Computers and Structures 87 (2009) 913929

Contents lists available at ScienceDirect

Computers and Structures

journal homepage: www.elsevier.com/locate/compstruc

Dynamics of spatial exible multibody systems with clearance

and lubricated spherical joints
Qiang Tian a, Yunqing Zhang a,*, Liping Chen a, Paulo Flores b,1
a
Center for Computer-Aided Design, School of Mechanical Science and Engineering, Huazhong University of Science and Technology Wuhan, Hubei 430074, PR China
b
Department of Mechanical Engineering, University of Minho, Campus de Azurem, 4800-058 Guimares, Portugal

a r t i c l e i n f o a b s t r a c t

Article history: A computational methodology for analysis of spatial exible multibody systems, considering the effects
Received 22 October 2008 of the clearances and lubrication in the system spherical joints, is presented. The dry contact forces are
Accepted 30 March 2009 evaluated through a Hertzian-based contact law, which includes a damping term representing the energy
Available online 25 April 2009
dissipation. The frictional forces are evaluated using a modied Coulombs friction law. In the case of
lubricated joints, the resulting lubricant forces are derived from the corresponding Reynolds equation.
Keywords: An absolute nodal formulation is utilized in exible body formulation. The generalized-a method is used
Clearance
to solve the resulting equations of motion. The effectiveness of the methodology is demonstrated by two
Spherical joints
Lubricated joints
numerical examples.
Reynolds equation Crown Copyright 2009 Published by Elsevier Ltd. All rights reserved.
Generalized-a method
Absolute nodal coordinate formulation

1. Introduction few years. Indeed, a number of theoretical and experimental works

devoted to the dynamics research of multibody mechanical sys-
Ideal kinematic joints in any mechanical system allow relative tems with realistic joints has been published recently [10]. Some
motion between the components. Due to the manufacturing toler- of them focus upon the systems in which only one joint is modeled
ances, wear or material deformations these joints are not perfect as a realistic joint. Liu et al. [11] developed a simple contact force
and inherently have some clearances. These clearances affect the formulation of the spherical clearance joints in multibody mechan-
dynamic response of the system and eventually lead to important ical systems, using the distributed elastic forces to model the com-
deviations between the projected behavior of the mechanisms and pliant of the surfaces in contact. Flores et al. [12] also presented a
their real outcome [14]. The contact-impact phenomenon associ- methodology to assess the inuence of the spherical joint clear-
ated with clearances directly affects the service life of a mecha- ances in spatial multibody systems. Both of these approaches are
nism, since it is usually the cause of vibrations, noise and wear, only valid for the case of dry contact between the socket and ball.
resulting in an increase of maintenance costs [57]. The attenua- Flores et al. [13] extended their work to include a general method-
tion of the impact response and of the vibration characteristics in ology for modeling lubricated revolute joints in constrained rigid
industrial machines is obtained by including treatment of joint multibody systems. Based on neural network and genetic algo-
clearances in their design. The presence and treatment of clearance rithms, Erkaya Uzmay [14,15] investigated the inuence of joint
in joints is a complex and important problem in the realistic mod- clearances on the mechanism path generation and transmission
eling of multibody systems. This is a key issue due to the demand angle. In this case, the joint clearances are modeled as a massless
for the proper design of the real joints in many industrial applica- virtual link. The thermal effect of high temperature on mechanisms
tions [810]. having joint clearances was studied by Bing and Ye [16]. They dem-
The serious consequences of the joint clearances on the dy- onstrated that by increasing the combined clearance of the revo-
namic behavior of the mechanical systems, along with the subject lute joint, a favorable reduction of the friction torque between
of the clearance physical representation, have stimulated and at- the bush and shaft can be obtained for normal operation conditions
tracted the attention of a large number of researchers over the last of a reheat-stop-valve mechanism. However, if the steam temper-
ature reaches to 650 C, the friction torque between the bush and
the shaft becomes larger and the sticking phenomenon exists. In
* Corresponding author. Tel./fax: +86 27 87547405.
[17], a methodology, based on the Archards model, was presented
E-mail addresses: tianqiang_hust@yahoo.com.cn (Q. Tian), zhangyq@hust.edu.cn
(Y. Zhang), pores@dem.uminho.pt (P. Flores). for studying and quantifying the wear phenomenon in revolute
1
Tel.: +351 253510220; fax: +351 253516007. joint clearances. The results of this study indicate that the wear

0045-7949/$ - see front matter Crown Copyright 2009 Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.compstruc.2009.03.006
914 Q. Tian et al. / Computers and Structures 87 (2009) 913929

phenomenon is not uniformly distributed around the joint surface, are based on the traditional nite element theory, in which the sys-
owing to the fact that the contact between the joint elements is tem is assumed to undergo small deformations and rotations. It has
wider and more frequent in some specic regions. been demonstrated that the traditional nite element method can
One of the most important issues for the dynamics research on not describe the behavior of exible multibody system with large
the multibody systems with joint clearances is how to model the deformation and rotation conguration with accuracy [30]. Sha-
contact-impact phenomenon. Many researchers have proposed dif- bana [31,32] proposed a simple procedure, the absolute nodal
ferent kinds of models such as KelvinVoigt model [5,6], the three coordinate formulation (ANCF), that employs the denition of the
modes model [18], the model based on the restitution coefcient slopes to dene the element coordinates instead of the innitesi-
and momentum method [19], but they all have their own disad- mal and nite rotations. As the locations and deformations of the
vantages. For example, for the KelvinVoigt model it is difcult material points on the nite element are dened in the global coor-
to quantify the parameters of the spring and damper elements dinate system, using the element shape functions and the nodal
and it inherently does not represent the nonlinear nature of the coordinates, the mass matrix within the system equations remains
contact-impact process. While the model based on the restitution constant and the centrifugal and Coriolis forces vanish [33]. Using
coefcient and momentum-based method assumes that no change the absolute nodal coordinate formulation, the beam, plate, and
in the realistic conguration between contacting bodies occurs shell elements can be treated as isoparametric elements and used
during the period of contact-impact, which is not true for joint to describe the exact modeling of the body dynamics. In addition,
clearances. Lankarani and Nikravesh [20] developed a continuous the description of the joint constraints and forces becomes much
contact force model with hysteresis damping for impact in multi- simpler based on the ANCF. The ANCF has been considered a
body systems. The model uses the general trend of the Hertz con- benchmark in exible multibody dynamics development [34,35],
tact law, in which a hysteresis damping function is incorporated and has been used by several other authors in different contexts
with the intent to represent the energy dissipated during the im- [3638].
pact. Based on this continuous contact model, Flores and Ambrsio In this paper, dynamic modeling and analysis of exible spatial
[1] studied the dynamic characteristics of multibody systems with multibody systems with clearance and lubricated spherical joints
joint clearances. More recently, this continuous contact-impact is investigated. The absolute nodal coordinated-based spatial beam
model has also been successfully used to carry out the wheel-rail element is used to derive the equations of motion for the exible
contact-impact dynamics research [21]. multibody systems with spherical joint clearances. A continuous
It is important to note that all the above mentioned studies contact-impact model is used to evaluate the intra-joint contact-
have been done based on the rigid body dynamics principle. The impact forces, in which the energy dissipation in the form of a hys-
deformations associated with the exibility of the bodies need to teresis damping is considered. The frictional effect is included
be studied while may lead to different dynamics characteristics, based on a modied Coulombs friction law. The lubrication effect
especially as far as the large deformations are concerned. Thus, in the spherical joint clearance is included by using the Reynolds
some studies have included the inuence of the exibility of the equation. To dissipate the spurious high-frequency responses and
bodies in the dynamic performance of multibody systems contain- to preserve the low-frequency responses, the generalized-a meth-
ing gaps in the joints. Dubowsky and Gardner [22] developed a dy- od is utilized in the simulations. The paper is organized as follows.
namic model for an elastic linkage mechanism with clearance In Section 2 the spatial ANCF-based beam element is briey revis-
connections by Lagrangian approach. This model is quite compre- ited. Full description of the spatial joints with clearance for the dry
hensive and complicated and needs to be simplied for possible contact situation is presented in Section 3. A model for the treat-
numerical implementation. Dubowsky et al. [23] applied the per- ment of lubricated spherical joints is provided in Section 4. Section
turbation method, treating the angular motion caused by clear- 5 deals with the computational strategy for the assembly and solu-
ances and elastic deformations as small quantities and neglecting tion of the equations of motion for the corresponding constrained
the high order and high-frequency responses of the small variables. exible spatial multibody systems. Section 6 presents two numer-
This method simplies the dynamics equations, but is not suitable ical examples, namely a spatial double pendulum and a spatial four
for the conditions of large elastic deformations or large clearances. bar mechanism, which incorporate ideal, clearance, and lubricated
Dubowsky and Moening [24] obtained a reduction in the impact joints. To verify the validity of the model for the stiff systems, the
force level by introducing exibility of the bodies. They also ob- results based on the absolute nodal coordinated-based model are
served a signicant reduction of the acoustical noise produced by compared with the results obtained by using the MSC.ADAMS soft-
the impacts when the system incorporates exible bodies. Kakizaki ware. Finally, in the last section the main conclusions from this
et al. [25] presented a model for spatial dynamics of robotic manip- study are drawn, in the light of the assumptions and procedures
ulators with exible links and joint clearances, where the effect of underpinning this research, and the perspectives for future re-
the clearance is taken into account in controlling the robotic sys- search are then outlined.
tem. Bauchau and his co-workers [26,27] also established a sys-
tematic approach considering the exibility of the bodies based
on the nite element method and lubrication conditions. Liu and 2. Spatial ANCF-based beam element
Lin [28] studied the dynamic performance of planar exible mech-
anisms with revolute joint clearances based on the nite element The main purpose of this section is to look at the most funda-
method. The lubrication effect in joints was also considered in this mental aspects of the spatial ANCF-based beam element. The abso-
study. Shiau et al. [29] presented a dynamic analysis of a 3-PRS ser- lute nodal coordinate formulation has been developed for
ies-parallel mechanism including exibility of the bodies, clear- modeling dynamic systems of large-displacement and large-rota-
ances and friction. They demonstrated that the dynamic response tion problems in exible multibody systems, because the conven-
of the system degrades as the joint clearances increase, and the tional nite element method deals with the small-displacement
joint contact forces increase with the joint clearances and friction problems. The rst spatial beam type nite element developed in
coefcient. Thus, proper modeling of joint clearances in multibody the context of the ANCF [39,40] was parameterized as a volume.
systems is required to achieve better understanding of the dy- It has been shown that this element suffers from some numerical
namic performance of the machines and mechanisms. problems such as shear or Poisson locking effects and, nowadays,
Almost all of the aforementioned studies, on the dynamics per- there are several implementations of this element that have
formance of the exible multibody system with joint clearances enabled prevention of these problems [41,42]. However, the
 Q. Tian et al. / Computers and Structures 87 (2009) 913929 915

