A Comparison in Terms of Accuracy and E) Ciency Between A MBS Dynamic Formulation With Stress Analysis and A Non-Linear FEA Code

INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING
Int. J. Numer. Meth. Engng 2001; 51:1033–1052
A comparison in terms of accuracy and e)ciency between a

MBS dynamic formulation with stress analysis and a
non-linear FEA code
J. Cuadrado∗;† , R. Guti2errez, M. A. Naya and P. Morer

Department of Mechanical Engineering; University of La Coruña; 15403 Ferrol; Spain
SUMMARY
To perform the mechanical design of a machine through computer-aided techniques, at least three main
di8erent products should be used: a CAD software, to model the parts of the machine; a MBS program,
to analyse the kinematics and dynamics of the whole system; and a FEA code, to determine the level of
stress and strain su8ered by each component. If it is true that CAD software is usually well connected
with the two other tools, the same does not happen in what respects to FEA–MBS interfaces. Moreover,
since both the large-amplitude motion and the elastic deformation are coupled, they cannot be solved
separately, and the usual practice consisting of ;rst analysing the machine motion assuming rigid bodies,
and then calculating stresses under the loads previously generated, is just an approximation. In order to
provide mechanical designers with a tool which makes easier and shorter the design-cycle, this paper
presents a comparison between the two options that are currently available to address the mentioned
problem: a dynamic MBS formulation which simultaneously solves motion and performs stress analysis
by considering >exible bodies; and a non-linear module of a FEA code, which takes into account large
displacements and ;nite rotations. The comparison is carried out in terms of accuracy and e)ciency
through four examples. The results lead to the conclusion that, for similar accuracy, the ;rst method
is largely more e)cient. Therefore, the interest of developing MBS commercial codes which integrate
motion calculation and stress analysis through the mentioned approach is envisioned, as long as they
would provide faster solutions. Copyright ? 2001 John Wiley & Sons, Ltd.
KEY WORDS: multibody systems dynamics; stress analysis; mechanical design; simulation tools
1. INTRODUCTION
Simulation tools are becoming more and more relevant in mechanical design as they al-
low for a reduction of the design cycle, thus leading to products at reduced costs and with
earlier presence in the market place. Very often when addressing machine design, the de-
signer is concerned about the kinematics and dynamics of their part-mobile products as well
∗ Correspondence to: J. Cuadrado, Department of Mechanical Engineering; University of La Coruña; 15403 Ferrol;
Spain
† E-mail: javicuad@cdf.udc.es
Contract=grant sponsor: The Galician Government; contract=grant number: XUGA16601A98
Received 3 December 1999

Copyright ? 2001 John Wiley & Sons, Ltd. Revised 25 October 2000
1034 J. CUADRADO ET AL.
as interested in the level of stress and strain incurred by their components. Automotive,
aerospace, robotics, biomechanical, heavy machinery and military industries are just some
examples.
There are two traditional ways to calculate the stresses su8ered by components of a
multibody system during its motion:
(a) The ;rst one, usually referred to as linear theory of elastodynamics, consists of initially
solving the motion of the mechanism through the use of a multibody dynamics code
assuming that members are rigid. This provides the loads and reactions that serve as input
for a FEA program, where the >exible condition of the elements is taken into account,
to obtain stresses and strains. This approach is an approximation, since both the large
amplitude motion and the elastic deformation are coupled and, therefore, they cannot be
solved separately. Moreover, data transference between programs becomes rather involved
as long as formats are usually di8erent. Detailed information about this method can be
found in Reference [1].
(b) The second one addresses the coupled problem making use of the non-linear dynamics
module of a FEA program, taking into account large displacements and ;nite rotations.
Most FEA codes are based on the incremental -nite element approach, illustrated in
Reference [2]. This method, although theoretically correct, seems to be very slow, as it
generates models with a large number of variables, leading to big problem sizes that
require an enormous computational e8ort.
In this paper, a third way is proposed as an alternative: the dynamics of the multibody
system are solved as rigid, but components whose stresses and strains are required, are mod-
elled as >exible bodies. Indeed, any method of those developed to solve the dynamics of
multibody systems with >exible bodies ( .exible multibody dynamics) provides a relation-
ship between the problem variables and the deformed con;guration of each >exible link.
Therefore, once positions are known at a certain point of time, local elastic displacements
of the body can be directly calculated and strains and stresses immediately derived from
them. Consequently, although this information is not explicitly obtained during a conven-
tional dynamic analysis, it can be calculated with very little e8ort for each time-step of
integration.
Several families of methods may be distinguished in what has been called >exible multi-
body dynamics: the -nite segment method, illustrated in Reference [3], in which the de-
formable body is assumed to consist of a set of rigid bodies connected by spring and=or
dampers; the large rotation vectors formulations, further explained in Reference [4] which
de;ne the element nodal co-ordinates with respect to the global co-ordinate system; the mov-
ing frame approach, clearly described in Reference [5], characterized by the use of a local
reference frame for each >exible body, where elastic deformations with respect to an unde-
formed con;guration are expressed; and the absolute nodal co-ordinate formulation, devel-
oped by Shabana [6], which employs absolute displacements and global slopes as the element
co-ordinates, in order to avoid in;nitesimal rotations and the subsequent linearization they
produce.
The method here presented falls into the moving frame approach family. Peculiarities of this
method may be pointed out through its comparison with other well-known methods belonging
to the same family as, for example, the modal reduction method proposed by Shabana [7].
Copyright ? 2001 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2001; 51:1033–1052
COMPARISON IN TERMS OF ACCURACY AND EFFICIENCY 1035
The main di8erences between both methods are listed below
Modal reduction Advantages of the

