Non-linear multi-body dynamic analysis for the study of clutch

torsional vibrations (judder)

D. Centea
a
, H. Rahnejat
b,
*
, M.T. Menday
c
a
Department of Mechanical Engineering, University of Bradford, Bradford, West Yorkshire BD7 1DP, UK
b
Wolfson School of Mechanical and Manufacturing Engineering, Loughborough University, UK
c
Powertrain Engineering, Research and Development Centre, Ford, UK
Received 9 April 1999; received in revised form 16 August 2000; accepted 29 August 2000
Abstract
This paper presents an investigation of the torsional vibration mode of the vehicular clutch system, referred to as
judder. This phenomenon takes place during the clutch engagement process on light trucks with Diesel engines. A non-
linear multi-body dynamic model of the clutch mechanism is presented in order to study the eect of various clutch
systems and driveline components on its vibration performance. The parameters include geometrical, physical, inertial
and force data of the clutch system of vehicles. The paper demonstrates that propensity to judder is related to the
various friction lining materials' characteristics. 2001 Published by Elsevier Science Inc. All rights reserved.
1. Introduction
Clutch take-up judder is a fore and aft vibration of a vehicle in the frequency range 520 Hz,
caused by the torsional vibrations of the driveline which occur during the clutch engagement
process, usually in the start-up process [14]. This phenomenon is distinct from the torsional
vibration of the drivetrain system during throttle tip-in from coast to drive condition or during
throttle tip-out from drive to coast condition, which signies the lowest rigid body motion of the
entire system. For the shue phenomenon, readers are referred to Refs. [57].
Judder is essentially regarded to be inuenced by frictional characteristics of the clutch system.
It is also related to the inertia of the driveline and to the variation of the coecient of friction l of
the friction disc linings with slip speed. The severity of clutch judder phenomenon is inuenced by
the way the vehicle is driven.
In recent years research eort on power train systems has been mostly concentrated upon
driveline vibrations. Clutch vibrations have been studied to a much lesser extent. Torsional
vibrations of the clutch system, leading to judder are induced by stick-slip at the clutch friction
disc to ywheel and pressure plate interfacial contacts.
Stick-slip is present to a varying degree in almost all actuators and mechanisms with fric-
tional contact, often leading to self-excited vibration. Karnopp [8] has presented a method for
www.elsevier.nl/locate/apm
Applied Mathematical Modelling 25 (2001) 177192
*
Corresponding author.
E-mail address: h.rahnejat@lboro.ac.uk (H. Rahnejat).
0307-904X/01/$ - see front matter 2001 Published by Elsevier Science Inc. All rights reserved.
PII: S 0 3 0 7 - 9 0 4 X ( 0 0 ) 0 0 0 5 1 - 2
representing and simulating stick-slip friction eects. He has analysed dierent friction models in
the stick-slip region, presented as the variation of the friction force F
f
with the relative interfacial
contact velocity v. In the model presented by Karnopp (see Fig. 1(a)), it is dicult to linearise the
dry friction law in the vicinity of v = 0 (i.e., the stiction region). Furthermore, the numerical
integration in the region where v = 0 is impossible to achieve due to the vertical step function
characteristic.
The model presented in Fig. 1(b) resolves the integration diculties encountered in the original
model (see Fig. 1(a)). In the stiction region the friction force becomes unique. The slope dened
within the stick region introduces a ``stiness'' which occurs after the engagement process in the
case of the clutch system. In a judder study one need not consider this eect.
Haessig and Friedland [9] have presented two new models for the stick-slip friction; the bristle
model and the reset integrator model. The bristle model represents friction as the interaction of
many bristles. It achieves a high degree of accuracy through an attempt to present the microscopic
physical process of contacting surfaces which can cause stick-slip friction. However, the model is
numerically inecient. The reset integrator model has been developed to reduce the computation
Nomenclature
C
k
kth constraint function
c eective drivetrain damping
dt variable integration time step
F
f
friction force
F
n
clamp force
F
q
k
generalised applied forces
i
gbx
rst gear ratio
i
dif
dierential gear ratio
J mass moment of inertial
[J[ Jacobian matrix
J
red
reduced mass moment of inertia
K kinetic energy
M applied torque
M
e
engine torque
M
f
friction torque
n number of constraint functions unless otherwise specied
q
k
kth generalised co-ordinate
R mean radius of friction lining
s slope of friction torque everywhere, unless in the Jacobian matrix, where it is a
scaling factor
t time
u
k
_ q
k
v slip speed
x; y; z translational components of motion
h relative angle between the friction disc and the hub in Eq. (5)
w; h; / Euler angles
k
k
Lagrange multiplier
l coecient of friction
x relative angular interfacial contact slip velocity
178 D. Centea et al. / Appl. Math. Modelling 25 (2001) 177192
time requirements of the bristle model, whilst retaining its characteristics. Instead of using the
bristles in the stick region, the reset integrator model represents the friction force as a function of
position by using a damping term which can be numerically integrated. The basics of the reset
integrator model proposed by Haessig and Friedland [9] and by the Karnopp stick-slip friction
model [8] are used in dening the friction torque.
Jarvis and Mills [10] have studied the vibrations induced by dry friction. They used an ap-
