Professional Documents
Culture Documents
1 s2.0 0045794989902551 Main
1 s2.0 0045794989902551 Main
A. CARDONA~and M. GLXADIN
L.T.A.S., Dynamique des Constructions MCaniques, Universite de Liege, Rue Ernest Solvay, 21,
B-4000 Liege, Belgium
Abstract-The formulation of time integration algorithms for mechanism analysis problems is discussed.
Particular aspects to be considered are the treatment of constraints and of the finite rotation associated
terms. It is shown that in order to time integrate constrained systems, the algorithmic damping at
infinite frequency is of utmost importance. It is demonstrated that the Newmark trapezoidal rule is
unconditionally unstable in the presence of constraints. The use of second order accurate dissipative
algorithms like the Hilber-Hughes-Taylor scheme is recommended. Finally, the integration of finite
rotation associated terms is performed by projecting the equations onto the tangent space to the rotation
group. In this way, well-known results of accuracy and convergence can be applied. Several examples
illustrating the covered topics are presented.
updated Lagrangian method (see Sets 3.1-3.5). apply standard algorithms to constramed dynamic
Some integration algorithms employ values at only systems, trying to rest within well-known results of
one time instant in order to determine the new convergence.
configuration, like the Newmark scheme. Then, their Gear [IO] and Petzold and Lotstedt [I I] presented
application to solve eqns (4) is straightforward, as a numerical theory for DAE systems. They showed
described in Sec. 3.6. Some other algorithms employ that the use of integrators developed for treating
values at various time instants in order to determine ordinary differential equations could lead to accept-
the new configuration (we will refer to them as able solutions when applied to differential/algebraic
multistep algorithms). The Hilber-Hughes-Taylor equations under mild conditions, and advocated the
scheme, for instance, employs forces evaluated at t, + , use of techniques based on backward differentiation
and at t, to compute the configuration at t, + ,. In formulas. The proposition of Gear and Petzold co-
order to make forces comparable, they should lie on incides in some sense with that of dynamicists.
the same tangent space; this is achieved by projection Equations (2) and (3) indicate that the numerical
onto the tangent space at t,+ ,, as pointed out in integration of these systems is strongly influenced by
Sec. 3.7. the properties at infinity of the particuiar integrator
In Sec. 4 we present a series of examples illustrating at hand; as is widely known, backward difference
the different theoretical findings of the paper. In formulas completely filter out the high frequencies, so
particular, the equations of motion for a pendulum that the scheme could lead to physically acceptable
and a spinning top are integrated by using the solutions.
Newmark and Hilber schemes. An example analyzing The widely accepted concern m structural dy-
motion of a squeezer mechanism is also shown. namics IS that algorithms of Integration should
provide at least a small dissipation at high
2. TIME INTEGRATIONOF CONSTRAINED frequencies [12]. We will show that this aspect is
DYNAMICSYSTEMS crucial when solving systems like (I), and that the
Newmark integrator without numerical dissipation
2.1. Constrained dynamic systems and d@erential/
(/I = f. y = i) presents a weak instability which is
algebraic equations
excited for all values of the time step, preventing the
In fact, the equations of motion for a constrained use of this algorithm in constrained systems (the
mechanical system do not form a system of ordinary scheme becomes unconditionally unstable!). The
differential equations (ODE system); they are a mixed introduction of a small numerical damping, via
set of algebraic and differential equations, called the Newmark-p damping scheme, stabilizes the inte-
a system of differential/algebraic equations (DAE gration at the cost of reducing the accuracy to only
system). first order.
One popular technique in mechanism analysis for Several unconditionally stable implicit algorithms
dealing with these equations consists of modifying have been developed in the last few years which
the constraints by introducing a stabilization provide both second order accuracy and numerical
term [5,6]. Then, the constraints are no longer exactly dissipation in the high frequency range. i.e. the
satisfied but they oscillate with given stabilization method of Houbolt. the method of Park. the
periods and damping constants about the exact verifi- a-method of Hilber and the cr,-method of
cation point, tending to it in the long-term. In this Bossak [12-141 (this list does not pretend to be ex-
way, the DAE system is transformed into an ODE haustive, since this held is still the object of active
system and standard ODE integrators can be applied research; see e.g. [15. 161).