rj,y
ri,y
j rj,x Socket
i
ri,z
ri,x rj,z
Ball
eij
j
ri t
Y
rj P
Q

X
Y ri n
rj rP
Z rQ
Fig. 1. Spatial ANCF-based beam element.

X
Z
numerical problems associated with this element can be avoided
Fig. 2. The socket and the ball of a spherical joint clearance in the ANCF-based
or, at least, reduced by increasing the number of elements used framework.
to discrete the exible part. In this work, the original element
developed by Shabana and Yakoub [39] is used as the kinematic
description. It is the same as that of others, its formulation is
clearance. Thus, let consider Fig. 2 in which the socket and the ball
simpler, and only differs from the new elements in the formulation
elements of a spherical joint clearance are shown.
of the elastic forces. Furthermore, it is quite easy to extend this
Based on the ANCF, node i indicates the center of the socket, and
method to other type spatial beam elements. Fig. 1 illustrates the
the center of the ball is located at node j. The vectors n and t rep-
ANCF-based spatial beam element.
resent the normal and tangential directions to the contacting sur-
As it is observed in Fig. 1, there are two nodes i and j in the ele-
faces, respectively. From Fig. 2, the eccentricity vector, eij, which
ment. The nodal coordinates vector e includes the node position
connects the centers of the socket and the ball, can be calculated
vector ri and rj, as well as the partial derivatives of the position vec-
as,
tor (ri,x, ri,y, ri,z, rj,x, rj,y and rj,z) with respect to the three local coor-
dinates or parameters of the element. Then the coordinate vector e eij rj  ri ex ey ez T 5
can be expressed as,
Thus, the magnitude of the eccentricity vector is given by,
e ei ej T ri ; ri;x ; ri;y ; ri;z ; rj ; rj;x ; rj;y ; rj;z T 1 q q
eij eTij eij e2x e2y e2z 6
Since for each node there are 12 nodal coordinates, then the coordi-
nate vector e can also be expressed as, The unit eccentricity vector n is aligned with the eccentricity vector,
 and, therefore, it can be written as,
@ri1 @r i2 @r i3 @r i1 @r i2 @r i3 @r i1 @ri2 @ri3
e r i1 ; ri2 ; r i3 ; ; ; ; ; ; ; ; ; ; eij
@x @x @x @y @y @y @z @z @z n 7
T eij
@r j1 @r j2 @r j3 @r j1 @r j2 @rj3 @rj1 @r j2 @r j3
rj1 ; r j2 ; r j3 ; ; ; ; ; ; ; ; ; : 2
@x @x @x @y @y @y @z @z @z With reference to Fig. 2, the penetration depth due to the contact
between the ball and the socket surfaces can be evaluated as,
The position vector in the global coordinate framework XYZ of an
arbitrary point of the element is given by, d eij  c 8
r Se 3 where c is the radial clearance given by c = RsRb, in which Rs and Rb
are the radius of the socket and the radius of the ball. In Fig. 2,
in which S is the element shape function. Finally, by using Newton
points P and Q indicate the contact points on socket and ball sur-
Euler formulation, the element equations of motion can be written
faces, respectively. Then, the position of these contact points can
as,
be expressed as,
Q e  Fe
Me e 4
R rp ri nRs 9
T
where Me v qS Sdv denotes the element constant mass matrix rQ rj nRb 10
(q is mass density and v denotes the volume), Fe is the element elas-
tic force vector which can be deducted by continuum mechanics ap- Since the contact between the socket and ball is an oblique
proach [43], and Qe represents the element external generalized eccentric collision, its contact-impact treatment involves both rel-
force vector. The interested reader in the details of the ANCF is re- ative normal velocity and relative tangential velocity. The velocity
ferred to the works by Shabana [3033,39,40]. of the contact points P and Q, expressed in terms of the global coor-
dinate system, can be obtained simply by differentiating Eqs. (9)
3. Spherical joints with clearance: dry contact model and (10) with respect to time, yielding,
r_ p r_ i nR
_ s 11
In order to consider the contact-impact forces into the equa- _rQ r_ j nR_ b 12
tions of motion for exible multibody systems based on the ANCF,
it is rst necessary to develop a mathematical model for joints with Then, the relative normal velocity can be expressed as,
916 Q. Tian et al. / Computers and Structures 87 (2009) 913929

v n r_ P  r_ Q T nn 13 numerical and experimental results quite good. Nevertheless,