paratus, comprising a disc and a cantilever. Using Lagrange's equations, they developed a
mathematical theory for the oscillation of the system. By means of numerical analysis, they
showed theoretically that variation of the coecient of friction with the relative velocity is in-
sucient to cause vibrations, and that the instability is due to the manner in which the motions of
the components take place. The relationship between the coecient of friction and the relative
velocity has been studied extensively by Armstrong-Helouvry [11]. He has proposed dierent
friction models; a combined kinetic and viscous friction model, a combined static, kinetic and
viscous friction model, and a negative combined kineto-viscous friction model which includes the
Stribeck eect. The models are mainly applied to the motions of machines, where stick-slip
phenomenon is particularly important.
Kani et al. [12] have proposed that judder is signicantly related to the lv characteristic
(where l is the coecient of friction and v is the slip speed) of an interface friction material. For a
constant clamp load they have shown that the dM=dn (where M is the torque and n is the relative
revolution speed) is directly proportional to dl=dv. They have found, using an experimental
tester, that the dl=dv has a negative gradient when judder occurs and that the value of dl=dv
depends on the type and the amount of lm formed on the friction surface. They concluded that
better characteristics can be obtained when a greater amount of superior quality carbon is in-
cluded.
Maucher [13] has also studied the basic principles governing vibrations that occur in the clutch
due to frictional behaviour of the clutch facing; the driveline damping, the clamp load, the mass
moment of inertia and the torsional spring rate of the drivetrain. He concluded that frictional
vibrations occur in the presence of low drivetrain damping values and a negative gradient of the
friction coecient, while the clamp load has a smaller eect. However, for reasons of comfort and
functional integrity of the clutch actuation system, the damping of the drivetrain cannot be in-
creased signicantly. Therefore, one way of reducing frictional vibrations in the clutch is to select
a friction material with a good friction coecient characteristic with slip speed.
Fig. 1. Stick-slip friction models (from [8]).
D. Centea et al. / Appl. Math. Modelling 25 (2001) 177192 179
Drexl [1] has devised a simplied linear driveline torsional model to study the judder problem
on a newly developed vehicle. His model included the inertial properties of the driveline com-
ponents, from ywheel to the rear road wheels, including the vehicle body inertia. He found that
when judder occurs, the model's lowest natural frequency was excited. His simulation results
show that a negative value for the variation of coecient of friction with slip speed excites the self-
excited oscillations, while a positive gradient of coecient of friction versus slip speed (or dif-
ferential speed) produces a dampening eect upon vibration response. His results can provide
information for a clutch to be designed with regard to minimisation of judder, but cannot
guarantee that the judder-related problems can be eliminated.
The torsional vibrations of the clutch were also studied by Newcomb and Spurr [2]. They stated
that although most of the published works show that judder has been generally attributed to a
particular type of variation of coecient of friction with slip speed, that this is not a necessary
condition for judder to occur.
A mathematical model, including torsional vibrations of the driveline, vehicle body fore-aft
vibrations and vertical vibrations is developed by Rabeih and Crolla [3] in order to investigate the
vibration behaviour of the system for steady and transient running conditions. They found that
the vehicle may be subjected to instability in the fore-aft and driveline system torsional motions
for certain values of the system parameters due to self excitation force induced by tyre-road
contact and friction during clutch engagement. They concluded that high values of system
damping tends to discourage self-excited vibrations, and that decreasing the gradient of friction
causes system instability and also decreasing the axle stiness has a signicant eect on the re-
duction of vibration levels of the driveline system.
Similar conclusions are presented by the same authors in Ref. [14], where the clutch judder and
shunt phenomena are discussed. Furthermore, they considered that it should be possible to
control the torsional vibrations instability during the engagement period, using some kind of
intelligent control system, linked to sensing information related to the system and taking control
to modify the response into a subjectively acceptable form.
Centea et al. [4] have developed a non-linear model for a cable operated clutch system to
include all assembly constraints and non-linear stiness and damping characteristics of the
system. This multi-body model includes the driveline and the vehicle body inertial components.
It also incorporates the non-linear clutch clamp load characteristics through the engagement
process as well as an interfacial stick-slip model, one that has been lacking in previous lin-
earised models. They showed that propensity to judder is enhanced by a combination of ef-
fects; the lv characteristics and an apparent loss of clamp load, a by-product of
manufacturing and assembly tolerances and/or driver clutch eort through an inappropriate
actuation speed. This paper details the mathematical approach undertaken in Ref. [4] and
