Computer Methods in Applied Mechanics and Engineering 89 (1991) 395-418

North-Holland

Rigid and flexible joint modelling in multibody

dynamics using finite elements
A. Cardona
Universidad National de1 Littoral, Argentina

M. Geradin
Universitt de Lit?ge, Belgium, and ONERA, Chatillon. France

D.B. Doan
Action de Recherche Concertke, Universitt de Liige, Belgium

Received 30 August 1990

Revised manuscript received 14 February 1991

This paper presents a finite element formulation of mechanical joints for multibody dynamics. The
augmented Lagrangian procedure is used to express constraints of both holonomic and non-holonomic
types. The method is first demonstrated on the holonomic case of the hinge joint. It is then used to
develop a wheel capable of radial deformation, slipping, breaking and coupled lateral/ torsional
deformations. Both joints are formulated with the objective of being integrated into a multiple purpose
finite element software for flexible mechanism analysis. Two application examples are presented to
demonstrate the effectiveness of the method: the dynamic deployment of a curved reflector in space
and the simulation of a nose landing gear during touch-down.

1. Introduction

A mechanism is a device to convert motion into another; it can be better defined as ‘an
assemblage of rigid or resistant bodies connected together for the purpose of transforming
motion’ [l]. The individual bodies that form a mechanism are called links, and the combina-
tion of two links in contact constitutes a kinematic pair or joint. A mechanism is formed when
at least one of the links of the kinematic chain is held fixed and any one of the other links can
move.
Mechanisms, or multibody systems, inherently possess constraints in the formulation of
their equations of motion: each kinematic pair imposes constraints on the relative motion
between the two bodies in contact through the pair.
Analytical dynamics usually classifies constraints into holonomic and non-holonomic.
Holonomic constraints are formulated as implicit functions of the generalized coordinates q,
and eventually of time:

@(q, t) =o.

00457825191 I$0350 @ 1991 Elsevier Science Publishers B.V. All rights reserved
396 A. Cardona et al., Rigid and flexible joint model&

They express a restriction on the number of degrees of freedom and, therefore, on the set of
possible configurations of the system. Non-holonomic constraints are formulated as implicit
functions of the generalized coordinates q, the generalized velocities 4 and time

Constraints of this kind are such that they cannot be integrated to be transformed to the
holonomic type. They bring a restriction on the behavior of the system, but not on the set of
possible configurations.
This paper focuses on the formulation of joints and, more generally, kinematic constraints
into a flexible mechanism analysis software based on the finite element concept.
The finite element context in which the constraints are formulated has the following
consequences:
1. By resorting to the concept of finite element, kinematic constraints and joints may be
treated in a very modular manner. Just as for structural elements, a fairly extensive library
of mechanical joints may be developed which may describe not only the classical lower
pairs, but also fairly complex non-holonomic kinematic pairs including additional effects
such as elasticity, friction, etc.
2. For stability reasons in the time integration of the dynamic response of multibody flexible
models, implicit methods of solution have been found to provide better performance, due
to the fact that overall motion is generally not influenced by the high frequency contents of
the elastic model.
3. In this implicit context, kinematic constraints contribute to the flexible mechanism model
exactly in the same way as structural elements, since they are formulated in terms of
internal forces on one hand and elementary tangent stiffness and damping matrices on the
other hand.
4. The differential-algebraic nature of the resulting system of equations governing the
dynamics of the flexible multibody system generates an additional important problem [2]:
the lagrangian multipliers associated with the kinematic constraints inject in the model
multiple parasitic eigenvalues with infinite frequency. The damping properties of the time
integration algorithm have to be adapted to this situation in order to provide good
computing performance.
5. The capabilities of flexible multibody dynamics may be greatly enhanced by resorting to a
dynamic substructuring technique whenever a detailed finite element model of some
members has to be included in the multibody model, provided that the substructure may be
assumed to behave in a geometrically linear manner in a convected frame.
The present paper is concerned only by the algebraic formulation of the joints into the finite
element context. The last two aspects of time integration and dynamic substructuring have
already been discussed by the authors in earlier papers [3,4].
This paper is organized as follows. The augmented Lagrangian method is briefly discussed
first to demonstrate how constraints of either holonomic or non-holonomic type may be taken
into account. We develop then the equations of constraint for two specific joints: the first one,
of the holonomic type, is a hinge; the second one is a wheel, a typical non-holonomic joint.
The particular case of the wheel is further improved by relaxing constraints and including
some additional effects: tire flexibility and slipping when the friction limit is exceeded. Both
joints are formulated with the objective of being integrated into a multiple purpose finite
element software for flexible mechanism analysis.
 A. Cardona et al., Rigid and flexible joint modelling 397

Two application examples are shown. Firstly, we present the dynamic analysis of a large
structure to be deployed in outer space. Secondly, the behavior of a landing gear during touch
down is analyzed in order to predict the effects of slipping, breaking and the coupling between
longitudinal and lateral vibrations in the system (‘shimmy’ effect).

