Professional Documents
Culture Documents
The manuscript was received on 26 May 2008 and was accepted after revision for publication on 17 February 2009.
DOI: 10.1243/14644193JMBD164
Abstract: One of the characteristics of floating frame of reference (FFR) formulations for flexible
multibody dynamics is the fact that the inertia terms are highly non-linear. At every time-step,
both the mass matrix and the velocity-dependent forces vector must be updated, and this can
become the most CPU intensive task. This work studies the efficiency of two different methods
for performing this operation, when applied to both a formulation in absolute coordinates and
another in relative coordinates.
The first method calculates the inertia terms by projecting the finite element (FE) mass matrix
into the generalized coordinates, by means of a variable projection matrix. The second one
calculates the inertia shape integrals at a preprocessing stage and uses them for obtaining the
inertia terms in a more efficient way, at the cost of a more involved implementation. Both methods
have been tested when used in combination with either the FFR absolute or relative formulation,
by simulating a vehicle with 12 flexible elements. The results show that the performance can be
considerably increased by means of the preprocessing method, especially in the case of large FE
models, whereas, for small models, the projection method can be more convenient due to its
simplicity.
Keywords: real-time, flexible multibody dynamics, floating frame of reference, inertia shape
integrals
JMBD164 © IMechE 2009 Proc. IMechE Vol. 223 Part K: J. Multi-body Dynamics
148 U Lugrís, M A Naya, A Luaces, and J Cuadrado
that, despite using component mode synthesis, the The formulation in relative coordinates is an O(n3 )
CPU-time depends on the size of the original FE semi-recursive formulation, which can be consid-
model, so that performance can be seriously affected ered as an extension of the composite rigid-body
when complex FE models are used. method [15] to the flexible case. The dynamic terms
In order to address this problem, the preprocessing are first obtained in an intermediate set of Cartesian
method uses the inertia shape integrals obtained in coordinates q, being then projected into the relative
a preprocessing stage [1, 7, 8] in such a way that the coordinates z by means of a recursive procedure [11].
CPU-time will only depend on the size of the reduced Therefore, the equations of motion have exactly the
model (i.e. the number of deformation modes used); same form seen in equation (1), but expressed in terms
this means that the mesh can be refined as much of z instead of q.
as needed without introducing any penalty to the In order to integrate the equations of motion, they
simulation time, but only to the preprocessing time. are first combined with the equations of the numerical
In the present work, the implementation of the integrator, leading to a non-linear algebraic system of
preprocessing method into two existing FFR formu- equations, with the dependent positions as unknowns,
lations, which already use the projection method, which is solved by means of a Newton–Raphson itera-
is first addressed. One of the formulations [9] uses tion [14]. Once convergence is attained into the time-
absolute coordinates, more specifically natural coor- step, the velocities and accelerations are corrected for
dinates [10], whereas the other one is based on them to satisfy the first and second derivatives of the
dependent relative coordinates [11]. The preprocess- constraints. This is carried out by means of a modi-
ing method has been already implemented in a formu- fied version [14] of the mass-orthogonal projections
lation in natural coordinates by Cuadrado et al. [12], introduced by Bayo and Ledesma [16].
although the modelling differs from that used in
reference [9], leading to very involved expressions for
the inertia terms. 3 FLEXIBLE BODY MODELLING
In order to assess the efficiency improvement
obtained by introducing the inertia shape integrals, The position of an arbitrary point of a deformed elas-
the same benchmark system used in reference [11] for tic body r can be obtained, as shown in Fig. 1, from
comparing absolute and relative coordinates is cho- the absolute position of the origin of its FFR r 0 , the
sen, i.e. the Iltis vehicle [13]. In the present work, three orthogonal unit vectors that define the axes of
the vehicle is modelled considering always 12 flexible the frame of reference u, v, and w, and the position
elements, which can be discretized using a variable of the point in local coordinates r̄, which can be in
number of finite elements. The same manoeuvre is turn decomposed into the undeformed local position
simulated several times, by using the projection and r̄ u and the elastic displacement r̄ f . If a rotation matrix
preprocessing methods, both in absolute and relative A is defined, containing the three unit vectors of the
coordinates, and with different levels of refinement of FFR as columns, the position r can be written
the FE models.