Proposed method method proposed method
Natural co-ordinates Reference point co-ordinates Smaller number of co-ordinates
Static and dynamic modes Dynamic modes Easier constraint introduction
Non-consistent velocities Consistent velocities No need of inertia shape integrals
2. THE MSD FORMULATION
The proposed approach to solve the dynamics of multibody systems consists of an improved
version of the index-3 augmented Lagrangian formulation with mass-orthogonal projections
given in Reference [8]. Description of the method includes the form of the dynamic equa-
tions and the integration scheme, as long as both are closely related. For the modelling,
fully-Cartesian co-ordinates, also known as natural co-ordinates, are used. Both rigid and
>exible links are modelled with this type of co-ordinates in a total compatible form. Further
explanation about these co-ordinates can be found in Reference [9].
The equations of motion are given by an index-3 augmented Lagrangian formulation in the
form
MqO + Tq + Tq ∗ = Q (1)
where M is the mass matrix, qO are the accelerations, q the Jacobian matrix of the constraint
equations, the penalty factor, the constraints vector, ∗ the Lagrange multipliers and
Q the vector of applied and velocity-dependent inertia forces. The Lagrange multipliers are
obtained from the following iteration process:
∗i+1 = ∗i + i+1 ; i = 0; 1; 2; : : : (2)
where the value of ∗0 is extrapolated from the Lagrange multipliers already calculated in
previous time steps (note that sub-index n indicates the time step, and sub-index i refers to
the iteration step within a time step)
(∗0 )n+1 = 2n − n−1 (3)
As integration scheme, the implicit single-step trapezoidal rule has been adopted. The
corresponding di8erence equations in velocities and accelerations are

2 2
q̇n+1 = qn+1 + q̇ˆ n with q̇ˆ n = − qn + q̇n (4)
St St

4 4 4
qO n+1 = qn+1 + qÔ n with qÔ n = − qn + q̇ + qO n (5)
St 2 St 2 St n
Dynamic equilibrium can be established at time step n + 1 by introducing the di8erence

Equations (4) and (5) into the Equations of motion (1), thus yielding
4
Mqn+1 + Tqn+1 (n+1 + n+1 ) − Qn+1 + MqÔ n = 0 (6)
St 2
For numerical reasons, the scaling of Equation (6) by a factor of St 2 =4 seems to be
convenient, thus yielding
St 2 T St 2 St 2
Mqn+1 + qn+1 (n+1 + n+1 ) − Qn+1 + MqÔ n = 0 (7)
4 4 4
or, symbolically f(qn+1 ) = 0.
In order to obtain the solution of this non-linear system, the widely used iterative Newton–
Raphson method may be applied

@f(q)
Sqi+1 = − [f(q)]i (8)
@q i
where
St 2
[f(q)] = MqO + Tq + Tq ∗ − Q (9)
4
and the approximated tangent matrix is

@f(q) St St 2 T
=M + C+ q q + K (10)
@q 2 4
where C and K represent the contribution of damping and elastic forces of the system provided
they exist.
A closer look at the tangent matrix reveals that ill-conditioning may appear when the time
step becomes small. It may be seen in Equation (10) that K and the constraint terms are
multiplied by St 2 , C by St and M is not a8ected by the step size. As a consequence when
St reaches small values, large round o8 errors will occur. In fact, it has been demonstrated
in Reference [10] that for an index-3 di8erential algebraic equation the tangent matrix has a
condition number of order 1=St 3 . Consequently, the method is bound to have round-o8 errors
for step sizes smaller than 10−5 , which lets a su)cient range for solving practical problems.
The procedure explained above yields a set of positions q n+1 that not only satis;es the Equa-
tions of motion (6), but also the constraint conditions = 0. However, it is not expected that
the corresponding sets of velocities and accelerations satisfy ˙ = 0 and O = 0, because these
conditions have not been imposed in the solution process. To overcome this di)culty, mass-
damping-sti8ness-orthogonal projections in velocities and accelerations are performed. It can
be seen that the projections leading matrix is the same tangent matrix appearing in Equation
(10). Therefore, triangularization is avoided and projections in velocities and accelerations are
carried out with just forward reductions and back substitutions.
If q̇∗ and qO ∗ are the velocities and accelerations obtained after convergence has been
achieved in the Newton–Raphson iteration, their cleaned counterparts q̇ and qO are calculated
from