2. Formulation of the constrained dynamic problem

Lagrange equations of motion for a dynamic system subjected to both holonomic and
non-holonomic constraints are stated as follows [S]:

(3)

Here, 22 is the Lagrangian of the unconstrained dynamic system and Q, are the non-
conservative forces on the system (they include, for instance, the friction forces which will be
later developed for some joints). Qb are the internal forces of constraint:
h, aqi/aqk if Qi is holonomic;
QL= (4)
hi aq/a(ik if @;is non-holonomic .

QL can be thought of as the generalized force that obliges the system to verify the imposed
constraints.
When employing an augmented Lagrangian approach, the constraint forces take also a
contribution from a penalty term. This penalty term adds some positive curvature in the range
space of aqlaqk, with a significant improvement of convergence [6]. Besides. this term
assures the positive definiteness of the displacements-associated submatrix of the Hessian
matrix, so the only safeguard to be implemented against the appearance of null pivots during
factorization is that terms associated to the Lagrange multipliers should be condensed after the
degrees of freedom participating in the constraint equation. The equations of motion now
have the following form:

(5)

with the constraint forces

(kh, - p@;) aQiilaqk if CD;is holonomic;

Q;= (6)
(kh, - p~Q)a@~/atj, if @;is non-holonomic .

k and p are, respectively, the scale and the penalty factors. Their values should be chosen so
as to adequately balance the equations. We normally compute them as a mean value of the
stiffness of the mechanism.
It is worthwhile noticing that the non-holonomic constraints can also be seen as derived
from a ‘pseudo-dissipation function’ 9,
398 A. Cardona et al., Rigid and flexible joint model@

(7)
which generates the ‘dissipation forces’

z =(k/i,
-pcq)
9) k
2
I
=k@, . (8)

In what follows, we will develop the equations of constraint for several joints. They will be
implemented by following a concept similar to that of a finite element, i.e., each joint is
treated as a separate matrix entity which adds relations between its local degrees of freedom
to the dynamic system. These relations correspond to the internal forces vector and the
tangent stiffness or tangent damping matrices in the finite element terminology.

a@.
with Bj, = 2 if Qi is holonomic , (9)
%
a@,.
1 with Bii = --G
a%
if Qi is non-holonomic ,

where B is the matrix of constraint gradients.

3. Finite element formulation of some particular joints

The next paragraphs describe the equations of constraint for some particular joints [7,8]
and their finite element implementation. We will assume that each joint ties two nodes A, B
belonging to the two links in contact, and that the links themselves can be either rigid or
flexible.
Let {PI, lu2,Pd and {5i7 52, e wo triads of orthogonal unit vectors attached to nodes
&:3> b t

A and B, respectively, at the initial configuration. Let us suppose that we have already
computed the configuration of the system at time t and that we want to compute the new
situation at time t + At. The former is called the reference configuration, and the latter one is
the current configuration.
Let {cc;, & pi} and {&I, gi, c;} be the triads mapped into the current configuration:

The rotation operators R,, R, give the current rotation at nodes A, B.

3.1. Revolute or hinge joint

The hinge joint is modeled by introducing five constraints: three imposing the equality of
positions at the nodes and two fixing the rotations about two directions (Fig. 1).
 A. Cardona et al., Rigid and flexible joint modelling 399

Fig. 1. Revolute joint.

x, = x, , Cc;-5;=0, /A;* g; =o. (llahc)

Equality of positions (lla) is directly imposed through a Boolean identification between the
corresponding degrees of freedom. The other two constraints are treated by the augmented
Lagrangian procedure, for which we have to compute the constraint gradients.
The orthogonality constraints (llb,c) are expressed in terms of nodal parameters

(12)

Their first variation is given by

where SO,, SO, are the variation of rotations in the material frame at nodes A, R. Then, the
constraint gradients matrix results:

