribution to driveline dynamics from driving torque to engine speed and wheel speed is the drive shaft.4 Models A number of driveline models were developed in Chapter 2.1 Inﬂuence from the Drive Shaft First. θt . the model will act as if there are two damped springs in series. The engine speed and the wheel speed data is used to estimate the parameters. but there are still deviations between model and measurement. 3. Besides ˙ ˙ ˙ the measured states (θm . The plots are typical examples that show that a major part of the driveline dynamics in the frequency range up to 6 Hz is captured with a linear mass-spring model with the drive shafts as the main ﬂexibility.9.4) k= kp i2 + kd f .2 Inﬂuence from the Propeller Shaft The model equations (2. The choices made in the modeling are now justiﬁed. Therefore.3) If the size of the three inertias are compared.38) describes Model 1 extended with the propeller shaft with stiﬀness kp and damping cp . The three inertias in the model are J1 J2 J3 = = = Jm + Jt /i2 t Jf Jw + 2 mrw (3. θw ). The data shown are from Trial 1.36) to (2. 3. Similar results are obtained from the other trials. also the transmission speed is plotted together with the model output engine speed scaled with the conversion ratio in the transmission (it ). by ﬁtting the models to measured data. where the driveline oscillations are well excited. The total stiﬀness of two undamped springs in series is kp i2 kd f (3. • The true angle diﬀerence (x1 = θm /it if − θw ) is unknown. the inﬂuence from the drive shaft is investigated by estimating the parameters and initial conditions of Model 1.24 Chapter 3 Field Trials and Modeling 3. In Figure 3. but the value estimated by the model has physically reasonable values.4. • The model output transmission speed (x2 /it ) ﬁts the measured transmission speed data well. the results from Trial 1 are shown.3).4. the load and the states describing the torsion of the ﬂexibilities are estimated by the models.

together with the measured data in solid.3.9 Model 1 estimated on data from Trial 1.5 ˙ x2 = θm 300 250 rad/s 200 150 100 ˙ x2 /it = θt 15 rad/s 10 5 ˙ x3 = θw 5 rad/s 4 3 2 1 14 16 18 20 22 Time. . The plots are typical examples of that a major part of the dynamics is captured by a linear model with drive shaft ﬂexibility.4 Models 25 x1 = θm /it if − θw 0. [s] 24 26 28 30 Figure 3. x3 ) in dashed lines. The top ﬁgure shows the drive shaft angle diﬀerence.5 rad 0 −0. and the bottom ﬁgures show the model outputs (x2 .

fm ft fw =1 1 1 + αs =1 = (3. and the unknown ﬁlter constant α can be estimated such that the model output (ym . The result of this is seen in Figure 3. and transmission speed.26 Chapter 3 Field Trials and Modeling whereas the total damping of two dampers in series is c= cp i2 cd f cp i2 + cd f (3. or sensor dynamics. there is good agreement between model and experiments for ˙ ˙ u = Mm − Mf r:m . . As mentioned before. and fw . with the model output from Model 1 with and without sensor ﬁltering. The conclusion is that the main part of the deviation between engine speed and transmission speed is due to sensor dynamics.12. 3. if . and yw .4. but there is a slight deviation between measured and estimated transmission speed. the bandwidth of the measured transmission speed is lower than the measured engine and wheel speeds. ft . The following sensor dynamics are assumed. Two natural candidates are additional mass-spring dynamics in the driveline. with the major diﬀerence explained by the sensor dynamics.10 shows the conﬁguration with Model 1 and sensor ﬁlter fm . yt . θm . Neither can backlash in the transmission explain the diﬀerence. In Figure 3.11 for Trial 1. initial condition. because then the engine and transmission speeds would be equal when the backlash is at its endpoint. This indicates that there ˙ is some additional dynamics between engine speed. The speed dependence of the transmission sensor dynamics is neglected. due to fewer cogs in the sensor.3 Deviations between Engine Speed and Transmission Speed As mentioned above. This eﬀect will increase with lower conversion ratio in the ﬁnal drive. yw ) ﬁts the measured data. after some comparison between sensor ﬁlters of diﬀerent order. yt . Figure 3. Now the parameters. and θw .5) The damping and the stiﬀness of the drive shaft in the previous section will thus typically be underestimated due to the ﬂexibility of the propeller shaft. The explanation is that there is a combined eﬀect.6) where a ﬁrst order ﬁlter with an unknown parameter α models the transmission sensor. It is assumed that the engine speed and wheel speed sensor dynamics are not inﬂuencing the data for frequencies up to 6 Hz. ˙ θt . The motivation for this is that the high stiﬀness of the clutch ﬂexibility (given from material data) can not result in a diﬀerence of a phase shift form. an enlarged plot of the transmission speed is seen. The outputs of the ﬁlters are ym . This deviation has a character of a phase shift and some smoothing (signal levels and shapes agree). The individual stiﬀness values obtained from parameter estimation are somewhat lower than the values obtained from material data. θm .

Linear Clutch (Model 2) The parameters and initial conditions of Model 2 are estimated with the sensor dynamics described above. but the ﬁgure shows that the drive shaft torsion have realistic values that agree with other experience. The true values of these torsions are not known.7. Result • If Model 1 is equipped with a ﬁrst order sensor ﬁlter for the transmission ˙ ˙ ˙ speed. the clutch angle torsion does not have realistic values.4 Inﬂuence from the Clutch So far the clutch has been assumed stiﬀ. Therefore. The model output velocities (θm . θt .13. θw ) show . However. However. 3.3. Model 2 and 3 on the other hand estimate a clutch angle diﬀerence. and the rest of the parameters are estimated. The model output ﬁts the data except for a number of time intervals where there are deviations between model and measured data (see Figure 3.4 Models 27 ym fm yt ft /it fw yw θm k Mm − Mf r:m c Jm + Jt /i2 + Jf /i2 i2 t t f θw rw m (cr1 + gsin(α)) 2 Jw + mrw Figure 3. The resulting clutch angle diﬀerence (x1 = θm − θt it ) and the drive shaft angle diﬀerence (x2 = θt /if − θw ) are seen in Figure 3. θt . which can be seen when comparing with the ˙ ˙ ˙ static nonlinearity in Figure 2.4. θw ) are estimated by the model. the value of the stiﬀness given by Scania is used and ﬁxed. these deviations will in the following be related to nonlinearities at low clutch torques. all three velocities (θm .10 Model 1 with sensor dynamics. A problem when estimating the parameters of Model 2 is that the bandwidth of 6 Hz in the data is not enough to estimate the stiﬀness kc in the clutch.12). and the drawback with the models considered so far is that they are unable to estimate the angle diﬀerence over the clutch that actually exists.

5 ˙ ym = x2 = θm 300 250 rad/s 200 150 100 ˙ yt = x2 /it (1 + αs) = θt /(1 + αs) 15 rad/s 10 5 ˙ yw = x3 = θw 5 rad/s 4 3 2 1 14 16 18 20 22 Time. The top ﬁgure shows the angle diﬀerence over the drive shaft. The main part of the deviation between engine speed and transmission speed is due to sensor dynamics. and the bottom ﬁgures show the model outputs (ym . .28 Chapter 3 Field Trials and Modeling x1 = θm /it if − θw 0.5 rad 0 −0.12. yw ) in dashed. together with the measured data in solid.11 Model as in Figure 3. See also Figure 3. [s] 24 26 28 30 Figure 3.9 but with sensor dynamics included. yt .

04 0.5 14 16 18 20 22 Time. on data from Trial 1. [s] 16 17 18 Figure 3. The parameters are estimated on data from Trial 1.3. output from Model 1 without sensor ﬁltering (dashed).02 14 16 18 20 22 24 26 28 30 x1 = θm − θt it 0.5 x2 = θt /if − θw rad 0 −0.11.01 −0.03 0.13 Clutch angle diﬀerence (top ﬁgure) and drive shaft angle diﬀerence (bottom ﬁgure) resulting from parameter estimation of Model 2 with sensor ﬁltering. [s] 24 26 28 30 Figure 3. 0.4 Models 29 280 260 240 220 200 180 160 140 120 100 rad/s 11 12 13 14 15 Time.01 0 −0. and output from Model 1 with sensor ﬁltering (dash-dotted).12 Enlargement of part of Figure 3.02 rad 0. Measured transmission speed (solid). but the plots show that the drive shaft angle has realistic values. The true values of these torsions are not known. .

5 14 16 18 20 22 Time. θt . In Figure 3. the ﬁgure shows that both angles have realistic values that agree with other experience. The true values of these torsions are not known as mentioned before. displayed in Figure 3. The interpretation of this is that the clutch model does not add information for frequencies under 6 Hz.30 Chapter 3 x1 = θm /it − θt θc1 rad 0 Field Trials and Modeling −θc1 14 16 18 20 22 24 26 28 30 0. Result • The model including a linear clutch does not improve the data ﬁt. but the plots show that they have realistic values. on data from Trial 1.11. the clutch static nonlinearity is ﬁxed with known physical values and the rest of the parameters are estimated. θw ) show no improvement compared to those generated by Model 1 with sensor dynamics.14. Nonlinear Clutch (Model 3) When estimating the parameters of Model 3.5 x2 = θt /if − θw rad 0 −0. The true values of these torsions are not known.2) are seen in Figure 3. ˙ ˙ ˙ The model output velocities (θm . The question is if . no improvement compared to those generated by Model 1 with sensor dynamics. The resulting clutch angle diﬀerence (x1 = θm − θt it ) and drive shaft angle diﬀerence (x2 = θt /if − θw ) after minimizing (3. [s] 24 26 28 30 Figure 3. displayed in Figure 3.12 it was seen that the model with the sensor ﬁltering ﬁtted the signal except for a number of time intervals with deviations. However. except for the sensor ﬁlter which is the same as in the previous model estimations.14 Clutch angle diﬀerence (top ﬁgure) and drive shaft angle diﬀerence (bottom ﬁgure) resulting from parameter estimation of Model 3 with sensor ﬁltering.11.

When estimating the parameters of the models investigated. is estimated. The assumption about sensor dynamics in the transmission speed inﬂuencing the experiments. • The model is able to estimate a clutch angle with realistic values. The reason is unmodeled dynamics at low clutch torques (Bj¨rnberg.4. Result • The model including the nonlinear clutch does not improve the data ﬁt for frequencies up to 6 Hz. The load can be recalculated to estimate road slope.5 Model Validity As mentioned before.3. o and Nielsen 1996). there is a problem with identifying the viscous friction components b. the unknown load l. the estimation of the states describing the torsion of the clutch and the drive shaft shows realistic values. In the parameter estimation. agrees well with the fact that the engine speed sensor and the wheel speed sensor have considerably higher bandwidth (more cogs) than the transmission speed sensor. and the calculated values agree well with the known values of the road slopes at Scania. the model deviates from the data. The parameters are estimated on the estimation data. Furthermore.7). The results are then evaluated on the validation data. Pettersson. 3. and these are the results shown in this chapter. This gives further support to model structure and parameters. which vary between the trials. The sensitivity in the model to variations in the friction parameters is low.4 Models 31 this is a result of some nonlinearity. Figure 3.15 that the deviation between model and experiments occurs when the clutch angle passes the area with the low stiﬀness in the static nonlinearity (see Figure 2. and the same model ﬁt can be obtained for a range of frictions parameters. . It is clear from Figure 3. • The estimated clutch angle shows that when the clutch passes the area with low stiﬀness in the nonlinearity.15 shows the transmission speed plotted together with the model output and the clutch angle torsion. the data sets are divided into two parts.

15 Clutch angle diﬀerence (top ﬁgure) and measured and estimated transmission speed (bottom ﬁgure) from estimation of Model 3 with sensor dynamics on data from Trial 1. [s] 16 17 18 Figure 3. .32 Chapter 3 Field Trials and Modeling x1 = θm /it − θt θc1 −θc1 rad 11 12 13 14 15 16 17 18 18 yt = x4 /(1 + αs) 16 14 ↑ ↑ ↑ 12 rad 10 8 ↑ 6 ↑ 4 11 12 13 14 15 Time. The result is that the miss ﬁt occurs when the clutch angle passes the area with the low stiﬀness (|θ| < θc1 ) in the static nonlinearity.

the mass-spring models with some sensor dynamics (Model 1 and Model 2) give good agreement with experiments. They are thus suitable for control design. The major deviations left are captured by the nonlinear eﬀects in Model 3 which makes this model suitable for verifying simulation studies in control design. In order for the model to ﬁt the data from all three measured velocities. drive shaft and clutch torsion. .5 Summary 33 3.5 Summary Parameter estimation of the models derived in Chapter 2 shows that a model with one torsional ﬂexibility and two inertias is able to ﬁt the measured engine speed and wheel speed.3. within the frequency regime interesting for control design. from a user perspective. Furthermore. The result is a series of models that describe the driveline in increasing detail by. the known road slopes are well estimated. in accordance with properties of the sensory system. Further supporting facts of the models are that they give values to the nonmeasured variables. The result. adding the eﬀect that seems to be the major cause for the deviation still left. Parameter estimation of a model with a nonlinear clutch explains that the diﬀerence between the measured data and the model occurs when the clutch transfers zero torque. that agree with experience from other sources. It is shown that all three velocities are ﬁtted. in each extension. is that. a ﬁrst order sensor ﬁlter is added to the model. By considering the diﬀerence between the measured transmission speed and wheel speed it is reasonably to deduce that the main ﬂexibility is the drive shafts.

