Sie sind auf Seite 1von 8

Globally Independent Coordinates

for Real-Time Vehicle Simulation


Radu Serban*
Department of Mechanical and
Environmental Engineering
University of California-Santa Barbara,
Santa Barbara, CA 93106
e-mail: radu@engineering.ucsb.edu

Edward J. Haug
Department of Mechanical Engineering
The University of Iowa
Iowa City, IA 52242
e-mail: haug@nads-sc.uiowa.edu

Models of the dynamics of multibody systems generally result in a set of differentialalgebraic equations (DAE). State-space methods for solving the DAE of motion are based
on reduction of the DAE to ordinary differential equations (ODE), by means of local
parameterizations of the constraint manifold that must be often modified during a simulation. In this paper it is shown that, for vehicle multibody systems, generalized coordinates that are dual to suspension and/or control forces in the model are independent for
the entire range of motion of the system. Therefore, these additional coordinates, together
with Cartesian coordinates describing the position and orientation of the chassis, form a
set of globally independent coordinates. In addition to the immediate advantage of avoiding the computationally expensive redefinition of local parameterization in a state-space
formulation, the existence of globally independent coordinates leads to efficient algorithms for recovery of dependent generalized coordinates. A topology based approach to
identify efficient computational sequences is presented. Numerical examples with realistic
vehicle handling models demonstrate the improved performance of the proposed approach, relative to the conventional Cartesian coordinate formulation, yielding real-time
for vehicle simulation. S1050-04720000404-9

Introduction

Numerical methods for multibody system simulation and analysis must resolve two conflicting requirements; 1 high fidelity of
the model for high accuracy of results, and 2 high computational
efficiency for low simulation time. In the past, a number of modeling approaches have been used to achieve one or the other of
these goals. Commercial analysis tools such as DADS and ADAMS that emphasize high fidelity are based on Cartesian coordinates, while more recent codes designed for real-time simulation
employ recursive formulations 1,2.
It is desirable that the same model be used for both fast and
high fidelity simulation. The Cartesian coordinate formulation
with Euler parameters for orientation 35 is well suited for
multibody system modeling, from the point of view of systematic
analysis, because of the simple structure of the resulting equations. The main disadvantage of formulations based on Cartesian
coordinates is the dimension of the problem to be solved. If Euler
parameters are used, the position and orientation of each body in
the model are defined by a set of seven parameters. Thus, there are
7n b coordinates for a system with n b bodies. If the mechanism
has n do f degrees of freedom, where n do f 7n b , there are 7n b
n do f 1 constraint equations. A large system of differentialalgebraic equations DAE must thus be solved, which is a computationally demanding task. State-space methods for solving the
equations of motion are based on a parameterization of the constraint manifold that provides a set of independent coordinates. In
these independent coordinates, the equations of motion can theoretically be reduced to a set of ordinary differential equations
ODE. For general mechanisms, such a parameterization of the
constraint manifold is possible only locally. Therefore the underlying ODE can be constructed only locally, and when the manifold parameterization is no longer valid, new independent coordinates must be selected. Independent coordinate selection is based
on matrix decomposition techniques, which are in general computationally expensive.
Coordinates that are independent for the entire range of admis*Address all correspondence to this author.
Contributed by the Design Automation Committee for publication in the JOURNAL OF MECHANICAL DESIGN. Manuscript received December 1998. Associate
Technical Editor: H. Lankarani