without trouble. This technique has the drawback of The Houbolt and Park methods do not permit
depending on arbitrary constants that influence the parametric control of the amount of dissipation
solution, and that the constraints are never exactly present, while the a-methods do. The spectral radii of
satisfied. the Park and Houbolt methods approach zero as
A second technique involves the algebraic elimi- h/T + x. as is typical of backward difference
nation of constraints previous to the integration of schemes. The z-methods are seen to own strong
the equation of motion. This technique was not damping in the high frequency regime, with typical
followed here since it is rather costly and it cannot be values of spectral radius at infinity ranging from
applied to all kinds of constraints [3,7]. 0.55 to I. All of them coincide with the Newmark
Park and coworkers [8,9] employed a staggered trapezoidal rule in the limit, when a = 0.
stabilized procedure to formulate a two-stage stag- The Houboh method affects the lower modes much
gered explicit algorithm. They obtained a specialized too strongly. Park’s method has good lower mode
algorithm for multibody dynamics systems, with damping properties; however. its period error is
some characteristics that made it quite different higher than the a-methods. Also, the relative period
from standard integration formulas for structural errors of the Houbolt and Park methods are greater
dynamics. However, some theoretical questions still than those of a-methods.
remain unanswered such as, for instance, a detailed The spectral properties of the algorithms are
stability analysis of the procedure. Our objective is to derived from the three eigenvalues of the amplifi-
Time integration of equations of motion in mechanism analysis 803
cation matrix. The spurious root plays an important and by considering that E 4 1 the eigenproblem (6) is
role in the behavior of the algorithms. For low transformed into the eigensystem
frequencies, it must be small or zero to guarantee a
good approximation. For high frequencies, the root
causes the desired dissipation. The spurious root of
the trapezoidal rule is zero over the entire frequency
range, while only the a-method of Hilber has ideal The matrix of constraint gradients B is assumed to
properties of the spurious root: zero in the low be well determined, in the sense that all rows are
frequencies and nonzero for the high range. linearly independent; then, the rank of B is m and
The asmethod has a nonzero spurious root in the there exist (n - m) linearly independent eigenvectors
low frequency range that affects its convergence with frequency CD:*= 0 [which correspond to the
properties [ 161. already mentioned (n - m) finite frequencies of (5)].
Another important measure of the performance of We will now compute the 2m eigenvectors of
unconditionally stable algorithms is called overshoot, nonzero frequencies. Since the mass matrix M is
a tendency to violently oscillate for a short time due positive definite, we can express, from (8), that
to certain initial conditions, or discontinuous loading.
Some schemes have been reported as behaving patho-
1&j, = --$M-‘Br(d;),. (9)
logically with respect to overshoot. This is the case in I
collocation methods and the Wilson-8 method [12].
After replacing the latter equation into (8), we get the
The algorithms considered here perform acceptably.
m-dimensional fourth order auxiliary eigensystem:
In particular, the Houbolt and Park algorithms pro-
duce no overshoot. The trapezoidal rule (Newmark
(c0;~1- BM-‘Br){4;}, = 0. (10)
with /I = $, y = i) does not evidence overshoot in the
computed displacements, but the computed acceler-
ations diverge at infinite frequency. The solution of (10) is given by the following 2m
Based on these general considerations we selected, eigenpairs:
from all those algorithms with numerical dissipation
and second order accuracy, the Hilber-Hughes- {(4C c,), (-4, r,), (4, v*), . . . (-tl:, &II>). (11)
(14)
804 A. CARDONA and M. GERADIN
!PrM*P = 1 (18)
where
(15)
11
M 0
M*=
(19)
0 1
2.3. Stability analysis of a constrained system
Let us now analyze the following homogeneous
Then, we are able to expand the solution of (16) in
linear system of motion equations:
the basis ‘P:
[: :I{}+[-;-:J{:}={:}
U6) {;}=!Q (3=Y% (20)
[ 01 00,1
..
1
1 0
I-
[ 0 0,
(22)
1.
a -b
-b 0,
..
Time integration of equations of motion in mechanism analysis 805
with form:
WfSWi vfBM_‘SM-‘Bv,
a, --rlf=
rlf
1
v:Bwi vfBM-‘Bvj
b,awT’ =q’ * yn (301
'li rt: A(Q,)
where
(25)
Z n+i= (z ,rr+~bi~,+,hZi’,,+,z~+,h~2n+,h2~~+~)r
This system leads to the eigenproblem
(32)
and with
The eigenfrequencies of (26) are
4b2A(fl,)+s2(a -&??+S&)
(27) A,, =
n;=(o,h)=-+ -m (34)
= 0. (29)
and the amplification matrices take the form:
The discrete solution of (29) is given through the
amplification matrix of the chosen integrator, in the &,+(A,+A_,)=A,
CAS 33/3-M
806 A. CARDONA
and M. GERADIN
A22-‘f(Ac+A_x,)=Aa. (35)
q,+,=q,,+k,+;((l
-Vk+Wn+,) m order to attain maximum accuracy in compu-
tations [13]. By placing them into eqn (39). we obtain:
1
A,=- (41)
(1 -a)2 1+LY
-4
-4 r’--2a - 1
I+a I
After some algebraic manipulations, the amplifi- A numerical computation shows that i, and I, have
cation matrix A is constructed such that: only one associated eigenvector.