r = r 0 + Ar̄ = r 0 + A(r̄ u + r̄ f ) (3)
Proc. IMechE Vol. 223 Part K: J. Multi-body Dynamics JMBD164 © IMechE 2009
Efficient calculation of the inertia terms in FFR formulations 149
mode shapes matrix X can be defined, containing in linearly on the generalized velocities q̇, so that it can
each column the displacements vector correspond- be expressed in product form as in equation (5), with
ing to one deformation mode. Accordingly, the modal the following B matrix
amplitudes can be grouped into a vector y, therefore
allowing to express the linear combination in matrix B = [I3 r̄1 I3 r̄2 I3 r̄3 I3 AX] (7)
form as
where I3 is the 3 × 3 identity matrix and each com-
r̄ f = Xy (4) ponent r̄i of the local deformed position can be
expressed as
The deformation modes X are chosen, as done
in [9, 11], according to the Craig–Bampton reduction r̄i = r̄ui + X i y, i = 1, 2, 3 (8)
method [3], using a set of static boundary modes and
fixed interface normal modes, although the methods In this equation, r̄ui and X i are the ith rows of r̄ u and X,
discussed in the present work are applicable to any respectively, i.e. the undeformed position and modal
Rayleigh–Ritz based reduction. displacements in the i direction.
JMBD164 © IMechE 2009 Proc. IMechE Vol. 223 Part K: J. Multi-body Dynamics
150 U Lugrís, M A Naya, A Luaces, and J Cuadrado
the following discrete form, when isoparametric FEs of isoparametric elements and, moreover, the blocks
are used become less sparse due to the rotation.
In practice, the projection matrix for non-
1 1
T = ṙ T ṙ dm = ṙ ∗T M∗ ṙ ∗ (12) isoparametric elements can be approximated as if they
2 V 2 were isoparametric, thus using the already described
B∗ matrix with three rows per node. Since it is not com-
where M∗ is the integral of NT N, which is nothing but patible with the original FE mass matrix, a custom one
the well-known FE mass matrix, and ṙ ∗ is a vector con- is used, obtained as the mass integral of NT N, being
taining the absolute velocities of all the FE nodes. In N the isoparametric interpolation functions. By apply-
the case of non-isoparametric elements, the angular ing this approximation, the CPU-time is significantly
velocities of the nodes must be added to the the vec- improved with almost no effect on accuracy.
tor of nodal velocities. Moreover, the nodal velocities
must be rotated to the local frame of the body, in order
to be consistent with the FE mass matrix. In isopara- 6 INERTIA SHAPE INTEGRALS METHOD
metric elements this is not necessary since the same
interpolation is used in the three directions of space, In the preprocessing method, the velocity projection
so that the interpolation matrices, and consequently for a generic point is applied first, and after taking q̇
the FE mass matrix, are invariant to rotation. out of the integral, the kinetic energy is obtained as
In the isoparametric case, a B∗ matrix can be
1 1 T
defined, containing the B matrices evaluated at all the T = ṙ ṙ dm = q̇
T T
B B dm q̇ (16)
n FE nodes, in such a way that 2 V 2 V
The velocity-dependent forces vector Q v can be In isoparametric FEs, the inertia terms calculated by
obtained by introducing this mass matrix into the using this method do not differ from those obtained
Lagrange equations. Taking into account that Bq q̇ = Ḃ with the projection method, since the only differ-
and that M∗ is symmetric, it can be found that ence is that the velocity transformation is moved from
∗
the last to the first step. In the projection method,
Q v = −B∗T M∗ Ḃ q̇ (15) the FE discretization is introduced, then the inte-
gration by means of the interpolation functions is
The procedure is, in short, the following: first the B∗ performed, and finally the velocity transformation is
∗
matrix and its derivative Ḃ must be assembled, cal- applied, whereas in the preprocessing method, the
culating B and Ḃ at all the FE nodes by using either velocity transformation is applied in the first place,
equation (7) or equation (11), depending on the for- then the integration step is carried out. In the case
mulation used. Then, the B∗T M∗ product is performed of simple bodies such as beams, the integration can
and used to obtain M and Q v . The projection method be performed by using the analytical functions of the
is very simple to implement but, as can be seen, the deformation modes, otherwise the FE discretization
CPU-time will depend on the size of the FE model, must be introduced.