other contact models are candidates to be used for the normal con-
Similarly, the relative tangential velocity is given by,
tact force and some more insight can be obtained from works that
v t r_ P  r_ Q  v n 14 have been developed independently of that utilized here. In partic-
ular the interested reader can nd in the works of Liu et al. [11],
Dividing the eccentricity eij by radial clearance c yields the eccen-
Kuwabara and Kono [51] and Ramirez et al. [52] good insights on
tricity ratio e expressed by,
the collision of spheres where the dissipative effects play an impor-
eij tant role.
e 15
c It is known that frictional forces act when contacting bodies
By differentiating Eq. (6) with respect to time, and dividing the re- tend to slide relative each other. These forces are tangential to
sult by radial clearance, the time rate of change of the eccentricity the surfaces of contact and are opposite to the sliding velocity.
ratio is expressed as, The tangential forces due to the friction phenomenon may be con-
sidered when the impact velocity has a relative tangential compo-
e_ ij nent, such as in the case of oblique eccentric collisions. Friction is a
e_ 16
c quite complex phenomenon which involves interaction between
It is known that modeling contact forces during an impact event the surfaces of contacting bodies and may lead to different friction
plays a key role in the dynamic analysis of mechanical systems regimes such as sliding and sticking [53]. The most fundamental
[44]. On the one hand, the contact force model must be computed and simplest frictional force model is the Coulombs friction law
by using a suitable constitutive law that takes into account mate- [54]. This model assumes that the frictional force between sliding
rial properties of the colliding bodies, geometric characteristics of bodies with respect to each other is proportional to the normal
the impacting surfaces and the impact velocity. On the other hand, contact force. The constant of proportionality is called the friction
the numerical method for the calculation of the contact forces coefcient. However, the denition of the Coulombs friction law
should be stable in order to allow for the integration of the poses numerical difculties when the relative tangential velocity
mechanical systems equations of motion. Various types of consti- is in the vicinity of zero. Thus, in the present work in order to avoid
tutive laws are suggested in the literature, being one of the more numerical difculties, the friction effect is included as a modied
prominent proposed by Hertz [45]. However, this law is purely Coulombs friction law [55] as,
elastic in nature and can not explain the energy loss during the im- vt
Ft cf cd F n 20
pact process. Lankarani and Nikravesh [20] overcame this difculty kv t k
by separating the normal contact force into elastic and dissipative
where cf is the friction coefcient, and dynamic correction coef-
components. Thus, in the present study, when impact between the
cient cd is given by,
socket and ball occurs, the normal contact forces are evaluated by 8
using the continuous contact model proposed by Lankarani and <0
> if kv t k 6 v 0
kv t kv 0
Nikravesh [20], which can be written as, cd v 1 v 0 if v 0 6 kv t k 6 v 1 21
"   # >
:
3 1  ce d_
2 1 if kv t k P v 1
F n Kdn 1 17
4 d_  in which, v0 and v1 are given tolerances for the relative tangential
velocity of the surfaces in contact. This dynamic correction factor
where ce is the restitution coefcient, d_  represents the initial im-
effectively prevents that the frictional force changes direction for al-
pact velocity, d_ denotes the relative penetration velocity, n is a spec-
most null values of the tangential velocity. Therefore, the great mer-
ied nonlinear index, and the stiffness coefcient K for two
it of the modied Coulombs law expressed by Eq. (20) is that it
spherical contacting bodies can be expressed by,
allows the numerical stabilization of the integration algorithm. This
 1=2 frictional force model does not account for other tribological phe-
4 Rb Rs
K 18 nomena like the adherence between the sliding contact surfaces.
3gb gs Rb Rs
Fig. 3 shows the frictional force and its components acting on
the socket and the ball. The spatial friction force can be expressed
in which gb and gs are the material parameters for ball and socket,
in its Cartesian coordinates as,
respectively, and can be given by,

1  v 2k
gk ; k b; s 19 Ft FtZ
Ek
vk and Ek represent the Poissons ratio and the material Youngs
modulus associated with the ball and the socket. Then, the resulting Rz Ftx
P
contact forces are included in the equations of motion of the system
F ty
as external generalized forces [46].
k Rn
It should be highlighted that the contact force model considered
here that accounts for the energy dissipation is found to be satis- Ry
Rx
factory for general mechanical contacts. Shivaswamy [47] demon-
strated experimentally that at low impact velocities, the energy
dissipation due to the internal damping is the main contributor Z
to energy loss. Moreover, the contact force model given by Eq.
(17) is only valid for low impact velocities, that is, speeds that
Y
are at the most one order of magnitude lower than the elastic wave
traveling velocity [48,49]. Recently, Flores et al. [50] carried out a
numerical and experimental investigation of multibody systems X
with joint clearances. They used the same contact-impact force
laws as in the present study, being the correlation between the Fig. 3. Representation of the frictional force components.
 Q. Tian et al. / Computers and Structures 87 (2009) 913929 917

Ft F tx F ty F tz T 22 sures, but this case is not realized in many applications due to

the uid incapacity to sustain signicant sub-ambient pressures
Form Fig. 3, the position vector Rn can be written as, [10].
R n Rx Ry Rz  T 23 For the case of lubricated spherical joints, the full general form
of the isothermal Reynolds equation can be stated as [57,58],
The direction of this vector can be determined by using the unit " ! !#
3 3
eccentricity vector n described earlier. Thus, if the frictional force is 1 1 @ h @p 1 @ h @p
sin h
acting at a point P, it is necessary to transform it into the force act- R2s sin h h 12l @h sin h / 12l @/
2

ing at node k, taking into account the corresponding moments that  h /

1  h h sin h @ U h @ U sin hU
 h @h
are generated in the process. With regard to Fig. 3, the three mo- h cos hU
Rs sin h @h @/ @h
ments generated by the friction force and that are applied to the 
node k can be written as, / @h @h
U 27
@/ @t
M X F ty Rz  F tz Ry 24
from which it is possible to obtain the following expressions,
M Y F tz Rx  F tx Rz 25
M Z F tx Ry  F ty Rx 26  h Rs xx sin / Rs xy cos /
2U 28
 / Rs xx cos / cos h  Rs xy sin / cos h Rs xz sin h
2U 29
Finally, based on the formulation presented in [56], these fric-
tional moments can be easily transformed into the generalized h c1  ex sin h cos /  ey sin h sin /  ez cos h 30
forces. where p denotes the lubricant pressure, r, h and / are the spherical
coordinates h is the lubricant thickness, l is the dynamic lubricant
4. Spherical joints with clearance: lubricated model viscosity, ex, ey and ez are the components of the eccentricity vector
in the X, Y and Z directions, and xx, xy and xz represent the angular
Generally, in order to avoid or at least minimize the ball-socket velocity components in the X, Y and Z directions.
direct contact and extend spherical joint lifetime, lubricants are of- Based on Eq. (15), the ex, ey and ez components can be evaluated
ten used for any practical application. Therefore, proper modeling by,
of lubricated spherical joints in multibody systems is necessary
to achieve a better understanding of their dynamic performance. e en ex ; ey ; ez T 31
In this section, a mathematical model for spherical joints with
clearance, taking into account the lubrication action, is presented. As mention above, the boundary condition for the hemispheri-
Fig. 4 shows a spherical joint in which the space between the ball cal socket lubricant pressure are dened by [58],
and socket wall is lled with a lubricant. The socket and ball cen- p

ters are denoted by i and j, respectively. Furthermore, a spherical P r; ; / 0 32