St St 2 T St St 2 St 2 T
M+ C+ q q + K q̇ = M + C+ K q̇∗ − t (11)
2 4 2 4 4 q
for the velocities, and

St St 2 T St St 2 St 2 T ˙ ˙ t)
M+ C+ q q + K qO = M + C+ K qO ∗ − (q q̇ + (12)
2 4 2 4 4 q
for the accelerations.
Sparse matrix technology has been implemented to solve all the linear sets of equations
that arise when applying this method. The fact that the leading matrix is symmetric and
positive-de;nite has been taken into account as well.
3. CONSIDERATION OF FLEXIBLE BODIES
The modelling of >exible bodies is carried out by a moving frame approach with component
mode synthesis for small elastic displacements. The global motion of each >exible body is
described as a superposition of the large-amplitude motion of a moving frame, rigidly attached
to a certain point of the body, and the small elastic displacements of the body with respect
to an undeformed con;guration, taken as reference. Any deformed con;guration of the body
is expressed as a linear combination of static and dynamic modes, in the sense of the mode
synthesis approach with ;xed boundaries. The static modes depend on the natural co-ordinates
de;ned at the joints of the body, while the number of internal, dynamic modes should be
decided by the analyst. A detailed description of the kinematics of this method can be found
in Reference [11].
However, the approach proposed in the above-mentioned reference for the dynamics has
not been adopted, as long as it showed to be too involved, particularly in what refers to the
form of the inertia terms. Therefore, the approximation suggested by Geradin and Cardona
[12] has been considered in the way proposed by Avello [13] when natural co-ordinates are
used for the modelling. This modi;ed approach provides the same level of accuracy while
notably simplifying the formulation of the mass matrix and velocity-dependent forces vector.
In what follows, the approach is brie>y described.
Figure 1 shows a general >exible body. The position of any point of the body can be
expressed as
r = ro + A(Uru + uU ) (13)
Figure 1. General >exible body.
where ro is the position of the origin of the local reference frame, A the rotation matrix, rUu
the position of the point in local co-ordinates for the undeformed con;guration of the body,
and uU the elastic displacement su8ered by the point, also expressed in local co-ordinates. If
the ;nite-element method is used, the elastic displacement can be written as
uU = NuU ∗ (14)
where N stands for the interpolation matrix which contains the interpolation functions, and
uU ∗ are the elastic displacements of the nodes. Substituting Equation (14) into Equation (13)
yields
r = ro + A(Uru + NuU ∗ ) (15)
At this point, the approach called co-rotational in Reference [12] is introduced. It implies
the following interpolation for the velocity of any point of the >exible body
ṙ = Nv∗ (16)
where v∗ are the velocities of the nodes. Obviously, expression given by Equation (16) for
the velocities is not consistent with expression given by Equation (15) for the positions, as
long as the former should be the derivative of the latter.
Based on Equation (16), the kinetic energy of the >exible body can be written in the form

1 1 1
T= T
ṙ ṙ dm = v∗T NT Nv∗ dm = v∗T MFEM v∗ (17)
2 v 2 v 2
where MFEM is the mass matrix of the ;nite element method. In order to express the kinetic
energy in terms of the problem variables, velocities of the nodes, v∗ , should be related with
the problem variables. Di8erentiating Equation (15) and evaluating the result at the nodes
gives
v∗ = ṙo + Ȧ(Uru + uU ∗ ) + AuU̇ ∗ (18)
and considering that unit vectors a, b and c are the columns of rotation matrix A,
 
 ṙo 

 ȧ  
 
   





 v1  I a 11 I a 12 I a 13 I A · · · · · · · · · 0 
 ḃ 


 
 
 


 .
. 
  .
. .
. .
. 
 ċ 


 .   . . .  
    
 uU̇ 1 

∗ .  . . . 
v = .. =  .. .. ..  . (19)

 . 
  
 . ..  .. 

 

 . 
 . ..  .  

 . 

 . . . 

 .



  
 . 

vp I ap1 I ap2 I ap3 I 0 · · · · · · · · · A   . 


 . 