1
UP; x 53 --RtBw x 5;)
B= (14)
[ e4(/4 x 5;) -G(P; x 5;) *

The internal forces vector and the tangent stiffness matrix for this element are computed
after replacement of (14) into (6) and (9).
Let us next analyze the physical meaning of the Lagrange multipliers. At equilibrium, the
unit vectors (pi, &, 54) form an orthonormal triad, so that they verify the identities
400 A. Cardona et al.. Rigid and flexible joint modelling

Then, in this case (13) takes the simplified form:

where I,!+, & are the rotation angles about axes p;.;, & at the joint. Then, the virtual work of
the constraint forces at equilibrium is written as

s=v = 69. Q’ = (kh; - p@l)6@; = kh, S& - kA, Se, , (17)

which shows that the Lagrange multipliers conjugated to Q1 and cP~give, respectively, the
internal moments at the joint about axes & and -pi, scaled by the constant k. We note also
that the constraint forces, evaluated at equilibrium, do not depend on the penalization factor p
but only on the Lagrange multipliers.

3.2. Rigid wheel element [9]

This particular type of kinematic joint element imposes the necessary constraints between
one node to which the center of a wheel of radius r and no thickness is attached, and a rolling
surface (Fig. 2). The wheel and the surface are assumed to be rigid, and no slip between
wheel and ground is allowed. It is also supposed that the rolling surface is planar and that the
contact is bilateral. Extension to unilateral contact is direct.
An analysis of the number of degrees of freedom of the joint shows that it has two degrees
of freedom: it can rotate about a direction normal to the plane and about its own axis and it
can translate along the driving direction, but these two latter freedoms are mutually
dependent. In a first approximation, we will suppose that although the wheel has no thickness,
its axis should be parallel to the plane. We will afterwards relax this constraint and allow small
inclination angles.
The constraints imposed by this element are, first, that the distance from the wheel center
to the ground equals the radius of the wheel:

@I= (x,4-xxp)*r)P-r=O, (18)

Fig. 2. Rigid wheel.

 A. Cardona et al., Rigid and jlexible joint modelling 401

where II, is a unit vector oriented upward and normal to the plane p, and xP gives the
coordinates of a point P of the plane. The second constraint imposes the parallelism between
the axis of the wheel and the ground:

where p; = R, p1 is a unit vector normally oriented to the wheel and pointing into the
direction of the angular velocity vector (see Fig. 2).
The third and fourth constraints are non-holonomic, since they involve non-integrable
relations between velocities. They are derived from the consideration of no slipping at the
contact point C. Let p2 be a unit vector pointing into the driving direction at the initial time
instant:

l-5 = Pl x r7p . (20)

Velocities at the point of contact are given in terms of the nodal parameters at the center of
the wheel:

xc = XA -AXb?,), m

where wA is the spatial angular velocity at node A. The condition of no slipping is imposed by
demanding that the components of the velocity at point C into the directions p,, p, equal
zero :

CD3
= (XA- r+?p)* /“$ = $4 ‘&-r”A*~px~,=o,
(22)
yP4= (XA- rdAr)p).ELZ=XA.JUIXr)p-rWA.lll=O,

where we have employed the equation of parallelism at the intial time instant qP * h, = 0 to
obtain (22b).
After differentiating the constraint equations, we get the arrays of constraint gradients B,
and B,,

0 PI
x?I,
VP

1 [ 1
Pl

B, = Bnh =
0 G(P? T,) ’ -Khr R:h x ~,)r ’

Since the equations of constraint (22) express the fact that the velocities into the lateral and
driving directions at the initial time instant should be zero, the Lagrange multipliers A, and A,
give the contact forces into the same directions, scaled by the constant k:

F,, = 4 > FCd= kh, .

These equations can be extended to the case of small inclination angles about the driving axis,
by simply eliminating the equation of parallelism. In such a case the velocity of the contact
402 A, Cardona et al., Rigid and flexible joint modelling

point is only approximately given by

Xc’& - rnc),
x (Yr)J) (25)

and the three equations of constraint (18) and (22a,b) express the wheel motion equations.
We remark that these equations are valid provided the deviation from normality between J.L~
and I,Q is small.