In order to get stability, we require that all powers
of the amplification matrix be bounded. The follow-
ing theorem by Gear [17] indicates the conditions to
be fulfilled for a matrix A of arbitrary dimension:
where
1 + ~(1 + a)25n
c:- PI -
K +/?aR* >
1 +(Y +Ya -BYxQ
+(P-z>o
P(1 + a)XQ
1
A= 1 + (1 + a)Q(2<y + /?a)
-y@
(.(gl+$y&*) I -Y $-a)@
(1 + r)((y - 1mn
If the matrix A has eigenvalues Iz,corresponding to Clearly, the term Dz3 is the one which causes the
sets of elementary divisors of rank m,, then: instability of the Newmark trapezoidal rule, when
I r3=-l.ThetermDr, = nlj3 IS
’ also responstble for
6) if all Irli)<: 1, all elements of A”+0 as n + 00; the slight overshooting behavior of the Hilber
(ii) if any l&l> 1, some elements of A” > O(n;) scheme [ 181. Figure 1 represents the evolution of ID,, I
when n ‘co; with n for different values of the constant a.
(iii) if all [Aild 1, and the largest mi such that By differentiation, we note that the maximum of
11,1=1 is m, then some elements of 1Dz31 occurs after n = (ln(1 + a)/(1 -a))-’ steps.
A”= O(nm-‘) when n + co. For a +O * n + co, the algorithm diverges. This
behavior is evidenced in the computed accelerations
Condition (iii) imposes that the maximum number of any numerical example (see example 4.1).
of unit roots with a common eigenvector should be With further analytical development, the Jordan
one for stability. The Newmark trapezoidal rule, form of the Newmark amplification matrix at infinity
which is recovered by setting a = 0 in (36), violates A, is obtained as the product of the matrices
this condition at n + cc, giving:
D = Y-‘A,Y (46)
a,-0 vT=(l -2 4)
with
a,= a,= -
1
1 v&=(0 0 1). (43) 10 0
For this reason, time integration of eqns (16) with
the Newmark trapezoidal rule gives unconditionally Y= [ -2 41 0 0
-0.25
unstable results for any value of the time step. and
Further insight into the short time behavior of the
Hilber algorithm can be gained by looking at the
powers of A. The Jordan matrix corresponding to (47)
matrix A is
I
0 0 0
The nth power of this matrix has the form
2(- 1) (-1) 0 . (48)
-4(-l)“-8n(-l)“-’ -4n(-l)“-’ 0
(45)
We see that both the initial conditions in displace-
ments and velocities will be amplified without bound
/
/’
/ - alpha - -B.Bl
during the computation of the accelerations; sign of the determinant distinguishes rotations
meanwhile, the computed velocities oscillate between from reflections, which are characterized by negative
bounds. values of the determinant.
Some investigations [19,201 showed that for simple Although R is a 3 x 3 matrix, the orthonormality
nonlinear dynamics examples with a suddenly soften- requirement leaves only three free parameters:
ing stiffness, an increase of energy appears in the
numerical model if the Newmark trapezoidal rule is R = R(Y,, Yy,, Y’;) (50)
applied with certain step sizes. It was also shown that
the energy growth can be suppressed by numerical where ( Yy,, Yy, I,) form the rotational vector, whose
dissipation. In that case, the instability was motivated components express the rotation about the direction
by a completely different mechanism: an accumu- n (Fig. 2):
lation of the error in the strain energy [15]. The
already mentioned weak instability of the Newmark Y =nY. (51)
integrator can also be excited by abrupt variations in
stiffness (see example 4.3) thus introducing new The exponential map gives the rotation operator in
elements that indicate that the Newmark trapezoidal terms of the rotational vector:
rule does not represent a good choice for nonlinear
dynamics. R=I+!$+&Y’+...=exp@) (52)
Finally, we remark that the energy method for the
analysis of stability of an integrator [ 121is not able to
detect this instability, since it looks at the computed where I is the skew-symmetric matrix formed by the
displacements and velocities of the oscillator to assess components of Y:
stability, and in this case only the accelerations
gr, = - E,lk yk
diverge.