2
coordinate system is adopted to model the lubrication force com-
Under the assumption of the pure squeeze condition, the lubri-
ponents, as it can be observed in Fig. 4. Thus, the position of an
cation force generated by the rotation motion can be neglected,
arbitrary point P located inside the joint boundaries can be deter-
and, consequently, the pressure eld can be expressed as,
mined by its spherical coordinates (r, h, /). Since lubrication force
has to be evaluated by integrating the Reynolds equation and be- 8 2
h i
< 3l2e_ Rs 1
1 if e0
cause the uid lubricants can not support negative pressure, the c e 1e cos h2
p 33
pressure eld is integrated only over the positive part by setting : 6l e_ R2s
c2
cos h if e0
the pressure in the remaining portion equal to zero. This corre-
sponds to Gmbels boundary conditions [57,58]. In fact, complete
Finally, the squeeze lubrication force can be obtained by inte-
or full lm does not take into account the cavitation phenomenon
grating the pressure,
and, consequently, contemplate the existence of negative pres-
Z 2p Z p
2
F squeeze R2s p cos h sin hdhd/
0 0
8 4_
h i
< 6pl2Rs e 13 ln1  e 2 1  1 if e0
c e e 1e 2e
34
: 4plR4s e_
c2
if e0
where l is the dynamic lubricant viscosity, being the remaining
parameters associated with the joint kinematics. As it can be seen
in Fig. 4, when e_ is in the same direction as the unit eccentricity vec-
tor n, the lubrication force can be calculated according to Eq. (34),
while when e_ is in the opposite direction of the unit eccentricity
vector n, the squeeze lubrication force is set to zero.
As the uid lm thickness can become very thin, that is, the ball
is very close to the socket surface, the squeeze force due to the lu-
bricant evaluated from Eq. (34) becomes very large (since e is very
close to 1). Consequently, the squeeze force will approach innity,
ultimately leading to simulation divergence. In order to solve this
problem, a model that smoothes the transition between the lubri-
cated and dry contact cases is considered, which is schematically
represented in Fig. 5. This model, which combines the squeeze-lm
action and the dry contact effect, was rst proposed in [59]. It
Fig. 4. Schematic representation of a lubricated spherical joint. should be noted that this approach ensures continuity in the joint
918 Q. Tian et al. / Computers and Structures 87 (2009) 913929

Force by the continuous high-frequency contact-impact forces. In addi-

tion, the spurious high-frequency response is also a product of
Squeeze Force Hybrid Force Model Dry impact Force the nite element discretization. The convergence for the solution
of the high-frequency modes is poor if the spurious high-frequency
responses could not be damped out. As a consequence, the high-
frequency responses produce articially extreme large impact
forces, which in turn lead to the divergence of the simulation.
Therefore, it is desirable for an algorithm to have a controllable
numerical dissipation in the high-frequency range, that is, the
low-frequency responses should be well-preserved and the high-
frequency responses should be damped in a controllable way. Thus,
according to the generalized-a method, it is possible to write [60],
1  am an1 am an 1  af q n1 af q
n; 0
a0 q 37
e0 e1 Eccentricity   
2 1
c qn1 qn hq_ n h  b an ban1 38
2
q_ n1 q_ n h1  can can1 39
where n denotes the nth iteration, b, c, af and am are the algorithm
parameters. In order to make the generalized-a algorithm second-
order accurate, to dissipate the spurious high- frequency responses
Fig. 5. Transition model between lubricated and dry contact models.
and to well-preserve the low-frequency responses, the parameters
in the algorithm can be determined by the following relations,

reaction force when the squeeze force model is switched to dry 1 af  am 2

b 40
contact force model. 4
Mathematically, the transition force model can be written as, 1
c af  am 41
8 2
> F if eij 6 c r
< squeeze af 42
eij c
F
ce0 eij
F squeeze F dry if c 6 eij 6 c e0 35 r1
> e0
:
e0
2r  1
F dry if eij P c e0 am 43
r1
where e0 and e1 are given eccentricity tolerances. The values of
in which r 2 0 1 , r = 1 will lead to an energy-preserving algo-
these parameters must be chosen carefully, since they depend
rithm and r = 0, the maximum energy will be dissipated by the
strongly on the clearance size. It should be noted that the clearance
algorithm [60]. Arnold and Brls [61] proposed a computational
used for the squeeze-lm model is not c, but it is c + e1 instead.
scheme based on the generalized-a method for constrained
mechanical systems. In the present study, and based on Arnold
5. Computational strategy to solve the equations of motion and Brls work, the BroydenNewton [62] approach is introduced
to the iteration process. The corresponding entire computational
This section deals with the computational strategy for the scheme is illustrated in the owchart of Fig. 6. In the process the fol-
numerical solutions of the equations of motion of spatial exible lowing relations are utilized,
multibody systems including spherical joints with clearance and
lubrication. As it has already been mentioned, based on the ANCF,
^ 1  am
b 44
the assembly of the elements can be carried out by traditional - 2
h b1  af
nite element method. The element nodal coordinates e can be eas- c
ily transformed into the exible multibody system generalized c^ 45
hb
coordinates q. Thus, based on the previous works on ANCF, the
equations of motions for constrained exible multibody systems which can make
can be expressed in a compact form as [33,43],
(
@q @ q_
^
Ib; Ic
^ 46
UTq k Fq Q q
Mq @q @q
36
Uq; t 0
satised. Here, the variable tol is the integration error tolerance, J is
where M is the system mass matrix, U(q,t) is the vector that con- the system Jacobian matrix, and G is given by,
tains the system constraint equations corresponding to ideal or per- " #
UTq k Fq  Q q
Mq
fect joints, Uq is the derivative matrix of constraint equations with G 47
respect to the system generalized coordinates q, vector k is Lagrang- Uq; t
ian multipliers associated with the constraints, and Q(q) denotes The Jacobian matrix, after some fundamental mathematical
the system external generalized forces. In the present study, vector operations, can be written as,
Q(q) includes the distributed gravitational force as well as the con- " #
tact-impact force, and F(q) contains the system elastic force vector. @G ^ UT k Fq  Q q
Mb UTq
q q q
J 48
It should be highlighted that the process of solving the equa- @qkT Uq 0
tions of motion in the case of exible bodies differs from the case
in which the bodies are considered as rigid and with joint clear- As observed from the owchart of Fig. 6, it is not necessary to
ances. This is due to the fact that in the above equations of motion evaluate the Jacobian matrix and the partial derivation term in
for exible multibody systems with joint clearances, the spurious each iteration because of the introduction of the BroydenNewton
high-frequency responses are stimulated, which is mainly caused iteration. In addition, the most cumbersome task is the evaluation
 Q. Tian et al. / Computers and Structures 87 (2009) 913929 919

Table 1
q n +1 = q n + hq n + h 2 (0.5 )a, q n +1 = q n + h(1 )a Total computation time for the Bathe and generalized-a methods expressed in
seconds.
n +1 = 0, a = ( f q n ma) / (1 m ) Integration time step 0.001 0.004 0.01 0.05

q n +1 = q n +1 + h 2 a, q n +1 = q n +1 + h a, q n +1 = 0
Bathe scheme 3995.8 1029.5 667.3 248.2
Generalized-a
r = 1.0 2004.8 527.4 Not converge
dif = 1 r = 0.8 2058.1 512.5 349.8 Not converge
r = 0.5 2042.5 518.9 376.9 152.2
r = 0.2 2055.2 517.1 391.5 187.2

Mqn+1 + ( Tq )n+1 + F ( qn+1 ) Q ( qn+1 )