extends its ndings by employing experimentally obtained lv interfacial characteristics in the
model.
2. Torsional vibrations of the clutch
The vibrations induced by an internal combustion engine are transmitted to the passenger
compartment through the engine mounts and through the driveline. The clutch system, being
mounted between the ywheel and the gearbox, inuences the driveline vibrations and nally the
vibrations and noise perceived by the driver.
The combustion process in engine cylinders induces a torsional uctuation of the rotational
speed of the crankshaft as well as bending of the shaft, which produces axial oating of the
180 D. Centea et al. / Appl. Math. Modelling 25 (2001) 177192
ywheel. The large inertia of the ywheel guards against the high frequency axial, radial and
torsional vibrations of the engine.
The energy necessary for the motion of a vehicle is transmitted by the engine to the
wheels through the ywheel, the clutch system and the driveline. The clutch takes the energy from
the ywheel and transmits it to the driveline. During the engagement process, the friction torque
acts upon the friction surfaces of the clutch as an engaging force for the driveline. A part of the
energy transmitted through the driveline is transformed into other forms of energy by positive
damping eects. If for some reasons the damping becomes negative, a part of the energy transmitted
by the clutch can induce self-excited torsional vibrations of the driveline which can induce judder.
Centea et al. [4] have shown that if the gradient of the coecient of friction with the slip speed
is negative, the damping of the driveline can be positive or negative. If
dl
dv
P
c
F
n
: (1)
Then the damping is positive and the system is stable.
However, if
dl
dv
<
c
F
n
(2)
then the damping coecient of the driveline becomes negative and the vibration system becomes
unstable. The system will exhibit self-excited vibrations at its lowest mode, leading to judder.
The results obtained experimentally in Refs. [1,12,13] corroborate these ndings that for the
negative values of the gradient of the coecient of friction with slip speed the vehicle is more
prone to judder.
3. Description of the clutch type
The clutch studied is a light truck clutch mounted in the gearbox bell housing between the
ywheel and the input shaft , as shown in Fig. 2.
The main parts of this clutch, used for studying the judder phenomenon are situated between
the ywheel and the diaphragm spring and are presented in Fig. 2. The pressure plate is
mounted by the clutch manufacturer to the clutch cover with straps . These straps keep the
plate and the cover rotating with the same speed and also, through their elasticity in the longi-
tudinal direction of the clutch, permit an axial displacement of the pressure plate against the
cover. The friction disc is free to oat between the ywheel and the pressure plate through a hub
spline-jointed to the input shaft of the gearbox. The friction disc is pressed between the pressure
plate and the ywheel by the clamp force F
n
, provided by the diaphragm spring . The clutch
engagement is obtained through the application of the clamp force, provided by the diaphragm
spring when the clutch is mounted onto the ywheel.
In the disengagement process, the force applied by the driver to the pedal is transmitted through
the pedal quadrant to one end of the cable. The other end of the cable is mounted through a
spherical type joint to the release lever . The motion of the cable is transferred to the release lever,
which rotates and pushes the release bearing against the diaphragm spring ngers. The dia-
phragm spring pivots on a fulcrum ring which is riveted onto the cover and the clamp force is
subsequently reduced. The cushion spring (in the friction disc) and the straps pull back the pressure
plate fromthe friction disc. The reducing friction torque permits a progressive braking of the torque
transmitted by the engine through the ywheel to the driveline. The engagement process is similar to
the disengagement process and occurs when the driver decreases the applied pedal force to zero.
D. Centea et al. / Appl. Math. Modelling 25 (2001) 177192 181
4. The clutch judder model
4.1. Inertial elements
The model of the clutch engagement, built in order to study the take-up judder problem, is a
multi-body non-linear dynamic model. The parts incorporated in the model are listed in Table 1.
The clutch pedal cable is dened by two separate parts; one is considered to be the connection
to the lever, called cable_lvr,and the other is the attachment to the clutch pedal quadrant and is
referred to as cable_qua. The translational motion of the two ends of the cable are coupled with a
1:1 ratio, using a motion coupling constraint function. The combination of translational and
rotational motions between the cable (at the lever end) and the housing cannot be directly rep-
resented by any type of joint. In order to represent the real displacements, a dummy part which
acts like a guide for the cable, called cable_guide, has been introduced between the cable and the
housing.
The diaphragm spring is not considered as a part. It is modelled as two non-linear forces
described in Section 4.3 and a coupler dened in Section 4.2.
The model contains the mass and inertia of all the parts. For the purpose of modelling the
engagement process and studying the torsional vibrations of the clutch, the mass is signicant
only for the parts subjected to translational motion. The masses of the parts considered xed to
the ground or having only a rotational motion are not used in the simulation. Their values are
stated only for information (specied in Table 1 in parentheses).