regardless of the number of mode shapes chosen, The integrals needed for obtaining the mass matrix
which is not desirable since it does not allow one to and the velocity-dependent forces vector can be effi-
take full advantage of the model reduction. ciently calculated by matrix and vector operations if
Non-isoparametric finite elements include infinite- some invariant matrices are extracted. The complete
simal rotations as nodal coordinates, along with the set of invariants consists of 16 mass integrals, includ-
displacements. This implies that the Bi matrix of a FE ing the undeformed positions, the mode shapes, and
node needs three extra rows for the angular velocities, several combinations of products among them. These
and all its 3 × 3 blocks must be rotated to the body integrals can be divided into two sets. The first set
frame, in which the FE mass matrix has been defined. consists of three mass integrals, where only unde-
Consequently, B∗ has twice as many rows as in the case formed positions appear so that they lead to inertia
Proc. IMechE Vol. 223 Part K: J. Multi-body Dynamics JMBD164 © IMechE 2009
Efficient calculation of the inertia terms in FFR formulations 151
terms associated to the undeformed body motion. The The complete set of undeformed geometry integrals
inertia shape integrals [1, 7, 8] are the remaining 13 and inertia shape integrals here defined, along with the
ones, which include the mode shapes so that they are generalized coordinates vector q and its time deriva-
used to obtain the variation of the inertia properties tive q̇, contain all the necessary information needed to
associated to deformation. calculate the mass matrix and the velocity-dependent
The first three integrals can be obtained from forces vector of a deformable body. In order to make
the undeformed geometry of the body, and include the procedure clearer, the inertia terms can be consid-
the mass of the body m, the static moment m̄ u , and the ered as divided into blocks, according to the structure
planar inertia tensor P̄u , all of them calculated in the of the generalized coordinates vector
local frame ⎡ ⎤ ⎧ ⎫
Mrr Mrθ Mrf ⎨ Q vr ⎬
M=⎣ Mθθ Mθf ⎦ , Q v = Q vθ (25)
m= dm, m̄ u = r̄ u dm, P̄u = r̄ u r̄ Tu dm ⎩ ⎭
V V V
sym. Mff Q vf
(19)
where the subindex r refers to the inertia associated to
The first integral does not need to be calculated unless the position of the frame of reference, θ to that of its
the mass of the body is unknown. The static moment rotation, and f to the inertia of the elastic coordinates.
m̄ u can also be directly obtained without integration,
if the mass and the undeformed position of the centre 6.1 Implementation in absolute coordinates
of gravity r̄ Gu are known
6.1.1 Mass matrix
m̄ u = r̄ u dm = mr̄ Gu (20)
V In order to obtain the mass matrix, the BT B product
can be developed
The planar inertia tensor can be derived from the
undeformed inertia tensor J̄u in case it is available
M = BT B dm
1
3 V
⎡ ⎤
P̄u = r̄ u r̄ Tu dm = (J̄u )ii I3 − J̄u = J̄u I3 − J̄u (21) I3 r̄1 I3 r̄2 I3 r̄3 I3 AX
V 2 i=1 ⎢ ⎥
⎢ r̄12 I3 r̄1 r̄2 I3 r̄1 r̄3 I3 r̄1 AX ⎥
= ⎢ ⎢ r̄22 I3 r̄2 r̄3 I3 r̄2 AX ⎥ dm
⎥
where J̄u being the moment of inertia of the unde- V ⎣ ⎦
sym. r̄32 I3 r̄3 AX
formed body with respect to the origin of the local
frame of reference. XT X
All the remaining integrals involve the mode shapes (26)
X; therefore, they will be used to obtain the variable In what follows, the evaluation of the different terms by
part of the inertia terms. Three kinds of these integrals means of the inertia shape integrals will be addressed
can be defined, generating a total of 13 constant matri- in detail.
ces. The first one is the integral of the mode shapes,
which results in a 3 × nm matrix, being nm the number
Mass terms associated to the reference coordinates.
of columns in X, i.e. the number of deformation modes
The first three blocks Mrr , Mrθ , and Mθθ contain the
chosen for the reduction of the FE model
inertia terms related to the motion of the frame of ref-
S = X dm (22) erence. They have the same physical meaning as in
V rigid body dynamics, although most of their terms are
now variable.