rn+1 =
(qn+1 , t ) on a PC with an Intel Pentium 1.5 GHz processor and 1 GB RAM.
The outcomes of the numerical simulations are listed in Table 1,
G
J=( )n+1 from which it can be observed how important is the choice of
[q ]T the method utilized, as well as the selection of the r-parameter
in the case of the generalized-a method, mainly for long time sim-
rn = rn+1 , = [q ]T = J 1rn+1 ulations and large time steps. In general, the generalized-a method
is more efcient from computational point of view, because the
qn+1 = qn+1 + q, qn+1 = qn+1 +q Newton iterative approach has to be carried out twice in the
Bathes scheme. Nevertheless, the Bathes scheme is more robust,
qn+1 = qn+1 + q, n+1 = n+1 + stable and quite interesting for long time simulations since larger
time steps can be used. The interested reader in the details of
Bathes scheme is referred to references [64,66,67]. However, when
dif = rn +1 < tol ? a multibody system includes joints with clearance, small time
steps have to be used, and special attention should be paid to de-
tect the precise instant of contact [68]. Take, for instance, a system
a = a + (1 f ) / (1 m )q n +1 in which before the rst impact, not only the bodies move slowly
relative to each other but also the complete system is moving with
Fig. 6. The numerical iteration procedure for the generalized-a method. an almost constant velocity. In this phase, the step size of the inte-
gration algorithm may become relatively large, being the global
of the tangent matrix of the elastic force. Therefore, with the intent motion of the system characterized by relatively large translational
to improve computation efciency in simulation, the invariant ma- and rotational displacements during a single time step. Therefore,
trix method [63] can be employed. if the numerical integration is not handled properly, the rst im-
It is important to mention that there are other alternative and pact between the colliding bodies may be observed with a high
robust methods to deal with the nonlinear dynamic problem, espe- penetration depth, and, consequently, the calculated contact forces
cially when large deformations and long time simulations are con- becomes articially large. Then, smaller time steps must be consid-
sidered, as it has been proposed by Bathe [64]. In fact, for long time ered, which increases the computation time, and, therefore, the
simulations the generalized-a method with large step can lead to generalized-a method is more efcient for these cases. Further-
some numerical problems, such as the divergence of the systems more, the inuence of the value of the algorithm r-parameter on
solution. In order to better understand the main implications of the outcomes is not signicant since for small integration time
the issue a simple pendulum is considered as an application exam- steps this effect can be neglected.
ple to compare the generalized-a and the Bathe methods. Fig. 7 With the intent to better understand this issue let the data pre-
shows the initial conguration of the simple pendulum which is sented in Fig. 8, where the plots of the X displacement of point B for
acted by a uniform distributed gravitational force [65]. The length the Bathe and generalized-a methods are depicted. The results are
of the pendulum is 1 m and the cross section is a square with obtained for different time steps and several values for the r-
0.02 m each side width. The Youngs elasticity modulus and Pois- parameter. By analyzing the curves of Fig. 8a and b, it can be ob-
sons ratio are 6e7 Pa and 0.3, respectively. The material density served that the Bathes scheme leads to more accurate results
is equal to 6000 kg/m3. Furthermore, the arm pendulum is discret- when compared with the data produced with the generalized-a
ized by 5 ANCF-based elements. A long time simulation equal to method. In fact, when the generalized-a method is considered with
20s was performed for both methods. In addition, 4 different time large time steps the systems response accuracy will be signi-
steps and 4 different values of the r-parameter of the generalized- cantly affected, which ultimately even can lead to a divergent
a method were considered in the analysis, being the admissible problem. This aspect is supported by similar analysis reported in
integration error equal to 1 e6. The simulations were performed the work by Bathe and Baig [66].

Y
2
g=9.81 m/s

X
A B

Fig. 7. Initial conguration of single pendulum.

920 Q. Tian et al. / Computers and Structures 87 (2009) 913929

Fig. 8. Displacement in the X direction of the point B.

Fig. 9. Evolution of the system total energy for the Bathe and generalized-a methods.

The inuence of the methods employed on the systems behav-

ior is also quite visible in the plots of the system total energy illus-
trated in Fig. 9. The system total energy is the sum of the elastic,
kinetic and potential energies evaluated in each time step. In the
curves shown in Fig. 9a and b, two different time steps are consid-
ered, namely, h = 0.001 s and h = 0.05 s. By observing Fig. 9, it can
be concluded that the generalized-a method causes a much more
energy dissipation. In fact, the Bathes scheme preserves the sys-
tem energy, especially for the case with larger time step. This as-
pect supports the accuracy lost plotted in Fig. 8b.

6. Results and discussion

6.1. Spatial double pendulum

In order to examine the effectiveness of the formulation pre- Fig. 10. Spatial double pendulum modeled by two exible bodies and two spherical
joints.
sented in this work, a double pendulum with the initial congura-
tion shown in Fig. 10 is studied here. The double pendulum is made
up of three bodies, the ground body and two arms. The two arms centers of mass of all bodies dominate the total potential system
are modeled as exible bodies. One perfect spherical joint connects energy and control the dynamic behavior of the system. The grav-
the rst arm to the ground, while a spherical joint clearance exists itational acceleration of 9.81 m/s2 is used in the study. The link
between the two arms, resulting in a spatial multibody system lengths are OA = 0.3 m and AB = 0.3 m. The cross section of the
with nine rigid body degrees of freedom. The system is released spatial beam element is a square with 0.01 m each side width.
from the initial position with null velocities and under the action The material Poissons Ratio is 0.3 and the coefcient of restitution
of gravity acting in the negative Y direction. Thus, the heights of of the contacting bodies is equal to 0.9. The spherical joint with
 Q. Tian et al. / Computers and Structures 87 (2009) 913929 921

clearance is characterized by a socket and ball radii equal to 0.02 m a rigid body. In order to study the inuence of the spherical joint
and 0.018 m, respectively. For the case of lubricated spherical joint, model (dry and lubricated formulations) on the global performance
the dynamic lubricant viscosity is chosen as 400 cP. The restitution of the double pendulum, kinematic and dynamic characteristics
coefcient ce is set as 0.9 and the specied nonlinear index n is set are presented and discussed in what follows. The results are com-
as 1.5. pared to those obtained with a simulation in which all joints are
A total simulation time equal to 2 s is considered. The simula- considered to be ideal or perfect. Finally, with the intent to validate
tions are carried out by using different models and with different this approach, the same double pendulum is simulated using the
elements and, consequently, different values of the material commercial MSC.ADAMS software, and the global results are ana-
Youngs modulus. For the material Youngs modulus E = lyzed and discussed.
2.07e11 Pa, the material density q is set as 7800 kg/m3, while for Fig. 11 shows the displacement in X direction of point B for dif-
E = 6e7 Pa, q = 6000 kg/m3. The system with E = 2.07e11 Pa is ferent joint models, for different number of elements used in the
named here as stiff system, since its behavior is similar to that of exible body discretization and for two different values of Youngs

Fig. 11. Inuence of the joint model, number of elements used in the exible body discretization and material Youngs modulus on the X displacement of point B.

Fig. 12. Transverse displacement of point B for different number of elements used in the exible body discretization and for different joint models.
922 Q. Tian et al. / Computers and Structures 87 (2009) 913929

modulus. By observing the results plotted in Fig. 11a and b, it can mentioned that the systems behavior for the case of E =
be concluded that the systems response is not affected by the joint 2.07e11 Pa is similar to that when the system is modeled with rigid
model, being the displacement of point B equal for the case of per- bodies.
fect joint and for the case of joint clearance. Another conclusion is Fig. 12 presents the plots of the transverse displacement of
that the number of elements used does not inuence the displace- point B for different number of elements used in the exible body
ment of point B, as Fig. 11b and c shown. Yet, when the system is discretization and for different joint models. It can be observed
modeled with the Youngs modulus equal to E = 6e7 Pa, its ampli- that for the point B, transverse displacements for the system with
tude response is larger than for the case of E = 2.07e11 Pa. These joint clearance are larger than for system with perfect joint. Fur-
differences are clearly visible in the plots of Fig. 11d. It should thermore, the transverse displacements for the exible system

Fig. 13. Ball center trajectories for the spherical clearance joint model.

Fig. 14. Animation sequence of the double pendulum for stiff and exible cases.
 Q. Tian et al. / Computers and Structures 87 (2009) 913929 923

are much larger than the transverse displacements for the stiff sys- the clearance size used this example. By analyzing Fig. 13, it can
tem. In addition, the effect of the Youngs modulus is also visible in be concluded that the penetration depth for the exible system
the outcomes plotted in Fig. 12. is larger than the penetration depth for the stiff system. It is also
For the case of spherical joint clearance model, the ball center possible to observe the different modes of motion between the ball
trajectories relative to the socket center are illustrated in Fig. 13. and socket, namely, the free ight motion, impact with rebound,
With the intent to keep the gures clear and readable, only data and continuous contact. In the free ight motion, the ball moves
relative to the rst 0.3 s of simulation are plotted. It should be freely inside the socket boundaries, that is, the ball and the socket
noted that radius of the ball shown in Fig. 13a and b is equal to are not in contact, hence, there is no joint reaction force. In the

Fig. 15. Displacement and velocity in the X direction of the point B (E = 6e7 Pa).