Y,= --fE,lkY,k. (53)
3. TIME INTEGRATION AND FINITE ROTATIONS
It can be shown that the series representation of R,
Owing to finite rotations, the dynamic equations eqn (52) can be put into the following closed form
take values on a differentiable manifold and not on expression [4]:
a linear space. as is usual. Then, standard time
integration algorithms cannot be directly used and a
special treatment should be considered. Simo and
R=l+
Vu-Quoc [21] proposed an implicit time stepping
algorithm that provides an extension of the classical
Newmark algorithm to the rotation group SO(3).
Their approach led to a configuration-dependent 3.2. Composite rotations
nonsymmetric tangent mass matrix which gave a Let R be the rotation operator mapping a rec-
quadratic rate of convergence. However, their formu- tangular Cartesian frame (0; E,, E,, E,} into the
lation is restricted, in principle, to the only algorithm orthonormal frame {0 ‘; t, , t,, t, }:
for which they have shown convergence in the paper.
We give here an alternative form of integrating t, = RE,. (55)
rotations, which can be applied to any time inte-
gration scheme. To illustrate the procedure, we Physically, it can be interpreted as a rigid body
consider the Newmark and Hilber-Hughes-Taylor rotation from the initial to the actual configuration,
schemes. We give, in Sets 3.6 and 3.7, the necessary and{O;E,,E,,E,}and{O’;t,,t,,t,}canbeviewed
modifications to these algorithms to integrate the as a material frame and a body-attached or moving
dynamic equations of systems involving large finite frame respectively.
rotations. Let us now consider an incremental rotation which
carries the actual frame {t,) to an updated configur-
3.1. The special orthogonal group-parametrization of ation {tj}. There are two ways of describing this
rotations incremental rotation: via a left translation or via a
Let SO(3) be the group of proper orthogonal linear right translation. The second approach, also called
transformations: material rotation, is followed in this paper. In it, the
total rotation from the initial basis is given by the
sO(3)={R: !l13+9131RrR=1,detR= +l}. (49) right-application of an incremental rotation operator
R,,, to the actual rotation R:
Geometrically, each element R of SO (3) represents a
finite rotation, so that SO(3) equipped with the R’ = RR(,,
matrix product (or composition of rotations) forms
the noncommutative group of rotations. The positive t; = RR,,, E,. (56)
Time integration of equations of motion in mechanism analysis 809
(62)
(63)
3.3. Derivatives of the rotation operator and where the unit vector n gives the direction of
rotation:
Linearized incremental rotations are obtained by
differentiating the rotation operator:
- d
DR·fJ=- I - -
Rexp(efJ)=RfJ. (57)
de ,~o
A derivation of this formula is given in [4].
e
The skew-symmetric tensor represents infinitesimal
or linearized material incremental rotations about 3.4. Angular velocities and accelerations
the eigenvector e associated with the only zero
eigenvalue: The tensor of angular velocities gives the variation
in time of rotations. It is expressed in terms of
material components of rotations as follows:
ee=o. (58)
The angular accelerations are computed by time- Time integration schemes are based on finite differ-
differentiating equation (67): ence and/or interpolation formulas of the primary
variables; then, these should lie on a vector space in
A = T(P)Bb + T(F)-yb. (68) order to allow us to perform the mentioned oper-
ations. We will consider the system of equations (72)
An expression of T(P) can be found in [4]. as being projected onto T,,.SO(3); in this way we
define a system of second order differential equations
3.5. Incremental rotations as unknowns (updated
with values on a vector space, for which standard
Lagrangian point of view) time integrators can be applied.
Dynamic nonlinear structural problems are formu- Next, we will develop the equations corresponding
lated in a sequential form: step-by-step algorithms to the Newmark scheme, in its predictor/multj-
are used to time integrate the nonlinear ordinary corrector form.
differential equations that constitute the nonlinear By noting that T(0) = 1, starting values for the
dynamic problem. Then, we are able to express the integrator are stated as follows:
problem to be solved in an incremental way: we know
a previous solution at a time instant t, and we want
to determine the increment necessary to obtain the
new admissible configuration.
We will treat rotations in an updated incremental
form. At each stage of the solution process we will pp,, n = A,. (73)
determine the incremental rotation necessary to carry
from the previous converged conjuration (or refer- Then, the predictor and corrector equations are
ence configuration) to the current one: written in the usual form:
(74)
exp(p) = exp(p,)exp@,,J (70)
6?Ir1=. G$ , = 0 (76)
A = T(Y’,,)‘P,“, + “Qpi,,)*Pmc* (7la-c) where 6P,,,c lies in the tangent space to %X3) at R,.