3.3. Wheel element with radial deformation and slipping

This joint element is generated by relaxing the constraints we imposed to make the rigid
wheel. We consider that the distance from the center of the wheel to the ground is a function
of the contact force. Also, we state that if the friction force exceeds the limit given by the
contact force times the friction coefficient, slipping will occur (the friction coefficient being a
function of the slipping velocity). Finally, we account for unilateral contact; that is to say, we
allow the wheel to separate from ground (Fig. 3).
The velocity of the contact point C is now given by

xc = x, - OA ’ hp) - hp > (26)

where the radius of the wheel is no longer considered constant but depends on the radial
contact force, and where oA is the spatial angular velocity of the wheel measured at node A.
The projection of the velocity of the contact point into the vertical direction is zero because
the rolling surface is infinitely rigid; then i = iA * I,I, and the velocity of the contact point can
be written in the following form:

xC=(l-~p@?+A-wA ‘h$,)- (27)

The variation of position of the contact point can be similarly computed:

(28)

Fig. 3. Wheel with radial deformation and slipping.

 A. Cardona et al., Rigid and flexible joint modelling 403

Let us next define the radial deformation

w = r. - Y = r. - (x, - XP) * I), . (29)

Its variation is

SW= -6x, - r), . (30)

The contact forces at xc give rise to the following expression of the internal energy variations,
which is split in two terms: a stick-slip term conjugated to the variation of xc, with
components into the driving and lateral directions (energy dissipated by friction), and the
radial elastic force conjugated to the deformation w (deformation energy):

6”lr=Sx,~F,,+6wFr. (31)

The radial contact force is a nonlinear function of the deformation w. We will assume for it a
polynomial law in the form

a,w + u2w2+ u3w3+ u4w4) w>o,

F,(w) = 0, wso. (32)

Clearly, whenever w < 0, the force is zero (no contact condition).

The friction force F,, is given by means of the regularized function

where
- p( 11iJ) is the friction coefficient, a known function of the modulus of the slipping velocity

-%k/l, E1’is t he regu1arizing function, with E the regularizing parameter. We have chosen
for 4 the function proposed in [lo], which takes the following form:

- e is a unit vector pointing into the direction of slipping:

e = Xc/ IIXc(( . (35)

The regularizing procedure for the dynamic friction problem can generate some difficulties: a
perfect zero slipping velocity between the contacting surfaces is not possible - when ‘stuck’ the
contact surfaces will creep with a relative velocity smaller than E; trying to impose a very small
E will imply the need to reduce substantially the steps of the time integration scheme,
particularly in transitions from ‘slip’ to ‘stick’. It is, however, a very simple and practical
means to introduce the friction effect without resorting to complicated schemes with activation
404 A. Cardona et al., Rigid and flexible joint modelling

and deactivation of constraints, which can in turn pose severe inconveniences in the time
integration procedure - i.e., violent oscillation of constraints and velocities in the transition
from ‘slip’ to ‘stick’.
After replacing (29)-(30) and (32)-(34) into (31), we obtain the 6-components vector of
internal forces Gint:

The tangent stiffness matrix is computed by further differentiation. To this purpose, we will
first compute the variation of the friction force W,,:

The variations on the right-hand side of the latter equation are computed by differentiating
(32), (34)-(35):

After replacing into (37) we obtain

SF,, = ((-Fr+’ - pF,qb’)e@e - $$ (l-e@e)) 6X, + p4Fi(e C3 7,) 6x, .

(3%

The variation of the slipping velocity is obtained from (27) giving

Replacing the latter equation into the expression of the friction force variation (39) we get

(41)

Differentiation of the internal forces vector gives

4, - q&w
AGint = Shq + CA4 = (42)
-d,Ao, + rR;i-&AF,, + R&F,,( vj,. AxA) *
 A. Cardona et al., Rigid and flexible joint modelling 405

After replacement of (41) into (42), we get the symmetric ‘tangent damping matrix’ C:

(43)

with
C,,=-Be-/W-s,@qJ,
C,, = rMe @(e X r),)) - PG,)% y (44)
C o. = +X-4(e X rl,) @ (e X rl,)) + PO - r), @ r7,NR, 7

and the non-symmetric tangent stiffness S:

(45)
where

The friction associated terms produce a non-symmetric tangent stiffness which is charac-
teristic of Coulomb friction; the only term that generates a symmetric matrix is that associated
to the radial deformation of the tire.