The parts subjected to torsional motion are characterised by their inertia. The inertia of the
dierential has to be reduced to the input shaft of the gearbox according to the rst gear ratio of
3.89 using the following relation
J
red
=
J
i
2
gbx
; (3)
Fig. 2. Main parts of the clutch used for studying the judder phenomena.
182 D. Centea et al. / Appl. Math. Modelling 25 (2001) 177192
where J is the inertia, J
red
the reduced inertia at the input shaft of the gearbox and i
gbx
is the rst
gear ratio.
The inertias of the road wheels and of the vehicle are also reduced to the input shaft using the
rst gear ratio i
gbx
of 3.89 and the dierential ratio i
dif
of 4.11, using the equation
J
red
=
J
i
2
gbx
+ i
2
dif
: (4)
For torsional dynamic analysis, the inertia of the gearbox is divided into two parts. The inertia of
the rst part, representing 1/3 of the total inertia of the gearbox, is included in the inertia of the
input shaft. Therefore, the part called gearbox has only 2/3 of the whole inertia of the gearbox.
The motions of inertial parts in the model introduce dierential equations of motion which are
formulated by the Lagrange's equation for constrained systems. The constraints, themselves,
introduce a number of non-linear algebraic functions.
4.2. Constraints
In multi-body formulation, constraint functions have to be formulated in order to assemble the
mechanism. For this purpose the constraint functions, in the form of joints and joint primitives
have to be chosen in a manner to ensure the designed functional assurance of the clutch and
driveline system. For the clutch connected to the studied driveline and subjected to torsional
vibrations in the engagement process, the constraints which have been chosen for the model are
described in Table 2. The constraints are provided between two inertial elements, referred to as
parts I and J. The constraint type indicates the type of joint. The number of constraints intro-
duced by each joint is given in the table.
For the purpose of modelling the torsional vibration of the clutch, the friction disc and the hub
are regarded as separate parts. The friction disc includes the rubbing surface on the ywheel side.
The rubbing surface on the pressure plate side is considered to be xed to the pressure plate.
Table 1
Inertial parts in the model
No. Part name Mass (kg) Inertia (kg m
2
) Ratio Reduced inertia (kg m
2
)
1 Crankshaft (10) 1 1 1
2 Flywheel (14.5) 0.25 1 0.25
3 Cover (1.6) 0.03 1 0.03
4 Pressure_plate (4.15) 0.04 1 0.04
5 Friction_disc (1.45) 0.065 1 0.065
6 Hub (1) 0.00001 1 0.00001
7 Shaft (1.5) 0.0025 1 0.0025
8 Gearbox (20) 0.002 1 0.002
9 Dierential (20) 0.045 3.89 0.003
10 Wheels (10) 2 3:89 + 4:56 0.0064
11 Vehicle (2900) 210 3:89 + 4:56 0.67
12 Housing (10)
13 Sleeve (0.5)
14 Bearing 0.2
15 Lever 1.5
16 Cable_lvr 0.2
17 Cable_guide ~0.1
18 Cable_qua ~0.2
19 Quadrant ~0.2
20 Pedal ~1.5
21 Ground
D. Centea et al. / Appl. Math. Modelling 25 (2001) 177192 183
Between the two friction elements there is a cushion spring with a non-linear stiness described in
Section 4.3.
The joints used in the model allow the same motion as in the actual clutch. To avoid creating
an over-constrained model some joints may dier from the actual assembly, but the combination
of modelled constraints ensures conformance to clutch system functional integrity. For instance,
the real joint between the lever and the housing is spherical. However, the translation on the
horizontal is also constrained by the curvecurve contact assurance constraints, described in
Section 5. Therefore, the modelled joint between the release lever and the housing, avoiding a
repeat constraint function is the cylindrical joint.
Translational motion occurs between the ywheel and the friction disc during the engagement
process along the main axis of the clutch, as well as a rotation about the same axis. The joint
which allows only these displacements is a cylindrical one. However, some of these constraints are
also imposed by the joints of the ywheel and of the hub. Therefore, the model uses an inplane
joint instead of a cylindrical joint between the ywheel and the friction disc in order to avoid
repeated redundant constraint functions.
The diaphragm spring allows the pressure plate lift when the release bearing is pushed along
input shaft through its contact with the articulating release lever. The relationship between the
two motions is dened by the clutch manufacturer in the pressure plate ``lift-o'' characteristics
(as shown for a typical clutch in Fig. 3). The vertical axis gives the clamp load. In the model, the
displacements of the two parts are connected by a coupler which is dependent on the diaphragm
spring design. The coupling ratio, representing the displacement of the bearing versus the
Table 2
Constraints in the multi-body model
No. Part I Part J Constraint type No. of constraints
1 Housing Ground Fix 6
2 Flywheel Housing Revolute 5
3 Friction_disc Flywheel Inplane 1
4 Hub Friction_disc Revolute 5
5 Hub Shaft Translational 5
6 Shaft Housing Revolute 5
7 Gearbox Shaft Revolute 5
8 Dierential Gearbox Revolute 5
9 Wheels Dierential Revolute 5
10 Vehicle Wheels Revolute 5
11 Cover Flywheel Fix 6
12 Pressure-plate Cover Translational 5
13 Pressure-plate Bearing Coupler 1
14 Bearing Sleeve Translational 5
15 Sleeve Housing Fix 6
16 Bearing Lever Curvecurve 2
17 Lever Housing Cylindrical 4
18 Cable_lvr Lever Cylindrical 4
19 Cable_lvr Cable_guide Translational 5
20 Cable-guide Housing Spherical 3
21 Cable_lvr Cable_pdl Coupler 1
22 Cable-pdl Ground Translational 5
23 Cable_pdl Quadrant Rack-pin 1
24 Quadrant Pedal Fix 6
25 Pedal Ground Revolute 5
26 Crankshaft Flywheel Fix 6
27 Motion Pedal Ground 1