In practice, eqn (76) is considered to be satisfied
3.6. One-step algorithms: the Newmark scheme whenever the inequality ]lG$“!,(1-z Eis satisfied, where
ilGf$, jj is any suitable norm of the residual vector
Let G be a system of second order differentia1 Glf’! i and E is a prespecified tolerance.
equations such that: The iteration increment A!#“” is determined by
requiring that the linearized equilibrium equations at
G: SO(3) x 9l+ +9X3. (72) the current state vanish:
where the dynamical pseudo-stiffness matrix is The term c, = T(P,,)-r G, is the projection of the
defined as follows: forces conjugated to increments in TI. _ ,SO(3) to
the space of forces conjugated to increments in
St&=iM+S
Bh* TRBSO (3). Now we are able to compare them directly
to the forces G,, 1, and the modified equation of the
HHT algorithm takes the form:
where M and S are the mass and stiffness matrices of
the problem at hand.
We note that the widely known properties of MA,+L+(l +aN%.tR+l,Qn+~)
accuracy and stability held by the Newmark schemes
are naturally extended to the integration in SO(3). In - a&,t@,,W = (1 + a%tn+l - a&,,. (81)
fact we should speak of a linearized stability prop-
The same technique of projection can be employed
erty, which provides necessary (but not sufficient)
to implement any other multistep algorithm.
conditions for stability in nonlinear systems [22].
By following this simple procedure, any standard
3.7. Multistep algorithms: the Hilber-Hughes-Taylor algorithm of time integration can be applied to
scheme integrate the rotation equations.
In this section, we discuss particular aspects for the
implementation of algorithms which consider terms
4. NUMERICAL EXAMPLES
evaluated at various time instants to determine the
new configuration. The term multistep is employed to 4.1. Simple pendulum
differentiate from algorithms that, in their actual
The first test-case involves the analysis of a simple
computer implementation, need values at only one
rigid pendulum, with a lumped mass at its free end.
time instant to determine the new configuration.
We compute the system response to an initial velocity
We have shown in Sec. 2.4 that the Newmark
distribution, as shown in Fig. 3. Figure 4 displays the
scheme develops a weak instability at /3 = $, y = 4
evolution in time of the vertical and horizontal
for constrained systems. Introducing numerical
position of the tip of the pendulum.
damping re-establishes the stability at the expense of
in Fig. 5, we show the computed vertical and
giving only first order accuracy. The Hilber-Hughes-
horizontal accelerations at the free end, when using
Taylor algorithm was proposed as an alternative
the Newmark scheme. We can appreciate the diver-
scheme, since it retains second order accuracy while
gent oscillations that characterize the weak instability
imposing a small numerical damping that stabilizes
of the integrator.
the integration.
Figure 6 shows the computed horizontal acceler-
The HHT scheme is obtained by retaining the same
ations when using the Hilber-Hughes-Taylor algor-
finite difference equations of the Newmark scheme
ithm. Two runs were made, the first by fixing the
[eqns (74) and (791, but by replacing the internal and
damping constant a to -0.05, and the second with
external forces in the equilibrium equation by a mean
a = -0.1. We see that the oscillations have almost
evaluated at t = t,+., giving in the linear case:
disappeared, the former case presenting more oscil-
lations than the latter one. In both cases, the oscil-
m%+,+(lfa)c4.+,-acq,+(l+a:)kq,+, lations are completely damped after some time steps.
In Fig. 7 we compare the accelerations for a = 0
- akq, = (1 + ax+, - af (79)
(the Newmark trapezoidal rule) and for a = -0.05.
We see that the Newmark results oscillate around the
where m, c and k are the mass, damping and stiffness
true solution, and that this one, in this simple case,
constants of a 1-DOF oscillator, and fis the external
could have been recovered by a filtering/smoothing
force.
post-processing.
We see that the equation of motion now involves
forces evaluated at two different time instants. When
treating rotations, we cannot directly compare forces T------T-----
--”
at t,, with those evaluated at t,+ ,, since they are
conjugated to different angle variations (see Sec. 3.5).
In effect, G, is conjugated to the variations of angular
increments with respect to R, _ 1 while G, + , takes as
a reference the rotation R, evaluated at t,. Then, in A
I
I
l-4
2
1.
order to make the integration by using an HHT-like
m; 1.
algorithm, we have to project the forces G, onto the Vo’ l.so5
space conjugated to Tp. SO (3): h =&OS
0.208
_I
_’
0.tae
T
/
cr
0.188 0.200 0.300 8.40% e.sea 0.680 0.78% ti..see 0.988 1.08
-0. req ,f’
1\ /’
-0.20% I. _ _ ,.