3.4. Flexible wheel element with lateral and torsional deformations [II]
The model we have just developed in Section 3.3 does not account for the deformation of
the pneumatic tire, which is of upmost importance when analyzing vehicle behavior. Owing to
this deformation, for instance, a pneumatic tire can roll without sliding even though it
possesses a component of velocity normal to the wheel plane.
A simplified model of tire deformation will be here assumed based on the contact point
theory [12]. In it, the deformation is characterized by two magnitudes: the slip angle +, and the
lateral displacement y. The slip angle is defined as the angle between the wheel plane and the
total velocity vector of the wheel center for a tire rolling without sliding. The lateral
displacement is the displacement of the contact point measured with respect to the wheel
plane.
Figure 4 illustrates the two deformation measures for a pneumatic tire rolling in pure yaw.
The effect of tire distortion is a system of distributed forces between the tire and the wheel,
whose net resultants are the cornering force F acting normal to the wheel plane, and the
aligning torque M, acting about the vertical wheel axis.
406 A. Cardona et al., Rigid and flexible joint modelling

Fig. 4. Pneumatic tire under pure yaw. Fig. 5. Contact point velocity in the flexible wheel.

More elaborated theories have been presented in the literature, some of them based on the
stretched string concept [13]. For the moment, we retain the simple contact point theory and
center our development efforts on establishing a full nonlinear geometric model between the
generalized displacements at the wheel center and the generalized deformations at the contact
point.
We will employ the theory of Moreland [14], later elaborated upon by Collins and Black
[15], that gives constitutive relations between the generalized forces (F M) and the general-
ized deformations ( y +t) for the pneumatic tire. It can be written in the form

{~}=Ao{;,}+4{;}~
t (47)
.where coefficients of the 2 x 2 matrices A,,, A, are determined from experience.
We will next determine the modifications to the model of the preceding paragraph in order
to account for the deformation of the pneumatic tire. Let us first establish the expression for
the velocities at the contact point. From Fig. 5, and by regarding the considerations of the
preceeding paragraphs, it is not difficult to verify that the following equation holds:

xc = XA
. - (1+ Giph.4 x (v, + Y/4> - (Jrl,) - (1+ 4%ig(l;P;) - (h,) (YP;) 7
(48)
where (1 + qt 6,) = R, is the small angles rotation induced by the slip angle J,$and pi = R, p,
is a unit vector normal to the wheel plane.
By considering that the vertical projection of ic should be zero, we note again that
 A. Cardona et al., Rigid and flexible joint modelling 407

i=x,* lJp.After replacing into (48), we obtain the following expression for the velocity of
the contact point:

where p; x qP is a unit vector pointing into the driving direction.

The internal energy variation, which includes the term due to the dissipation by friction,
now contains a contribution from the tire deformation:

67’” = 6x,. - F,, + 6w Fr + Sy F + S+( M . (50)

After replacement of (30), (49) into the latter expression, we get the internal forces vector of
the flexible wheel:
%l Ffr - _?,Fr
so, q-w - 6, - YPl + @fYrlP@ PI)Ff,
6V=6q-Gint = sy * (51)
F+ 6,W x rip - P.;>+,, .
1 wt 11 M+Y(P;x rl,)*F,r 1
Further differentiation allows to obtain the 8 x 8 stiffness and damping matrices. To this end,
the tire behavior model (47) should also be differentiated.
We finally remark that the pneumatic tire contact theory we have employed is a highly
simplified one (some authors have signaled that its coefficients would depend on the rolling
speed, an item that could have been incorporated into the model without any major trouble).
Its main drawback is the difficulty in determining the constants in matrices A,, Al. Neverthe-
less, we point out that the model we have developed accounts for the full six degrees of
freedom at the wheel center, that mass and inertia properties can be included at any point in
the kinematic chain, and that friction/slipping effects are also considered. The model could be
further improved by following a similar theoretical outline, using more elaborated theories for
tire behavior.

4. Examples

4.1. Dynamic analysis of the deployment of a curved reflector in outer space

The first example is a module of an experimental structure of a deployable reflector
designed by Aerospatiale. The part of the structure under study is composed by 6 modules
disposed side by side to form an hexagon (Fig. 6). One cell has a triangular cross-section with
three longerons and six diagonals. Battens are articulated at the middle length to permit
folding. The diagonals are made of a cable equipped with a winder, so that their final length
defines the shape of the parallelograms and builds the curvature of the reflector surface. The
battens and diagonals are hinged to the longerons.
408 A. Cardona et al., Rigid and flexible joint model&g

Fig. 6. Global view of the reflector model.