constraints 113
184 D. Centea et al. / Appl. Math. Modelling 25 (2001) 177192
displacement of the pressure plate, can be obtained graphically from the characteristics provided
by the clutch manufacturer. In the current model, a typical value of 4.6 for the coupling ratio is
used. Note that the rst part of the characteristics is caused by the cover deection due to its
stiness. In this part the displacement of the bearing produces a very small displacement of the
pressure plate which is ignored in the clutch judder model.
The release lever articulates in the clutch bell housing. At its mid span, the lever has a at face
which remains in contact with the cylindrical surface of the release bearing at all times. This is
represented by a curvecurve constraint function.
The cable is connected to the pedal via the pedal quadrant. When the pedal is actuated, the
cable surrounds the quadrant. This is represented by a rack-pin type coupling constraint.
The model species two motions: the release motion of the pedal and the rotation of the y-
wheel. The release motion is transmitted in the engagement process to the quadrant, cable, lever,
release bearing, pressure plate and to the friction disc material which, in the model, is attached to
the pressure plate. It starts at the position, where the pedal is totally depressed. The displacement
of the pedal takes 5 s, allowing a translational displacement of the pressure plate by 4.5 mm. The
engagement starts only in the last 0.7 mm of the pressure plate travel, when the cushion spring is
compressed and the induced clamp load produces the friction torque.
Before the engagement, the rotation of the ywheel is imposed by the engine torque. If no
compliance is considered between the engine crankshaft and the ywheel, the rotational velocity
of the ywheel can be considered to be constant. This is of course not the case in practice as
reported by Boysal and Rahnejat [15,16] and Rahnejat [17]. With this assumption, the engine
torque M
e1
is necessary only to overcome the damping torque c
1
_
h. Since in the model no damping
is included between the engine and the ywheel no engine torque needs to be considered before
the engagement process. Hence, the rotation of the ywheel can be dened by the initial angular
velocity of the ywheel, with a value of 1000 rpm (for typical judder conditions).
After the engagement process, the rotation of all the parts between the engine and vehicle
(considered as the inertia connected to the tyres) is imposed by the engine torque.
The aim of the model is to study the take-up judder, so it does not need to include the process
that occurs after engagement. In order to account for a typical 20% reduction in the ywheel
speed during the engagement process, the model considers that the inertia of the crankshaft is
large, so that the combined inertia of the crankshaft and ywheel is higher than the total inertia of
the driveline.
4.3. Force elements in the model
The model is subjected to forces and torques listed in Table 3.
Fig. 3. Clutch system characteristics.
D. Centea et al. / Appl. Math. Modelling 25 (2001) 177192 185
The clamp load is produced by the compression of the diaphragm spring, and acts on both
ywheel and pressure plate surfaces by means of the cushion spring. The relationship between the
compression of the cushion spring and the force created is dened by a non-linear stiness. This
non-linear stiness characteristic is usually provided by the clutch manufacturer.
In order to model the cushion spring stiness, a spline curve is employed. The model of the
cushion spring characteristics, therefore, uses a matrix obtained from the points on the curve. The
corresponding force developed in the model during the translation of the pressure plate is shown
in Fig. 3. The characteristics show that in the model the force is developed only in the last 0.7 mm
of the pressure plate travel that occurs when the pedal is released from the maximum travel,
representing the real compression of the cushion spring.
The diaphragm spring can be modelled as two translational forces: a non-linear force that is
applied between the pressure plate and the cover, and a non-linear force that is applied between the
release bearing and the cover. The straps have a constant translational stiness, given in Table 3.
The release force characteristics show the relationship between the displacement of the pressure
plate and the release force developed by the diaphragmspring ngers. The release characteristics, as
well as the pressure plate lift-o characteristics, have two distinct regions. In the rst region, the
release force increases from an initial value of 50 N to its maximum value which is around 1000 N.
The displacement of the release bearing has practically no inuence on the displacement of the
pressure plate. The former's displacement is caused by the stiness of the diaphragm spring's n-
gers, as well as by the displacement of the contact point between the diaphragmspring and the cover
multiplied by a lever ratio. This is due to the cover stiness (given in Table 3). Since this phe-
nomenon does not aect the friction torque, the release load is therefore, not included in the model.
The main shaft of the gearbox, the driveshaft, the rear axle halfshafts and the tyres are dened
as linear torsional spring dampers. Typical values of these are presented in Table 3. As no
damping values for these structural elements are included in the model, higher take-up torsional
vibration amplitudes but at the same frequency as in the real system are expected.
The characteristics of the springs mounted in the friction disc are usually provided by the
clutch manufacturer. The characteristics show two levels of stiness. In order to represent these,
the model includes a torsional torque which is dependent on the relative angle h between the
friction disc and the hub and is dened as follows:
M = k
1
+ h