!
-0.30% i
/ HORIZ. DISPL.
‘k
-0.40; Y,
*..
.c-_ __*/’ 0,;’
I
-0.50%
20.0 Ixc.
15.0
f
-5.00
-10.8
-15.8
-20.0
-25.8
-38.0
Fig. 5. Horizontal accelerations of the tip of the pendulum with the Newrnark scheme; B = f, y = f.
Fig. 6. Horizontal accelerations of the tip of the pendulum with the Hilber scheme.
Time integration of equations of motion in mechanism analysis 813
Fig. 7. Horizontal accelerations of the tip of the pendulum. Comparison between the trapezoidal rule
(a = 0) and the Hilber scheme (a = 0.05).
Fig. 8. Spinning top in a gravity field g. Fig. 9. Evolution of the spinning top.
814 A. CARDONA and M. GERADIN
3.00 4
2.50
2.00
1.58
1.00
0.50e
-0.50
-1.W
-1.58
-2.08
-2.58
-3.08
150.
125.
100.
75.0
50.0
25.8
-25.0
-50.0
-75.0 \
-100. - \
‘~ ome z
-125. _ ,: -
.. ,,
‘- . . _A”
-150.
8.08 ‘JIT
?
Fig. 12. Time evolution of the component normal to the body axis of the material angular velocity.
Time integration of equations of motion in mechanismanalysis 815
-58.8
-106.
-158.
Fig. 13. x-acceleration of the disk center. Continuous line: a = -0.05; dash and points: a = 0,
Figure 14 displays the same component of acceler- Hilber scheme, with a = -0.05, gives the better
ations for the cases a = -0.05 and /I = 0.275625, accuracy followed by the Newmark trapexoidal rule,
y = 0.55 (Newmark+ damping scheme). The inte- while the least accurate results are those given by the
grator constants are 6xed so that the amount of Newmark+ damping scheme. We think that the
numerical damping is the same for both analyses. We Newmark trapezoidal rule departs from the correct
see that the accelerations are damped by following trajectory because the oscillations begin to perturb
almost the same pattern, but there exists a progressive the results.
separation between the responses, signaling that the
low frequency response is not the same for both of
them. Figure 15 displays a comparison, for the three 4.3. Squeezer mechanism [24]
integrations, of the trajectory of the disk center The squeezer mechanism shown in Fig. 16 is a
projected onto the x-y plane. The exact solution planar mechanism formed by seven links and a spring
corresponds to the passage of the disk center through (see Table I). Links are articulated between them and
the origin of coordinates. We see that in this case the to the foundation by means of hinge joints.
48.8
38.0
20.8
18.8
-10.B
-20.1
-3e.e
-4a.a
-5e.e
Fig. 14. x-acceleration of the disk center. Contjnuo~ line Q = -0.05; dashed line: /I = 0.275625,~ = 0.55.
816 A. CARLWNA
and M. GE~UD~N
0.800.
_ 0.600
0.500
_ 0.400.
0.300.
0.200.
0. ’ 00 , /’
$9 u
P
-0.8 -Ia
-.__-_ -8149 - _w 0.200 0.400 0.600 0.800
-0.100
Fig. 15. Trajectory of the disk center on the x-y plane. Continuous line: u = -0.05; dashed line:
/? = 0.275625, y = 0.55; dash and points: a = 0.
z x
The prestressed spring H acts on the mechanism computed at node 5 for two analyses: the first by
through link A and provides a reaction to the input using the Hilber scheme (a = -0.1) and the second
solicitation which is materialized by a time varying employing the Newmark trapezoidal rule. We
moment applied at node 2. Since the mechanism only see a slight phase difference between them.
is entirely contained in the x-y plane, the hinge Discrepancies are greater when plotting accelerations
joints are simply modeled by means of Boolean at the same node: Figs 19 and 20 plot the evolution
assignments [3]. Links are modeled by using beam
elements, in which flexibility effects are neglected by
setting a high value to the Young modulus. Table 1. Links forming the squeezer mechanism
Figure 17 displays the system configurations in Link NOdas Link Nodes
time, for a simulation performed by using the A 12-10-13-29 E 27-29
Hilber-Hughes-Taylor scheme with a set to -0.1 B 5-29 F 22-19-17
and with a total of 400 time steps made from t = 0 c: 2-5 G 22-24-27
to t = 0.02. Figure 18 compares the positions in time
6 17-29 H @p&9) 13-14
Tie integration of equations of motion in ma&anism analysis 817
LX
Fig. 17. System evolution. Dashed line: initial caption; ~ntin~~ line: lotion at t = 0.005,
0.0075 and 0.01.