34 Chapter 3 Field Trials and Modeling .

The formulation of performance output and controller structures used in the rest of the thesis are given in Section 4. The diﬀerence can be formulated in control theoretic terms e.g. forming the main contribution of this chapter. However. If the driveline were rigid. these diﬀerences are illustrated for driveline models. the choice would not matter. since the sensor outputs would diﬀer only by a scaling factor. it will be demonstrated that the presence of torsional ﬂexibilities implies that sensor choice gives diﬀerent control problems.3. e. transmission speed sensor. or wheel speed sensor. with respect to sensor location (Spong and Vidyasagar 1989). engine speed sensor. even though its relevance for control characteristics.g.2. an investigation about how these properties transfers when using more complicated controller structures like LQG/LTR is made. there are signiﬁcant torsional resonances in a driveline. by saying that the poles are the same. Ljung 1988). The driveline model equations in Chapter 2 are written in state-space form in Section 4. but the zeros diﬀer both in number and values. Besides formulating the control problem in this chapter.1. The issue of sensor location seems to be a little studied topic (Kubrusly and Malebranche 1985. Control of resonant systems with simple controllers is known to have structural properties e.g. In Section 4. there is one architectural issue that will be given special attention. In Section 4. Active control of these resonances is the topic of the rest of this thesis.4. Chapters 5 and 6 treats two diﬀerent problems. 35 . There are diﬀerent possible choices in driveline control between using diﬀerent sensor locations. This part is based on the material in Pettersson and Nielsen (1995). as mentioned before.4 Architectural Issues for Driveline Control As seen in the previous chapters.

6) . angle diﬀerences. Possible physical state variables in the models of Chapter 2 are torques.36 Chapter 4 Architectural Issues for Driveline Control 4. The state space representation is x = ˙ Ax + Bu + H l (4. c/iJ1 A = −k/iJ1 −(b1 + c/i2 )/J1 k/J2 c/iJ2 −(c + b2 )/J2 0 0 0 B = 1/J1 . B. In this work.1. x.1) where A. the angle diﬀerence of each torsional ﬂexibility and the angle velocity of each inertia is used as states. H.3) (4.2) ˙ = θw = rw m (cr1 + gsin(α)) 0 1/i −1 . and l are deﬁned next for the linear Models 1 and 2 in Chapter 2. the diﬀerence between the driving torque and the friction torque. H = 0 −1/J2 (4.5) 2 mcr2 rw bt /i2 t bw + + bf /i2 i2 t f 3 cw Aρrw + according to Deﬁnition 2. State-space formulation of Model 1: x1 x2 x3 l giving = = θm /it if − θw ˙ θm (4. and angle velocity of any inertia.1 State-Space Formulation The input to the open-loop driveline system is u = Mm − Mf r:m . State-space formulation of Model 2: x1 x2 x3 x4 x5 θm − θt it θt /if − θw ˙ = θm ˙ = θt ˙ = θw = = (4.4) where i = it if J1 J2 b1 b2 = = = = Jm + Jt /i2 + Jf /i2 i2 t t f 2 Jw + mrw (4.

and l deﬁned in (4. B. In the case of a vehicular driveline the normal sensor alternative is an inductive sensor mounted on a cog wheel measuring the angle. (4. A. H= 0 0 0 0 −1/J2 0 −1 0 cd /if J2 −(b3 + cd )/J3 (4.2) to (4. This disturbance is considered because the ﬁring pulses in the driving torque can be seen as an additive disturbance acting on the input. A second state disturbance n is a disturbance acting on the input of the system.1. and are seldom used in a production vehicular applications. H.1.4. The output of the process is deﬁned as a combination of the states given by the matrix C in y = Cx + e (4. The state-space description is x = Ax + Bu + Bn + H l + Hv ˙ with x.8) 4.10) .9) 4. = = = = = 0 0 1/J1 0 0 −it 1/if cc it /J1 −(cc i2 + b2 + cd /i2 )/J2 t f cd /if J3 .1 Disturbance Description The disturbance l can be seen as a slow-varying part resulting from the rolling resistance and the road slope plus and additive disturbance v.1 State-Space Formulation 37 A is given by the matrix 0 0 1 0 0 0 −kc /J1 0 −cc /J1 kc it /J2 −kd /if J2 cc it /J2 0 kd /J3 0 and B= where J1 J2 J2 b2 b3 according to Deﬁnition 2.8). Sensors that measure torque are expensive.2 Measurement Description For controller synthesis it is of fundamental interest which physical variables of the process that can be measured.2. as mentioned before.7) Jm Jt + Jf /i2 f 2 Jw + mrw bt + bf /i2 f 3 bw + cw Aρrw + cr2 rw (4.

14) where. the following are deﬁned (corresponding to a sensor on θm and θw for Model 1). The control signal u is a linear function of the states (if they are all measured) or else the state estimates. In this work. where e is a measurement disturbance. . r represents the commanded signal with the gain l0 .1 Plant and controllers Fr and Fy .38 Chapter 4 Architectural Issues for Driveline Control D l+v H r Fr (s) u B x ˙ 1/s x M z A C Fy (s) e y Figure 4.13) The control problem can be seen as in Figure 4. The unknown controllers Fr and Fy are to be designed such that that the performance output (4.13) meets its requirements (deﬁned later).11) (4. Cm Cw = = (0 1 0) (0 0 1) (4. ˙ ˙ Especially.12) 4. u ˆ = l0 r − Kc x (4. This combination of the states is described by the matrices M and D in the following way z = M x + Du (4. which are obtained from a Kalman ﬁlter. In this thesis controllers will be designed as state-feedback controllers extensively except for a few simple examples.2 Controller Formulation The performance output z is the combination of states that has requirements to behave in a certain way. and therefore.1. the output of the process is one/some of the state variables deﬁning an angle velocity. only angle velocity sensors are considered. and Kc is the statefeedback matrix.

these return ratios are scalar and thus equal. This is because there are unequal number of sensors and control signals.1 gives Fy (s) Fr (s) = Kc (sI − A + BKc + Kf C)−1 Kf = l0 1 − Kc (sI − A + BKc + Kf C)−1 B (4.23) When only one sensor is used. deﬁned as ˙ z = θw = Cw x (4. which characterizes the closed-loop behavior at the plant output and input respectively GFy = C(sI − A)−1 BFy Fy G = Fy C(sI − A)−1 B (4.3 Some Feedback Properties The performance output when controlling the driveline to a certain speed is the velocity of the wheel.21) Two return ratios results. In Figure 4. when extending to more sensors.20) (4. two diﬀerent control problems results.3 Some Feedback Properties 39 Identifying the matrices Fr (s) and Fy (s) in Figure 4. 4.22) (4. This is however not considered in this work. The open-loop transfer functions from u to engine speed Gum has three poles and . it is important with the type of investigation about the structural properties made in this chapter.17) (4.4.19) N (4.2 a root locus with respect to a ˙ ˙ P-controller gain is seen for two gears using velocity sensor θm and θw respectively.24) ˙ ˙ When studying the closed-loop control problem with a sensor on θm or θw . v. and e to the control signal u are given by Gru Gvu Geu = = I − Kc (sI − A + BKc )−1 B l0 r −1 −1 (4.15) The closed-loop transfer functions from r. LQG/LTR is not directly applicable to driveline control with more than one sensor as input to the observer.18) Kc (sI − A + Kf C) N − Kc (sI − A + BKc ) N −Kc (sI − A + BKc )−1 BKc (sI − A + Kf C)−1 N = Kc (sI − A + BKc )−1 BKc − I (sI − A + Kf C)−1 Kf The transfer functions to the performance output z are given by Grz Gvz Gez = = = (M (sI − A)−1 B + D)Gru M (sI − A + BKc ) BKc (sI − A + Kf C) +M (sI − A + BKc )−1 N + DGwu (M (sI − A)−1 B + D)Gvu −1 −1 (4.16) (4. Therefore.

and θw (right ﬁgures).2. but is stable for ˙m gains.2 Root locus with respect to a P controller gain. for gear 1 (top ﬁgures) ˙ ˙ and gear 8 (bottom ﬁgures) with sensor on θm (left ﬁgures). as can be seen in Figure 4.40 Chapter 4 Architectural Issues for Driveline Control ˙ Gear 1 and θm feedback ˙ Gear 1 and θw feedback 6 4 2 Imag 0 −2 −4 −6 −10 −5 0 Imag 6 4 2 0 −2 −4 −6 −10 −5 0 ˙ Gear 8 and θm feedback ˙ Gear 8 and θw feedback 15 10 5 Imag 0 −5 −10 −15 −20 −10 Real 0 Imag 15 10 5 0 −5 −10 −15 −20 −10 Real 0 Figure 4. but the resonance in the drive shaft ˙ makes the wheel speed oscillate. When the engine speed sensor is used. two poles must go to inﬁnity which makes the system unstable. Guw on the other hand has one zero. The system goes unstable when the θw gain is increased. and the gain is increased. Hence. the engine speed behaves well when the gain is increased.3 demonstrates the problem with resonances that occurs with increasing gain for the two cases of feedback. Figure 4. while the rings represents the open-loop ˙ zeros. When using θw feedback it is diﬃcult to increase . and the closed-loop system is stable ˙ for all gains. When the velocity ˙ sensor θm is used. all θ two zeros. and the same poles. the relative degree is one. (Remember that θw is the performance output and thus desirable to use. The cross represent the open-loop poles.) The same eﬀect can be seen in the step response when the P controller is used. the relative degree of Gum is one and Guw has a relative ˙ degree of two. This means that when θw feedback is used.

giving a resonant system. However.3 only depend on the relative degree. since the poles moves closer to the imaginary axis. θm feedback Amplitude 0. The reason for using .4. increased rise time results in instability.4 Driveline Control with LQG/LTR ˙ Wheel speed.5 2 2. θm feedback Amplitude 2 1. as seen in the previous section. but an oscillating wheel speed. and are thus parameter independent. increased gain results in a well behaved engine With θ speed.2 and 4.5 Time.4 Driveline Control with LQG/LTR Diﬀerent sensor locations result in diﬀerent control problems with diﬀerent inherent characteristics. the bandwidth. [s] 3 3. The topic of this section is to show how this inﬂuences control design when using LQG/LTR. θw feedback 41 1.5 5 Figure 4.5 1 0. ˙m feedback (bottom ﬁgures).5 0 0 ˙ Wheel speed.5 4 4.5 0 80 60 40 20 0 ˙ Engine speed. and therefore a more detailed analysis is performed in the following sections.3 Step responses using a P controller with diﬀerent gains on Model 1 ˙ with gear 1.5 Amplitude 1 0. this observation may depend on feedback structure. With θw feedback (top ﬁgure). The characteristic results in Figures 4. 4.5 1 1.

These have the same number of poles but diﬀerent number of zeros as mentioned before. in this principle study. Using (4.16) to (4. Sw + Tw = 1 ˙ Feedback from θm ˙ The following transfer functions results if the sensor measures θm Grz Gnu = = Guw Fy Fr 1 + Gum Fy 1 1 + Gum Fy (4.30) ˙ These transfer functions corresponds to a conﬁguration where θm is the output (i. where (4. The following transfer functions are deﬁned Sm = 1 . ˙m ).42 Chapter 4 Architectural Issues for Driveline Control LQG/LTR. and it is also an easy method for obtaining robust controllers.21) are used together with the matrix inversion lemma Grz Gnu = = Guw Fy Fr = T w Fr 1 + Guw Fy 1 = = Sw 1 + Guw Fy (4.31) . as usual. ˙ It is desirable to have sensitivity functions that corresponds to y = θm and ˙ z = θw .28) it is natural to deﬁne T m by y=z=θ Tm = Guw Fy Guw = Tm 1 + Gum Fy Gum (4. 4. Then among others the following transfer functions results. open-loop transfer functions Gum and Guw results. ˙ Feedback from θw ˙ A natural feedback conﬁguration is to use the performance output. the sensitivity function and the complementary sensitivity function.e. Also.26) where n is the input disturbance.29) (4. as usual.27) The diﬀerence between the two feedback conﬁgurations is that the return diﬀerence is 1 + Guw Fy or 1 + Gum Fy . θw . Two diﬀerent closed-loop systems results depending on which sensor that is used.1 Transfer Functions ˙ ˙ When comparing the control problem with using θm or θw as sensors.28) (4. is that it oﬀers a control design method resulting in a controller and observer of the same order as the plant model. 1 + Gum Fy Tm = Gum Fy 1 + Gum Fy (4.4.25) (4. The transfer functions Sw and Tw are.