k
2
+ h if h < h

;
M = k
1
+ h if h

< h < 0;
M = k
1
+ h if 0 < h < h

;
M = k
1
+ h

k
2
+ h if h > h

;
(5)
Table 3
Force and compliance elements in the model
No. Part I Part J Stiness
type
Stiness Nm/
deg-N/mm
Gear ratio Reduced stiness
Nm/deg-N/mm
1 Hub Friction disc Torsional Non-linear 1 Non-linear
2 Gearbox Shaft Torsional 150 1 0.150
3 Dierential Gearbox Torsional 43.5 3.89 140
4 Wheels Dierential Torsional 270 3:89 + 4:56 0.868
5 Vehicle Wheels Torsional 700 3:89 + 4:56 2.222
6 Pressure-plate Friction disc Axial Non-linear
7 Pressure-plate Cover Axial 50
8 Pressure-plate Cover Axial Non-linear
9 Bearing Cover Axial Non-linear
10 Cover Housing Axial 32 000
186 D. Centea et al. / Appl. Math. Modelling 25 (2001) 177192
where M is the torque, k
1
the torsional stiness of the torsional springs (situated between the
friction disc and the hub) on the negative side of the characteristic curve, when the angle varies
between zero and h

, and k
2
is the torsional stiness on the negative side of the characteristic
curve when the angle is smaller than zero. k
1
and k
2
are the corresponding values on the positive
side of the characteristic curve, h

and h

are the angles where the characteristic changes. The

values of all four stiness components and both angles are dened as input values for the model.
For the description of the friction torque M
f
one has to take into consideration the integration
errors that occur when the relative angular velocity between the ywheel and the friction disc, x,
becomes zero or negative. To alleviate this integration problem a friction model based on Kar-
nopp [8] is employed. This consists of dening a slope which corresponds to the stiction region, as
previously described. The slope s applies when the relative angular velocity between the ywheel
and the friction disc is above a certain value, dened by x
0
. Noting that the maximum value of the
friction torque cannot be higher than the engine torque M
e
, the function used for the analysis is
dened as
M
f
= min(M
e
; 2 + R + l + F
n
) if x < x
0
;
M
f
= s + x if x
0
< x < 0;
M
f
= 0 if x = 0;
M
f
= s + x if 0 < x < x
0
;
M
f
= min(M
e
; 2 + R + l + F
n
) if x > x
0
;
(6)
where M
f
is the friction torque, M
e
the engine torque, x the relative angular velocity between the
ywheel and the friction disc, R the mean friction contact radius, l the coecient of friction and
F
n
is the clamp load. The friction torque, given by the clamp load R + l + F
n
, is multiplied by 2
because there are two friction surfaces. The friction radius R, the coecient of friction l and the
maximum engine torque M
e
are input data to the model.
The slope s is dened as
s =
M
f
[
x=x
0
x
0
: (7)
The disadvantage of using this function is that the slope s, which depends on the stiction limit of
the relative angular velocity behaves like a torsional stiness so that after coupling of the friction
disc and the ywheel the common velocity exhibits an oscillatory response. The amplitude of this
depends on the value of x
0
.
4.4. Friction characteristics
Dierent lv characteristics have been measured experimentally for four types of friction lining
materials. The specic material types cannot be stated due to commercial condentiality. A
polynomial t of the following form has been employed:
l = c
4
v
3
c
3
v
2
c
2
v c
1
; (8)
where l is the coecient of friction, c
1
; c
2
; c
3
and c
4
the coecients and v is the relative velocity
between the two surfaces at the friction radius, measured in m/s. In the current model, the value of
the coecient of friction is introduced in the friction torque function as being dependent on the
relative angular velocity between the ywheel and the friction disc and the mean friction radius R.
Table 4 below lists the aforementioned coecient in Eq. (8) for the four tested friction lining
materials.
D. Centea et al. / Appl. Math. Modelling 25 (2001) 177192 187
4.5. Degrees Of Freedom in the system model
The number of Degrees Of Freedom (DOF) of the multi-body model is obtained, using the
GrueblerKutzbach expression as
DOF = 6(parts 1)

constraints: (9)
The system has 21 parts, listed in Table 1. The number of constraints in the model is 113, listed in
Table 2. Therefore, using Eq. (9), 7 degrees of freedom result. These are the angular displacements
of the ywheel, friction disc, hub, gearbox, dierential, wheels and the fore-aft motion vehicle
body. The crankshaft, clutch cover and pressure plate have the save displacement as the ywheel.
The translational displacement of the actuation trains formed by the pedal, quadrant, cable, lever,
bearing and pressure_plate is not an independent motion because it is governed by motion
statement dened in Table 2 at position 27.
5. Method of formulation
The model is represented by a set of dierentialalgebraic equations (DAE). The dierential
equations of motion for each part in generalised co-ordinates, q
k
; k = 16, are obtained by
Lagrange's equation for constrained systems as
d
dt
oK
o_ q
k
_ _

oK
oq
k
F
q
k

n
k=1
k
k
oC
k
oq
k
= 0; (10)
where
q
k

k=16
= x; y; z; w; h; /
T
:
The dierential equations can be reduced to rst-order. This has been proven to yield an identical
solution to the original constrained DAE [18]. Thus:
F
q
k
= u
k
_ q
k

n
k=1
#
k
oC
k
oq
: (11)
The n constraint functions for the dierent joints, joint primitives, couplers and motions listed in
Table 2 are given as holonomic or non-holonomic constraint functions as
C
k
; _ q
k
oC
k
oq
k
_ _
T
= 0: (12)
The set of DAE are represented in a matrix form as
[J[u
k
; q
k
; k
k
; #
k

T
= F
q
k
_ _
T
; (13)
Table 4
Coecients for lv characteristics of dierent linings
Friction material c
1
c
2
c
3
c
4
A )0.00020 0.0051 )0.0310 0.3663
B )0.00010 0.0034 )0.0224 0.3961
C )0.00010 0.0036 )0.0229 0.3850
D )0.00001 0.0009 )0.0146 0.6032
188 D. Centea et al. / Appl. Math. Modelling 25 (2001) 177192
where [J[ is the Jacobian matrix, u
k
; q
k
; k
k
; #
k

T
the required solution vector in all small time
steps, dt, and F
q
k

T
is the vector of applied forces. The Jacobian matrix is of the following
form:
[J[ =
s
dt
oK
o_ q
k

oK
oq
k
_ _
oC
k
ok
k
_ _
oC
k
oq
k
_ _
[0[
oC
k
oq
k
; _ q
k
o(oC
k
=oq
k
)
oq
k
_ _
[0[
_