 . 
  
uU̇p
where p is the number of nodes of the body, I the 3 × 3 identity matrix, and
a11 = rUux1 + uUx1 a12 = rUuy1 + uUy1 a13 = rUuz1 + uUz1
(20)
ap1 = rUuxp + uUxp ap2 = rUuyp + uUyp ap3 = rUuzp + uUzp
Elastic displacements can be expressed as superposition of static and dynamic modes,

U i i +
ns nd
uU ∗ = U j j (21)
i=1 j=1
where ns and nd are the number of static and dynamic modes, respectively, U i are the static
modes, i are their amplitudes, U j are the dynamic modes, and j are their amplitudes.
Substituting Equation (21) into Equation (19) leads to
 

 ṙo 

 ȧ 

 

 
 

1
U 1 · · · A 1
U n A 1
U 1 · · · A
Un 1  ḃ 

I b11 I b12 I b13 I A 
 

. .. ..
s
.. .. 
d

 ċ 

.  

. . . . .   ˙ 1 

 . . . . .  .
v =
∗
. . .
. .
. .
. .  .
.  .  = Bq̇ (22)
  
 .. .. .. .. ..   ˙ns 
. . . . .  






U p U p
U p
U p 
 ˙ 

I bp1 I bp2 I bp3 I A1 · · · Ans A1 · · · And   1 


 . 
 . 
 . 


˙  
n d
where U js is a 3 × 1 vector
U ri is a 3 × 1 vector containing static mode i evaluated at node r,
containing dynamic mode j evaluated at node s, and

U ix1 i +
U iy1 i +
ns nd ns nd
b11 = rUux1 + U jx1 j b12 = rUuy1 + U jy1 j
i=1 j=1 i=1 j=1

ns nd
b13 = rUuz1 + U iz1 i +
U jz1 j

i=1 j=1
(23)

U iyp i +
ns nd ns nd
bp1 = rUuxp + U ixp i +
U jxp j
bp2 = rUuyp + U jyp j
i=1 j=1 i=1 j=1
ns
nd
bp3 = rUuzp + U izp i +
U jzp j

i=1 j=1
The vector appearing on the right-hand side of Equation (22), q̇, is the vector of the
derivatives (velocities) of the body variables q, and then, substituting the value of v∗ given
by this equation into Equation (17), the kinetic energy of the body can be written as
T = 12 q̇T BT MFEM Bq̇ = 12 q̇T Mq̇ (24)
which means that the mass matrix of the general >exible body is given by the expression
M = BT MFEM B (25)
Application of Lagrange equations with the kinetic energy of the >exible body given by
Equation (24) leads to calculate the vector of velocity-dependent inertia forces
Qv = −BT MFEM Ḃq̇ (26)
Therefore, it can be seen that, unlike their predecessors in Reference [11], the ;nal expres-
sions for the inertia terms obtained through the above-mentioned approach, given by Equations
(25) and (26), are compact and relatively easy to evaluate.
4. NUMERICAL EXAMPLES
In this work, the problem of determining the level of stress undergone by the components of
a mechanical system during its motion has been addressed through two di8erent methods:
(a) A multibody-system dynamic analysis, just considering the >exibility of those bodies
whose stresses are of interest. This method will be referred to as multibody-system dy-
namics (MSD). The corresponding code has been developed by the authors as an imple-
mentation of the proposed formulation, already described.
(b) A dynamic analysis performed by a non-linear module of a FEA program which takes
into account large displacements and ;nite rotations. This method will be referenced as
Finite Element Analysis (FEA). For this purpose, commercial code COSMOS=M 1.75 A
has been used.
The objective is to demonstrate that the MSD method, proposed by the authors, enables to
achieve a similar level of accuracy to that provided by the FEA method. Moreover, it will be
seen that the MSD method is largely more e)cient. To this aim, comparison between both
formulations in terms of accuracy and e)ciency is established for four examples. All the
calculations have been performed on a SGI Indigo2 IMPACT with one processor R4400SC
at 200 MHz and 2 Mb of secondary cache memory.
4.1. Flexible link with a bang–bang torque pro-le at the articulated joint
This ;rst example, shown in Figure 2(a), consists of a beam pinned at one end to the ground,
which starts from the rest and undergoes the torque depicted in Figure 2(b). Gravity e8ects are
neglected. Physical properties of the beam are: mass density 8000 kg=m3 , modulus of elasticity
2 × 1011 N=m2 , length 1.5 m, cross-sectional area 10−4 m2 , moment of inertia 10−10 m4 .
For the MSD method, the modelling of the beam has been carried out as illustrated in
Figure 3. At the pinned end, point p1 and unit vectors v1 and v2 are de;ned, thus constituting
the local reference frame of the body. In this case, point p1 is ;xed. At the free end, point p2 is
de;ned, whose local displacement in v2-direction activates static mode . To better represent
the deformed con;guration of the beam, dynamic modes 1 and 2 are also considered. They
Figure 2. (a) Pinned-free beam; (b) Bang-bang torque.
Figure 3. Modelling of a >exible pinned-free beam with the MSD method.
Figure 4. History of the vertical co-ordinate of the beam with MSD and FEA methods.
are the two ;rst natural modes of vibration of the beam with ;xed boundaries (points p1 and
p2, and unit vectors v1 and v2), which means that, for their calculation, left end must be
clamped and right end must be pinned.
As a consequence, the vector of problem variables results in
qT = {v1x v1y v2x v2y 1 2 p2x p2y } (27)
where , 1 , 2 are the amplitudes of the static and dynamics deformation modes, respectively.
Therefore, the total number of variables is 9, with only 4 independent.
For the FEA method, the beam has been modelled by making a mesh of ten two-dimensional
beam elements (BEAM2D), all of them identical, with nodes of three degrees of freedom:
two displacements in the plane of the beam and the corresponding slope. As the node placed
at the pinned end of the beam can only experiment rotation, the total number of variables
rises to 31 for this method.
A simulation of 2 s is performed through both methods. The time-step in both cases is
0.001 s. As a measurement of the motion obtained in each case, the history of the vertical
co-ordinate of the free end of the beam is shown in Figure 4. Regarding the stress ;eld of
the beam, Figure 5 represents the history of the normal stress at the upper point of the middle
section of the beam, also calculated with both methods.
Figure 5. History of the maximum normal stress at the middle section of the
beam with MSD and FEA methods.
Figure 6. Modelling of a double pendulum with the MSD method.
4.2. Double-pendulum under gravity e8ects