If the mode shapes X are multiplied by the three
The first block Mrr is a constant 3 × 3 diagonal
components of the undeformed position r̄ui and
matrix, representing the translational inertia of the
integrated, three more 3 × nm constant matrices are
body
obtained
S = r̄ui X dm, i = 1, 2, 3
i
(23) Mrr = I3 dm = mI3 (27)
V V
The remaining nine matrices, of size nm × nm , include The second block Mrθ contains the mass terms that
the integrals of the products between the three direc- couple the translational and rotational inertia of the
tions of the mode shapes reference frame
Sij = X Ti X j dm, i, j = 1, 2, 3 (24) Mrθ = [r̄1 I3 r̄2 I3 r̄3 I3 ] dm (28)
V V
It must be noted that only six of these integrals need to and its calculation requires the integration of the three
be calculated, since Sji is equal to the transpose of Sij . components of the deformed local position r̄. The
JMBD164 © IMechE 2009 Proc. IMechE Vol. 223 Part K: J. Multi-body Dynamics
152 U Lugrís, M A Naya, A Luaces, and J Cuadrado
integral of r̄ is by definition the static moment m̄ in the most involved of the mass matrix. If the rotation matrix
deformed configuration, which, after decomposing r̄ A is taken out of the integral in each block, the remain-
into its constant and variable terms and integrating, is ing integrals are those of r̄i X. It can be observed that
easily found to be these integrals are analogous to the inertia shape inte-
grals Si , but, in this case, the factors that multiply the
m̄ = m̄ u + Sy (29) mode shapes are the components of the deformed
local position; so they might be named Sid
Therefore
Mrθ = [m̄1 I3 m̄2 I3 m̄3 I3 ] (30) Sd = r̄i X dm, i = 1, 2, 3
i
(37)
V
In the centre of the mass matrix the block Mθθ is and after some manipulations, they result as
found. It contains the rotational inertia of the frame of ⎡ T i1 ⎤
reference, and it is itself formed by nine diagonal 3 × 3 y S
blocks Sid = Si + ⎣y T Si2 ⎦ , i = 1, 2, 3 (38)
⎡ 2 ⎤ y T Si3
r̄1 I3 r̄1 r̄2 I3 r̄1 r̄3 I3
Mθ θ = ⎣ r̄22 I3 r̄2 r̄3 I3 ⎦ dm (31) therefore leading to the following expression for Mθ f
V sym. r̄32 I3
⎡ 1⎤
ASd
In this case, the terms to be integrated are the compo- ⎢ ⎥
nents of the deformed planar inertia tensor P̄ Mθf = ⎣AS2d ⎦ (39)
3
ASd
P̄ij = r̄i r̄j dm, i, j = 1, 2, 3 (32)
V Mass terms associated to the elastic coordinates. The
Each r̄i r̄j product can be calculated by first decompos- last nm × nm block of the mass matrix is constant, as
ing r̄i and r̄j into their constant and variable parts, then it happened to the mass inertia associated to the ori-
integrating, leading to gin of the reference frame r 0 . The value of this block,
according to equation (17), is equal to XT AT AX. Since
j
P̄ij = (P̄u )ij + (S ij + S i )y + y T Sij y, i, j = 1, 2, 3 (33) the vectors of the frame of reference u, v, and w have
been defined as unit orthogonal vectors, the rotation
j
where S ij and S i are the jth row of Si , and the ith row of matrix A is orthogonal, and this makes the product AT A
Sj respectively. This result can be used to assemble the identically equal to I3 independently on the orienta-
rotational inertia submatrix tion of the frame, thus making this block constant. This
⎡ ⎤ block, once eliminated the orientation dependency,
P̄11 I3 P̄12 I3 P̄13 I3 is nothing but the well-known modal mass matrix,
Mθ θ = ⎣ P̄22 I3 P̄23 I3 ⎦ (34) widely used in the structural dynamics field. It can be
sym. P̄33 I3 obtained as
Mass terms coupling the reference coordinates to the XT X dm = S11 + S22 + S33 (40)
elastic coordinates. The first four blocks of the last V
These are the only terms that depend both on the ori- It is observed that this product includes the velocity of
entation and on the deformation state and are the deformation ẏ. This means that, if this velocity is zero,
Proc. IMechE Vol. 223 Part K: J. Multi-body Dynamics JMBD164 © IMechE 2009
Efficient calculation of the inertia terms in FFR formulations 153
i.e. the body is behaving like a rigid one, there exist no total mass of the body
velocity-dependent inertia forces, which is consistent
with what happens in the absolute formulation when
M̄rr = I3 dm = mI3 (47)
a rigid body is modelled by using a point and three unit V
vectors.