Fig. 16. Transverse displacement of point B for different joint models.

Fig. 17. Eccentricity for different spherical clearance joint models.

924 Q. Tian et al. / Computers and Structures 87 (2009) 913929

Fig. 18. Ball center trajectories for the spherical lubricated joint model.

relative to each other exists. This mode ends when the ball and
socket separate from each other and the ball enters in free ight
mode [12].
Fig. 14 shows the global trajectory described by point B for the
spherical clearance model and considering the stiff and exible sit-
uations. From Fig. 14a and c, it can be observed that trajectory of
point B is smoother for the stiff case compared with the one for
the exible case. The sequence simulation is enlarged in Fig. 14b
and d, from which the detail impact process between the socket
and the ball surfaces can be observed.
The above results were obtained using the spherical clearance
model without taking into account the lubrication effect. Thus, in
what follows, the spherical lubrication model presented in Section
4 and summarized in Eq. (35), is considered in the dynamic simu-
lation of the double pendulum. In Fig. 15 the main differences in
the displacement and velocity in X direction of point B for dry
Fig. 19. MSC.ADAMS double pendulum model with a spherical joint clearance.
and lubricated models is presented. Since the value of clearance
is low, the differences at the displacements level are not signi-
cant. However, the velocities for the dry clearance case present
impact mode, which occurs at the termination of the free ight some peaks associated with the contacts that occur between the
motion, impact forces are applied to the system. This mode is char- ball and socket surfaces. Obviously, for the lubricated model, the
acterized by an abrupt discontinuity in the kinematic and dynamic velocities are smoother and closer to the perfect joint response
responses, and a signicant exchange of momentum occurs be- due to the damping effect of the lubricant. The same conclusions
tween the two impacting bodies. At the termination of the impact can be drawn from Fig. 16 in which the transverse displacement
mode, the ball can enter either in free ight or in the following of point B is plotted. In fact, the dry impact force produces larger
mode. In the continuous contact or following mode, the ball and transverse displacements than those obtained with perfect kine-
the socket are in permanent contact and a rolling or sliding motion matic joints.

Fig. 20. Displacement of point B in the X direction for different joint models.
 Q. Tian et al. / Computers and Structures 87 (2009) 913929 925

Fig. 17 shows the plots for the eccentricity variable when dry pact between the ball and socket walls because of the presence of
and lubricated spherical joint models are utilized. It can be ob- the lubricant. The same phenomenon can be observed in Fig. 18,
served that, for the case of lubrication model, the magnitude of which presents the ball center trajectories inside the socket bound-
eccentricity is smaller than for the dry situation. This is due to aries. Again, with the intent to keep the gures clear and readable,
the fact that in the lubrication model, there is no direct contact-im- only data relative to the rst 1.0 s of simulation are plotted. It
should be highlighted that this phenomenon associated with the
presence of the lubricant is a benet from the mechanical system
performance point of view and parts life.
Finally, in order to validate the approach presented in this pa-
per, the same double pendulum was simulated using the MSC.A-
DAMS software for the stiff case (E = 2.07e11 Pa). The option to
simulate the stiff case only is due to the fact that this software is
a multi-rigid-body package. Fig. 19 shows the ADAMS model of
the double pendulum system, in which the spherical joint clear-
ance is modeled by a two massless spheres and corresponding con-
straints and contact-impact condition are set. One of the balls has
0.02 m of radius and is connected to the arm OA, the other ball
with a radius 0.018 m is linked to arm AB, such that the two balls
together compose a spherical joint clearance. In order to keep the
analysis simple and short, only the results for the displacement
of point B in the X direction are plotted, as it is illustrated in
Fig. 20. The observed correlation validates the methodology pre-
sented in this study.

6.2. Spatial four bar mechanism

In this section, a spatial four bar mechanism is used as an illus-

Fig. 21. Spatial four bar mechanism which includes a spherical joint clearance. trative example to demonstrate how the spherical joint clearance

Fig. 22. Displacement and velocity in the Y direction of the point B (E = 2.07e11 Pa).

Fig. 23. Joint reaction forces produced at the spherical joint clearance.
926 Q. Tian et al. / Computers and Structures 87 (2009) 913929

models affect the behavior of the mechanism. Fig. 21 represents fore, this four bar mechanism has a total of ve rigid body degrees
the initial conguration of the four bar mechanism and includes of freedom. The system is driven by a variable torque applied to the
the main geometric denitions. The kinematic joints of this multi- crank arm OA.
body system include two ideal three-dimensional revolute joints, The spherical joint modeled with clearance is characterized by
connecting the ground to the links OA and BH, and one perfect the socket and ball radii equal to 0.030 m and 0.029 m, respec-
spherical joint that connects the links AB and BH. A spherical joint, tively. The material Poissons ratio is set to 0.3 and the coefcient
with a given clearance, interconnects the links OA and AB. There- of restitution of the contacting bodies is equal to 0.9. The specied

Fig. 24. Displacement and velocity in the Y direction of the point B (E = 6e7 Pa).

Fig. 25. Eccentricity for difference spherical clearance joint models.

Fig. 26. Ball center trajectories for the spherical clearance joint model.
 Q. Tian et al. / Computers and Structures 87 (2009) 913929 927

Fig. 27. Ball center trajectories for the spherical lubricated joint model.

In order to study the inuence of the spherical clearance model

in the global behavior of the spatial four bar mechanism, some
kinematic and dynamic characteristics, corresponding to the rst
four seconds of the simulation, are presented and discussed in
what follows. The results are compared to those obtained with a
simulation in which all kinematic joints are considered to be ideal
or perfect. The simulations are carried out by using different joint
models and with different material properties. For the material
Youngs modulus E = 2.07e11 Pa, the material density q is set as
7800 kg/m3, while for E = 6e7 Pa, q = 6500 kg/m3. As in the previ-
ous example, here the system with E = 2.07e11 Pa is named as stiff
system.
Fig. 22 presents the evolution of displacement and velocity of
point B in the Y direction obtained in the simulations. These results
are for different joint models and for the stiff case. In general, there
are no signicant differences in the displacement curves, as it can
Fig. 28. MSC.ADAMS four bar mechanism model with a spherical joint clearance. be observed in Fig. 22a. In sharp contrast, the velocity proles indi-
cate visible differences, namely in regard to the joint clearance
model. Due to the impact between the ball and socket the velocity
nonlinear index n is set as 1.5. For the case of lubricated spherical presents jumps clearly associated with those impacts. However,
joint, the dynamic lubricant viscosity is chosen as 400 cP. The when the joint is modeled as a lubricated joint, the results are com-
mechanism consists of four bodies, being three of them modeled parable to those obtained with a perfect kinematic joint. This is due
as rigid. The links OA and BH are considered rigid with their mass to the fact that the lubricant acts as a dissipater or damper, avoid-
centers located at mid points G and F, respectively. The inertia ing the direct contact-impact between the ball and socket surfaces.
properties of each rigid link body are set as Ig1 = Ig2 = The evolution of the contact and lubrication forces over the sim-
3.12  105 kg m2 and I11 = I12 = In1 = In2 = 8.93  104 kg m2. The ulations is presented in the diagrams of Fig. 23, from which the
link AB is modeled as a exible body discretized by ve elements, precise instants where the large and sudden peak forces occur
and the cross section of the spatial beam element is square shape can be observed. Furthermore, it can be drawn that for the dry
with 0.02 m side width.

Fig. 29. Displacement of point B in the X direction for different model.