The second example is a double-pendulum that starts from the rest in horizontal position,
and falls under gravity action. Both links are identical and their physical properties are: mass
density 2000 kg=m3 , modulus of elasticity 7 × 1010 N=m2 , length 1.5 m, cross-sectional area
1:2 × 10−3 m2 , moment of inertia 4 × 10−7 m4 . In this case, the objective is to simulate 5 s of
motion of the system, determining the stresses su8ered by the ;rst link along the time.
For the MSD method, the modelling of the double pendulum is illustrated in Figure 6. As
only stresses at the ;rst link are of interest, it is modelled as >exible, while the second link is
modelled as rigid. Modelling of the >exible body is identical to that described in the previous
example.
Hence, the vector of problem variables is the following:
qT = {v1x v1y v2x v2y 1 2 p2x p2y p3x p3y } (28)
which means that the total number of variables is 11, with only 5 independent.
For the FEA method, each link has been modelled in an identical way of that explained
above for the previous example, imposing in this case the identity of nodal displacements at
the connection between both links. Therefore, the total number of variables is 64 with only
62 independent.
The simulation of 5 s is carried out through both methods using a time-step of 0.001 s.
Figure 7 shows the history of the vertical co-ordinate of the free end of the double-pendulum.
Figure 7. History of the vertical co-ordinate of the end of a double

pendulum with MSD and FEA methods.
Figure 8. Maximum normal stress at the middle section of the ;rst link of a double-pendulum.
In Figure 8, the maximum normal stress at the middle section of the ;rst link due to bending
e8ects is represented.
It might be useful to see the behaviour of the MSD method if both links are considered as
>exible. It has been previously said that, to take the maximum bene;t from this method, the
idea is to model as >exible only those bodies whose stresses are of interest for the analyst.
Acting in this way, the number of variables is kept small and, therefore, the calculation is
faster. However, as shown in Figure 8, stress correlation between MSD and FEA methods can
be notably improved. This improvement is achieved when both links are taken as >exible in
the MSD method, as illustrated in Figure 9. The time spent for the calculation is multiplied
by a factor of 3.7. In consequence, the following statement can be established for the MSD
method: if high accuracy is needed, more bodies should be modelled as >exible; if speed is
needed, the number of >exible bodies should be limited to those whose stresses are of interest.
Figure 9. Stress comparison if both links are considered >exible in MSD method.
Figure 10. Slider-crank mechanism under triangular force pro;le.
Figure 11. Links of the slider-crank mechanism.
4.3. Slider-crank mechanism