The different blocks of the velocity-dependent The terms that couple ṡ and ω contain the integral of
forces vector can be obtained after evaluating −r̃, which is the skew-symmetric matrix associated to
the product BT Ḃq̇ and performing the integration. the static moment of the deformed body with respect
The forces associated to the reference coordinates to the global origin of coordinates, m
are straightforward if the inertia shape integrals are
already known. The first block includes the integral of
M̄rθ = −r̃ dm = −m̃ (48)
the mode shapes V
Q vr = −2ȦSẏ (43) This integral can be easily derived from the static
moment of the deformed body in local coordinates m̄,
and the second needs the integrals of the deformed which can be calculated by using equation (29), and
local position components times the mode shapes then expressed in global coordinates by rotation and
translation
⎡ ⎤
ȦS1d
⎢ 2⎥ m = mr 0 + Am̄ (49)
Q vθ = −2 ⎢ ⎥
⎣ȦSd ⎦ ẏ (44)
ȦS3d The integral of the terms related to rotation is the
inertia tensor, expressed in global coordinates, and
The last term Q vf is the integral of −2XT AT ȦXẏ, which calculated with respect to the global origin
can be obtained as
M̄θθ = −r̃ r̃ dm = J (50)
Q vf = −2[u T v̇(S12 − S21 ) + u T ẇ(S13 − S31 ) V
JMBD164 © IMechE 2009 Proc. IMechE Vol. 223 Part K: J. Multi-body Dynamics
154 U Lugrís, M A Naya, A Luaces, and J Cuadrado
and the product is developed, it can be found, after to the absolute position r̃, which is equal to
some manipulation, that the skew-symmetric matrix associated to the static
⎡ ⎤ moment. The second term is directly obtained from
A 3 r2 X − A 2 r3 X
the integral of the modes S. Therefore, the translational
M̄θ f = r̃AX dm = ⎣A 1 r3 X − A 3 r1 X⎦ dm (52)
forces result
V V A 2 r1 X − A 1 r2 X
Q̄ vr = ω̃m̃ω − 2ȦSẏ (60)
Naming Sia the integral of the ith component of the
absolute position times the mode shapes matrix, this
The second section are the rotational inertia forces,
can be rewritten as
⎡ ⎤ where the deformed inertia tensor in absolute coordi-
A 3 S2a − A 2 S3a nates is recognized in the first term
⎢ ⎥
M̄θ f = ⎣A 1 S3a − A 3 S1a ⎦ (53)
A 2 S1a − A 1 S2a Q̄ vθ = −ω̃Jω − 2 r̃ ȦX dm ẏ (61)
V
The Sia
integrals can be obtained by means of the and the second one is calculated as done for the
following expression integral of r̃AX in equation (53), but using Ȧ instead
of A
3
j ⎡ ⎤
Sia = ri X dm = r0i S + Aij Sd dm, i = 1, 2, 3 Ȧ 3 S2a − Ȧ 2 S3a
V j=1 ⎢ ⎥
r̃ ȦX dm = ⎣Ȧ 1 S3a − Ȧ 3 S1a ⎦ (62)
(54) V
Ȧ 2 Sa − Ȧ 1 Sa
1 2
Proc. IMechE Vol. 223 Part K: J. Multi-body Dynamics JMBD164 © IMechE 2009
Efficient calculation of the inertia terms in FFR formulations 155
JMBD164 © IMechE 2009 Proc. IMechE Vol. 223 Part K: J. Multi-body Dynamics
156 U Lugrís, M A Naya, A Luaces, and J Cuadrado
5 x 0.3 2.5
2.0 Absolute
0.5
Relative
5.0 10.0 5.0 5 x 1.0 15.0 2
0.5
0
0 1 2 3 4 5 6 7 8
Time (s)
12 Projection
Chassis height (m)
2 Preprocessing
10
CPU-time (s)
1.5 8
6
1
4
0.5 2
0 5 10 50 100 5 10 50 100
0 1 2 3 4 5 6 7 8 Elements per bar Elements per bar
Time (s)
Fig. 7 CPU-time versus number of FEs
Fig. 5 Time history of the z coordinate of the local origin
of the chassis
relative formulations benefit from the use of prepro-
cessing, especially in the case of large FE models,
the only difference that the preprocessing avoids to where the projection method needs a significantly
repeat operations by calculating them before the exe- larger amount of time.