Fig. 18. Positions at node 5. Continuous line: Hikr (a = -0.1); dashed line: Newmark trapezoidal rule.
6.0eE 6
5.seE 6
S.86E 6
4.SBE 6
4.BBE 6
3.505: 6
3.e6E 6
2.S6E 6
t.%6E 6 :: i... L.
:
l.BBE 6 *.
;
“,., ,,i., !
5.8SE 5 i: i:,_,..,_,,,....
___
,._di
-S.68E I ...”
i .,...
,.....:: ..... . ..,._.,.,. ..,,_ .... .
E 70 f .ACC.
3.88 . ^
I...I
-1.88 _ _ : .
-2.88 ,
-3.88
-4.88
A
0.888 u
8.688
a.488
8.288
T
P
2.0eE-2
-0.288
I
-8.488
I
I
-0.600
Fig. 21. x-positions at node 5. Continuous line: Hilber (a = -0.1); dashed line: Newmark- damping
scheme.
13.8 4U
/’
12.0 /
/”
11.0 - - /’
/1.
le.0
9.ee
:
8.00
,I’
7.88 - - ,/.A:
6.ee
5.e0
4.88
3.08
2.60
1 .ee
I ---- P
Fig. 22. Angle of rotational node 2. Continuous line: Hilber (a = - 0.1); dashed line: Newmark- damping
scheme.
Time integration of equations of motion in mechanism analysis 819
6.eeE 6 KC. _ -
5.58E 6
5.00E 6
4.5eE 6 _
i
4.00E 6
3.5eE 6 5
II
3.eeE 6 II
2.5eE 6
2.eeE 6
1.56E 6
i.eeE 6
5.eeE 5
T
-E-2
-5.eeE !
Fig. 23. x-accelerations at node 5. Continuous line: Hilber (a = - 0.1); dashed line: Newmark-/? damping
scheme.
of the x-acceleration for both analyses. We see that terns is to employ a second order accurate dissipative
the Newmark results diverge owing to the weak algorithm like, for instance, the Hilber-Hughes-
instability phenomenon. Taylor scheme.
We remark that in this example, no Lagrange Different aspects concerning the time integration of
multipliers are introduced in the model; the instability equations involving large finite rotations have been
is excited by the large sudden variations of stiffness discussed. We showed that these equations can be
that appear, for instance when links B and C are accurately solved by appropriately projecting the
aligned. concerned quantities onto the tangent space to the
If one tries to solve the same problem by using rotation group SO(3). In this way, all magnitudes lie
the Newmark-fi damping method, the computed in the same vector space and the well known results
response is completely in error as presented in Fig. 21. of convergence of the employed integrator can be
What has happened is that since the method provides applied.
only first order accuracy, the computed response was Some examples illustrating the different topics
such that the spring overpowered the accumulated covered in the paper have been presented, showing a
inertia of the system, and the input axis turned back good agreement with the theoretical predictions.
at t = 0.11, as shown in Fig. 22. However, we see in
Fig. 23 that the oscillations are almost completely
REFERENCES
damped out, as when the Hilber scheme has been
used. 1. A. Cardona and M. Geradin, A beam finite element non
linear theory with finite rotations. Inr. J. Numer. Meth.
5. CONCLUDING REMARKS
Engng 26, 2403-2438 (1988).
2. A. Cardona and M. Geradin, A superelement formu-
The implementation of time integration algorithms
lation for mechanism analysis. Compur. Meth. appl.
for solving multibody dynamics systems was dis- Mech. Engng (submitted).
cussed. Two principal aspects were treated: the time 3. A. Cardona and M. Geradin, On the formulation of
integration of constrained systems, and the time joints for mechanism analysis. LT.4S Report, University
integration of systems in which large finite rotations of Lidge, Belgium (1988).
4. M. Geradin, K. C. Park and A. Cardona, On the
are modeled. representation of finite rotations in spatial kinematics.
A stability analysis of constrained dynamic systems LTAS Report, University of LiZge, Belgium (1988).
was presented. It was found that the Newmark 5. J. Baumgarte, Stabilization of constraints and integrals
trapezoidal rule cannot be used to integrate con- of motion in dynamical systems. Compur. Merh. uppl.
Mech. Engng 1, 1-16 (1972).
strained system equations, since this scheme has a
6. P. E. Nikravesh, Some methods for dynamic analysis of
weak instability which in this case is excited for all constrained mechanical systems: a survey. In Computer
values of time step, resulting in an unconditionally Aided Analysis and Opti.&ation of Me&mica1 S$em
unstable algorithm. It is also concluded that this Dynamics (Edited by E. J. Haua). NATO ASI Series
algorithm is not suitable for nonlinear dynamic prob- Vbl. F9, pi. 351-36b. Springer, -&lin (1984).