The motorisation for deployment is provided by the elastic energy stored at rotation springs
located at the hinges in the middle battens. Figure 7 shows the torque versus angle law
describing the behavior of the motor hinges. It is highly nonlinear and provides locking at the
final angle of deployment.
Figure 8 displays the time evolution of the rotation angle and the corresponding angular
velocity at one of the twelve motorized hinges of the system. The time of locking appears very
clearly in this diagram.
Figure 9 displays the time evolution of the length of one of the 24 diagonals of the structure
and the resulting axial stress. It shows clearly the shock induced by the complete uncoiling of
the cable.

GLE

Fig. 7. Motorization law: torque versus hinge angle

 .i..

-+

..i.

.i..

i.
x.
..i. _..
+

“7.

.(..

Fig. 8. (a) Hinge angular displacement versus time. (b) Hinge angular velocity versus time.
 4

Fig. 9. (a) Diagonal length versus time. (b) Axial stress in the diagonal.
 A. Cardona er al., Rigid and flexible joint modelling 411

Figure 10 displays six successive configurations computed up to complete deployment at

time t = 3.65 s.
This simplified model has 756 DO$ and mean bandwidth of 151; 210 time steps are used to
time integrate the equations of motion. The CPU time consumption up to the first locking is
15 hours 50 minutes on an APOLLO DN-4000 workstation.
We remark that the computational effort is largely influenced by the strategy adopted to
solve impacts at the joints, since the time step is restricted by convergence at the shock
instants. The approach currently adopted to solve shocks is based on the hypothesis of
momentum conservation. This approach provides excellent computational efficiency since the
discontinuity of velocities during impact is explicitly computed. Unlike a shock capturing
algorithm in which the stop and locking device is represented as a high stiffness component, it
allows to pursue the simulation beyond complete locking of the structure. At the same time,
this approach leaves the integration scheme unperturbed and therefore allows to increment
the time step to reasonable values.

t q 3.22 s t = 3.43 s
-__

Fig. 10. Successive configurations from 0 to 3.65 s.

412 A. Cardona et al., Rigid and flexible joint modelling

4.2. Simulation of the landing of a nose gear

The landing gear under study is represented in Fig. 11. It is modelized by a set of beams, a
spring-damper that represents the shock absorber and a flexible wheel. The undeformed wheel
radius is rO= 0.3225 m. The load is represented by a rigid body at the top node, with a mass
m = 6.44 * lo3 Kg. All mechanical joints are of hinge type. The variation of the friction
coefficient p in terms of the slipping velocity is shown in Fig. 12. At touch-down, the vertical
speed of the landing gear is U, = -3.05 m/s and its horizontal speed is u, = 60m/s. The
parameters of the relationship between the radial contact force and radial deformation are
provided by the manufacturer [16]. Two simulations are presented in parallel in Figs. 13-17:
case (a) corresponds to a flat rolling surface, while in case (b) the latter is supposed to have a
sinusoidal profile.
Touch-down is immediately followed by two bounces, the first rebound occurring between
t = 0.45 s and t = 0.8 s and the second one between t = 1.1 s and t = 1.2 s. After t = 1.8 s, the
wheel rolls without slipping. The lateral and radial deformations of the tire, y and W, take
their maximal values between the rebound. This behavior may be explained by the non-
linearity of the radial stiffness of the wheel. The lateral force and the radial contact force vary
as y and W. The fact that the locked configuration of the landing gear possessesa large lateral
stiffness permits to annihilate the lateral oscillations. In the case of the sinusoidal ground, the
oscillations around the ‘equilibrium’ vertical position of the wheel center arise from the
ondulation of the ground. The strongest efforts on the landing gear appear during the impact
period. After t = 2.5 s, a braking torque is applied onto the wheel center. An upper limit is set
for the braking torque in order to avoid the undesirable slipping during the braking phase.
This limit is depending on the translational speed of the wheel center with respect to the
ground. The variation law of the friction coefficient during braking is estimated by an
exponential law in terms of the translational speed u:

where k and p,, are experimental data.

b
H

s-o

‘1 Hd

w H
1
X
_-.-.-

Fig. 11. Landing gear finite element model. Fig. 12. Friction coefficient versus slipping velocity.
 ;.

:
/
3
.;
.,..