Journal of Mechanical Design

sible motion, called globally independent coordinates GIC, are


particularly attractive for multibody system simulation. The existence of such coordinates eliminates the need for local independent coordinate selection to parameterize the constraint manifold.
It must be noted, however, that for a general mechanical system, a
full set of GIC may not exist. Consider for example the case of a
particle constrained to move on a circle. A globally independent
coordinate that can describe the evolution of the one degree of
freedom of this mechanism is the angle between the position
vector and a fixed axis. The equations of motion can thus be
written as ordinary differential equations in . However, the case
of a particle constrained to move on a sphere yields a fundamental
difficulty in defining GIC. Indeed, as a consequence of the fact
that there are no vector fields on the sphere without singularities
6,7, a sphere cannot be globally parameterized by two independent coordinates without singularities. Therefore, solving for the
motion of a particle on a sphere requires a local parameterization
of the surface. This simple example shows that globally valid
independent generalized coordinates do not always exist in multibody dynamics.
For the special case of vehicle multibody systems, information
on functional intent and design criteria enables definition of a
subset of generalized coordinates that remain independent for the
entire admissible range of motion. These coordinates are dual to
compliant elements of the vehicle system, such as suspension
springs, shock absorbers, and control elements such as the steering rack. A set of GIC can therefore be obtained by combining
these additional coordinates with Cartesian coordinates related to
the position and orientation of the chassis. In this setting, improved efficiency in simulation is obtained by 1 avoiding the
computationally expensive local parameterization of the constraint
manifold and 2 from more efficient dependent coordinate recovery algorithms.
Note that the proposed approach retains the simplicity and
structure coming from a Cartesian formulation, while the improved efficiency in simulation is obtained without affecting the
accuracy of the simulation. As numerical experiments show,
simulations of any given accuracy can be performed up to 5 times
faster when compared to traditional generalized coordinate partitioning methods.

Copyright 2000 by ASME

DECEMBER 2000, Vol. 122 575

Equations of Motion Using GIC

The equations of motion of a multibody system using GIC are


formulated in this section. The variational form of the equations
of motion 5 in Cartesian coordinates qR n is

qT MqQA 0

(1)

q q0

(2)

for all q such that

s
(3)

(4)

ss

q u ,v

T T

with v R
independent coordinates and uR
dependent coordinates. Therefore, z vT ,wT T is the set of independent
coordinates.
Assume that the dependent coordinates are uniquely determined
by Eqs. 3 and 4; i.e.,
mp

u
0
u

det w 0

(7)

Then, the variational form of the equations of motion can be rewritten as

qT MqQA 0

(8)

for all q and w such that


q q0
q q w w0

(9)

The Lagrange multiplier theorem implies existence of Lagrange


multipliers and such that

qT MqQA qT qT qT qT wT wT 0 (10)
for all q and w. Using the fact that q0 is admissible and w
T
is arbitrary in Eq. 10, it follows that w
0. Since w is
nonsingular, 0.
Returning to Eq. 10, with q arbitrary, the equations of motion are
Mq qT QA

(11)

qT


q
QA
w

qT


q
QA

w

576 Vol. 122, DECEMBER 2000

(17)

Proposition 2.1. If globally independent coordinates satisfying the conditions of Eqs. (6) and (7) exist, then the equations of
motion can be expressed in ODE form as
zQ

(18)

with
z vT ,wT T

T
T
uv
uu 1
w
1
u M M u v

Mvu Tv u1 T Muu u1 w

(16)

In this form, it can be seen that as long as the conditions in the


Constrained Dynamic Existence Theorem 5 hold, this matrix is
nonsingular and the DAE of Eqs. 11 and 15 is solvable.
The condition of Eq. 6 insures that the nonlinear equations of
Eq. 13 can be solved for the dependent generalized coordinates
u. For special classes of multibody systems, such as vehicle systems, the dependent coordinate recovery procedure can take advantage of the system topology. As presented in Section 4, existence of dual coordinates results in a decoupling of the nonlinear
equations to be solved.
Theoretically, the existence of GIC results in an ODE form of
the equations of motion. However, for practical reasons in modeling and computation, the implicit form given by Eq. 16 is
used. The advantages of being able to solve ordinary differential
equations do not justify the amount of computation involved in
explicitly generating these equations. The theoretical ODE form is
given by the following proposition:

MvvMvu u1 v Tv u1 T MuvMuu u1 v

(14)

qq qq

(15)
qq qq2 qq ww

ww
ww

which is the same as if no additional dual coordinates are used.