928 Q. Tian et al. / Computers and Structures 87 (2009) 913929

contact situation the magnitude of the joint reaction forces are than those obtained with rigid system model. In addition, results
greater than for the lubricated case, as it would be expected. demonstrate that the direct ball-socket impact can be avoided
Fig. 24 shows the inuence of the joint model on the displace- when a uid lubricant is introduced in the joint. This leads to a per-
ment and velocity of the point B in the Y direction for the exible formance of the system with lubricated joint closer to the perfect
system case. As it can be seen, the joint model clearly inuences joint case. In order to verify and validate the methodology pre-
the system response, mainly for the dry joint clearance model. In sented in this paper, some results are compared to those obtained
a similar way to the stiff case, the system performance with lubri- by using the MSC.ADAMS software, but only for the stiff system
cated joint is closer to that of the perfect joint. case. The simulation results indicate that the methodology for
Fig. 25 shows the plots for the eccentricity variable when a dry modeling and dynamics analysis of spatial exible multibody sys-
clearance and lubricated spherical model are used. It can be ob- tems with spherical joint clearance and lubrication condition based
served that the eccentricities for the lubricated model are smaller on the absolute nodal coordinate method is valid and effective.
than those obtained by dry contact model. The effect of Youngs
modulus is also visible in Fig. 25. The same phenomenon can be Acknowledgments
seen in Figs. 26 and 27 which present the ball center trajectories
inside the socket boundaries. For Fig. 26, with the intent to keep We would like to thank Prof. Daniel Garca-Vallejo from Univer-
the gures clear and readable, only data relative to the rst 0.3s sity of Seville, Spain, and Dr. Johannes Gerstmayr from Johannes
of simulation are plotted, while for Fig. 27 only data relative to Kepler University of Linz, Austria, for their great help with the
the rst 1s of simulation are plotted. In order to keep the gures ANCF method. Finally, we would also like to acknowledge the
clear and readable, only data relative to the rst instants of simu- anonymous reviewers for their insightful comments and sugges-
lation are plotted. For the lubricated joint model, the ball center tions on an earlier draft of this paper. The research is supported
moves closer to its center than for the dry contact model due to by a key National Natural Science Foundation of China
the presence of the lubricant. (60736019) and a National Natural Science Foundation of China
Finally, in a similar way to the previous example, the spatial (60874064).
four bar mechanism was simulated in the commercial MSC.ADAMS
software for different situations. Similar to the rst example, References
Fig. 28 shows the ADAMS model of the spatial four bar mechanism,
in which the spherical joint clearance is modeled by a two massless [1] Flores P, Ambrsio J. Revolute joints with clearance in multibody systems.
Comput Struct 2004;82:135969.
balls and corresponding constraints and contact-impact condition [2] Liu C-S, Zhang K, Yang R. The FEM analysis and approximate model for
are set. In order to keep the analysis simple and short, only the out- cylindrical joints with clearances. Mech Mach Theory 2007;42:18397.
comes for displacement of point B in the Y direction are plotted, as [3] Crowthera AR, Singha R, Zhangb N, Chapman C. Impulsive response of an
automatic transmission system with multiple clearances: formulation,
it is illustrated in Fig. 29. The correlation observed allows to vali- simulation and experiment. J Sound Vib 2007;306:44466.
date the methodology presented in the present work. [4] Bouzidane A, Thomas M. An electrorheological hydrostatic journal bearing for
controlling rotor vibration. Comput Struct 2008;86:46372.
[5] Dubowsky A, Freudenstein F. Dynamic analysis of mechanical systems with
clearances, Part 1: formulation of dynamic model. J Eng Ind 1971;93(1):3059.
7. Conclusions [6] Dubowsky S, Freudenstein F. Dynamic analysis of mechanical systems with
clearances, Part 2: dynamics response. J Eng Ind 1971;93(1):3106.
[7] Miedema B, Mansour WM. Mechanical joints with clearance: a three-mode
In this paper, a general and comprehensive methodology for the
model. J Eng Ind 1976;98(4):131923.
dynamic modeling and analysis of spatial exible systems with [8] Turvey GJ, Wang P. An FE analysis of the stresses in pultruded GRP single-bolt
clearances and lubricated joints has been presented. The absolute tension joints and their implications for joint design. Comput Struct
nodal coordinate method is used to formulate the dynamics of ex- 2008;86:101421.
[9] Srivastava N, Haque I. Clearance and friction-induced dynamics of chain CVT
ible multibody systems. The equations of motion incorporate the drives. Multibody Syst Dyn 2008;19(3):25580.
contact-impact forces due to the collisions of the bodies that con- [10] Flores P, Ambrsio J, Claro JCP, Lankarani HM. Kinematics and dynamics of
stitute the joint clearances, as well as the lubrication forces pro- multibody systems with imperfect joints: models and case
studies. Berlin: Springer; 2008.
duced when the uid lubricant action is considered. In the joint [11] Liu C-S, Zhang K, Yang L. Normal force-displacement relationship of spherical
model with clearance, the intra-joint contact forces are evaluated joints with clearances. J Comput Nonlinear Dyn 2006;1(2):1607.
according to a modied Hertzian-based contact law, which in- [12] Flores P, Ambrsio J, Claro JCP, Lankarani HM. Dynamics of multibody systems
with spherical clearance joints. J Comput Nonlinear Dyn 2006;1(3):2407.
cludes a damping term to account for the energy dissipation. The [13] Flores P, Ambrsio J, Claro JCP, Lankarani HM, Koshy CS. Lubricated revolute
friction phenomenon is included through a modied Coulombs joints in rigid multibody systems, Nonlinear Dynamics 2009;56(3):27795.
friction law. In the case of lubricated joints, the resulting lubrica- [14] Erkaya S, Uzmay I. A neural-genetic (NNGA) approach for optimising
mechanisms having joints with clearance. Multibody Syst Dyn
tion forces are derived by employing the corresponding Reynolds 2008;20(1):6983.
equation. Furthermore, suitable models for spherical joint clear- [15] Erkaya S, Uzmay I. Determining link parameters using genetic algorithm in
ances for dry and lubricated cases are embedded into the multi- mechanisms with joint clearance. Mech Mach Theory 2009;44(1):22234.
[16] Bing S, Ye J. Dynamic analysis of the reheat-stop-valve mechanism with
body systems equations of motion. The generalized-a method is
revolute clearance joint in consideration of thermal effect. Mech Mach Theory
used to solve the system equations of motion with constant system 2008;43(12):162538.
mass matrix based on the absolute nodal coordinate method. To [17] Flores P. Modeling and simulation of wear in revolute clearance joints in
improve computation efciency, the invariant matrix method is multibody systems. Mech Mach Theory 2009;44(6):121122.
[18] Miedema B, Mansour WM. Mechanical joints with clearance: a three-mode
used. model. J Eng Ind 1976;98(4):131923.
The effectiveness of the approach presented here is demon- [19] Hunt KH, Crossley FR. Coefcient of restitution interpreted as damping in
strated throughout the dynamic simulation of two spatial multi- vibroimpact. J Appl Mech 1975;7:4405.
[20] Lankarani HM, Nikravesh PE. A contact force model with hysteresis damping
body systems that include perfect, clearance and lubricated for impact analysis of multibody systems. J Mech Des 1990;112:36976.
joints, namely, an open-loop system, the spatial double pendulum, [21] Pombo JC, Ambrsio JAC. Application of a wheel-rail contact model to railway
and a closed-loop system, the spatial four bar mechanism. The dynamics in small radius curved tracks. Multibody Syst Dyn 2008;19:91114.
[22] Dubowsky S, Gardner TN. Design and analysis of multilink exible
inuence of the exibility of bodies and the joint models is inves- mechanisms with multiple clearance connections. J Eng Ind 1977;99:8896.
tigated, as the main contribution of the paper, not studied litera- [23] Dubowsky S, Deck JF, Costello H. The dynamic modeling of exible spatial
ture. In general, the results evidence that the contact forces machine systems with clearance connections. J Mech Transmissions Automat
Des 1987;109:8794.
obtained by using the relative exible system model are smaller
 Q. Tian et al. / Computers and Structures 87 (2009) 913929 929