The third example, shown in Figure 10, is an slider-crank mechanism that starts from the rest
in horizontal position, and undergoes the triangular force pro;le also shown in Figure 10.
The shape and dimensions (m) of the links are illustrated in Figure 11, and their physi-
cal properties are: mass density 7850 kg=m3 , modulus of elasticity 2:1 × 1011 N=m2 , Poisson
modulus 0.3. The simulation lasts 1 s and stresses su8ered by the connecting rod along the
time should be determined.
For the MSD method, the modelling of the slider-crank mechanism may be seen in
Figure 12. As only stresses at the coupler are of interest, it is modelled as >exible, while the
Figure 12. Modelling of an slider-crank mechanism with the MSD method.
Figure 13. (a) Axial static mode; (b) Bending static mode.
crank is modelled as rigid. Modelling of the >exible body is very similar to that described in
the previous example, but some remarks should be done.
As it can be seen in Figure 12, an additional static mode has been considered: the axial
static mode 1 , controlled by the displacement of point p3 along the coupler line. In the two
previous examples, this axial static mode had been neglected, as long as axial deformations are
usually much smaller than bending deformations, but introduce high frequencies that make the
numerical integration becomes more di)cult. However, in this case axial stresses are dominant,
so that they cannot be neglected and the axial static mode 1 should be considered.
On the other hand, it has been shown in Figure 11 that, unlike the two previous examples,
in this case the links of the mechanism are no longer represented by beams. Therefore, the
static and dynamic modes depicted in Figure 12 are just a simpli;cation of the true modes.
Figures 13 and 14 illustrate, respectively, the authentic static and dynamic modes.
For this example, the vector of problem variables is the following:
qT = {p2x p2y v1x v1y v2x v2y 1 2 1 2 p3x } (29)
which means that the total number of variables is 11, with only 5 independent.
Figure 14. First- and second-bending dynamic modes.
Figure 15. FEA model of the slider-crank mechanism.
For the FEA method, the mechanism has been modelled using a mesh of 341 plane stress
quadrilateral elements (PLANE2D), each of them featuring four nodes with two degrees of
freedom (two orthogonal displacements in the plane), leading to a total number of 351 nodes.
Figure 15 shows the complete model of the mechanism. The pinned crank origin can only
experiment rotation, so that displacements of the corresponding node has been prevented.
Moreover, in order to keep the slider motion constrained to the horizontal line, vertical dis-
placement of the corresponding node has been prevented too. Finally, to impose the revolute
joint between both links, displacements of the nodes belonging to them at that point have
been made identical by means of two constraint equations. Therefore, the total number of
variables rises to 699 with 697 independent.
The simulation of 1 s is performed through both methods. Best time-steps are 0:00075 s for
the MSD method and 0:0001 s for the FEA method. As a measurement of the motion obtained
Figure 16. History of the horizontal co-ordinate of the slider with MSD and FEA methods.
Figure 17. Normal stress at the centre of the middle section of the connecting rod.
in each case, the history of the horizontal co-ordinate of the slider is shown in Figure 16,
while stress analysis results are represented by the history of the normal stress at the centre
of the middle section of the connecting rod, illustrated in Figure 17.
4.4. Four-bar mechanism with assembly defect

The fourth example is a four-bar mechanism, shown in Figure 18. The angular velocity !
of the left crank is kinematically guided at a constant rate of 1 rad=s. Gravity e8ects are
neglected. Bar lengths are: AB = 0:12 m; BC = 0:24 m; CD = 0:12 m; AD = 0:24 m. Typically,
the axes of rotation of the four revolute joints are orthogonal to the plane of the mechanism,
and then it moves as a planar mechanism. However, in this case, the axis of rotation of joint
C is at a 5◦ angle, with respect to, the plane normal to simulate an assembly defect in the
Figure 18. Four-bar mechanism with Figure 19. Modelling of the four-bar mechanism
assembly defect. with the MSD method.
mechanism. If the bars were rigid, no motion would be possible as the mechanism would
lock. For elastic bars, motion becomes possible, but generates large internal forces. Physical
properties of the bars are: mass density 3000 kg=m3 , modulus of elasticity 7 × 1010 N=m2 ,
Poisson modulus 0.33. The section of the bars is circular with a diameter of 5 mm. For this
example, the objective is to simulate 10 s of motion, determining the stresses su8ered by the
coupler along the time.
For the MSD method, the modelling is illustrated in Figure 19. Although, as said in the
previous paragraph, only stresses at the coupler are to be computed, this time the three links
should be modelled as >exible; otherwise, the mechanism could not move. As shown in
Figure 19, the mechanism has been de;ned, following the general rules of the method, by
points p1 to p4, with p1 and p4 ;xed, unit vectors v0 to v8, with v0 and z-components of
v1, v2, v7 and v8 ;xed, and angle ’, which makes easier to introduce the kinematic guidance
at the left crank. Bars have been considered as beams, and their deformed con;gurations
represented by ;ve static modes (three for the point de;ned at the opposite end of the local
reference frame of the link, and two for the unit vector at the same location), and four dynamic
modes (the two ;rst natural bending modes in each plane). Therefore, the total number of
variables is 54, with only 25 independent.
For the FEA method, the two cranks have been modelled by making a mesh of ;ve three-
dimensional beam elements (BEAM3D), all of them identical, with nodes of six degrees of
freedom: three displacements and three rotations. For the coupler, 10 elements of the same
type have been used, as long as its length is twice the length of the cranks. To consider the
kinematic guidance of the left crank, rotation of the node de;ned at joint A has been imposed
along the time. To de;ne revolute joints at points A and D, ;ve degrees of freedom of the
corresponding nodes have been prevented, leaving free the rotation around z-axis only. For
revolute joints at points B and C, the same technique cannot be applied, as long as their
orientation is expected to change during the motion. To overcome this problem, modelling of
joints B and C has been carried out as illustrated in Figure 20: two extra-nodes have been
de;ned for each link at the connection points and, afterwards, identity of displacements has
been imposed between each pair of nodes. In this way, three points along the joint axis of
each neighbour link are made coincident at all times, thus assuring revolute behaviour of
joints B and C. Therefore, the total number of variables for this method rises to 176, with
only 158 independent.
Figure 20. Modelling of revolute joints B and C for the FEA method.
Figure 21. Motion of the four-bar mechanism with assembly defect: (a) z-displacement
of point B; (b) z-displacement of point C.
The simulation of 10 s is performed through both methods. Best time-steps are 0:01 s for
the MSD method and 0:001 s for the FEA method. As a measurement of the motion obtained
in each case, Figure 21 shows the z-displacement of points B and C with both methods. Stress
analysis results are depicted in Figure 22, where torsional and bending moments at the middle
section of the coupler are represented for MSD and FEA methods.
Figure 22. Torsional and bending moments at the middle section of the coupler:
(a) for MSD method;(b) for FEA method.
Table I. E)ciency of MSD and FEA methods.
Example CPU time CPU time Ratio