Define the composite generalized coordinate vector

with the constraint equations of Eqs. 3 and 4.


Recording the vector of unknown accelerations and Lagrange
multipliers, Eq. 16 can be rewritten as

(6)

and that additional coordinates w are uniquely defined as function


of q by Eq. 4 i.e.,

(5)

n do f p

(13)

The equations of motion of Eqs. 11 and 15 can now be written


as the differential-algebraic equations

and that, at a given configuration, the remaining independent coordinates can be obtained by partitioning the generalized coordinates q as

The acceleration constraint can thus be written in the form

The multibody system thus has n do f nm degrees of freedom.


Suppose there exists a set of GIC wR p , with pn do f , defined
by p equations of the form
w,q 0

(12)

The Jacobian of with respect to s is

q 0

det

q
u

w
z

where z vT ,wT T , and the vector of extended constraints

where m independent kinematic constraint equations are

T
w
u1 T Muu 1
u w

QvMvu u1 Tv u1 T QuMuu u1
T
w
u1 T Qu Muu u1

Transactions of the ASME

In the above expressions, the generalized mass matrix M and the


generalized force vector Q are partitioned as
M

Muu

Muv

Mvu

Mvv

Qu
Qv

Proof: The variational form of the equations of motion of Eqs.


8 and 9 can be rewritten as
uT ; vT

Muu

Muv u
Qu
v 0
vv v

Q
M

Mvu

(19)

for all u, v and w such that


u u v v w w0

(20)

Also, the extended acceleration constraint of Eq. 15 can be written as


uu vv ww

(21)

Using the condition of Eq. 6, u and u can be computed from


Eqs. 20 and 21, respectively, as

u u1 v v w w
u u1 vv ww

(22)

Substituting the above relations into Eq. 19, it follows that


T
vT Tv u1 T ;I wT w
u1 T ;0

Muu
Mvu

Muv u1 vv ww

Qu

0
Qv
v
Mvv
(23)

Using the fact that v and w are arbitrary, the condition of Eq.
23 is equivalent to the following two relations:

Tv u1 T MuuMvu u1 vv ww
Tv u1 T MuvMvv v
Tv u1 T QuQv

(24)

dependent coordinates is used, selection of the subset v of independent coordinates in the GIC formulation is not expensive.
Moreover, these coordinates may remain independent for large
ranges of motion. For example, in the case of vehicle systems, the
vector v contains only generalized coordinates associated with
position and orientation of the chassis.

3 Selection of Independent Coordinates for Vehicle


Systems
Controllability requirements justify the selection of dual coordinates associated with compliant and control elements of the vehicle, such as suspension force elements and the steering rack, as
GIC. Indeed, controllability of the vehicle system requires that all
dependent coordinates are uniquely defined for given values of
suspension and control variables. This is equivalent to the condition that Eq. 13 is solvable for u; i.e., Eq. 6.
In modeling vehicle systems in which Cartesian coordinates
with Euler parameters for orientation are used, the chassis introduces problems in terms of defining GIC. As a free body, the
chassis has six degrees of freedom. If the orientation of the chassis
is unrestricted it follows that no subset of three chassis Euler
parameters can be used as GIC. However, for a typical vehicle
system, only the yaw angle is unrestricted while the pitch and roll
angles evolve in a relatively small range. Unfortunately, as the
following proposition shows, even under such assumptions a singular configuration can be reached.
Proposition 3.1. If the yaw angle of the chassis is unrestricted, there is no set of three globally independent chassis Euler parameters, no matter what the initial orientation.
Proof: In general, a singular configuration for a set of orientation parameters is that in which the given set cannot be uniquely
determined from the orientation matrix. A set of three Euler parameters fails to be independent when the fourth Euler parameter
becomes zero. In the following it is shown that, for any initial
orientation of the chassis, any of the four Euler parameters can be
made zero by a rotation about the global z axis.
Let the initial orientation be described by the set of Euler paT
rameters pi e i0 ,ei T and the corresponding orientation matrix
be