7. R. A. Wehage and E. J. Haug, Generalized coordinate
lems since this weak instability can also be excited by partitioning for reduction dimension in analysis of
sudden variations of stiffness. It is shown that one of constrained dynamic systems. J. Mech. Design 104,
the better choices to solve constrained dynamic sys- 247-255 (1982).
820 A. CARDONAand M. GERADIN
8. K. C. Park and J. C. Chiou, Evaluation of constraint B,-method and comparison with other dissipative algor-
stabilization procedures for multibody dynamics ithms in structural dynamics. Compur. Meb. appl.
systems. Proc. 28th Structural Dynamics and Materials Mech. Engng 67, 87-l 10 (1988).
Conference, AIAA Paper No. 87-0927. DV. _. 769-773 17. C. W. Gear, Numerical Initial Value Problems in
(1987). Ordinary Dzfirentiai Equations. Prentice-Hall, Engle-
9. K. C. Park, J. C. Chiou and J. D. Downer, A compu- wood Cliffs, NJ (1971).
tational procedure for large rotational motions in multi- 18. H. M. Hilber and T. J. R. Hughes, Collocation, dissi-
body dynamics. Proc. 2gh Structural Dynamics and pation and ‘overshoot’ for time integration schemes in
Materials Conference, AIAA Paper No. 88-2416, structural dynamics. Earthquake Engng Sfruct. Dyn. 6,
pp. 1593-1601 (1988). 99117 (1978).
10. C. W. Gear, Differential-algebraic equations. In Com- 19. T. J. R. Hughes, Stability, convergence and growth and
puter Aided Analysis and Optimization of Mechanical decay of energy of the average acceleration method
System Dynamics (Edited by E. J. Haug), NATO ASI in nonlinear structural dynamics. Comput. Struct. 6,
Series Vol. F9, pp. 323-334. Springer, Berlin (1984). 313-324 (1976).
11. L. Petzold and P. Lotstedt, Numerical solution of 20. K. C. Park, Evaluating time integration methods for
nonlinear differential/algebraic systems from physics linear dynamic analysis. In Finite Element Analysis of
and engineering. In Inn&ative Methods for Nonlz’near Transient Nonlinear Structural Behavior (Edited by
Problems (Edited bv W. K. Liu. T. Belvtschko and K. C. T. Belvtschko. J. R. Osias and P. V. Marcal). Aodied
Park), pp.’ 155-163. Pineridge’Press, Swansea (1984). Mechanics Symposia Series, pp. 35-58. ASME: ‘New
12. T. J. R. Hughes, Algorithms for hyperbolic and York (1975).
parabolic-hyperbolic problems. In The Finite Element 21. J. C. Simo and L. Vu-Quoc, On the dynamics m space
Method. Linear Static and Dynamic Finite Element of rods undergoing large motions-a geometrically
Analysis, Ch. 9, pp. 490-569. PrenticeHall, Englewood exact approach. Comput. Meth. appl. Mech. Engng 66,
Cliffs, NJ (1987). 1255161 (1988).
13. H. M. Hilbcr, T. J. R. Hughes and R. L. Taylor, 22. T J. R. Hughes, Analysis of transient algorithms with
Improved numerical dissipation for time integration particular reference to stabihty problems. In Compu-
algorithms in structural dynamics. Earthquake Engng tational Methods for Transient Analysis (Edited by
Struct. Dyn. 5, 283-292 (1977). T. Belytchsko and T. J R. Hughes). North-Holland,
14. W. L. Wood, M. Bossak and 0. C. Zienkiewicz, An Amsterdam (1983).
alpha modification of Newmark’s method. In?. J. 23. J. B. Jonker, Dynamics of spatial mechanisms with
Namer. Merh. Engng 15, 1562-1566 (1981). flexible links, Reporf N. 804, Department of Mechanical
15. C. Hoff and P. J. Pahl, Development of an implicit Engineering, Delft University of Technology (1984).
method with numerical dissipation from a generalized 24. D. R. A. Giles, A Comparison of Three Problem-oriented
single-step algorithm for structural dynamics. Compuf. Simulatzon Programs for Dynamic Mechanzcal Systems.
Meih. appl. Mech. Engng 67, 367-385 (1988). University of Waterloo, Ontario (1978).
16. C. Hoff and P. J. Pahl, Practical performance of the