[24] Dubowsky S, Moening MF. An experimental and analytical study of impact [46] Nikravesh PE. Computer aided analysis of mechanical systems. Englewood
forces in elastic mechanical systems with clearances. Mech Mach Theory Cliffs (NJ): Prentice-Hall; 1988.
1978;13:45165. [47] Shivaswamy S. Modeling contact forces and energy dissipation during impact
[25] Kakizaki T, Deck JF, Dubowsky S. Modeling the spatial dynamics of robotic in multibody mechanical systems. Ph.D. dissertation. Wichita, USA: Wichita
manipulators with exible links and joint clearances. J Mech Des State University; 1997.
1993;115:83947. [48] Shivaswamy S, Lankarani HM. Impact analysis of plates using quasi-static
[26] Bauchau OA, Bottasso CL. Contact conditions for cylindrical, prismatic, and approach. J Mech Des 1997;119(3):37681.
screw joints in exible multibody systems. Multibody Syst Dyn [49] Schiehlen W, Seifried R. Three approaches for elastodynamic contact in
2001;5:25178. multibody systems. Multibody Syst Dyn 2004;12(1):116.
[27] Bauchau OA, Rodriguez J. Modelling of joints with clearance in exible [50] Flores P, Koshy CS, Lankarani HM, Ambrosio J, Claro JCP. Numerical and
multibody systems. Int J Solids Struct 2002;39:4163. experimental analysis of planar multibody systems with revolute clearance
[28] Liu TS, Lin YS. Dynamic analysis of exible linkages with lubricated joints. J joints. In: Bottasso CL, Masarati P, Trainelli L, editors. Proceedings of
Sound Vib 1990;141(2):193205. ECCOMAS multibody dynamics 2007 international conference on advances
[29] Shiau T-N, Tsai Y-J, Tsai M-S. Nonlinear dynamic analysis of a parallel in computational multibody dynamics. Italy: Milano; 2007. p. 17 p. 2528
mechanism with consideration of joint effects. Mech Mach Theory June.
2008;43:491505. [51] Kuwabara G, Kono K. Restitution coefcient in a Collision between two
[30] Shabana AA. Flexible multibody dynamics review of past and recent spheres. Jpn J Appl Phys 1987;26(8):12303.
developments. Multibody Syst Dyn 1997;1:189222. [52] Ramirez R, Poschel T, Brilliantov N, Schwager T. Coefcient of restitution of
[31] Shabana AA. An absolute Nodal coordinates formulation for the large rotation colliding viscolelastic spheres. Phys Rev E 1999;60(4):446572.
and deformation analysis of exible bodies. In: Report No. MBS96-1-UIC, [53] Ahmed S, Lankarani HM, Pereira MFOS. Frictional impact analysis in open loop
University of Illinois at Chicago; 1996. multibody mechanical system. J Mech Des 1999;121(1):11927.
[32] Shabana AA. Denition of the slopes and absolute nodal coordinate [54] Greenwood DT. Principles of dynamics. Englewood Cliffs (NJ): Prentice Hall;
formulation. Multibody Syst Dyn 1997;1:33948. 1965.
[33] Shabana AA. Dynamics of multibody systems. 3rd ed. Cambridge, New [55] Ambrsio JAC. Impact of rigid and exible multibody systems: deformation
York: University Press; 2005. description and contact models. In: Schiehlen W, Valsek M, editors.
[34] Eberhard P, Schiehlen W. Computational dynamics of multibody systems Proceedings of the NATO-ASI on virtual non-linear multibody systems, vol.
history, formalisms, and applications. J Comput Nonlinear Dyn 2006;1:312. II, 2002. p. 1533.
[35] Yoo WS, Dmitrochenko O, Yu D. Review of nite elements using absolute nodal [56] Sopanen JT, Mikkola AM. Description of elastic forces in absolute nodal
coordinates for large-deformation problems and matching physical coordinate formulation. Nonlinear Dyn 2003;34:5374.
experiments. In: ASME 2005 international design engineering technical [57] Goenka PK. Effect of surface ellipticity on dynamically loaded spherical and
conferences and computers and information in engineering conference, cylindrical joints and bearings. Ph.D. dissertation. Ithaca, New York: Cornell
California, Long Beach, DETC2005-84720. University; 1980.
[36] Garca-Vallejo D, Mikkola AM, Escalona JS. A new locking-free shear [58] Pinkus O, Sternlicht SA. Theory of hydrodynamic lubrication. New
deformable nite element based on absolute nodal coordinates. Nonlinear York: McGraw-Hill; 1961.
Dyn 2007;50:24964. [59] Flores P, Ambrsio J, Claro JP. Dynamic analysis for planar multibody
[37] Gerstmayr J, Matikainen MK, Mikkola AM. A geometrically exact beam mechanical systems with lubricated joints. Multibody Syst Dyn
element based on the absolute nodal coordinate formulation. Multibody Syst 2004;12:4774.
Dyn 2008;20(4):35984. [60] Chung J, Hulbert G. A time integration algorithm for structural dynamics with
[38] Dmitrochenko O, Mikkola AM. Two simple triangular plate elements based on improved numerical dissipation: the generalized-a method. J Appl Mech
the absolute nodal coordinate formulation. J Comput Nonlinear Dyn 1993;60:3715.
2008;3:041012128. [61] Arnold M, Brls O. Convergence of the generalized-a scheme for constrained
[39] Shabana AA, Yakoub RY. Three-dimensional absolute nodal coordinate mechanical systems. Multibody Syst Dyn 2007;18(2):185202.
formulation for beam elements: theory. J Mech Des 2001;123:60613. [62] Broyden CG. A class of methods for solving nonlinear simultaneous equations.
[40] Yakoub RY, Shabana AA. Three-dimensional absolute nodal coordinate Math Comput 1965;19(92):57793.
formulation for beam elements: implementation and applications. J Mech [63] Garca-Vallejo D, Mayo J, Escalona JL. Efcient evaluation of the elastic forces
Des 2001;123:61421. and the Jacobian in the absolute nodal coordinate formulation. Nonlinear Dyn
[41] Schwab AL, Meijaard JP. Comparison of three-dimensional exible beam 2004;35:31329.
elements for dynamic analysis: nite element method and absolute nodal [64] Bathe KJ. Conserving energy and momentum in nonlinear dynamics: a simple
coordinate formulation. In: ASME 2005 international design engineering implicit time integration scheme. Comput Struct 2007;85:43745.
technical conferences and computers and information in engineering [65] Zhang Y, Tian Q, Chen L, Yang J. Simulation of a viscoelastic exible multibody
conference, California, Long Beach, DETC2005-85104. system using absolute nodal coordinate and fractional derivative methods.
[42] Sugiyama H, Gerstmayr J, Shabana AA. Cross-section deformation in the Multibody Syst Dyn 2009;21(3):281303.
absolute nodal coordinate formulation. In: ASME 2005 international design [66] Bathe KJ, Baig MMI. On a composite implicit integration procedure for
engineering technical conferences and computers and information in nonlinear dynamics. Comput Struct 2005;83:251324.
engineering conference, California, Long Beach, DETC2005-84524. [67] Bathe KJ. On reliable nite element methods for extreme loading conditions.
[43] Shabana AA. Computational continuum mechanics. New York: Cambridge In: Ibrahimbegovic A, Kozar I, editors. Extreme man-made and natural hazards
University Press; 2008. in dynamics of structures. Springer-Verlag; 2007. p. 71102.
[44] Flores P, Ambrsio J, Claro JCP, Lankarani HM. Inuence of the contact-impact [68] Flores P, Ambrsio J. On the contact detection for contact analysis in
force model on the dynamic response of multi-body systems. Multibody Syst multibody, systems. Proceedings XXII ICTAM 22nd international congress of
Dyn 2006;220(1):2134. theoretical and applied mechanics. Adelaide, Australia: Adelaide Convention
[45] Hertz H. On the contact of solids, 1881. On the contact of rigid elastic solids Centre; 2008. August 2529, 2p.
and on hardness, 1882 [Jones DE, Schott GA, Trans.]. Miscellaneous Papers,
MacMillan and Co. Ltd., London; 1896. p. 14683.