When the return ratio behaves like a ﬁrst order system.27) is Sm + T m Gum =1 Guw (4. This conﬂicts with the design goal of having . one pole has to be moved to inﬁnity. Limitations on Performance The relations (4. Instead. without damping and load. It could be expected that a higher control ˙ signal is needed for θw feedback in order to move the poles towards inﬁnity. Structural Properties of Sensor Location The controller (4. ˙ ˙ being combinations of two sensor locations. as in (4.31). a good alternative is to have relative degree one in GFy . Deﬁnition 4. T m = Guw /Gum for these frequency domains. and two models with the same structure. There are (2n − 1) poles. have the same denominators.32) will be the fundamental relations for discussing design considerations. The impact of the ratio Guw /Gum will be analyzed in the following sections. Design without pre ﬁlter (Fr = 1) is considered.4.27).2 Design Example with a Simple Mass-Spring Model Linear Quadratic Design with Loop Transfer Recovery will be treated in four cases. The relative degree of Gum Fy is thus 2. but with diﬀerent parameters. and the relative degree of Guw Fy is 2n.4 Driveline Control with LQG/LTR 43 The functions Sm and T m describe the design problem when feedback from θm is used. and Gw/m = Guw /Gum is the dynamic output ratio. The section covers a general plant with n inertias connected by k − 1 torsional ﬂexibilities.31) is the modiﬁed complementary sensitivity function.27) and (4. and when ˙ ˙ using θw there are no zeros. When using Gum Fy . 2n − 1 poles have to be moved to inﬁnity. When combining (4. When considering design. and when using Guw Fy . The number of zeros depends on which sensor that is used. and the location of the poles are the same for the diﬀerent sensor locations.15) has a relative degree of one. then T m will not be equal to one. Thus. implying inﬁnite gain margin and high phase margin.1 T m in (4. the corresponding relation to (4.32) If Sm is made zero for some frequencies in (4. the transfer functions Gum and Guw . When using feedback from θm there are (2n − 2) zeros.4. and with unit conversion ratio. and a relative degree of 1 and (2n − 1) respectively. also the closed-loop transfer function behaves like one.30) and (4. in order for the ratio to resemble a ﬁrst order system at high frequencies. 4.32). θm or θw .

a high frequency gain roll-oﬀ rate of 20(2n − 2) dB/decade. Gw/m (0) = 1 (if the conversion ratio is equal to one).0974.220.44 10 Chapter 4 Architectural Issues for Driveline Control 0 −10 Gain (dB) −20 −30 −40 −50 −60 −1 10 10 0 10 Frequency (rads^−1) 1 10 2 Figure 4. k = 2.660. there is a trade-oﬀ when using θw feedback. Gw/m has a relative degree of 2n − 2 and thus.4. l = 0. b) J1 = 0. Furthermore.0280. The shape of Gw/m can be seen in Figure 4. l = 0. . where qm is the roll-oﬀ ˙ rate of Gum Fy . Hence.0244. b2 = 0. Tw will have the same roll-oﬀ rate as Guw Fy . a steep roll-oﬀ rate for the closed-loop system in order to attenuate measurement ˙ noise.0974. b1 = 0. Example 4. k = 5. Structure of Gw/m We have in the previous simple examples seen that the relative degree and the zeros are important. The dynamic output ratio contains exactly this information and nothing else. J2 = 0. c = 0. the dynamic output ratio gives the closed-loop transfer function T m a high frequency gain roll-oﬀ rate of qm + 20(2n − 2) dB/decade.80. b2 = 0. since the relative degree of Gum When using θ is one. Parametric properties of Gw/m Typical parametric properties of Gw/m can be seen in the following example. For low frequencies the dynamic output ratio has gain equal to one. Hence. J2 = 0. ˙m feedback.1 Two diﬀerent plants are considered: a) J1 = 0.70. there is no trade-oﬀ. b1 = 1. with labels according to the state-space formulation in Section 4. When using θw feedback. c = 0.1.50.566.4 Gw/m for a) (solid) and b) (dashed).

this Kc is used for recovery. The design in Example 4 is shown in Figure 4. and R = 1. Kf = Pf C T V −1 and the closed-loop transfer functions S and T show satisfactory performance. where the bandwidth is lower in order to suppress the peak in Gw/m . Ma . are adjusted until the return ratio (4. and Cma results. W is adjusted (and thus Fy (s)) such that Sm and Tm ˙m ) show a satisfactory performance. and recover the satisfactory return ratio obtain previously. the bandwidth in T m is chosen such that the peak is suppressed. Kf . and that the desired bandwidth is obtained.6 and 4. If the Riccati equation AT Pc + Pc A − Pc BR−1 B T Pc + C T QC = 0 (4.4. Ba . Kc = ρC is obtained in the limit. and |S| ≤ 1. The state-space realization Aa . Note that the roll-oﬀ rate of Tw is 20 dB/decade.33) Pf AT + APf − Pf C T V −1 CPf + BW B T = 0 The covariances W and V . The remaining poles move towards inﬁnity (compare to Section 5. This means that there is inﬁnite gain margin. Depending on the shape of Gw/m (and thus θ for middle high frequencies. Cwa . is derived using a Riccati equation (Maciejowski 1989) (4.34) C(sI − A)−1 Kf . The parameters of the dynamic output ratio are thus important in the LQG step of the design. and to guarantee stability. ˙ Design for θw feedback.5 shows such an example. of v and e respectively.4 Driveline Control with LQG/LTR 45 LQG Designs Integral action is included by augmenting the state to attenuate step disturbances in v (Maciejowski 1989). the relative degree is one.5. In Figures 4. A simplistic LTR can be obtained by using Kc = ρC and increasing ρ. LTR The next step in the design process is to include Kc . . If there is a resonance peak in Gw/m . W is adjusted (and thus Fy (s)) such that Sw and Tw show a satisfactory performance. The Nyquist locus remains outside the unit circle centered at −1. the recovered closed-loop transfer functions. 2n − 1 poles move towards the open system zeros. When using the combined state feedback and Kalman ﬁlter. and control signal are seen. the return ratio is GFy = C(sI − A)−1 BKc (sI − A + BKc + Kf C)−1 Kf . corrections in W must be taken such that T m achieves the desired bandwidth. Note also the diﬀerence between Sw and Sm . Furthermore.35) √ is solved with Q = ρ. and a phase margin of at least 60◦ . ˙ Design for θm feedback. The Kalman-ﬁlter gain. As ρ is increased.1). Loop Transfer Recovery. Figure 4. Nyquist locus. ˙ θm feedback in b).7.

but have diﬀerent relative degree and diﬀerent zeros.e. a). There is a trade-oﬀ when choosing an appropriate ρ. The problem that the performance output signal is not the same . and T (right ﬁgures). b). together with a steep roll-oﬀ rate. There is no trade-oﬀ when choosing ρ. A low ρ gives good attenuation of measurement noise and a low control signal. The dynamic output ratio.5 Summary Control and damping of torsional oscillations in vehicular drivelines is an important problem. These functions have the same poles. and W = 50 (θm . ˙ ˙ Feedback from θw in solid curves. It is possible to achieve good recovery with reasonable stability margins and control signal.46 Chapter 4 Architectural Issues for Driveline Control a) 0 Gain (dB) Gain (dB) 0 a) −20 −20 −40 −40 −60 10 0 10 b) 2 −60 10 0 10 b) 2 0 Gain (dB) Gain (dB) 0 2 0 −20 −20 −40 −40 −60 −60 10 10 Frequency (rads^−1) 10 10 Frequency (rads^−1) 0 2 Figure 4. but in order to have good stability margins. W = 5 · 104 (θm . 4. T m is seen in right ﬁgures in dash-dot curves. W = 15 (θw . and a 20 dB/decade roll-oﬀ rate in Tw for a wider frequency range. exactly captures these diﬀerences and nothing else. Diﬀerent sensor locations give diﬀerent transfer functions. a high ρ must be chosen. ˙ Recovery for θm feedback. and feedback from θm in dashed curves. This gives an increased control signal. the relative degrees are thus dominant in determining the LTR step of the design. W = 5 · 102 (θw . Gum or Guw . b). Gw/m . a).5 Closed-loop transfer functions S (left ﬁgures). ˙ Recovery for θw feedback. The structural properties i.

and feedback from θm in dashed ˙m design ρ = 106 curves.6 Closed-loop transfer functions S (left ﬁgures). dominate in the LTR step.4. and for the θw design ρ = 104 . T m is seen in right ﬁgures in dash-dot curves. Feedback from θw in solid curves. and 1011 is used in both a) and b). Both structural and parameter dependent aspects of sensor location have been characterized. and T (right ﬁgures) ˙ ˙ after recovery. 108 . i. whereas structural properties. . as the measured output signal is handled by introducing a modiﬁed complementary sensitivity function.e.5 Summary 47 a) 0 Gain (dB) Gain (dB) 0 a) −20 −20 −40 −40 −60 10 0 10 b) 2 −60 10 0 10 b) 2 0 Gain (dB) Gain (dB) 0 2 0 −20 −20 −40 −40 −60 −60 10 10 Frequency (rads^−1) 10 10 Frequency (rads^−1) 0 2 Figure 4. sensor location. In LQG/LTR. being modiﬁed with Gw/m . parameter dependent properties dominate in the LQG step of the design. For the θ ˙ (a) and ρ = 105 (b) is used.

is also shown in the Nyquist plots.1 0 −0.2 0.48 Chapter 4 Architectural Issues for Driveline Control a) 0. Feedback from θw in solid curves.1 0 Real Axis Gain (dB) Imag Axis 50 0 10 0 a) 50 Gain (dB) Imag Axis 0 10 b) 5 0 10 10 Frequency (rads^−1) 0 5 Figure 4.1 −0. A dash-dotted circle with radius one and centered at -1.2 0.2 −50 −0. and feedback from θm in dashed curves. and 10 is used in both a) and b).1 b) 0. .2 −50 −0.7 Nyquist plot of return ratio (left ﬁgures) and Fy /(1 + Guw Fy ) (right ˙ ˙ ﬁgures).2 −0. and for the θw design For the θ 4 8 11 ρ = 10 .2 −0.1 −0.1 0 −0. 10 . ˙ ˙m design ρ = 106 (a) and ρ = 105 (b) is used.

In todays electronically controlled engines. Beaumount. going uphill or downhill) gives a stationary 49 .1 RQV Control RQV control is the traditional diesel engine control scheme steaming from the mechanical centrifugal governor. the speed control problem keeping RQV characteristics is formulated in Section 5. Pettersson and Nielsen 1995). The sections following study the problem using available computationally powerful methods like LQG/LTR. This driving character is important to maintain when extending speed control with active damping. the RQV scheme is still used for controlling the fuel amount to the engine. The RQV control gives a speciﬁc character to the driving feeling e. and a nonzero load (e.2. inﬂuence from disturbances. known as vehicle shuﬄe (Mo. Sensor location.5 Speed Controller Design and Simulations Driveline oscillations may occur in diﬀerent modes of operation. 5. Active damping in two modes will be treated in this and next chapter. Traditional RQV control is explained in Section 5. since the driver wants the engine to behave as with the mechanical governor. when going uphill and downhill. The ﬁrst problem is wheel speed oscillations following a change in accelerator pedal position. and load estimations are treated. Traditionally in diesel powered trucks. RQV control is essentially a P controller with the accelerator as reference value and a sensor measuring the engine speed. and Powell 1996.g. Thereafter.g.1. the relation between the accelerator pedal and the amount of fuel metered by the diesel pump is governed by a system called RQV control. The RQV controller has no information about the load. used to control the diesel pump (Bosch 1993).

the high frequency roll-oﬀ rate is steeper for Model 2 than for Model 1.2. the rise time and the overshoot is increasing. The RQV controller is described by ˙ u = u0 + Kp (ri − θm ) (5.1 Consider the truck modeled in Chapters 2 and 3 traveling at a speed of 2 rad/s (3. A diesel engine can only produce torque in a certain range.) When the controller gain is increased.2) where r is the reference velocity given by the driver. z = θw .1 shows the RQV control law (5. when the gain is increased. Furthermore. Model 2 adds a second resonance peak from the clutch.2) is extended such that a large control signal adds to the cost function. 5.1 Mathematical Problem Formulation A ﬁrst possible attempt for speed control is a scheme of applying the engine torque to the driveline such that the following cost function is minimized T T →∞ lim 0 (z − r)2 (5. since the plant model is linear. This chapter deals with the development of a controller based on Model 1. the engine speed behaves well. but the ﬂexibility of the driveline causes the wheel speed to oscillate. 5. Let the new desired velocity be r = 2.2) can be made arbitrarily small if there are no restrictions on the control signal u. as deﬁned in Chapter 4. Note that the transfer function from the load l to the performance output z is the same for the two models. and r is the reference velocity. (5. If a control law is to minimize the cost function. Figure 5. u0 is calculated such that the stationary level is the same for the three gains.2 shows the transfer functions from load and measurement disturbances v and e to the performance output. there is a trade-oﬀ between short rise time and little overshoot. Example 5.3 rad/s. The resonance peak in the transfer functions is increasing when the controller gain is increased. RQV control is demonstrated in the following example.1) applied to Model 1 with three gains Kp . In Figure 5. the wheel speed z is seen for Models 1 and 2.2 Problem Formulation ˙ The performance output for the speed controller is the wheel speed. but not the load since this is unknown to the RQV controller. In the plots.3.6 km/h) with gear 1 and a total load of 3000 Nm (≈ 2 % road slope). and therefore. Hence. Figure 5.50 Chapter 5 Speed Controller Design and Simulations error. then (5. . The constant u0 is a function of the speed. when the RQV controller is used. Kp is the controller gain.1) where i = it if is the conversion ratio of the driveline. Furthermore. (Otherwise there would be a gain dependent stationary error.

4 2.1 2 1. [s] 5 6 7 8 1 2 3 4 5 6 7 8 Control signal u Figure 5.5) . B. Controller gains Kp = 8.1 RQV control (5.4) = λx θw + λl l l The stationary point is obtained by solving Ax + Bu + Hl = 0 for x and u.2 Problem Formulation ˙ Engine speed θm 51 140 Speed. l) = i = δ x θw + δl l (5.2 2. dashed and dash-dotted lines respectively. l) = (5. The stationary point z = r is reached if a control signal u0 is used. and Kp = 85 are shown in solid.9 0 2000 Torque. [rad/s] 135 130 125 120 115 0 1 2 3 4 5 6 7 8 ˙ Wheel speed θw (performance output) Speed.3) l 1 0 ˙ θw ˙ ˙ (b1 i2 + b2 )/i 1/i u0 (θw .2) to (4. Kp = 25. Increased gain results in a well behaved engine speed. This torque ˙ is a function of the reference value r and the load l.1) of Model 1. [rad/s] 2. [Nm] 1500 1000 500 0 0 1 2 3 4 Time.5. H. (5.5). but an oscillating wheel speed. where A. and x is given by (4.3 2. For a given wheel speed θw and load l the driveline has the following stationary point b2 /k 1/k ˙ θw ˙ ˙ 0 x0 (θw .

for instance. • The control signal can not exceed umin = −300 Nm or umax = 2300 Nm. By using these equations. The controller that minimizes (5. Furthermore. as mentioned before. The resonance peaks increase with increasing gain. The extended cost function is given by T T →∞ lim 0 (z − r)2 + η(u − u0 (r. • The inﬂuence from load and measurement disturbances on the performance output. the following demands should be considered. the cost function can be written such that a control signal u that deviates from the stationary value u0 (r. since the load l is included and thus compensated for. it is desirable that the stationary error characteristic for the RQV controller is maintained in the speed controller.1.6).2 Closed-loop transfer functions Gvz and Gez when using the RQV control law (5.3. A stationary error comparable with that of the RQV controller can be achieved by using only a part of the load l in the criterion (5. However.6) where η is used to control the trade-oﬀ between rise time and control signal amplitude.6) has no stationary error. . and Kp = 85 (dash-dotted). Load disturbances result from. road roughness or impulses from towed trailers. Kp = 25 (dashed).52 Chapter 5 Speed Controller Design and Simulations Closed-loop transfer function Gvz −80 Gain [dB] −100 −120 −140 −2 10 10 −1 10 0 10 1 10 2 10 3 Closed-loop transfer function Gez Gain [dB] −50 −100 −150 −2 10 10 −1 10 10 Frequency [rad/s] 0 1 10 2 10 3 Figure 5. should be minimized. l) adds to the cost function. wheel speed.1) for the controller gains Kp = 8 (solid). l))2 (5. as will be demonstrated in Section 5.

u0 ).e. The problem formulation (5.6) will be treated in two steps. Model 1 is shown in solid and Model 2 is shown in dashed. First without RQV behavior i.3 Transfer functions from control signal u and load l to performance output z. using the load l.8) .6) is in this section solved with LQG technique. This is done by solving a Riccati equation.3 Speed Control with Active Damping and RQV Behavior Guz for Model 1 and 2 53 0 −50 Gain [dB] −100 −150 −200 −60 −80 Gain [dB] −100 −120 −140 −2 10 Glz for Model 1 and 2 10 −1 10 10 Frequency [rad/s] 0 1 10 2 10 3 Figure 5.6) in terms of the linearized variables.6). The linear model is ∆x = A∆x + B∆u ˙ where ∆x = ∆u x0 = = x − x0 u − u0 x0 (x30 . 5. The derived feedback law is a function of η which is chosen such that a feasible control signal is used.3 Speed Control with Active Damping and RQV Behavior The problem formulation (5. and then extending to RQV behavior. This is done by linearizing the driveline model and rewriting (5.7) is aﬃne since it includes a constant term l.1) x = Ax + Bu + Hl ˙ (5. The model (4. l) (5. A state-feedback matrix is derived that minimizes (5. The model is linearized in the neighborhood of a stationary point (x0 .9) (5.5. The modeled clutch gives a second resonance peak and a steeper roll-oﬀ rate.

l))2 (M ∆x + r1 )2 + η(∆u + r2 )2 (5.6).11) = with T →∞ lim 0 r1 r2 = = M x0 − r u0 − u0 (r. l) (5. the cost function (5.18) . they must be stable in order to solve the Riccati equation (Maciejowski 1989). Instead the following are used r1 ˙ r2 ˙ = = −αr1 −αr2 (5. l) where the stationary point (x0 . The cost function is expressed in terms of ∆x and ∆u by using (5.10) (5.9) T T →∞ lim 0 T (M (x0 + ∆x) − r)2 + η(u0 + ∆u − u0 (r.16) By using these equations. xr = (∆xT r1 r2 )T 0 T T →∞ (5. B) with models of the constants r1 and r2 .14) which with a low α indicates that r is a slow varying constant. Ar = −α 0 0 0 0 0 −α B Br = 0 .13) (5. The augmented model is given by 0 0 A 0 0 0 0 . Since these models will not be controllable.4). Note that the linear model is the same for all stationary points.10) a Riccati equation is solved.10) can be written in the form lim xT Qxr + R∆u2 + 2xT N ∆u r r (5.15) (5. This can be done by augmenting the plant model (A. The problem is to devise a feedback-control law that minimizes the cost function (5.54 Chapter 5 Speed Controller Design and Simulations u0 = u0 (x30 .12) In order to minimize (5. Then the constants r1 and r2 must be expressed in terms of state variables.17) 0 with Q = (M 1 0)T (M 1 0) + η(0 0 0 0 1)T (0 0 0 0 1) N = η(0 0 0 0 1)T R = η (5.3) and (5. Therefore the model r1 = r2 = 0 cannot be used because the ˙ ˙ poles are located on the imaginary axis. u0 ) is given by (5.

The result is seen in Figure 5.3 Speed Control with Active Damping and RQV Behavior 55 The cost function (5.25) where η = 5 · 10−8 and α = 0. With this controller the phase margin is guaranteed to be at least 60◦ and the amplitude margin is inﬁnity (Maciejowski 1989).4). The LQG controller can be changed such that a load diﬀerent from zero gives a stationary error.230x30 + 4470r + 0. and only when using a cruise controller the stationary error should be zero.19) becomes ∆u = −Kc xr = − Kc1 Kc2 Kc3 ∆x − Kc4 r1 − Kc5 r2 (5.3) and (5. The constant βl range from βl = 0 which means no compensation for the load.0001 are used.9) and (5.24) δl − Kc4 M δl + λl Kc1 Kc2 Kc3 x (5.12) the control law is written u = K0 x30 + Kl l + Kr r − with K0 Kr Kl = = = Kc1 Kc2 Kc3 Kc4 + Kc5 λx Kc1 Kc2 Kc3 δx − Kc4 M δx + λx − Kc5 λx (5.5.21) By using (5. and the stationary error is zero if the load is known.4.3. Also the overshoot is less when using LQG control. The LQG feedback law (5.19) (5.36 x (5. There is however a demand by the driver that the load should give a stationary error.25) is a function of the load. λx . as seen in Figure 5.4. 5. and therefore a stationary error will be present when the load is diﬀerent from zero. The rise time of the LQG controller is shorter than for the RQV controller.23).10) is minimized by using ∆u = −Kc ∆x with T Kc = Q−1 (Br Pc + N T ) (5.1 Extending with RQV Behavior The RQV controller has no information about the load l. The controlled driving torque makes the engine speed oscillate.125l − 7620 0.20) where Pc is the solution to the Riccati equation AT Pc + Pc Ar + R − (Pc Br + N )Q−1 (Pc Br + N )T = 0 r The control law (5.23) where δx .1 the controller gains becomes u = 0. δl .22) (5. and λl are described in (5. The driving torque is controlled such that the oscillations in the wheel speed are actively damped. This is done by using βl l instead of the complete load l in (5.0347 2. to . When this control law is applied to Example 5.

[rad/s] 140 130 120 0 1 2 3 4 5 6 7 8 ˙ Wheel speed θw (performance output) Speed.56 Chapter 5 Speed Controller Design and Simulations ˙ Engine speed θm 150 Speed. the speed controller with active damping is extended with a stationary error comparable with that of the RQV controller. x2 = θm . With active damping. resulting in a well behaved wheel speed. By adjusting βl . The compensated LQG control law becomes u = K0 x30 + Kl βl l + Kr r − Kc1 Kc2 Kc3 x (5.4 2.3 rad/s) is compared to the compensated LQG controller (5.9 0 2000 Torque.1) with Kp = 25 is seen in dashed lines.5. RQV control (5. [rad/s] 2. This is not possible if only .4 Inﬂuence from Sensor Location The LQG controller investigated in the previous section uses feedback from all ˙ ˙ states (x1 = θm /it if − θw .3 2.1 for three values of βl . and x3 = θw ).1 2 1.26) applied to Example 5.4 Model 1 controlled with the LQG control law (5.25). 5. [s] 5 6 7 8 1 2 3 4 5 6 7 8 Control signal u Figure 5. βl = 1 which means fully compensation of the load and no stationary error. the engine speed oscillates.26) In Figure 5. the RQV controller with its stationary error (remember the reference value r = 2. [Nm] 1500 1000 500 0 0 1 2 3 4 Time.2 2.

0.3 2. The LQG controller achieves the same stationary as the RQV controller by adjusting βl .27) with K0 .23) then becomes u = K0 x30 + Kr r + Kl l − Kc1 Kc2 Kc3 x ˆ (5.05 2 1. The observer gain is calculated using LTR technique. The output matrix C is either equal to Cm (4.4 Inﬂuence from Sensor Location 57 2.12) when measuring the wheel speed.2 2.1) in dashed line.5 Example 5.29) where Pf is found by solving the Riccati equation Pf AT + APf − Pf C T V −1 CPf + W = 0 (5. and the LQG controller with stationary error (5. The sensor either ˙ ˙ measures the engine speed θm or the wheel speed θw . Kr . 1. and Kl given by (5.25 Speed. or Cw (4.15 2.1 2. .35 2.95 0 1 2 3 4 Time.28) (5.5.11) when measuring the engine speed. [s] 5 6 7 8 Figure 5. one sensor is used.30) The covariance matrices W and V corresponds to v and e respectively. which is the case considered in this work.5. [rad/s] 2.1 controlled with the RQV controller (5. Then two diﬀerent observer problems results depending on which sensor location that is used. In this section an observer is used to estimate the rest of the states. The LQG feedback law (5.26) with βl = 0. The estimated states x are given by the ˆ Kalman ﬁlter ˙ ∆x ˆ Kf x = A∆ˆ + B∆u + Kf (∆y − C∆ˆ) x T −1 = Pf C V (5.24).

Hence. or Cw for wheel speed feedback. By using the matrix inversion lemma (4. This is obtained also when using the observer by increasing ρ towards inﬁnity. There is a resonance peak in Gvz when using feedback from the engine speed sensor.35) with C either as Cm for engine speed feedback.20). 5.15) as Fy (s) = Kc (sI − A + BKc + Kf C)−1 Kf (5. This is done by selecting V = 1 (5.34) becomes (Gvz )cl = Gvw 1 + Guw Fy (5.34) mean the output of the system.36) . However. When using LQG with feedback from all states. which is not present when feedback from the wheel speed sensor is used. and cl stands for closed ˙ ˙ loop.31) W = ρBB T C = Cm or Cw ρ = ρm or ρw and solving (5.0 (5. the observer dynamics will be included in the transfer functions from disturbances both to z and u.6 shows how the performance output and the control signal are aﬀected by the load disturbance v.9◦ . For Example 5.30) for Kf .33) where the aim has been to have at least 60◦ phase margin. these transfer functions are not aﬀected by the sensor location. The observer dynamics is cancelled in the transfer function from reference value to performance output and control signal.58 Chapter 5 Speed Controller Design and Simulations Loop-Transfer Recovery (LTR) is used to recover the properties achieved in the previous design step when all states are measured. i.20) is rewritten as (Gvz )cl = Gvz + Fy (Guy Gvz − Guz Gvy ) 1 + Guy Fy (5.4. ˙ For the speed controller (z = θw ). The controller Fy is given by (4. am = ∞ 1014 ⇒ ϕw = 59.e. Equation (5. The reason to this can be seen when studying the transfer function Gvz in (4. the phase margin ϕ is at least 60◦ and the amplitude margin a is inﬁnity as stated before.29) and (5.34) where Gab means the transfer function from signal a to b.1 Inﬂuence from Load Disturbances Figure 5.5◦ .32) (5. aw = 35. either θw or θm . The signal y in (5.1 the following values are used ρm ρw = = 5 · 105 ⇒ ϕm = 60.

**5.4 Inﬂuence from Sensor Location
**

Closed-loop transfer function Gvz

−80 Gain [dB] −100 −120 −140 10

−2

59

10

−1

10

0

10

1

10

2

10

3

**Closed-loop transfer function Gvu
**

0

Gain [dB]

−50

−100

−150 −2 10

10

−1

10 10 Frequency [rad/s]

0

1

10

2

10

3

Figure 5.6 Closed-loop transfer functions from load disturbance v to performance ˙ output z and control signal u. Feedback from θw is shown in solid and feedback ˙ ˙m is shown in dashed lines. With θm feedback the transfer functions have a from θ resonance peak, resulting from the open-loop zeros. when the sensor measures the wheel speed, and (Gvz )cl = Gvw + Fy (Gum Gvw − Guw Gvm ) 1 + Gum Fy (5.37)

when the sensor measures the engine speed. Hence, when using the wheel speed sensor, the controller is cancelled in the numerator, and when the engine speed sensor is used, the controller is not cancelled. The optimal return ratio in the LQG step is Kc (sI − A)−1 B (5.38)

Hence the poles from A is kept, but there are new zeros that are placed such that the relative degree of (5.38) is one, the phase margin is at least 60◦ , and the gain margin is inﬁnite. In the LTR step the return ratio is Fy Guy = Kc (sI − A − BKc − Kf C)−1 Kf C(sI − A)−1 B (5.39)

When ρ in (5.31) is increased towards inﬁnity, (5.38) equals (5.39). This means that the zeros in the open-loop system C(sI − A)−1 B are cancelled by the controller. Hence, the open-loop zeros will become poles in the controller Fy . This means that

60

Chapter 5

Speed Controller Design and Simulations

**Closed-loop transfer function Gez
**

0 Gain [dB]

−50

−100

−150 −2 10

10

−1

10

0

10

1

10

2

10

3

**Closed-loop transfer function Geu
**

100 Gain [dB]

50

0 10

−2

10

−1

10 10 Frequency [rad/s]

0

1

10

2

10

3

Figure 5.7 Closed-loop transfer functions from measurement noise e to perfor˙ mance output z and control signal u. Feedback from θw is shown in solid and ˙m is shown in dashed. The diﬀerence between the two feedback feedback from θ principles is described by the dynamic output ratio. The eﬀect increases with lower gears. the closed-loop system will have the open-loop zeros as poles when using the engine speed sensor. This means that the Gvz will have the poles −0.5187 ± 3.0753i which causes the resonance peak in Figure 5.6.