cution time. This is not true for non-isoparametric
elements, but approximating them as isoparametric
when using the projection method yields practically 8 CONCLUSIONS AND CRITERIA OF USE
the same results. Figures 5 and 6 compare the vertical
trajectories of the local origin of the chassis and the From the obtained results it can be said that, for a given
centre of the front left wheel, obtained with both the set of deformation modes, the use of inertia shape inte-
absolute and the relative formulations. grals preprocessing always improves the performance
As can be seen in Table 1 and Fig. 7, if the projec- with respect to the B∗ matrix projection. Preprocess-
tion method (B) is used, the CPU-time varies when ing is more efficient even for simple FE models such
the resolution of the FE mesh is modified whereas, as those described in the present work, where only
in case preprocessing (P) is used, the CPU-time (not two-node beam elements in relatively small numbers
including preprocessing) remains constant regard- are used; in case of large models including higher-
less of the number of FE. Both the absolute and the order elements, the difference between projection and
Proc. IMechE Vol. 223 Part K: J. Multi-body Dynamics JMBD164 © IMechE 2009
Efficient calculation of the inertia terms in FFR formulations 157
preprocessing would become more significant, since 2 Hurty,W. C. Dynamic analysis of structural systems using
preprocessing keeps the CPU-time dependent only on component modes. AIAA J., 1965, 3(4), 678–685.
the number of deformation modes, whereas in the pro- 3 Craig, R. R. and Bampton, M. C. C. Coupling of sub-
jection method the number of arithmetic operations structures for dynamic analyses. AIAA J., 1968, 6(7),
1313–1319.
grows with the size of the FE model.
4 Agrawal, O. P. and Shabana, A. A. Dynamic analysis of
Apart from the higher difficulty of implementation,
multibody systems using component modes. Comput.
the only drawback of the method could be the prepro- Struct., 1985, 21(6), 1303–1312.
cessing time but, in practice, it is negligible, especially 5 Bathe, K. J. Finite element procedures, 1998 (Prentice
if compared to the calculation of the mode shapes by Hall, Englewood Cliffs, New Jersey).
solving the FE system, since all the integrals can be 6 Géradin, M. and Cardona, A. Flexible multibody dyna-
obtained by direct matrix multiplication after integrat- mics: a finite element approach, 2001 (John Wiley & Sons,
ing the interpolation functions. In the present work, New York).
the preprocessing has been carried out in MATLAB, 7 Wallrapp, O. Standardization of flexible body modeling
and it takes less than 0.02 s for an A–arm with 100 in multibody system codes, part I: definition of standard
elements per bar (i.e. 201 elements). input data. Mech. Struct. Mach., 1994, 22(3), 283–304.
8 Sugiyama, H., Shabana, A. A., Omar, M. A., and Loh, W.
The projection method, on the other hand, is much
Development of nonlinear leaf spring model for multi-
easier to implement, and the only input data it needs
body vehicle systems. Comput. Methods Appl. Mech.
from FE software are the mass and mode shapes matri- Eng., 2006, 195(50–51), 6925–6941.
ces, along with the undeformed local positions of the 9 Cuadrado, J., Gutiérrez, R., Naya, M. A., and Morer, P. A
nodes. This might make it more convenient for certain comparison in terms of accuracy and efficiency between
applications where the implementation time is more a MBS dynamic formulation with stress analysis and a
relevant, as long as the size of the FE models is not too non–linear FEA code. Int. J. Numer. Methods Eng., 2001,
large. It can be also a good choice for simulating geo- 51(9), 1033–1052.
metrically non-linear beams, since it enables to easily 10 García de Jalón, J. and Bayo, E. Kinematic and dynamic
introduce the foreshortening effect [18] in the kine- simulation of multibody systems: the real–time challenge,
matic model, leading to a very accurate and efficient 1994 (Springer–Verlag, Berlin).
11 Lugrís, U., Naya, M. A., González, F., and Cuadrado, J.
method to capture geometric stiffening in FFR formu-
Performance and application criteria of two fast formu-
lations [19]. In this case, the preprocessing method
lations for flexible multibody dynamics. Mech. Based Des.
would become pointless, since the foreshortening ren- Struct. Mach., 2007, 35(4), 381–404.
ders the mode shapes, and consequently the inertia 12 Cuadrado, J., Cardenal, J., and García de Jalón, J. Flexible
shape integrals, variable. mechanisms through natural coordinates and compo-
With respect to the comparison between the for- nent synthesis: an approach fully compatible with the
mulations in absolute and relative coordinates, the rigid case. Int. J. Numer. Methods Eng., 1996, 39(20),
latter seems to have an advantage when using the pre- 3535–3551.
processing method, despite its more involved inertia 13 Frik, S., Leister, G., and Schwartz, W. Simulation of the
terms, achieving a CPU-time around five times faster IAVSD road vehicle benchmark bombardier Iltis with
than the formulation in absolute coordinates. The FASIM, MEDYNA, NEWEUL and SIMPACK. In Multibody
computer codes in vehicle system dynamics, 1993 (Swets
manoeuvre has a duration of 8 s, so that the relative
and Zeitlinger, Amsterdam).
method with preprocessing can simulate the full vehi-
14 Cuadrado, J., Cardenal, J., Morer, P., and Bayo, E. Intelli-
cle ten times faster than real-time. It is also observed gent simulation of multibody dynamics: space–state and
that the improvement with respect to the preprocess- descriptor methods in sequential and parallel computing
ing method is always more noticeable in the case of environments. Multibody Syst. Dyn., 2000, 4(1), 55–73.
the formulation in relative coordinates; for instance, 15 Featherstone, R. Robot dynamics algorithm, 1987
in the case of 100 elements per bar, the simulation (Kluwer Academic Publishers, Norwell, Massachusetts).
in absolute coordinates is three times faster when 16 Bayo, E. and Ledesma, R. Augmented Lagrangian and
using preprocessing, whereas in relative coordinates mass–orthogonal projection methods for constrained
it becomes up to ten times faster. multibody dynamics. Nonlinear Dyn., 1996, 9(1–2),
113–130.
17 Cuadrado, J., Dopico, D., González, M., and Naya, M. A.
ACKNOWLEDGEMENT A combined penalty and recursive real–time formulation
for multibody dynamics. J. Mech. Des., 2004, 126(4), 602–
This research has been sponsored by the Spanish MEC 608.
(Grant no. DPI2006–15613–C03–01). 18 Mayo, J., Domínguez, J., and Shabana, A. A. Geo-
metrically nonlinear formulations of beams in flexible
multibody dynamics. J. Vibr. Acoust., 1995, 117, 501–509.
REFERENCES 19 Lugrís, U., Naya, M. A., Pérez, J. A., and Cuadrado, J.
Implementation and efficiency of two geometric stiff-
1 Shabana, A. A. Dynamics of multibody systems, 1998 ening approaches. Multibody Syst. Dyn., 2008, 20(2),
(Cambridge University Press, Cambridge). 147–161.
JMBD164 © IMechE 2009 Proc. IMechE Vol. 223 Part K: J. Multi-body Dynamics