, ,,

Fig. 13. Vertical position of the wheel center versus time.

 ?x”S
c
I (a)

I
I)SD00 =T-
-
i
i
b-+J
- 1 ,,..

.,.

.I

: ..’ :
.._...............; ..:__ .;....

i....

j.... .:..

p
j... ..j.
....._. ..;.. I.-

Fig. 14. Lateral deformation of the wheel

 -

Yz

,...

,....

Fig. 15. Radial contact force.

416

I ., .:

: L

: ...........

...... : ............... .............

......... .... . ........ I ....... j ......... : ... .......... : ........

........ ..: .....

Fig. 16. Angular velocity of the wheel.

 (a)

Fig. 17. Slipping velocity of the wheel.

418 A. Cardona et al., Rigid and flexible joint modelling

5. Concluding remarks

This paper presented a general systematic methodology to model joints in three-dimension-

al kinematics of mechanisms. Either rigid or flexible joints can be included in multibody
dynamic analysis by following this method. Both holonomic and non-holonomic constraints
were treated, and the constraint equations were generated by using an augmented Lagrangian
technique. A rigid hinge and a wheel were developed using this technique.
Elasticity and friction effects in joints have also been considered by the inclusion of
non-conservative loads into the Euler-Lagrange equations of the dynamic system. A flexible
wheel was developed in this way, incorporating the Moreland model for tire behavior to the
joint.
The joints here discussed form part of the vast library of joints of the multiple purpose
software for mechanism analysis developed at the University of Liege. Reference [17]
describes this library in greater detail.
Two examples of application of these joints have been shown. First, the dynamic analysis of
the deployment of a large structure in outer space. Second, the dynamic analysis of a landing
gear. These examples illustrate the powerfulness to simulate motion and dynamics of complex
mechanical systems of the developed software.

References

[l] R.S. Hartenberg and J. Denavit, Kinematic synthesis of linkages (McGraw-Hill, New York, 1964).
[2] M. Geradin, K.C. Park and A. Cardona, On the representation of finite rotations in spatial kinematics, LTAS
report, University of Liege, Belgium, 1988.
[3] M. Geradin and A. Cardona, Time integration of the equations of motion in mechanism analysis, Comput. &
Structures 33 (3) (1988) 801-820.
[4] M. Geradin and A. Cardona, A superelement formulation for mechanism analysis, LTAS report, University
of Liege, Belgium, 1988; Comput. Methods in Appl. Mech. Engrg., submitted.
[5] L. Meirovitch, Fundamentals of analytical dynamics, in: Methods of Analytical Dynamics (McGraw-Hill, New
York, 1970) Chapter 2, pp. 45-100.
[6] A.J. Morris, ed. Foundations of structural optimisation. A unified approach (Wiley, New York, 1982).
[7] J. Angeles, Spatial Kinematic Chains (Springer, Berlin, 1982).
[8] P.E Nikravesh, Computer aided analysis of mechanical systems, Aerospace and Mechanical Engineering
Department, The University of Arizona, Tucson, 1984.
[9] A.L. Schwab, The wheel in 3D and other special finite elements for kinematic and dynamic analyses of
multibody systems, Laboratory for Engineering Mechanics, University of Technology, Delft, Netherlands,
1985.
[lo] J.T. Oden and J.A.C. Martins, Models and computational methods for dynamic friction phenomena, Comput.
Methods Appl. Mech. Engrg. 52 (1985) 527-634.
[ll] D.B. Doan, M. Geradin and A. Cardona, Modelisation dune roue flexible dans le module MECANO, LTAS
report VF-65, University of Liege, Belgium, 1989.
[12] L.C. Rogers and H.K. Brewer, Synthesis of tire equations for use in shimmy and other dynamic studies, J.
Aircraft 8 (1971) 689-697.
[13] B. Von Schlippe, Shimmying of a pneumatic wheel, NACA TM-1365 (August 1954) 125-147.
[14] W.J. Moreland, The story of shimmy, J. Aeronaut. Sci. 21 (12) (1954).
[15] R.L. Collins and R.J. Black, Experimental determination of tire parameters for aircraft landing gear shimmy
stability studies, AIAA Paper No. 68-311 (1968).
[16] Messier-Hispano-Bugatti, Internal technical notes, 1988.
[17] A. Cardona and M. Geradin, On the formulation of joints for mechanism analysis, LTAS report, University
of Liege, Belgium, 1988.