5.4.2

Inﬂuence from Measurement Disturbances

The inﬂuence from measurement disturbances e is seen in Figure 5.7. The transfer functions from measurement noise (4.21) can be rewritten with the matrix inversion lemma as Guz Fy (5.40) (Gez )cl = − 1 + Guy Fy The complementary sensitivity function is deﬁned for the two sensor alternatives as Tw = Then (Gez )cl = ˙ −Tw with θw feedback (5.42) Guw Fy Gum Fy , Tm = 1 + Guw Fy 1 + Gum Fy (5.41)

5.4 Inﬂuence from Sensor Location

61

0

−20

−40 Gain [dB] −60 −80 −100 −2 10

10

−1

10 10 Frequency [rad/s]

0

1

10

2

10

3

Figure 5.8 The dynamic output ratio Gw/m for gear 1 (solid), gear 7 (dashed), and gear 14 (dash-dotted). (Gez )cl = −Tm Guw ˙ = Tm Gw/m with θm feedback Gum (5.43)

where the dynamic output ratio Gw/m was deﬁned in Deﬁnition 4.1. For Model 1 the dynamic output ratio is Gw/m = cs + k i(J2 s2 + (c + b2 )s + k) (5.44)

where the state-space description in Chapter 4 is used. Especially for low frequencies, Gw/m (0) = 1/i = 1/it if . The dynamic output ratio can be seen in Figure 5.8 for three gears. When ρ in (5.31) is increased towards inﬁnity, (5.38) equals (5.39). Then (5.42) and (5.43) gives (5.45) (Gez )cl,m = (Gez )cl,w Gw/m ˙ ˙ where cl, m and cl, w means closed loop with feedback from θm and θw respectively. The frequency range in which the Tm = Tw is valid depends on how large ρ in (5.31) is made. Figure 5.9 shows the sensitivity functions Sw = 1 1 , Sm = 1 + Guw Fy 1 + Gum Fy (5.46)

and the complementary sensitivity functions Tw and Tm (5.41) for the two cases of feedback. It is seen that Tm = Tw is valid up to about 16 Hz. The roll-oﬀ rate at

The load is hard to model correctly since it is a function of road slope. and Gum has a relative degree of one. Therefore.48) . The augmented model is x4 = l.62 Chapter 5 Speed Controller Design and Simulations 10 0 −10 −20 Gain [dB] −30 −40 −50 −60 −70 −2 10 10 −1 10 10 Frequency [rad/s] 0 1 10 2 10 3 Figure 5. When only ˙ one velocity is measured.4. Tw has a steeper roll-oﬀ rate than Tm . 5.3 Load Estimation u = K0 x30 + Kr r + Kl ˆ − l The feedback law with unknown load is Kc1 Kc2 Kc3 x ˆ (5. The dash-dotted lines correspond to the case with all states known.47) where ˆ is the estimated load. Guw has a relative degree of two.9 Sensitivity function S and complementary sensitivity function T . lines correspond to θ higher frequencies diﬀer between the two feedback principles. Hence. the diﬀerence in Gez depending on sensor location is described by the dynamic output ratio Gw/m . This is due to that the open-loop transfer functions Guw and Gum have a diﬀerent relative degree. the model used in the l Kalman ﬁlter is augmented with a model of the load. this eﬀect increases with lower gears. Therefore. the solid lines correspond to θw feedback. and the dashed ˙m feedback. The diﬀerence in low frequency level is equal to the conversion ratio of the driveline. However it can be treated as a slow varying constant. with x4 = 0 ˙ (5. In order to estimate the load.

5.5 Simulations

63

r

Vehicle

Model 3: (2.49) to (2.51)

˙ z (θw ) ˙ ˙ y (θw or θm )

Controller

Design based on Model 1 Control law (5.27) Observer (5.28)

Figure 5.10 Simulation conﬁguration. As a step for demonstrating feasibility for real implementation, Model 3 is simulated with the controller based on Model 1. This gives ˙ ˆ ˆ x = Al xl + Bl u + Kf (y − Cl xl ) ˆ with xl ˆ Al = 0 Bl The feedback law is u = K0 x30 + Kr r − Kc1 Kc2 Kc3 −Kl xl ˆ (5.53) = B 0 x ˆ , ˆ l A 0 , 0 (5.50) (5.49)

=

0 0 , −1/J2 0 C 0

(5.51)

Cl =

(5.52)

5.5

Simulations

An important step in demonstrating feasibility for real implementation is that a controller behaves well when simulated on a more complicated vehicle model than it was designed for. Here, the control law based on the reduced driveline model is simulated with a more complete nonlinear model, derived in Chapter 2. The purpose is also to study eﬀects from diﬀerent sensor locations as discussed in Section 5.4. The simulation situation is seen in Figure 5.10. The nonlinear Model 3, given by (2.49) to (2.51), is used as vehicle model. The steady-state level for Model 3 is calculated by solving the model equations for the equilibrium point when the load and speed are known. The controller used is based on Model 1, as seen in the previous sections. The wheel speed or the engine speed is input to the observer (5.28), and the control law (5.27) with βl = 0 generates the control signal.

64

Chapter 5

Speed Controller Design and Simulations

The simulation case presented here is the same as in Example 5.1, i.e. a velocity step response. The stationary point is given by ˙ θw = 2, l = 3000 ⇒ x0 = 0.0482 119 2.00 , u0 = 109 (5.54)

˙ where (5.3) and (5.4) are used, and the desired new speed is θw = 2.3 rad/s. At steady state, the clutch transfers the torque u0 = 109 Nm. This means that the clutch angle is in the area with higher stiﬀness (θc1 < θc ≤ θc2 ) in the clutch nonlinearity, seen in Figure 2.7. This is a typical driving situation when speed control is used. However, at low clutch torques (θc < θc1 ) the clutch nonlinearity can produce limit cycle oscillations (Bj¨rnberg, Pettersson, and Nielsen 1996). This o situation occurs when the truck is traveling downhill with a load of the same size as the friction in the driveline, resulting in a low clutch torque. This is however not treated here. At t = 6 s, a load impulse disturbance is simulated. The disturbance is generated as a square pulse with 0.1 s width and 1200 Nm height. In order to simulate the nonlinear model, the diﬀerential equations (2.49) to (2.51) are scaled such that the ﬁve diﬀerential equations (one for each state) have about the same magnitude. The model is simulated using Runge Kutta (45) (Simulink 1993) with a low step size to catch the eﬀect of the nonlinearity. Figures 5.11 to 5.13 show the result of the simulation. These should be compared with the same control law applied to Model 1 in Figure 5.4. From these plots it is demonstrated that the performance does not critically depend on the

2.35 2.3 2.25 Speed, [rad/s] 2.2 2.15 2.1 2.05 2 1.95 0

1

2

3

4

5 Time, [s]

6

7

8

9

10

Figure 5.11 Wheel speed when controlling Model 3 with the LQG control law ˙ (5.27) derived from Model 1. The solid line corresponds to θw feedback and feed˙m is seen in dashed line. At t = 6 s, an impulse disturbance v acts on back from θ the load. The design still works when simulated with extra clutch dynamics.

5.5 Simulations

65

1600 1400 1200 1000 Torque, [Nm] 800 600 400 200 0 −200 −400 0

1

2

3

4

5 Time, [s]

6

7

8

9

10

Figure 5.12 Control signal corresponding to Figure 5.11. There is no diﬀerence between the two sensor alternatives in the step response at t = 1 s. However, the load impulse (at t = 6 s) generates a control signal that damps the impulse disturbance when feedback from the wheel speed sensor is used, but not with engine speed feedback.

Angle difference, [rad]

θc1

0

1

2

3

4

5 Time, [s]

6

7

8

9

10

Figure 5.13 Clutch angle diﬀerence corresponding to Figure 5.11. The inﬂuence from the clutch nonlinearity can be neglected, because the area with low stiﬀness (θc < θc1 ) is never entered.

In conclusion. An increased controller gain results in more wheel speed oscillations while the engine speed behaves well. To solve the criterion. The area with low stiﬀness in the clutch nonlinearity (θc < θc1 ) is never entered. This eﬀect increases with lower gears. there is no active damping of wheel speed oscillations. even though there are sensor choices. the use of active damping signiﬁcantly improves the behavior for both sensor cases. The design still works if the extra dynamics are added.66 Chapter 5 Speed Controller Design and Simulations simpliﬁed model structure. Simulations show that the performance of the design. a linear driveline model with drive shaft ﬂexibility. based on the simpliﬁed model. RQV control is the traditional way speed control is performed in diesel engines. The load impulse disturbance is better attenuated with feedback from the wheel speed sensor. which is a veriﬁcation of the behavior that was discussed in Section 5. A major contribution in this chapter is a formulation of a criterion for speed control with active damping of wheel speed oscillations and a stationary error giving RQV behavior. the formulation is natural. RQV control gives a certain driving character with a load dependent stationary error when going uphill or downhill. The diﬀerence between these two is described by the dynamic output ratio. Load disturbances are thus better attenuated with this feedback conﬁguration. 5. An investigation of the inﬂuence from diﬀerent sensor locations on the control design shows that when using LQG/LTR the open-loop zeros are cancelled by the controller.6 Summary Speed control with active damping and RQV behavior has been proposed in this chapter. When feedback from the wheel speed sensor is used. Two diﬀerent closed-loop transfer functions result. with a nonlinear clutch characteristics. . Further. Measurement disturbances are better attenuated when the engine speed sensor is used. and parameters estimated from experiments are used. because the open-loop transfer function has a resonant zero couple. This results in undamped load disturbances when engine speed feedback is used. no resonant open-loop poles are cancelled. With engine speed feedback this is critical. than when using the wheel speed sensor.4. resulting in vehicle shuﬄe. depending on feedback conﬁguration. works well for a more complicated model. Further evidence supporting this is seen in Figure 5. it allows eﬃcient solution. and there is a simple tuning of the amount of RQV feeling.13. It is shown that this zero couple becomes poles of the transfer functions from load disturbances to wheel speed. With RQV.

Following that. and thus seriously increasing the total time needed for a gear shift. whereafter neutral gear is engaged. the new gear is engaged. If the automatic gear shifting system is to work with a clutch and a manually shifted transmission.6 Gear-Shift Controller Design and Simulations Traditionally a gear shift is performed by disengaging the clutch.e. changing its velocity. i. The observation that the vehicle is free rolling. when obtaining gear shift conditions shows that the desired control goal is not a stationary point. This is because the system has to wait until satisfactory gear shift conditions are reached. since there is no torque. neutral gear is engaged when the transmission transfers zero torque. the vehicle is free rolling which may be serious with heavy loads and large road slopes. In todays traﬃc it is desired to have an automatic gear shifting system. which has to be handled. In this case the engine is controlled such that the torque in the transmission is zero. 67 . • The gear shift is performed with a microprocessor controlling the clutch and the shift event. and engaging the clutch again. where the complete shift action is controlled by a microprocessor. • The gear shift is performed without using the clutch (Orehall 1995). The engine speed is then controlled to the propeller shaft speed (scaled with the new conversion ratio). It is clear that driveline oscillations is an important performance limiting factor if they are not damped out. shifting to a new gear. one of the following strategies can be taken. One reason this is not acceptable is that. When using the second approach. engaging neutral gear. and then the speed controller controls the driveline to the speed demanded by the driver.

1.68 Chapter 6 Gear-Shift Controller Design and Simulations This chapter is devoted to study a new idea that the transmission torque can be estimated and controlled to zero while having active damping. A cog wheel is mounted at the end of the input shaft which is connected to a cog wheel mounted on the output shaft.1. The inﬂuence from load and measurement disturbances should be minimized. the gear-shift controller should still drive the transmission torque to zero.2 Transmission Torque The performance output z for the gear-shift controller is the torque transmitted between the cog wheels in the transmission.4. The conversion ratio between these are it . A model of the transmission is developed and the torque transmitted in the transmission is modeled as a function of the states and the control signal in Section 6. Inﬂuence from sensor location and simulations are presented in the sections following. as mentioned in Chapter 2.5. θm = θ c . Some ﬁrst primitive attempts are then discussed in Section 6. 6.3. The output shaft is also connected to bearings with the viscous friction component bt2 . and its treatment in Section 6.1 Problem Formulation The gear-shift controller is the controller that drives the transmission torque to zero.2.1) (6.12) Mc = Mt .1 Transmission Torque for Model 1 ¨ Jm θm = Mm − Mf r:m − Mc (6. A more detailed study of the transmission is depicted in Figure 6. the formulation of the gear-shift control criterion in Section 6. Here.1) together with (2.2) 6.3) (6. in light of the simplistic attempts in Section 6. The control signal is restricted to be in the interval between umin = −300 Nm and umax = 2300 Nm. while damping oscillations. The problem formulation is further discussed in Section 6. the input shaft is connected to bearings with a viscous friction component bt1 .2. The time it takes for a gear shift should be possible to optimize. If a gear shift is commanded when the driveline is oscillating.4) By using (2. A key result in this chapter is.3. 6. Two equations describe the inputs and outputs of the transmission ¨ Jt1 θc ¨ Jt2 θt = = ˙ Mt − bt1 θc − z ˙ it z − bt2 θt − Mp (6.

5).7) ˙ ˙ −k(θm /it if − θw )/it if −c(θm /it if − θw )/it if From this it is possible to express the performance output as a function of the control signal u and the state variables x.1 The performance output for Model 1 is z = M x + Du with .25) ¨ (Jm + Jt /i2 + Jf /i2 i2 )θm t t f = ˙ Mm − Mf r:m − (bt /i2 + bf /i2 i2 )θm t t f −k(θm /it if − θw )/it if ˙ ˙ −c(θm /it if − θw )/it if (6.6.1) is expressed in terms of engine speed ¨ ˙ (Jm + Jt1 )θm = Mm − Mf r:m − bt1 θm − z (6. The cogwheels are connected to the input and output shaft respectively. equation (6.6) which together with u = Mm − Mf r:m gives ˙ u − bt1 θm − z = Jm + Jt1 Jm + Jt /i2 + Jf /i2 i2 t t f ˙ Mm − Mf r:m − (bt /i2 + bf /i2 i2 )θm t t f (6.2 Transmission Torque 69 Transmission Jt1 bt1 θc Mt θt Jt2 bt2 Mp Input shaft Output shaft Figure 6.5) ¨ To describe the performance output in terms of state variables. θm (which is not a state variable) is replaced with (2. Deﬁnition 6. according to the state-space description (4.1 Transmission with two cogwheels with conversion ratio it .2) to (4.

the equation for the transmission is ¨ ˙ (Jt1 i2 + Jt2 )θm = i2 Mc − it Mp − (bt1 i2 + bt2 )θm t t t (6. clutch.1) and (6. Assumption 6.16) .15) (6.4) is used.11) (6. corresponding to neutral gear.12) gives Jt1 bt1 = = Jt 1 + i2 t bt 1 + i2 t (6. Then a model including the engine. and the input shaft of the transmission results.8) D = 1− Jm + Jt1 J1 where the labels from (4.12) For the rest of this chapter the following assumption about the parameters in the transmission is used.11) and (6. If z is eliminated in (6.5) are used.2) and (6. in which the performance output z is equal to zero. The other parameters are estimated in Chapter 3.9) By comparing this with the equation describing the transmission in Chapter 2.15) ¨ ˙ (6. the estimated combinations of parameters from Model 1 are J1 b1 = = Jm + Jt /i2 + Jf /i2 i2 t t f bt /i2 t + bf /i2 i2 t f (6. One way of estimating these unknowns would be to decouple Model 1 into two models.8) are Jm + Jt1 and bt1 . Trials with neutral gear would then give a possibility to estimate the unknowns. (2. The unknown parameters in (6.14) In Chapter 3.70 Chapter 6 Gear-Shift Controller Design and Simulations MT = (Jm +Jt1 )k J1 i Jm +Jt1 (b1 + c/i2 ) J1 +J − (JmJ1 it1 )c − bt1 (6.13) (6. In the derivation of Model 1 in Chapter 2 the performance output z is eliminated.1 Jt1 = Jt2 and bt1 = bt2 .10) Jt θm = i2 Mc − bt θm − it Mp t the following equations relating the parameters are obtained Jt bt = = i2 Jt1 + Jt2 t i2 bt1 + bt2 t (6. Then (6.

2 Transmission Torque 71 From (6.18) For low gears (it large).25) MT = .2.1) by equation (2.2 Transmission Torque for Model 2 The performance output expressed for Model 2 is given by replacing Mt in (6.39) ˙ ˙ Mc = Mt = kc (θm − θt it ) + cc (θm − θt it ) (6.45) ¨ (Jt + Jf /i2 )θt f = ˙ ˙ it kc (θm − θt it ) + cc (θm − θt it ) ˙ −(bt + bf /i2 )θt − f 1 if (6.20) 6. and since Jf and bf are considerably less than J1 and b1 .19) (6.24) leading to Deﬁnition 6.17) A combination of (6. the following assumptions are used Jm + Jt1 bt1 i2 t 1 + i2 t i2 t ≈ b1 1 + i2 t ≈ J1 (6.22) ˙ ˙ kd (θt /if − θw ) + cd (θt /if − θw ) (6.14) and (6.23) (6.6.2 The performance output for Model 2 is z = M x with J i2 kc (1 − t12 t ) J Jt1 it kd J2 if J i2 cc (1 − t12 t ) J Jt1 i2 2 t (it cc + b2 + cd /i2 ) − cc it f J2 − Jt12iitfcd J − bt1 it (6.16) gives bt1 bt1 = (6.15) Jm + Jt1 can be derived Jm + Jt1 = = Jt i2 t 2 2 2 = Jm + 1 + i2 (J1 − Jm − Jf /it if ) 1 + it t 1 i2 1 t + J1 − Jf 2 Jm 1 + i2 1 + i2 if (1 + i2 ) t t t Jm + bt i2 t 2 2 2 = 1 + i2 (b1 − bf /it if ) 1 + it t (6.21) Then the performance output is ˙ ˙ ˙ ¨ z = kc (θm − θt it ) + cc (θm − θt it ) − bt1 it θt − Jt1 it θt This is expressed in terms of state variables by using (2.13) and (6.

30) (6.33) The parameters not estimated in the deﬁnition above are approximated in the same way as for the performance output for Model 2. θw ) Jt1 i2 t J2 Jt1 it kd J2 if J i2 cc (1 − t12 t ) J Jt1 i2 2 2 t J2 (it cc + b2 + cd /if ) − cc it Jt1 it cd − J2 if 1− − bt1 it (6. θm .28) (6. θt .8).48). given by (2.31) = = i2 t (J2 − Jf /i2 ) f 1 + i2 t i2 t (b2 − bf /i2 ) f 1 + i2 t (6.3 Transmission Torque for Model 3 The performance output for Model 3 is derived in the same way as for Model 2.2.3 The performance output for Model 3 is ˙ ˙ ˙ ˙ ˙ z = (Mkc . In Chapter 3.72 Chapter 6 Gear-Shift Controller Design and Simulations with states and labels according to to the state-space description (4. (6. (6.6) to (4. J2 b2 = = Jt + Jf /i2 f bt + bf /i2 f (6. .26) (6. and (6.32) where Mkc is the torque transmitted by the clutch nonlinearity.21) is replaced by ˙ ˙ Mc = Mt = Mkc (θm − θt it ) + cc (θm − θt it ) (6. Then the performance output is deﬁned as Deﬁnition 6. 6.13) . Jt1 and bt1 can be derived as Jt1 bt1 which are approximated to Jt1 bt1 ≈ ≈ i2 t J2 1 + i2 t i2 t b2 1 + i2 t (6. θt /if − θw . the following combinations of parameters from Model 2 are estimated.14) . with the diﬀerence that (6.27).29) since Jf and bf are considerably less than J1 and b1 .26).27) From (6.

25) during Trial 1 are shown from the parameter estimation of the linear Models 1 and 2. The diﬀerence at higher frequencies is a result from the clutch which gives a second resonance peak for Model 2.3 shows the performance output in the frequency domain.1 Unconstrained Active Damping A ﬁrst attempt is to study the performance output.8) and (6. 6. the roll-oﬀ rate of Model 2 is steeper than for Model 1. z = M x + Du. [s] 10 15 Figure 6. (6.8).2 the performance output (6.3.34) . [Nm] 100 0 −100 −200 −300 −400 0 5 Time. with M and D given by (6. If u is chosen as u = −D−1 M x z = 0 is guaranteed. The solid line corresponds to Model 1 and the dashed line corresponds to Model 2. Figure 6. A control law can be derived since z includes the control signal and D is scalar.2 Transmission torque z from parameter estimation of Model 1 and Model 2 on data from Trial 1.3 Preliminary Trials 73 500 400 300 200 Torque.6. and the main reason to this is the diﬃculties to estimate the viscous damping coeﬃcients described in Chapter 3. to study gear-shift control. The low frequency level diﬀers between the two models.3 Preliminary Trials Two preliminary trials will be performed in this section. Furthermore. Comparison In Figure 6. 6.

34). x30 = 3. it would be desirable to use an control law which also considers that the control signal must be in a certain interval. u0 = 138 (6. To deal with this situation. while the engine speed is oscillating. Model 1 is shown in solid and Model 2 is shown in dashed.4 km/h) with gear 1 and a total load of 3000 Nm (≈ 2 % road slope). The modeled clutch adds a second resonance peak and a steeper roll-oﬀ rate.00 .34) generates a control signal that is impossible for the engine to generate. Unconstrained active damping (6. The stationary point is obtained by using (5.1 Consider the truck modeled in Chapters 2 and 3 traveling at a speed of 3 rad/s (5.35) In Figure 6. A second attempt is to explicitly handle the expected vehicle behavior (free rolling) . control signal u. since the speed is decreasing.0511 178 3. and wheel speed is seen when the control signal is chosen as in (6.3) and (5.3 Transfer functions from control signal u and load l to transmission torque z. l = 3000 ⇒ x0 = 0.4).4 the resulting transmission torque z. engine. 6.2 Undamped Gear-Shift Condition The previous approach is not realizable because of the unrealistic control signal. The wheel speed decreases linearly. Example 6. This means that the vehicle is free rolling which can be critical if lasting too long. It can be noted that despite z = 0 is achieved this is not an stationary point.74 Chapter 6 Gear-Shift Controller Design and Simulations Guz for Model 1 and 2 0 Gain [dB] −50 −100 −20 −40 Gain [dB] −60 −80 −100 −2 10 Glz for Model 1 and 2 10 −1 10 10 Frequency [rad/s] 0 1 10 2 10 3 Figure 6. Unconstrained active damping is achieved which obtains z = 0 instantaneously.3.

34) drives the transmission torque to zero.8 2.25) and (2. z.6.5 1 1. [Nm] 40 30 20 10 0 −10 0 0.4 Unconstrained active damping of Model 1. . Then by ˙ ˙ using θm = θw i. the control law (6.5 0 0. [s] 2 170 165 160 155 150 145 0 2.7 3 Figure 6. [Nm] Torque.1 180 175 [rad/s] [rad/s] 0. This control law is thus derived by considering a stiﬀ driveline.5 1 1. [s] 2 2.3 Preliminary Trials Transmission torque z Engine torque u 75 60 50 Torque.5 2 ˙ Engine speed θm ˙ Wheel speed θw 3. the diﬀerential equation.26). when the transmission torque. The wheel speed decreases linearly. describing the stiﬀ driveline is ¨ ˙ (J1 i + J2 /i)θw = u − (b1 i + b2 /i)θw − l/i (6. The oscillations in the transmission torque are damped with an unrealizable control signal. and solving for z = 0. is zero. and eliminating ˙ ˙ the torque transmitted by the drive shaft.5 Time.36) This equation is developed by using Model 1 in (2.36) results. k(θm /i − θw ) + c(θm /i − θw ).6 2. (6.5 Time. but without using active damping.9 2. By using the labels according to Chapter 4. At t = 1 s.5 1 1.5 2 0 −5000 −10000 −15000 0 0.5 1 1.

The transmission torque is estimated and controlled to zero with active damping.4 Gear-Shift Control Criterion Neither of the two approaches in the previous section solve the problem satisfactory. prevents the control signal from having large deviations from the undamped gear-shift condition ushif t . .37) Combining (6. If the driveline is stiﬀ.38) J1 i2 + J2 J1 i2 + J2 J1 i2 + J2 The control signal to force z = 0 is given by solving (6. The performance of this approach is worse if the driveline is oscillating at the time for the gear shift. the point at which the cost function is zero is no stationary point. The criterion is formulated such that active damping is obtained with a control law whose deviation from the undamped gear-shift condition (6. l))2 ˙ (M x + Du)2 + η(u − µx θw − µl l)2 (6.76 Chapter 6 Gear-Shift Controller Design and Simulations Equation (6.5 shows Example 6. since the speed of the vehicle will decrease despite z = 0 and u = ushif t . and at the same time. ˙ ushif t (θw . The idea is formulated as a cost criterion which uses a combination of the two previous approaches.36) and (6.37) gives the performance output for the stiﬀ driveline. z = (1 − (Jm + Jt1 )i (Jm + Jt1 )i (Jm + Jt1 )i2 ˙ )u − (bt1 i − (b1 i2 + b2 ))θw + l (6. but the oscillations introduced are not damped. l) µx µl = = = ˙ µx θw + µl l with (Jm + Jt1 )i (Jm + Jt1 )i2 −1 (b1 i2 + b2 ))(1 − ) (6.40) = T →∞ lim 0 The controller that minimizes this cost function damps oscillation (since the ﬁrst parenthesis is minimized). This control law achieves z = 0 with a realizable control signal. Let the cost function be T T →∞ lim 0 T ˙ z 2 + η(u − ushif t (θw . Furthermore.39). 6.40). The tradeoﬀ is controlled by η. In this section a new idea for gear-shift control is formulated.38) for u while z = 0.39) adds to the cost function.1 applied to Model 1 controlled with the undamped gear-shift condition (6. there is no diﬀerence between the two parenthesis in (6.5) expressed in terms of wheel speed is ˙ ¨ z = u − bt1 iθw − (Jm + Jt1 )iθw (6. Therefore. the time needed to fulﬁll the gear-shift condition is not optimized.39) 2+J J1 i J1 i2 + J2 2 (Jm + Jt1 )i (Jm + Jt1 )i2 −1 − (1 − ) J1 i2 + J2 J1 i2 + J2 (bt1 i − Figure 6.