T
T
w
w
u1 T Muu u1 vv ww
u1 T Muvv
T
w
u1 T Qu

(25)

zQ
, with
These expressions can be written in matrix form as M
and Q
defined as above. This concludes the proof.
the matrices M
As noted before, in general, it may not be possible to define a
number of dual coordinates equal to the number of degrees of
freedom. Therefore, some of generalized coordinates q, denoted
by v, must be included in the set of independent coordinates z.
However, unlike state-space methods such as the generalized coordinate partitioning method 8, QR method 9, and null-space
method 10, where numerical partitioning into dependent and in-

Ai j

cos2

1
a i31 2 sin2
2
2 2

a i31 a i32 sin2

a i33 sin cos


2
2
2

a i31 a i33 sin2 a i32 sin cos


2
2
2

Journal of Mechanical Design

a i13

i
Ai a 21

a i22

a i23

a i31

a i32

a i33

e i0j cos
2

(26)

a i31

i
ei j a 32 sin
2
a i33

(27)

The transformation matrix is therefore

1
a i32 2 sin2
2
2 2

a i32 a i33 sin2

a i12

Next, consider a rotation of angle about the global z axis. Since


a unit vector along this axis is a i31 ,a i32 ,a i33 T , Euler parameters
corresponding to this transformation are

a i31 a i32 sin2 a i33 sin cos


2
2
2
cos2

a i11

a i31 sin cos


2
2
2

a i31 a i33 sin2

a i32 sin cos


2
2
2

a i32 a i33 sin2 a i31 sin cos


2
2
2
cos2

1
a i33 2 sin2
2
2 2

(28)

DECEMBER 2000, Vol. 122 577

4 e i3 te i0 2 0

The orientation matrix in the final configuration is

A j Ai Ai j

j
a 11
j
a 21
j
a 31

j
a 12
j
a 22
j
a 32

j
a 13
j
a 23
j
a 33

(29)

and the corresponding Euler parameters are given by


2

e 0j

tr A j 1
4

j
12a 11
tr A j

j
12a 22
tr A j

j
12a 33
tr A j

e 1j
e 2j
e 2j

4
(30)

Using Eqs. 26 and 28, and the fact that Ai is orthonormal; i.e.,
T
Ai Ai I, the diagonal elements of A j are obtained as
j
a 11
a i11cos a i12a i33a i13a i32 sin
j
a 22
a i22cos a i23a i31a i21a i33 sin

(31)

j
a i33
a 33

In terms of the Euler parameters corresponding to the initial configuration, the diagonal elements of A j become

j
a 11
2 e i0 e i1

j
a 22
2 e i0 e i2

1
cos 2 e i0 e i3 e i1 e i2 sin
2
1
cos 2 e i0 e i3 e i1 e i2 sin
2
2

j
a 33
2 e i0 e i3

1
2

(32)

The trace of the matrix A j is then


2

tr A j 2 e i0 e i3 )cos 4e i0 e i3 sin 2 e i0 e i3 1
(33)
while the Euler parameters corresponding to the final configuration are
2

e 0j
2

e 1j
2

e 2j
2

e 3j

1
2
2
2
2
e i0 e i3 cos 2e i0 e i3 sin e i0 e i3
2
1
2
2
2
2
e i1 e i2 cos 2e i1 e i2 sin e i1 e i2
2
1
2
2
2
2
e i2 e i1 cos 2e i2 e i1 sin e i2 e i1
2
1
2
2
2
2
e i3 e i0 cos 2e i3 e i0 sin e i3 e i0
2

