a modular approach to 3-d impact computation with frictional contact
TRANSCRIPT
A modular approach to 3-D impact computation withfrictional contact
P. Ladeveze *, H. Lemoussu, P.A. Boucard
Laboratoire de Mechanique et Technologie Cachan, ENS de Cachan, CNRS, Univ. Paris VI 61, Avenue du President Wilson, 94235
Cachan Cedex, France
Received 30 November 1998; accepted 17 July 1999
Abstract
An approach suitable for 3-D impact computation with a large number of unilateral friction contact surfaces is
presented. Applications concern connection re®ned models for which the structure components are assumed to be
elastic. It is a modular approach based on a mixed domain decomposition method and the LATIN method. The
resolution process is iterative and operates over the entire time-space domain. Several examples are ®nally analyzed to
show the method's capability of describing shocks. Ó 2000 Civil-Comp Ltd. and Elsevier Science Ltd. All rights re-
served.
Keywords: Impact; Frictional contact; Computation
1. Introduction
The analysis of complex assemblies, subjected to
shock leads to serious diculties especially when the
global behaviour of the structure is very non-linear be-
cause of a large number of unilateral friction contact
surfaces, on which the contact ¯uctuates. Even within a
static framework, industrial computational codes imply
prohibitive computation times in such cases.
The aim of the present study is to propose a new
approach that provides a satisfactory analysis of the
shock response of assemblies, or shock attenuator. Such
an approach would focus, for example, on the major
phenomena occurring during the two-stage separation
of a launcher, when the shock, initiated by the explosion
of a cord cutter, in¯uences the assemblies that serve to
install the on-board devices. Hence, this study has been
part of the pyrotechnic shock investigation group con-
ducted by the National Center for Space Studies
(CNES), based in Evry (France).
The approach involved consists of the continuation
of previous works conducted in statics for which the
computational time has been divided by 50 for several
industrial 3-D connection problems presenting many
contact surfaces, regarding industrial ®nite element
codes [1,3,5].
The ®rst step, already used for static problem [1,5], is
decomposition of a structure into elastic substructures
linked together by interfaces exhibiting a distinct be-
haviour adapted with the phenomena encountered.
Moreover, the dialogue existing between the two entities
is mixed and performed with a velocity±force duality,
which is very well suited to the dynamic framework. The
interfaces contain all non-linear information, such as
unilateral contact with friction [2]. This represents a
meso-modellisation of the structure which, adapted to
parallel computers, provides a high level of modularity
and ¯exibility in the problem's description.
The second step is to use a non-incremental compu-
tational strategy and called the Latin method, where the
non-linearities are concentrated at the interfaces. The
relevant principles are described in Ref. [5]. This method
is an iterative one and operates over the entire space
domain and the entire studied time interval. The added
Computers and Structures 78 (2000) 45±51
www.elsevier.com/locate/compstruc
* Corresponding author.
0045-7949/00/$ - see front matter Ó 2000 Civil-Comp Ltd. and Elsevier Science Ltd. All rights reserved.
PII: S0 04 5 -7 94 9 (00 )0 0 09 4 -8
step herein concern the resolution of the global prob-
lems, whereby the constancy of the matrices present
throughout the iterations is used. