Non-Linear Modal Rolling Tyre Model For Dynamic Simulation With Adams F. Mancosu, D. Da Re Pirelli Pneumatici Spa - D. Minen MDI Italy
Non-Linear Modal Rolling Tyre Model For Dynamic Simulation With Adams F. Mancosu, D. Da Re Pirelli Pneumatici Spa - D. Minen MDI Italy
Non-Linear Modal Rolling Tyre Model For Dynamic Simulation With Adams F. Mancosu, D. Da Re Pirelli Pneumatici Spa - D. Minen MDI Italy
Presented at European Adams users’ conference – November 18th –19th 1998 Paris
Interfacing a Finite Element program with a i-markers on Flex Body
Large Displacement Multibody Dynamics
program achieves two very desirable goals:
2
The dynamic characterization was necessary to comparing the amplitude of the response at the
describe the frequency-dependent properties, in eigenfrequencies with respect to the static
terms of elastic and hysteretic behavior. The response.
reinforcing materials, such as carcass and belts, The steady state analysis is a linearized response
have been also experimentally tested to achieve to harmonic excitation based on the physical
both the positive part of the stress-strain curve degrees of freedom of the model. The response
and the negative part, using a special can be achieved through the direct solution of the
methodology and device set up by Pirelli. In equations in matrix form, where the stiffness and
figure 3 is shown an example of the dependence damping matrices are updated each frequency
of the elastic modulus from the frequency. since the definition of the isotropic linear
viscoelasticity allows evaluating stiffness and
damping matrices as function of the frequency.
TREAD COMPOUND PROPERTIES While the response in this analysis is for linear
6.4
vibrations, the prior static response is strongly
6.2 nonlinear and the initial stress effects (stress
6
stiffening) are included in the steady state
response.
[MPa]
5.8
The finite element model can give any kind of
ELASTIC MODULUS [MPa]
5.6 data about any nodes of the model itself but it is
5.4 a huge model, unable to be used for time domain
5.2
dynamic simulations. As a consequence a strong
10 20 30 40 50 60 70 80 90 100 110
FREQUENCY [Hz]
120 130 140 150 160 170 180 190 200
reduction of the physical degrees of freedom is
needed. Since the tyre acts as an interface
Fig. 3: variations of the elastic modulus of a
between the road and the hub it is advisable to
tread compounds with the frequency.
retain just the degrees of freedom of the nodes
involved in the foot print area and the degrees of
The reinforcing material was implemented inside
freedom of the rim, which has to be linked to the
the rubber element as additional stiffness as
hub.
function of the properties of the cords
themselves. The rubber compound properties
have been implemented using a hyperelastic
formulation (Mooney Rivlin).
The finite element model is made up of 20
thousands of brick elements (linear shape
functions, 8 nodes each element) that give
something like 80 thousands degrees of freedom.
The model was generated in an axisymmetric
way and it has been demonstrated that requires at
least 48 wedges to be able to properly evaluate
the eigenfrequency up to about 150 Hz. In this
specific case an 80 wedges model was built to
investigate higher frequencies, up to 400 Hz. The
higher is the maximum frequency of interest the Fig.4: Example of mode shape, which occurs at
finer should be the mesh of the model and the 131.4 Hz.
longer is the computational time. In figure 4 is
shown an example of mode shape at 131.4 Hz. The reduction of the degrees of freedom can be
To represent the real static working conditions of made, once known the stiffness matrix and all the
the tyre, it was carried out a preliminary set of applied loads, through a rearrangement of the full
analysis like the mounting on the rim, the stiffness matrix. This procedure uses no
inflation and the deflection with a specified approximations. The nonlinear response, due to
vertical load. These analyses show strong non- the materials and the contact between tread and
linearity due to the contact kinematics and to the road surface, which occurs during the static pre-
materials. After these analysis some steady state analysis, has been defined as pre-load prior to the
analysis was performed in order to calculate the elimination of the degrees of freedom. As a
modal damping associated to each eigenvalue, consequence it has took into account the
3
stiffening behaviour by pre-loading in a The tread elements will enter and leave the
geometrically nonlinear analysis: stress stiffening contact patch with a law that depends on the
effects has been included when the rearranged pitch sequences.
stiffness matrix was created.
While the reduced stiffness matrix uses no In this case, the tread has been discretized with
approximations, the mass matrix needs more “n” elements and their deformation and position
degrees of freedom than just the ones, which in the contact patch is considered only at discrete
refer to the nodes, involved into the contact time intervals. For example, in the tangential
patch. Therefore some eigenmodes amplitude direction, at a time interval ∆t, the position of the
was restrained as additional degrees of freedom: tread element increases by ∆s, and the
in this way the choosing of the retained degrees deformation by ∆u. Denoting with Vr the rolling
of freedom is no longer critical in order to get an speed, the following relationships subsist:
accurate mass matrix representation.
∆s = Vr∆t
3.2.2 The brush model: rolling-tyre.
Discretisation of the tread pattern ∆u = Vsx∆t
The road unevenness excites the tyre dynamics,
and slipping phenomena in the contact patch can where Vsx is the slip velocity.
arise. For the simulation of these phenomena in For the generic tread element i in the contact
the foot print area, in lateral and tangential zone, the position and the deformation change
directions, a discrete brush model has been during the rolling:
implemented. This model takes into account the
main characteristics of the tread: slip stiffness, si (t + ∆t) = si (t) + ∆s
pitch sequences and size of the contact zone.
ui (t + ∆t) = ui (t) + ∆u
The discrete brush model is represented as
individual elastic elements linked to the tyre belt. The deformation multiplied for the tread element
Tangential and lateral slip velocities can result stiffness Ccpi gives the local longitudinal force:
from the tyre dynamics, and consequently the the sum of these local forces over the contact
tread elements will have a shear deformation. patch gives the total longitudinal force at a time t
The maximum deformation of the tread elements :
is limited by the friction coefficient between the
tyre and the road. n
Nevertheless the values of longitudinal and Fc = ∑ Ccpi ⋅ ui
lateral slip, due to the roughness of the road, are i =1
pretty small; therefore all brush elements in the
contact patch adhere to the road surface. The tread elements stiffness Ccpi has been
determined by an internal Code (user subroutine
The elastic forces, which arise in the contact written in Fortran and implemented in ADAMS),
zone, depend on the slip stiffness, in tangential that is able to calculate those stiffness directly
and lateral directions, and the shear deformation. analyzing the tread pattern in the contact zone.
These forces are function of the contact patch The same procedure applies to the tangential
size, the tread pattern and the elastic modulus of force to evaluate the lateral force. In this case,
the tread compound. the tread element deformation is due to the slip
In particular the size and shape of the contact velocity Vsy in the lateral direction.
depend on the tyre structure; the stiffness of the Therefore the simple modal model can simulate
tread elements is affected from the pitch the rolling conditions when linked to the discrete
sequences of the tread pattern. Finally, the slip contact model. In fact, the modal model
stiffness of the tread elements also depends on reproduces only the eigenfrequencies and
the tread compound. eigenmodes of the non-rolling tyre. All the
dynamic effects affected from the forward
According to the Lagrange method, we follow velocity are neglected, such as, for example, the
the tread elements from the entry to the exit of increase of the first longitudinal mode relative
the contact. In fact, just the elements in contact damping with the velocity of the of the tyre
with the road will generate forces during rolling.
4
closed to 35 Hz. As well known, the relative estimates of dynamic behavior. We may reduce
damping variation depends on the slip our solution set by partitioning the f-set into the a
phenomena in the contact patch. (analysis)-set and the o-set (omitted degrees of
For these reasons a discrete brush model (fig. 5) freedom):
has been used as tyre-road interface of the modal
model. M aa M ao u&
&a Baa Bao u&a
M + +
oa M oo u&
&o Boa Boo u&o
(2)
ω K aa K ao u a Pa
∆u K =
oa K oo u o Po
Vsx
Xi Though equally valid for non-superelement
Vx (residual structure only) models, eq. (2) is most
Xe
ξ a
familiar in the superelement context, with the a-
set frequently referred to as the exterior, or
boundary, degrees of freedom and the o-set, the
Fig.5: discrete brush model interior.
or,
4. Component mode synthesis u o = G oa u a + u oo (4)
The following chapters describe how the flexible
body is taken into account in the ADAMS
environment Note that the solution for the u o interior degrees
of freedom consists of two parts: the
4.1 The CMS approach (Component Goa u a response to boundary displacements, and
Modes Synthesis)
A brief review of dynamic reduction techniques the u oo , or fixed-boundary solution to interior
in FEA code follows: this section will present loading. The static condensation results in eq.
Guyan Reduction, Generalized Dynamic (4) suggest a framework for approximating the
Reduction (GDR) and Component Modal coupled dynamic equations in (2).
Synthesis (CMS).
Let’s define the f-set, the degrees of freedom A consistent way of presenting the various
which are unconstrained (free) in dynamic approximate dynamic reduction techniques is
analysis. These are what remain of the g-set (all through the use of symmetric transformations.
of the structural, or grid, degrees of freedom) We can introduce the transformation:
after removal of the s-set (degrees of freedom
eliminated by single point constraints) and the m-
set (degrees of freedom eliminated by multipoint {u } = uu
f
I aa
=
a o u a
I oo u oo
(5)
constraints). The f-set equations of dynamic
o Goa
equilibrium are
which is just the static condensation in matrix
M ff u&
&f + B ff u&f + K ff u f = Pf (1) form. Eq. (5) and its time derivatives can be
used to transform eq. (2), which, ignoring
In dynamic analysis, we often have more finite damping, is:
element data than is necessary to obtain adequate
5
′ M ao
M aa ′ u&
&a K aa o ua Pa′ "component" modes.1 Use of the coordinate
M ′ M ′ u& +
o = (6)
Koo uoo Po
transformation (9) yields a set of compact,
oa &
oo o o stiffness-uncoupled, equations of a-set dynamic
equilibrium. Since the basis vector set of (9) is
linearly independent, one possible solution
The advantage of eq. (6) is that the dynamic technique is to first orthogonalize the set and
reduction techniques in FEM code can all be then solve the resulting uncoupled equations.
conveniently explained in terms of their
corresponding u oo approximations.
4.2 FEA data conversion to Adams
Guyan Reduction simply assumes: interface
4.2.1 Integration of component flexibility in
u&
& &&
o = Goa u a (7)
ADAMS
A high level goal when implementing flexible
bodies in ADAMS was that a flexible body could
or, u&
&o ≡ 0 . The resultant upper partition of eq.
o
be integrated into a mechanism in a way similar
(6) can thus be immediately solved for the a-set to a rigid body and interact with the mechanism
degrees of freedom. through ADAMS joints and forces.
Generalized Dynamic Reduction uses Early in the development cycle, the need for
Component Mode Synthesis became evident.
approximate mode shapes to approximate the u oo Attempts to model the effect of attachments to
degrees of freedom (Experimentally obtained the flexible body using only component
mode shapes could be used just as well.). eigenvectors required an extremely large number
Component Modes offer a further logical of eigenvectors to be considered. While the
extension by using the o-set eigenvectors to ADAMS implementation of modal flexibility is
approximate u oo . We can simply write: general enough to accept any kind of mode
shape, the FEM code interface has been set up to
export the computationally-determined
u a I aa o u a component modes.
= (8)
u o Goa φ oq u q
Though c-set degrees of freedom are provided in
component modes synthesis, they’ve typically
not been used when generating modes for input
The q-set has been introduced to represent to ADAMS/Flex. (Though their use holds
generalized degrees of freedom in dynamic promise.) Since the resulting simplification
analysis. The q-set is included as a partition of yields the familiar Craig-Bampton modes, this
the a-set, and the t-set partition of the a-set is section will refer to them as such. There have
used to more clearly identify the "total" physical been certain challenges however:
degrees of freedom on the boundary, hence:
1.) Embedded in the Craig-Bampton modes are 6
u a′ I tt o u t rigid body modes. Since ADAMS provides its
= = [Γ ]{u a }
φ oq u q
(9) own large displacement rigid body motion, these
u o G ot modes need to be removed from modal basis.
6
controlled to ensure numerical integration ADAMS solver, the subset that corresponds to
robustness. those nodes where connections have been made
or where forces are being applied. This allows
3.) Craig-Bampton constraint modes can not be the solver to satisfy boundary conditions at
safely disabled without imposing an unacceptable connections and to project the applied load on
constraint effect between the boundary degrees- the mode shapes.
of-freedom.
4.2.2 Used FEA-ADAMS Interface
Orthogonalizing the Craig-Bampton basis of The used interface is based on specific ABAQUS
modes eliminated all of these problems. All instructions and an external translator utility
modes get an associated frequency and the rigid function. After ABAQUS run, the provided
body modes show up as zero frequency modes output (.fil) contained generated superelement
and can be easily disabled. A user can choose to data for the flexible component,; it has been used
enable or disable modes for the dynamic a good beta version of Abaqus translator ( By
simulation on a mode-by-mode basis. HKS) that converts this output into a form
Let’s start to describe of how inertia, stiffness, suitable for ADAMS.
damping and mode shapes of the flexible body Output quantities for each superelement include:
are handled in ADAMS. • Node data
Inertia: • Element connectivity
The mass matrix of a flexible body in ADAMS is • Constraint data
not constant. As the flexible body is deformed, • Physical mass
the center of mass shifts, and the inertia tensor • Modal mass and stiffness
changes. These effects are accounted for in • Component modes
ADAMS by formulating the mass matrix in terms The above allows a complete characterization of
of inertia invariants which are computed in a pre- the mass and stiffness properties of a part in
processor. The large translational DOFs, the terms of its modal components (and, of course,
large rotational DOFs and the modal DOFs are physical mass), as well as graphical display of
coupled through this mass matrix. This is the the part itself (via grid and element data) within
mechanism by which spinning gives rise to ADAMS.
deformation, etc.
ADAMS uses a special flexible body description
Stiffness: file called the Modal Neutral File (MNF) to
The generalized stiffness from the finite element communicate with a variety of Finite Element
analysis is diagonalized and used directly by Programs. It has been used a beta version of a
ADAMS. translator (by HKS). The translator extracts node
locations, element connectivity, nodal mass
Damping: information mode shapes and the corresponding
ADAMS uses modal damping specified generalized mass and stiffness from the .fil file
separately by the user as a fraction of critical and deposits this information in the MNF. In
damping. Damping can be specified on a mode- addition to formatting the MNF the routine
by-mode basis. Users are encouraged to use provides the orthogonalization of the component
damping to control modal response. In other modes
words, it is recommended that rather than Moreover it has been used a MDI toolkit: a set of
disabling a mode, because it is assumed to lie library functions suitable for reading, modifying
outside the frequency range of interest, that the and writing the MNF platform: the User can
mode should instead be critically damped. This modify the original .mnf file performing mesh
will eliminate the dynamic response of this mode simplification and computing the inertia
while allowing ADAMS access to it to satisfy invariants.
boundary conditions.
4.3 Next improvements
Mode shapes: The FLEX_BODY in ADAMS has deformations
After using the mode shapes to compute the that are described via a superposition of mode
inertia invariants, the modes do not contribute in shapes. In the current implementation, the modes
their entirety to the ADAMS simulations. Only a
subset of each mode shape is passed to the
7
that make up the modal basis of the configuration shape, with a linearized stiffness
FLEX_BODY are assumed to corresponding to the one at the end of step 3.
have been obtained by linearizing the flexible
body about an unstressed configuration, as is 5 Adams implementation of the sub-
customary in linear finite element analysis. models
This implementation is obviously a limitation for
5.1 Flexible Body
users of nonlinear finite element analysis codes
In order to implement into ADAMS the Pirelli
like ABAQUS. An ABAQUS user is likely to
Flexible Tyre model, it has been developed a
wish to linearize the nonlinear finite element
first release of a customized ADAMS/View
model about an operating point which is different
interface, including macro and command files.
from the undeformed position. In this particular
From the mnf file (after fil2mnf routine),
case, as the tyre comes into contact with the
ADAMS reads all modal data to build up a
ground, it reaches a state of nonlinear
Flexible body, but it doesn’t read the Ids of the
deformations which ABAQUS can use as an
interface nodes. Providing an external file which
operating point for the linearization.
contains the hardpoints (nodes) list of the
The source of this limitation is that a
Flexible Body, the implemented macro
linearization about a stressed state contains an
automatically creates markers on nodes (hub and
associated modal preload which currently the
foot-print area) and all the objects for Road-Tyre
MNF format does not accomodate and
interface: dummy parts, joints, motions, impact
ADAMS/Solver does not account for.
forces, state variables, design variables,
The following ingredients are missing to allow
measures, etc.
ADAMS and ABAQUS to correctly
A modified Flexible Body dialogs box (see fig.
communicate and process flexible components
6) permits the User to select the Node list for the
represented by modes obtained in a stressed
Flexible Tyre and to decide where to attach the
state:
Tyre (ground, chassis, etc.)
In order to get kinematics information from
1. ABAQUS must export node locations to the
ADAMS analysis and to correctly define the
MNF that correspond to the deformed
brush model of the tyre an equivalent rigid tyre
locations of the nodes, not the input
(having equal mass and inertia information,
configuration;
location and orientation) has been created. It is
2. The MNF format must be enhanced so that it
joined to the chassis through a Revolute joint
can account for generalized forces
plus a Motion or an Applied Torque, in
associated with the deformed configuration
dependence of type of the performed analysis
and ABAQUS must compute these forces
(deflection, cleat tests, running on uneven road,
and store them in the MNF;
etc.).
3. The ADAMS/Solver must be enhanced to
The resulting model after the execution of the
account for this preload.
macro is represented in figure 7: it is possible to
see the rigid part coaxial with the flexible body,
Note that any future removal of this limitation
the markers in the footprint, the forces and
will probably keep a requirement that the
motions applied on the markers.
linearization is performed at an operating point
that corresponds to a static equilibrium.
To overcome this limitation, the current
simulation in FE is done accordingly to what is
described in section. This means the Tyre model
goes through the following sequence of
calculation steps:
1. inflation at the nominal pressure;
2. squeezing against a rigid surface which
represents the road;
3. static concentrated loads on contacting
nodes based on resulatnt forces coming from
2
At this point the Mode extraction is requested.
The model is ouput in the undeformed
8
can decide the deflection time and the deflection
amplitude.
6 Models Validations
6.1 Generality
It has been modeled a 195/65 R15 tyre size.
Fig.6: customized dialogs box for the automatic
The below paragraphs show the different models
creation of the Flexible Tyre Model
validations: at first the proposed tyre model static
and modal validations, the
The following paragraphs show the validation of
the model running over: a) single obstacle, b)
single obstacle, c) real road.
9
Deflection curves 6.3 Single obstacle
5000 The obstacle is a single triangular obstacle high
4500 15 mm and long 20 mm.
4000 The simulation has followed the “basic curve “
3500
approach.
3000
Figures 11 and 12 show the experimental data,
ABAQUS
2500
ADAMS i.e. the basic curves: vertical and longitudinal
2000
force variations in fig. 5.5, and Rolling radius
1500
variation in fig. 5.6. Those data refer to a vertical
1000
load of 450 kg.
500
0
0 2 4 6 8 10 12 14 16 18 20
500
mm
400
300
vertical force
-200
longitudinal force
Comparison between the Eigenfrequency calculated by ABAQUS and ADAMS
-300
160
-400
140
-500
position [m]
120
100
Fig.11: vertical and longitudinal forces at 425
80
ABAQUS Kg (basic curves measured at 3 km/h)
ADAMS
60
40
20
20
15
0 10
rolling radius variations [mm]
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
N° Modo 5
0
0 0.05 0.1 0.15 0.2 0.25 0.3
-10
obtained by Abaqus and ADAMS.
-15
Footprint nodes free. -20
-25
position [m]
180
Fig.12: rolling radius variations at 425 Kg
160
(basic curves measured at 3 km/h)
140
120
100
80
ABAQUS
ADAMS
Figures 13 and 14 report the Adams simulation
60
and experimental results, regarding the cleat test
40
at fixed hub and at 30 km/h.
20
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Number of modes
10
1500
7 Conclusions
calculated
1000 experimental An automatic procedure to create a modal model
tyre is presented. The linking of this model with
500
a multibody vehicle model is very simple
vertical force [N]
1000
0
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16
-1000
-2000
-3000
-4000
time [sec.]
11