(34)

i
It is to be shown that if e m
0, then there exists an angle such
j
that e m 0, for any m0,1,2,3. The case of the first Euler parameter is shown in detail. The other three cases can be treated in a
similar fashion.
The condition e 0j 0 becomes
2

e i0 e i3 cos 2e i0 e i3 sin e i0 e i3 0

(35)

if e i3 0, Eq. 35 has the solution cos 1, or (2k1) ,


for k an integer. If e i3 0, then cos /20 and, using the substitution ttan /2, Eq. 35 becomes
578 Vol. 122, DECEMBER 2000

(36)

which has the solution


, or 2
for
k an integer.
Thus, for any initial orientation of the chassis, there exists an
angle such that a rotation of angle about the global z axis
makes any of the four Euler parameters zero. Therefore, if the
yaw angle of the chassis is unrestricted, no set of three Euler
parameters remains independent for the entire range of orientations. This concludes the proof.
The above proposition implies that, in the case of a vehicle
system modeled with Euler parameters for orientation, a set of
GIC equal in number to the number of degrees of freedom cannot
be defined. Considering the chassis as base body, the subset of
independent coordinates v contains the three chassis position coordinates and three of the four chassis Euler parameters. If the
chassis local reference frame is initially aligned with the global
reference frame, the first Euler parameter e 0 can be used as a
dependent coordinate. The remaining Euler parameters e
e 1 ,e 2 ,e 3 T will remain independent until the yaw angle approaches . At that point, the Euler parameter with largest absolute value can be selected as a dependent coordinate.
te i0 /e i3

4 Topology
Recovery

Analysis

for

arctan (ei0/ei3)2k,

Dependent

Coordinate

In addition to the immediate advantage of avoiding numerical


repartitioning, additional efficiency enhancements are obtained
from the fact that optimal computational sequences for dependent
coordinate recovery can be defined using the GIC formulation.
Once values of independent generalized coordinates and velocities
are obtained by integration, dependent states and velocities are
recovered from the nonlinear constraint equations, usually using a
Newton-Raphson or Newton-like method. The dimension of this
nonlinear system of equations is large, especially when Cartesian
coordinates with Euler parameters for orientation are used. Availability of GIC associated with force and/or control elements in the
multibody system results in a decoupling of the nonlinear constraint equations into smaller sets of independent equations. Since
the number of equations being solved has an important influence
on the computational effort in solving sets of nonlinear equations,
it follows that recovery of dependent coordinates can be performed more efficiently by solving the lower dimensional subsets
of equations. Even greater efficiency is realized when these solutions are obtained in parallel on a multiprocessor computer.
To obtain this decoupling, topology analysis of the multibody
system is used. A graph associated with the mechanical system is
constructed, in which bodies are represented by vertices, while
joints and constraints are represented by edges. From the point of
view of the dependent coordinate recovery algorithm, the vertices
are thus a representation of the unknowns and the edges are a
representation of the equations. In order to identify decoupled
subsets of equations, weights are introduced for both vertices and
edges in the associated graph. The vertices are weighted by the
number of dependent coordinates associated with the corresponding body. Edges are weighted by the number of equations in the
associated constraints. The weight of the graph is defined as the
difference between the sum of the weights of the vertices and the
sum of the weights of the edges. Note that the set of nonlinear
equations has a unique solution only if the weight of the graph is
zero; i.e., the number of unknowns is equal to the number of
equations.
A subgraph is defined as a collection of edges and all vertices
adjacent to the edges in that collection. The weight of a subgraph
is defined in the same way as for the entire graph. When using
Cartesian coordinates, constraint equations are written only in
terms of the coordinates of the adjacent bodies. Therefore, a subgraph with zero weight corresponds to a decoupled subset of the
nonlinear constraint equations. Such a subgraph is called decoupled subgraph.
Transactions of the ASME

Fig. 3 Identification of computational sequences

2 The weight of an edge corresponding to a joint is equal to the


number of constraint equations introduced by that joint.
3 The weight of an edge corresponding to a dual constraint is
equal to the number of defining equations for that dual coordinate.
4 The weight of a loop corresponding to an Euler parameter
normalization constraint is one.

Fig. 1 HMMWV 10-body model

As an example, consider the 10 body model of the HMMWV


High Mobility Multipurpose Wheeled Vehicle of Fig. 1. In the
Cartesian formulation, the HMMWV is described by 70 generalized coordinates, if Euler parameters are used, with 10 Euler parameter normalization constraints and 59 constraints corresponding to the 15 joints. Thus, there are 11 degrees of freedom, six for
the chassis, one for each suspension assembly, and one for the
steering mechanism. For this system, eleven independent coordinates can be identified as follows:
1 Six parameters defining the position and orientation of the
chassis.
2 One dual coordinate for each suspension assembly, the
length of the coaxial coil spring and shock absorber.
3 The steering rack displacement.
Eight of these eleven coordinates are globally independent. Only a
subset of three chassis orientation parameters may not be valid for
some configuration of the vehicle.
The graph representation of this mechanism is given in Fig. 2-a.
Five defining equations for the dual coordinates are appended to
the graph representation as edges, the dotted lines in Fig. 2-b.
Euler parameter normalization constraints must also be represented in the associated graph. Since such a constraint involves
coordinates of only one body, they are represented as loops in the
associated graph, as shown in Fig. 3. Next, weights are introduced
for both vertices and edges in the graph, using the following rules:
1 The weight of each body is equal to the number of dependent
coordinates associated with that body. In this case, the chassis has weight 1, since one Euler parameter is dependent,
while the remaining bodies have weight seven.

The resulting weighted graph is presented in Fig. 3-a.


The first step is to recover the dependent chassis Euler parameter by solving the nonlinear equation whose corresponding decoupled subgraph is highlighted in Fig. 3-a. All coordinates associated with the chassis are now known and the weighted graph of
Fig. 3-b can be constructed. Five decoupled subgraphs can be
identified in this graph. Such a decoupled subgraph is formed by
bodies 1 chassis and 4 front right lower control arm as vertices
and all constraints between these two bodies as edges the subgraph highlighted in Fig. 3-b. Dependent coordinates corresponding to the four lower control arms and to the steering rack are
recovered by solving five sets of seven nonlinear equations in
seven unknowns. Once a decoupled subset of equations is solved,
its corresponding subgraph can be replaced by an edge with zero
weight. The procedure is then applied to the reduced graph shown
in Fig. 3-c. Four more decoupled subgraphs are now identified
such as the one highlighted in Fig. 3-c. They correspond to four
decoupled sets of constraint equations. Thus, the remaining dependent coordinates are obtained by solving four sets of 7 nonlinear equations in 7 unknowns. The same computational sequences
can be used in recovering the dependent velocities, once the dependent coordinates are known.

Numerical Experiments

As an example of dynamic simulation using GIC, the 10 body


HMMWV model, discussed in Section 4, is presented. The vehicle is initially at rest at y2.5m. With an initial speed of 5
m/s, the vehicle is driven for 4 s over the road profile of Fig. 4.
Results of the simulation are presented in Fig. 5.
In order to highlight the improved computational efficiency of
the GIC formulation this method is compared to the usual generalized coordinate partitioning GCP technique 8. The topology
based linear solver presented by Serban et al. 11 was used to
compute generalized accelerations. The following three cases are
compared:
1 Globally independent coordinate method, using the topology
based linear solver GIC

Fig. 2 Graph representation of the HMMWV 10-body model

Journal of Mechanical Design

Fig. 4 Road profile

DECEMBER 2000, Vol. 122 579

Table 1 HMMWV 10-body model. DDABM integrator.

Fig. 5 Vertical position of chassis, rear left wheel, and front


left wheel

Fig. 8 HMMWV 10-body, AB3 integrator, CPU time

Table 2 HMMWV 10-body model. AB3 integrator.

Fig. 6 HMMWV 10-body. DDABM integrator, CPU time

Fig. 7 HMMWV 10-body, DDABM integrator, speed-up

2 Generalized coordinate partitioning method, using the topology based linear solver GCP1
3 Generalized coordinate partitioning method, using sparse linear solver ma47 from the Harwell library for acceleration
computation GCP2
The last case is given here as a reference, since it is the method
currently used in commercial multibody software, such as DADS.
For off-line simulation, the accuracy of the solution is of primary interest. Therefore, variable step size integration methods
are used. Efficiency comparisons are performed using a variable
order, variable step size Adams-Bashforth-Moulton integrator
DDABM from the SLATEC library 12. Computational times
and corresponding speed-ups are shown in Figs. 6 and 7 and pre580 Vol. 122, DECEMBER 2000

Fig. 9 HMMWV 10-body, AB3 integrator, speed-up

sented in Table 1. For all three methods considered, the overall


CPU times are shorter than those for the constant step size integrator, while the speed-up obtained using the GIC formulation is
lower.
The goal of this study is to develop a multibody system formulation that permits at the same time fast and accurate dynamic
simulation. In order to assess the real-time capabilities of the proTransactions of the ASME

Table 3 Timing profile. HMMWV 10-body model.

A Acceleration computation total


B Function evaluations in acceleration computation
C Linear solver in acceleration computation
D Dependent coordinate recovery total
E Function evaluations in dependent coordinate recovery
F Linear solver in dependent coordinate recovery

posed method, the three methods previously presented GIC,


GCP1, and GCP2 are compared using a constant step size third
order Adams-Bashforth AB3 integrator. For integration time
steps ranging from 10 ms to 1 ms, CPU times are plotted in Fig. 8
and presented in Table 2. The corresponding speed-ups obtained
with the GIC formulation, relative to generalized coordinate partitioning, are shown in Fig. 9. It can be seen that the computational effort for one integration step is about 4.5 ms. This suggests
that real-time simulation can be achieved as long as the integration step size is 4.5 ms or greater.
For the experiment considered, the generalized coordinate partitioning algorithms use the same subset of independent coordinates throughout the entire simulation. Therefore, in this particular
case, the improved efficiency of the GIC formulation, as compared to the GCP1 algorithm in the previous examples, is due
entirely to the new dependent coordinate recovery algorithm.
However, as the next example will show, this is not the case for a
more complicated simulation scenario, when the generalized coordinate partitioning algorithm requires frequent redefinitions of
locally independent coordinates. The speed-up obtained for this
portion of the numerical simulation is best seen in Table 3, which
presents the timing profile for one step of numerical integration
for each of the three algorithms considered. For this experiment,
the 10-body HMMWV model is simulated for 4 s, using a third
order Adams-Bashforth integrator with a time step of 5 ms. The
proposed computational sequence for dependent coordinate recovery is faster than the conventional method by a factor of 6.1, with
an overall speed-up of 4.9.
Finally, an example of a complete vehicle model with subsystems is presented, to demonstrate the real-time capabilities of
the proposed formulation. DADS full tire model 13 is linked to
the HMMWV 10-body model previously described. A one dimensional torque-angular velocity powertrain model is used, as well
as a kinematic steering model. Due to the high-frequency behavior
induced by the tire model, a dual rate integration scheme is used,

in which the multibody dynamics is considered the slow part integrated with a time step of 6ms and the fast subsystems tires
and powertrain are integrated with a time step n ratio6 times
smaller. The vehicle is initially at x200 and y50 with a yaw
angle 0.5. With the normalized control inputs acceleration
pedal position and steering input evolving as in Fig. 10a, the
path followed by the vehicle is shown in Fig. 10b. On an SGI
computer with R10000 processors @200 MHz, it takes 26.23s to
complete a 30s simulation. As a comparison, the same simulation
in DADS with integration tolerances set at 105 is completed in
423.3s.

Conclusions

In conclusion, advantages of the proposed approach can be


summarized as follows:
1 No numerical partitioning of the generalized coordinates is
performed. Local selection of independent coordinates is reduced
to a limited number of coordinates the chassis Euler parameters
and is performed analytically.
2 Topology analysis is used to define efficient computational
sequences to recover dependent generalized coordinates.
3 In the proposed formulation, driver inputs are easily defined,
since variables related to control elements are in the set of globally independent coordinates. Such is the case for the steering
rack of a vehicle. In the GIC formulation, the rack displacement is
explicitly defined, whereas in the usual Cartesian formulation, an
additional constraint must be introduced.
Numerical examples involving realistic vehicle models show that
a speed-up of approximately 5.0 can be obtained, when comparing
the proposed approach to the classical GCP method. In all the
examples presented, improved efficiency is obtained for simulations of the same accuracy as the GCP method. Real-time simulation can thus be achieved with models that preserve the simplicity and sparsity structure characteristic of models based on
Cartesian coordinates with Euler parameters for orientation.

Acknowledgment
This research was supported by the US Army Tank-Automotive
Command TACOM, through the Automotive Research Center
Department of Defense contract number DAAE07-94-R094.

References

Fig. 10 Real-time simulation of the HMMWV 10-body with subsystems

Journal of Mechanical Design

1 Bae, D.-S., and Haug, E. J., 1987, A Recursive Formulation for Constrained
Mechanical System Dynamics. Part I: Open-Loop Systems, Mech. Struct.
Mach., 15, pp. 359382.
2 Bae, D.-S., and Haug, E. J., 1987, A Recursive Formulation for Constrained
Mechanical System Dynamics. Part II: Closed-Loop Systems, Mech. Struct.
Mach., 15, pp. 481506.
3 Wittenburg, J., 1977, Dynamics of Systems of Rigid Bodies, Teubner, B. G.,
ed., Stuttgart.

DECEMBER 2000, Vol. 122 581

4 Nikravesh, P. E., and Chung, I. S., 1982, Application of Euler Parameters to


the Dynamic Analysis of Three-Dimensional Constrained Mechanical Systems, ASME J. Mech. Des., 104, pp. 785791.
5 Haug, E. J., 1989, Computer-Aided Kinematics and Dynamics of Mechanical
Systems, Volume I: Basic Methods, Allyn and Bacon, Needham Heights, Massachusetts.
6 Hirsch, M. W., 1976, Differential Topology, Springer-Verlag, New York, NY.
7 Prasolov, V. V., 1995, Intuitive Topology, American Mathematical Society,
Providence, R.I.
8 Wehage, R. A., and Haug, E. J., 1982, Generalized Coordinate Partitioning
for Dimension Reduction in Analysis of Constrained Dynamic Systems,
ASME J. Mech. Des., 104, pp. 247.

582 Vol. 122, DECEMBER 2000

9 Kim, S. S., and Vanderploeg, M. J., 1986, QR Decomposition for State


Space Representation of Constrained Mechanical Dynamic Systems, ASME
J. Mech., Transm., Autom. Des., 108, pp. 183188.
10 Liang, C. G., and Lance, G. M., 1987, A Differentiable Null Space Method
for Constrained Dynamic Analysis, ASME J. Mech., Transm., Autom. Des.,
109, pp. 405411.
11 Serban, R., Negrut, D., Haug, E. J., and Potra, F. A., 1997, A Topology
Based Approach for Exploiting Sparsity in Multibody Dynamics in Cartesian
Formulation, Mech. Struct. Mach., 25, No. 3, pp. 379396.
12 Vandevender, W. H., and Haskell, K. H., 1982, The SLATEC Mathematical
Subroutine Library, SIGNUM Newsletter, 17, No. 3, pp. 1621.
13 CADSI, 1995, Coralville, IA., DADS Reference Manual, Revision 8.0.

Transactions of the ASME

Das könnte Ihnen auch gefallen