Undamped oscillations in the transmission torque increase the time needed to fulﬁll the gear-shift condition. The derived feedback law is a function of η which is chosen such that a feasible control signal is used. by solving a Riccati equation. a gear shift is commanded.5 Gear-Shift Control Design The new idea for gear-shift control is in this section given eﬃcient treatment by solving (6.5 Gear-Shift Control Design Transmission torque z Engine torque u 77 200 60 Torque.40) in terms of the linearized variables. A state-feedback matrix is derived that minimizes (6.5 1 0.40).3.5 2 1.39).9) in Section 5. [Nm] Torque.8) and (5. 6. At t = 1 s. This is done by linearizing the driveline model and rewriting (6.5 3 150 [rad/s] [rad/s] 2 4 Time. [Nm] 2 4 6 8 40 20 0 −20 −40 0 150 100 50 0 −50 −100 −150 0 2 4 6 8 ˙ Engine speed θm ˙ Wheel speed θw 200 3.5 Model 1 controlled with the undamped gear-shift condition (6.5 100 50 0 0 0 0 2 4 Time. The linearized driveline model is given by (5.6. The speed dependent realizable control signal drives the transmission torque to zero. [s] 6 8 Figure 6. using available software. The . [s] 6 8 2.40) for a control law by using LQG technique.

9) T T →∞ lim + = with η(∆u − µx ∆x3 + u0 − µx x30 − µl l)2 T T →∞ 0 (M ∆x + D∆u + M x0 + Du0 )2 lim 0 (M ∆x + D∆u + r1 )2 + η(∆u − µx ∆x3 + r2 ) (6.21).48) Kc1 Kc2 Kc3 x (6.45) where Pc is the solution to the Riccati equation (5.41) can be written in the form T T →∞ lim 0 xT Qxr + R∆u2 + 2xT N ∆u r r (6. This was done in (5.44) (6.41) r1 r2 = = M x0 + Du0 u0 − µx x30 − µl l (6.16).42) The constants r1 and r2 are expressed as state variables.13) to (5.49) .78 Chapter 6 Gear-Shift Controller Design and Simulations cost function is expressed in terms of ∆x and ∆u by using (5.47) 1 − Kc4 D − Kc5 Kc1 Kc2 Kc3 − Kc4 M Kc5 (6.43) is minimized by the state-feedback gain T Kc = Q−1 (Br Pc + N T ) (6. B) with models of the constants r1 and r2 .42) gives u = K0 x30 + Kl l − with K0 Kl where Γ is Γ= = = λx λl δx δl µx µl Γ Γ (6. the cost function (6. by augmenting the plant model (A.46) ∆u = −Kc xr = − Kc1 Kc2 Kc3 ∆x − Kc4 r1 − Kc5 r2 using (6.43) with Q = (M 1 0)T (M 1 0) + η(0 0 − µx 0 1)T (0 0 − µx 0 1) N = (M 1 0)T D + η(0 0 − µx 0 1)T R = D2 + η The cost function (6. The resulting control law is (6. By using these equations.

A realizable control signal is used such that the transmission torque is driven to zero. and µ given by (5. [s] 3 4 160 140 120 100 0 3.0207 −1.50). active damping is obtained with a realizable control signal.2521 x (6. With this controller the phase margin is guaranteed to be at least 60◦ and the amplitude margin is inﬁnity (Maciejowski 1989).6 Model 1 controlled with the LQG-control law (6.40). δ. By solving the gear-shift criterion (6. [Nm] 100 0 −100 −200 ˙ Engine speed θm ˙ Wheel speed θw 200 180 [rad/s] [rad/s] 1 2 Time. while oscillations are actively damped. and (6.3). The control law is a function of η which is chosen such that the control signal is feasible.5 2 1. solving the gearshift criterion (6. with λ.37 · 10−4 x30 − 0.5 0 1 2 Time.5 Gear-Shift Control Design Transmission torque z Engine torque u 79 200 60 50 Torque.6.0327l − 4. The result is seen in Figure 6.1 the controller gains becomes u = 2. a gear shift is commanded. (5. . [s] 3 4 Figure 6.0001 are used.5 3 2. When this control law is applied to Example 6.40).39). [Nm] 40 30 20 10 0 −10 0 1 2 3 4 −300 0 1 2 3 4 Torque.50) where η = 0. At t = 1 s.6.2123 0.03 and α = 0.4).

This is obtained also when using the observer by increasing ρ towards inﬁnity.47) becomes u = K0 x30 + Kl l − Kc1 Kc2 Kc3 x ˆ (6.48).7 shows how the performance output and the control signal are aﬀected by the load disturbance v. aw = 2.6 Inﬂuence from Sensor Location The LQG controller investigated in the previous section uses feedback from all ˙ ˙ states (x1 = θm /it if − θw . The closed-loop transfer function Gvu is given by (Gvu )cl = − Fy Gvy 1 + Fy Guy (6.8) is not equal to zero.55) ⇒ ϕw = 74. which is the case considered in this work. x2 = θm . 6. The observer gain is calculated using LTR technique. However. and the amplitude margin a is inﬁnity. In this section an observer is used to estimate the rest of the states.4.56) .4 it was shown that for the speed controller. as stated before. The unknown load can be estimated as in Section 5. When using the LQG with feedback from all states.52) (6.3.1 the following values are used ρm ρw = = 104 ⇒ ϕm = 77.51) with K0 and Kl given by (6.84 ◦ where the aim has been to have at least 60◦ phase margin. The estimated state x is given by the Kalman ˆ ﬁlter ˙ ∆x ˆ Kf x = A∆ˆ + B∆u + Kf (∆y − C∆ˆ) x T −1 = Pf C V (6.30). and x3 = θw ). am = 2. also the transfer function DGvu should be added to (5. the phase margin ϕ is at least 60◦ .6.82 10 11 (6. The LQG feedback law (6. This is not possible if only one sensor is used.34).80 Chapter 6 Gear-Shift Controller Design and Simulations 6. these transfer functions are not aﬀected by the sensor location. Hence.53) where Pf is found by solving the Riccati equation (5.3 . Hence.3◦ . The same equations are valid for the gearshift controller except the diﬀerence that the D matrix in (6.1 Inﬂuence from Load Disturbances Figure 6. The observer dynamics is cancelled in the transfer function from reference value to performance output and control signal. the dynamics will be included in the transfer functions from disturbances to both z and u. In Section 5. The sensor either ˙ ˙ measures the engine speed θm or the wheel speed θw . as for the speed controller. the resonant open-loop zeros become poles of the closed-loop system when feedback from the engine speed sensor is used. Two diﬀerent observer problems results depending on which sensor location that is used. For Example 6.54) (6.

. According to (5. For θm feedback.6.41).2 Inﬂuence from Measurement Disturbances The inﬂuence from measurement disturbances e are seen in Figure 6.7 Closed-loop transfer functions from load disturbance v to performance ˙ output z and control signal u. resulting from the open-loop zeros. Thus. 6.59) Guz Fy 1 + Guy Fy (6.6 Inﬂuence from Sensor Location Closed-loop transfer function Gvz 81 −20 −40 Gain [dB] −60 −80 −100 −2 10 10 −1 10 0 10 1 10 2 10 3 0 −50 Gain [dB] −100 −150 −200 −2 10 Closed-loop transfer function Gvu 10 −1 10 10 Frequency [rad/s] 0 1 10 2 10 3 Figure 6. this means that a resonance peak is present in the transfer functions from v to performance output z and control signal u. Feedback from θw is shown in solid and feedback ˙ ˙m is shown in dashed lines. the closed-loop transfer function from v to z has the open-loop ˙ zeros as poles. With θm feedback the transfer functions have a from θ resonance peak. Hence.40) the closed-loop transfer function from e to z is (Gez )cl = − Then (Gez )cl (Gez )cl = = Guz ˙ with θw feedback Guw Guz ˙ −Tm with θm feedback Gum −Tw (6.58) (6.6.57) with Tw and Tm from (5. the closed-loop transfer function from v to u also has the controller Fy in the numerator.8.

as discussed in Section 5.9 shows the sensitivity functions (5. The dynamic output ratio Gw/m was deﬁned in Deﬁnition 4.4. because that Guw has a relative degree of two.60) ˙ ˙ where cl.44). Hence.w Gw/m (6. The frequency range in which the Tm = Tw is valid depends on how large ρ in (5.31) is made. Feedback from θw is shown in solid and ˙m is shown in dashed. Tw has a steeper roll-oﬀ rate than Tm .4.59) gives (Gez )cl. This is due to that the open-loop transfer functions Guw and Gum have a diﬀerent relative degree. Therefore. as discussed in Section 5. and Gum has a relative degree of one. Then (6. Tm = Tw as was discussed in Section 5. The diﬀerence between the two feedback feedback from θ principles are described by the dynamic output ratio. and is given by (5. w means closed loop with feedback from θm and θw respectively. When ρ in (5.4.m = (Gez )cl. Figure 6. the diﬀerence in Gez depending on sensor location is described by the dynamic output ratio Gw/m .82 Chapter 6 Gear-Shift Controller Design and Simulations 100 50 Gain [dB] 0 −50 −100 −2 10 Closed-loop transfer function Gez 10 −1 10 0 10 1 10 2 10 3 100 Closed-loop transfer function Geu Gain [dB] 50 0 −50 −2 10 10 −1 10 10 Frequency [rad/s] 0 1 10 2 10 3 Figure 6. this eﬀect increases with lower gears. The roll-oﬀ rate at higher frequencies diﬀer between the two feedback principles.58) and (6. It is seen that Tm = Tw is valid up to about 2 Hz.31) is increased towards inﬁnity. The eﬀect increases with lower gears.46) and the complementary sensitivity functions Tw and Tm (5.41) for the two cases of feedback. The diﬀerence in low frequency level is equal to the conversion ratio of the driveline.1. m and cl. .8 Closed-loop transfer functions from measurement noise e to perfor˙ mance output z and control signal u.

the feasibility of the gear-shift controller is studied by simulation on a more complicated vehicle model than it was designed for. given by Exam˙ ple 6. in Section 5.e.1.33). When only ˙ one velocity is measured. the equation for the transmission torque is calculated using (6. In the simulations. wheel speed θw = 3 rad/s. In the second simulation. The . the relationship between the model parameters in the transmission is given.7 Simulations 83 20 0 −20 Gain [dB] −40 −60 −80 −100 −120 −2 10 10 −1 10 10 Frequency [rad/s] 0 1 10 2 10 3 Figure 6. The eﬀects from sensor placement are also studied in accordance with the discussion made in Section 6. Three simulations are performed with the same parameters. The wheel speed or the engine speed is input to the observer (6.6. and the control law (6.6.9 Sensitivity function S and complementary sensitivity function T . The steady-state level for Model 3 is calculated by solving the model equations for the equilibrium point when the load and speed are known.5.51) generates the control signal. In Assumption 6.7 Simulations As in the case of the speed controller. the solid lines correspond to θw feedback. a gear shift is commanded at t = 2 s. according to Figure 6. lines correspond to θ 6.1. Model 3 is given by Equations (2.52). The controller used is based on Model 1. as seen in the previous sections. The oscillations are a result of a sinusoid disturbance acting on the control signal. (i.51). The dash-dotted lines correspond to the case with all states known.10. the driveline is oscillating prior to the gear shift. and the dashed ˙m feedback. The ﬁrst simulation is without disturbances. By these. and load l = 3000 Nm).49) to (2. The control design is simulated with the nonlinear Model 3.

. The disturbance is generated as a square pulse with 0. than from the engine speed sensor.6. where the design is tested on Model 1. the diﬀerential equations (2. In the simulation. neither of the sensor alternatives reaches z = 0.49) to (2.84 Chapter 6 Gear-Shift Controller Design and Simulations r Vehicle Model 3: (2.1 s width and 1200 Nm height.2.6. The model is simulated using Runge Kutta (45) (Simulink 1993) with a low step size to catch the eﬀect of the nonlinearity. This plot should be compared with Figure 6. The model errors between Model 1 and Model 3 are handled better when using the wheel speed sensor.11 the simulation without disturbances is shown. The design still works if the extra nonlinear clutch dynamics is added. In Figure 6.49) to (2.1 z (6. which is a veriﬁcation of the discussion in Section 6. As a step for demonstrating feasibility for real implementation. The result is that the performance of the controller is not aﬀected by the oscillations.33) ˙ ˙ y (θw or θm ) Controller Design based on Model 1 Control law (6. In Figure 6.51) Assumption 6. This is due to the low frequency model errors discussed in Section 6.51) Observer (6.10 Simulation conﬁguration.51) are scaled such that the ﬁve diﬀerential equations (one for each state) have about the same magnitude. The result is that the performance does not critically depend on the simpliﬁed model structure.13 shows the simulation with load disturbance. Model 3 is simulated with the controller based on Model 1. However.52) Figure 6. there are diﬀerent results depending on which sensor that is used. In order to simulate the nonlinear model. Figure 6.12 the simulation with driveline oscillations prior to the gear shift is shown. third gear shift is simulated with a load impulse at t = 3 s. The disturbance is better damped when using feedback from the wheel speed sensor.