No. MSD (s) FEA (s) FEA=MSD
1 5 220 44.0
2 10 894 89.4
3 112 14 992 133.8
4 48 2646 55.1
5. EFFICIENCY
So far, results of motion and stresses have been obtained through both methods being com-
pared, but nothing has been said in what respects to e)ciency. Table I presents the CPU
times spent by MSD and FEA methods to solve the four examples previously described.
Table I shows that MSD method is largely more e)cient than FEA method. In fact, as it can
be seen in the last column of the table, the method proposed by the authors is, approximately,
two orders of magnitude faster. Moreover, the di8erence in speed between both methods is
greater when the number of bodies of the system goes up or when the links are more complex.
It has been said above that only bodies whose stresses are of interest should be modelled
as >exible when applying the MSD method and, obviously, this fact gives advantage to the
proposed method. However, even in the case that all the bodies are modelled as >exible, like
in the last example, the di8erence in e)ciency between both methods remains at the same
level. Evidently, the four examples analysed are academic, complexity of their links is reduced
and the number of bodies keeps small in all the cases. Therefore, the gain in e)ciency that
could be provided by the MSD method with respect to the FEA method in big-size realistic
multibody systems can be enormous.
Although complete information about commercial codes implementation is never available,
reasons for the substantial di8erence in e)ciency found between the two methods compared
in the paper might be the following:
(a) The problem size is kept moderate when applying the MSD method, as long as natural
co-ordinates plus deformation modes are used as variables. However, the use of nodal
variables by the FEA formulation leads to huge problem sizes. This e8ect is more acute
due to the fact that the MSD method only models as >exible those links whose stresses
are of interest, while the FEA method is forced to consider >exibility of all bodies in the
system.
(b) High frequencies in the solution are eliminated with the MSD approach, since modal
reduction is used. Hence, numerical integration of the dynamic equations becomes easier
than with the FEA method, which should face a curlier response of the system.
(c) Linear extrapolation of Lagrange multipliers proposed for the MSD method provides quick
convergence of the Newton–Raphson iteration process performed at each time step. Fur-
thermore, mass–damping–sti8ness–orthogonal projections contribute to keep velocities and
accelerations clean along the integration procedure, thus improving too the convergence
ratio of the algorithm.
6. CONCLUSIONS
Based on the results, the following conclusions may be established:

• For similar accuracy, the proposed MSD method is largely more e)cient than the FEA
approach, near two orders of magnitude faster. Reasons for this fact might be the di8erence
in problem size between both modelling techniques, the presence of higher frequencies
when applying the FEA method, and the excellent performance achieved by the numerical
integration procedure of the MSD formulation.
• The aforementioned trend is increased when the number of bodies of the mechanical system
and=or their complexity grows, as long as the di8erence in problem size between both
methods goes up for more realistic and sophisticated systems.
• The two previous conclusions do not mean that FEA codes are not longer useful. In fact,
the MSD method needs the help of FEA codes at its preprocessing stage. What this paper
suggests is that perhaps FEA codes are not the best choice when trying to determine stresses
in components of a multibody system as motion takes place.
• Moreover, FEA is an excellent tool when the physics of the problem become very involved,
like in cases with impacts, cracks, non-linear material properties, etc. However, customized
software tools based on formulations like the proposed MSD method could be of great
interest for many industrial design applications of multibody systems, which handle standard
components that hardly present any special complexity.
REFERENCES
1. Lowen GG, Chassapis C. The elastic behavior of linkages: an update. Mechanism and Machine Theory 1986;
21:33– 42.
2. Belytschko T, Hsieh BJ. Nonlinear transient ;nite element analysis with convected co-ordinates. International
Journal for Numerical Methods in Engineering 1973; 7:255–271.
3. Huston RL, Wang Y. Flexibility e8ects in multibody systems. Computer-Aided Analysis of Rigid and Flexible
Mechanical Systems. Pereira MS, Ambrosio JAC (eds). Kluwer Academic Publishers: Dordrecht, 1994;
351–376.
4. Geradin M, Cardona A, Doan DB, Duysens J. Finite element modeling concepts in multibody dynamics. In
Computer-Aided Analysis of Rigid and Flexible Mechanical Systems, Pereira MS, Ambrosio JAC (eds). Kluwer
Academic Publishers: Dordrecht, 1994; 233–284.
5. Shabana AA. Computer implementation of >exible multibody equations. In Computer-Aided Analysis of Rigid
and Flexible Mechanical Systems, Pereira MS, Ambrosio JAC (eds). Kluwer Academic Publishers: Dordrecht,
1994; 325–349.
6. Shabana AA. Finite element incremental approach and exact rigid body inertia. ASME Journal of Mechanical
Design 1996; 118(2):171–178.
7. Shabana AA. Dynamics of Multibody Systems. Wiley: New York, 1989.
8. Bayo E, Ledesma R. Augmented Lagrangian and mass-orthogonal projection methods for constrained multibody
dynamics. Nonlinear Dynamics 1996; 9:113–130.
9. Garc2Va de Jal2on J, Bayo E. Kinematic and Dynamic Simulation of Multibody Systems—The Real-Time
Challenge. Springer: New York, 1994.
10. Brenan KE, Campbell SL, Petzold LR. The Numerical Solution of Initial Value Problems in Di8erential-
Algebraic Equations. Elsevier: Amsterdam, 1989.
11. Cuadrado J, Cardenal J, Garc2Va de Jal2on J. Flexible mechanisms through natural co-ordinates and component
synthesis: an approach fully compatible with the rigid case. International Journal for Numerical Methods in
Engineering 1996; 39:3535–3551.
12. Geradin M, Cardona A. Eight VPI & SU Symposium on Dynamics and Control of Large Space Structures,
Blacksburg, USA, May 6–8, 1991.
13. Avello A. Simulaci2on Din2amica Interactiva de Mecanismos Flexibles con Pequeñas Deformaciones. Ph.D.
Thesis, Universidad de Navarra, Spain, 1995.

A Comparison in Terms of Accuracy and E) Ciency Between A MBS Dynamic Formulation With Stress Analysis and A Non-Linear FEA Code

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A Comparison in Terms of Accuracy and E) Ciency Between A MBS Dynamic Formulation With Stress Analysis and A Non-Linear FEA Code

Uploaded by

Copyright:

Available Formats

INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING

Int. J. Numer. Meth. Engng 2001; 51:1033–1052

A comparison in terms of accuracy and e)ciency between a

J. Cuadrado∗;† , R. Guti2errez, M. A. Naya and P. Morer

Contract=grant sponsor: The Galician Government; contract=grant number: XUGA16601A98

Received 3 December 1999

The main di8erences between both methods are listed below

Modal reduction Advantages of the

2. THE MSD FORMULATION

MqO + Tq  + Tq ∗ = Q (1)

∗i+1 = ∗i + i+1 ; i = 0; 1; 2; : : : (2)

(∗0 )n+1 = 2n − n−1 (3)

Dynamic equilibrium can be established at time step n + 1 by introducing the di8erence

for the velocities, and

3. CONSIDERATION OF FLEXIBLE BODIES

Figure 1. General >exible body.

Elastic displacements can be expressed as superposition of static and dynamic modes,

Figure 2. (a) Pinned-free beam; (b) Bang-bang torque.

Figure 3. Modelling of a >exible pinned-free beam with the MSD method.

qT = {v1x v1y v2x v2y  1 2 p2x p2y } (27)

Figure 6. Modelling of a double pendulum with the MSD method.

4.2. Double-pendulum under gravity e8ects

Figure 7. History of the vertical co-ordinate of the end of a double

Figure 10. Slider-crank mechanism under triangular force pro;le.

Figure 11. Links of the slider-crank mechanism.

4.3. Slider-crank mechanism

Figure 12. Modelling of an slider-crank mechanism with the MSD method.

qT = {p2x p2y v1x v1y v2x v2y 1 2 1 2 p3x } (29)

Figure 14. First- and second-bending dynamic modes.

Figure 15. FEA model of the slider-crank mechanism.

4.4. Four-bar mechanism with assembly defect

Table I. E)ciency of MSD and FEA methods.

Example CPU time CPU time Ratio

Based on the results, the following conclusions may be established:

You might also like

MqO + Tq + Tq ∗ = Q (1)

∗i+1 = ∗i + i+1 ; i = 0; 1; 2; : : : (2)

qT = {v1x v1y v2x v2y 1 2 p2x p2y } (27)

qT = {p2x p2y v1x v1y v2x v2y 1 2 1 2 p3x } (29)