_
_

_
: (14)
The solution procedure is fully described in Refs. [17,19].
Fig. 4. Engagement process for friction material A.
Fig. 5. Engagement process for a material similar to A but with a constant coecient of friction l = 0:3663.
D. Centea et al. / Appl. Math. Modelling 25 (2001) 177192 189
6. Simulations results and discussion
Fig. 4 shows the engagement process for the friction material A. Note that as the engagement
process occurs the ywheel rotational speed is reduced due to the increased coupled inertial. The
driven inertia (the drivetrain to the rear wheels and including the vehicle body) exhibits an in-
crease in its angular velocity with an initial oscillatory behaviour, known as judder. The period of
oscillations is 7 Hz. The amplitude of oscillations has a lower value in Fig. 5, where a theoretical
material with a constant coecient of friction with the same value of c
4
is used. This indicates
some damping in the model for the friction material A, which is inherent in the experimentally
obtained data. Therefore, this material reduces the risk of judder.
Fig. 6. Engagement process for material B.
Fig. 7. Engagement process for material C.
190 D. Centea et al. / Appl. Math. Modelling 25 (2001) 177192
Fig. 6 shows the self-excited vibrations that occur in the engagement process using the friction
material B. Between the four friction materials tested in the model this material indicates the
worst judder problem, owing to a negative slope for lv characteristics. The initial oscillations at
7 Hz give way to a growing amplitude and the introduction of other self-excited frequencies. It is
also observed that the damping eect is only but very gradual. When such conditions occur, the
increased amplitude of oscillation is felt by the driver as a tactile sensation under the foot, often
accompanied by a fore and aft motion of the vehicle body.
Figs. 7 and 8, corresponding to the use of friction lining materials C and D, exhibit acceptable
take-up oscillatory behaviour. In fact both these materials exhibit positive gradients for lv
characteristics. These results corroborate on-vehicle testing at Ford and are in-line with the
ndings in Refs. [1,4,10,13].
Acknowledgements
The authors wish to express their gratitude for the nancial support extended to this research
project by the Ford Motor Company.
References
[1] H.J. Drexl, Clutch judder-causes and counter measures, Technische Konferenz SITV 90 (1990) 746.
[2] T.P. Newcomb, R.T. Spurr, Clutch judder, in: International Automobile Congress of FISITA, 1972, pp. 1/161/18.
[3] E.M.A. Rabieh, D.A. Crolla, Coupling of driveline and body vibrations in trucks, in: SAE Conference
Transactions on International Trucks and Bus Meeting and Exposition, vol. 1203, Detroit, 1996, pp. 1726.
[4] D.N. Centea, H. Rahnejat, M.T. Menday, The inuence of interface coecient of friction upon propensity to
judder in automotive clutches, Proc. Instn. Mech. Engrs., Part D: J. Automobile Engrg. 213 (D2) (1999).
[5] R.A. Krenz, Vehicle response to throttle tip-in/tip-out, in: Proceedings of the Surface Vehicle Noise and Vibration
Conference, SAE Transactions, Paper No. 850967, 1985, pp. 4551.
[6] D.J. Fothergill, N. Swierstra, The application of non-linear displacements modelling techniques to an automotive
driveline for the investigation of shunt, VDI Berichte, Nr. 1007, 1992, pp. 163179.
Fig. 8. Engagement process for material D.
D. Centea et al. / Appl. Math. Modelling 25 (2001) 177192 191
[7] A. Farshidianfar, M. Ebrahimi, H. Rahnejat, M.T. Menday, Low frequency torsional vibration of vehicular
driveline systems in shue, in: H. Rahnejat et al. (Eds.), Multi-body Dynamics: Monitoring & Simulation
Techniques-II, Professional Engineering Publishing (IMechE), 2000, pp. 269282.
[8] D. Karnopp, Computer simulation of the stick-slip friction in mechanical dynamical systems, Trans. ASME, J.
Dynamic Systems, Measurement and Control 107 (1985) 100103.
[9] D.A. Haessig, B. Friedland, On the modelling and simulation of friction, Trans. ASME, J. Dynamic Systems,
Measurement and Control 113 (1991) 354362.
[10] R.P. Jarvis, B. Mills, Vibrations induced by dry friction, Proc. Instn. Mech. Engrs. 178 Pt. 1 (32) (196364) 847
866.
[11] Armstrong-Helouvry, Stick slip and control in low-speed motion, Trans. IEEE Automatic Control 38 (10) (1993)
1831496.
[12] H. Kani, J. Miyake, T. Ninomiya, Analysis of the friction surface on clutch judder, JSAE Review, Technical Notes
13 (1) (1992) 8284.
[13] P. Maucher, Clutch chatter, in: Proceedings of the Fourth International Symposium on Torsional Vibrations in
the Drive Train, Baden-Baden, Germany, 1990, pp. 109124.
[14] E.M.A. Rabieh, D.A. Crolla, Intelligent control of clutch judder and shunt phenomena in vehicle drivelines, Int.
J. Vehicle Design 17 (3) (1996) 318332.
[15] A. Boysal, H. Rahnejat, ADAMS generic engine: a tribo-dynamic analysis tool, in: Proceedings of the ADAMS
International Users' Conference in Multi-body Dynamics, Yipsilanti, USA, 1996, pp. 119.
[16] A. Boysal, H. Rahnejat, Torsional vibration analysis of a multi-body single cylinder internal combustion engine
model, J. Appl. Math. Modelling 21 (1997) 481493.
[17] H. Rahnejat, Multi-body Dynamics: Vehicles, Machines and Mechanisms, Professional Engineering Publications
(IMechE), UK, July 1998; Society of Automotive Engineers (SAE), USA, July 1998.
[18] K.E. Brenan, S.L. Campbell, L.R. Petzold, Numerical Solution of Initial-value Problems in DierentialAlgebraic
Equations, North-Holland, New York, 1989.
[19] N.V. Orlandea, A study of the eect of the lower index method on ADAMS sparse tableau formulation for the
computational dynamics of multi-body mechanical systems, Proc. Instn. Mech. Engrs., Part K: J. Multi-body
Dynamics 213 (K1) 19.
192 D. Centea et al. / Appl. Math. Modelling 25 (2001) 177192