[s] 6 8 −50 −100 −150 10 −200 −250 −300 0 30 20 0 −10 0 2 4 Time.7 Simulations 85 70 Transmission torque z 150 100 50 0 Control signal u 60 50 40 Torque. and from the engine speed sensor is seen in dashed.6. and from the engine speed sensor is seen in dashed. [Nm] 200 150 100 50 Torque. . [s] 6 8 Figure 6. The design still work when simulated with extra clutch dynamics. [Nm] Torque. Feedback from the wheel speed sensor is seen in solid.12 Simulation of Model 3 with observer and control law based on Model 1. The conclusion is that the control law works well despite initial driveline oscillations. [s] 6 8 2 4 Time.11 Simulation of Model 3 with observer and control law based on Model 1. Feedback from the wheel speed sensor is seen in solid. [Nm] 2 4 Time. [s] 6 8 Figure 6. [Nm] 0 −50 −100 −150 −200 −250 −300 0 50 40 30 20 10 0 −10 0 2 4 Time. Transmission torque z Control signal u 90 80 70 60 Torque.

works well for a more complicated model with a nonlinear clutch characteristics. This results in undamped load disturbances when engine speed feedback is used.13 Simulation of Model 3 with observer and control law based on Model 1. The proposed solution handles the fact that the gear-shift condition is not a stationary point. there can be problems with a low frequency level that gives a stationary error. Based on a model of the transmission torque. from diﬀerent sensor locations. that actively damps driveline oscillations. 6. load disturbances are better attenuated with feedback from the wheel speed sensor. This diﬀerence in level is a result of the diﬃculty to estimate the driveline friction parameters. Simulations show that the performance of the design.86 Chapter 6 Gear-Shift Controller Design and Simulations Control signal u 200 Transmission torque z 80 70 60 50 Torque. . The conclusion is that the load disturbance is better attenuated when using feedback from the wheel speed sensor. and from the engine speed sensor is seen in dashed. However. it is possible to solve the criterion for a control law that minimizes the cost function. on the control design results in the same conclusion as in Chapter 5. Feedback from the wheel speed sensor is seen in solid. An impulse disturbance is acting on the load at t = 3 s. [s] 6 8 Figure 6. When using a driveline model with drive shaft ﬂexibility. a criterion for a gear-shift controller is obtained. Therefore. [Nm] 40 30 20 10 0 −10 −20 0 −150 −200 −250 −300 0 Torque. based on the simpliﬁed model.8 Summary Driveline oscillations is a limiting factor in gear shifting with engine control. An investigation of the inﬂuence. When using LQG/LTR the open-loop zeros are cancelled by the controller. [Nm] 2 4 Time. [s] 6 8 0 −50 −100 150 100 50 2 4 Time. The control law is derived with LQG/LTR technique.

The diﬀerence between these two is described by the dynamic output ratio. actively damped transmission-torque control works well also in the case of existing initial oscillations.6. disturbances occuring during the control action are actively damped. This eﬀect increases with lower gears. depending on feedback conﬁguration. and thus reducing the time needed for a gear shift.8 Summary 87 Measurement disturbances are better attenuated when the engine speed sensor is used. . than when using the wheel speed sensor. Furthermore. In conclusion. Two diﬀerent closed-loop transfer functions result.

88 Chapter 6 Gear-Shift Controller Design and Simulations .

is obtained. This controller is to be used with a new automatic gear shifting system. Extra clutch dynamics is not able to explain more of the experiments for low frequencies. which is important since the vehicle is free rolling when in gear-shift condition. A key contribution is the observation that a linear model with stiﬀ clutch and drive shaft ﬂexibility is able to explain the measured engine speed and wheel speed. The major contribution of this thesis is a novel gear shifting strategy based on modeling of the transmission torque. 89 .7 Conclusions The driveline is a fundamental component in a vehicle. Therefore. which is veriﬁed by simulations on a model with a nonlinear clutch characteristics. A second important contribution is the extension of the traditionally used RQV controller. The proposed solution oﬀers a possibility to optimize the time needed for a gear shift. With this controller the performance and driveability is improved since vehicle shuﬄe is reduced. the formulation is natural. it allows eﬃcient solution. and design of a criterion for a controller that drives this torque to zero. and there is currently a strong trend in improving performance by adding functionalities in driveline management systems. and there is a simple tuning of the amount of RQV feeling. the linear model is concluded to be a basis for control design. Furthermore. Furthermore. A criterion for a controller that actively damps wheel speed oscillations with a stationary error characteristic for the RQV controller. utilizing engine controlled gear shifting without using the clutch. A basis for these results is the modeling conclusions drawn from experiments and modeling using a heavy truck. neutral gear can successfully be engaged also when facing critical load disturbances and initial driveline oscillations.

and increases with lower gears. The major part of the diﬀerence is explained by a simple sensor model. Diﬀerent sensors give the same open-loop poles. due to well damped open-loop zeros. Measurement disturbances are better attenuated when the engine speed sensor is used.90 Chapter 7 Conclusions Another important observation from the experiments is the explanation of the diﬀerence between the measured engine speed and transmission speed. An investigation of the inﬂuence from diﬀerent sensor locations on the control design shows that when using LQG/LTR. The diﬀerence is explained by the dynamic output ratio. A common architectural issue in driveline control is the issue of sensor location. . There are thus issues to be considered in sensor choice. than when using the wheel speed sensor. but diﬀerent zeros. but the overall conclusion is that the proposed strategies improve performance and driveability in both speed control and gear-shift control. Parameter estimation of a nonlinear model shows that the deviations still left occur when the clutch transfers zero torque. load disturbances are better damped with feedback from the wheel speed.

Gustafsson (1993). A. T. a. Academic Press. Automatica 24. Automatica 21. User’s Guide. To be presented at Reglerm¨tet ’96 in o Lule˚ Sweden. Automotive Handbook. (1992). Stuttgart. J. Bosch (1993).. 117–128. J. Malebranche (1985). T. Pettersson. Nielsen (1996). Meriam. 91 . System Identiﬁcation Toolbox. (1988). Volume 2 of Engineering Mechanics. Pettersson. G. Dynamics.Bibliography Bj¨rnberg. M. Department of Electrical Engineering. (1995).. and H. Technical Report LiTH-ISY-R-1506. Germany: Robert Bosch GmbH. Kubrusly. M. D. H. Addison-Wesley. Maciejowski. Gillespie. and L. Control Theory 1984-1986. L. Henriksson. Fundamentals of Vehicle Dynamics. 573–583.a survey. SAE International. L. J. Ljung. Link¨ping o University. M. Ljung. especially air drag and rolling resistance. Nonlinear driveline oscillao tions at low clutch torques in heavy trucks. Liversidge. Multivariable Feedback Design. L. MathWorks. (1952). (1989). and L. and F. Sensors and controllers location in distributed systems . Kraige (1987). Backlash and Resilience within Closed Loop of Automatic Control Systems. C. An investigation of the longitudinal dynamics of a car. Inc. John Wiley & Sons.

Vidyasagar (1989). and M. M. M. Switzerland . W. Preprint of the IFAC-Workshop on Advances in Automotive Control. Robot Dynamics and Control. John Wiley & Sons. Tozawa (1992). and Y. MathWorks. (1995). A. Pettersson. and N. C. Nielsen (1995). Orehall. . K. Powell (1996). Y.92 Bibliography Mo. J. Ascona. Truck and Commercial Vehicle International ’95. (1993). O. Spong. C. Inﬂuence of powertrain torsional rigidity on NVH of 6x4 trucks. Scania opticruise: Mechanical gearchanging with engine control. Sensor placement for driveline control. SAE Paper 922482. SAE Paper 960046. Simulink (1993). Suzuki. and L. Active control of driveability. Nwagboso. Beaumount. Automotive Sensory Systems. Inc. L. Chapman & Hall. N. User’s guide..

reference signal Control signal Performance output State vector Sensor output State disturbance. velocity Measurement disturbance Input disturbance Load Angle Road slope Symbols J i k Mass moment of inertia Conversion ratio Torsional stiﬀness 93 .Notations Variables r u z x y v e n l θ α Radius.

94 Notations c b m cr1 . cr2 cw ρa Aa Fa Fr M A B C H D G Gw/m S T Kc Kf ϕ a Torsional damping Viscous friction component Vehicle mass Coeﬃcients of rolling resistance Air drag coeﬃcient Air density Vehicle cross section area Air resistance force Rolling resistance force Torque. performance output state matrix State-space matrix Input state matrix Output state matrix Load state matrix Performance output control signal matrix Transfer function Dynamic output ratio Sensitivity function Complementary sensitivity function State-feedback matrix Observer gain Phase margin Amplitude margin Subscripts m c t p f d w fr 0 t1 t2 Engine Clutch Tranmission Propeller shaft Final drive Drive shafts Wheel Friction Stationary value Transmission input Transmission output .

- D375A-5 Catalogue.pdf
- Auto Clutch Control
- JT15 Right Angle Gearbox,Direction Change Gearbox, Gearbox Input and Output Directions,Miniature 90 Degree Gearbox 1 to 1 Ratio,90 Deg Gear Boxes 1-1 Ratio,1 to 1 Ratio Gearbox
- September 2014 GEARS
- Simulation and Control of an Automotive Dry Clutch
- TM 9-1786B Power Train, Track Suspension and Equipment for 13-Ton High Speed Tractor M5 1944
- Chapter5-Clutch & Gear box New.pdf
- PRINT Deliverable 3 Print out google dioc.doc
- Del 2 E 180-195 Gear Boxes Operational Manual
- 6F35 gearspdf
- T5SwapFAQ
- File 88
- NewCLUTCH
- FALLSEM2012-13_CP0839_13-Jul-2012_RM01
- George مهمة جدا
- May/Jun 2012
- BHS Prospekt DKX USA 072014 Web
- GearExample-v6
- METHOD FOR OPERATING AN AUTOMATIC START_STOP SYSTEM IN A VEHICLE - Patent application.pdf
- 050555 En
- GR00002600-23A.pdf
- Eclipse Bicycle Co. v. Farrow, 199 U.S. 581 (1905)
- gxmm
- Daewoo Excavadora Hidráulica SOLAR 130W-V
- Scarab Merlin XP - Uni-drive Technical Specifications
- Turboflex P-1940-BB.pdf
- Abstract (2)
- Husquarna 2012 TC-TE-TXC-SMR 449-511
- octavia.pdf
- Pasquali Parts - INDEX TO PASQUALI PARTS MANUAL

Skip carousel

- RUKUS October 2009
- 2014 January/February GEARS Buyers Guide issue
- GEARS September 2013
- GEARS December 2013
- GEARS Oct/Nov 2014 show issue
- August 2012
- GEARS September 2012
- Dana Driveshaft Installation
- GEARS January-February 2016
- GEARS April 2014
- Gears Magazine March 2013
- GEARS Buyer's Guide 2015 Issue - January/February 2015
- GEARS March 2015 - Annual Torque Converter Issue
- April 2013 Gears Magazine
- October/November 2009
- tmpA0FB.tmp
- Analysizing the effect of eliminating belt drive in lathe Gear Box by directly engaging of gears
- April 2010
- December 2011
- Automatic Pallet Carrier for Small Workshop
- Manitou M Series (EN-US)
- A Review Paper on Development of Automatic Gear Shifting Mechanism
- Gears August 2013
- July 2011
- tmpE735.tmp
- June 2014 GEARS
- 46306_1990-1994
- GEARS May 2015
- May/June 2010
- March 2010

Skip carousel

- Nouis Technologies v. Polaris Industries et. al.
- March 2010
- December 2011
- August 2012
- GEARS September 2012
- July 2012
- GEARS April 2014
- GEARS March 2015 - Annual Torque Converter Issue
- April 2013 Gears Magazine
- GEARS May 2015
- October/November 2009
- May/June 2010
- Paulette M. Ibbitson, Administratrix, Estate of Roger R. Ibbitson, Deceased v. The Ramsey Manufacturing Co. Holland Supply Company Advanced Industries, Inc. Everett Spurlin Erection Company, the Fort Worth Tower Co., Inc. Tommy Moore, Inc., Defendants/third Party Direct Channels of Defiance, Inc., Third Party, 802 F.2d 458, 3rd Cir. (1986)
- Automatic Gear Changer in Two Wheeler Using Pic
- April 2010
- December 2010
- Static and Dynamic Analysis of Composite Clutch Plate
- September 2011
- September 2009
- Gears August 2013
- October/November 2013
- July 2013
- Conti, Patricia and Conti, Richard, Husband and Wife v. Ford Motor Company and Winner Ford, J/s/a v. Richard Conti. Appeal of Ford Motor Company, 743 F.2d 195, 3rd Cir. (1984)
- Gears May 2016
- Design and Analysis of Clutch Plate for Automatic Single Plate Clutch
- Military Searchlight Director (1932)
- September 2010
- Apollo Experience Report Guidance and Control Systems CSM Service Propulsion System Gimbal Actuators
- May/Jun 2012
- GEARS April 2015

Sign up to vote on this title

UsefulNot usefulClose Dialog## Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

Close Dialog## This title now requires a credit

Use one of your book credits to continue reading from where you left off, or restart the preview.

Loading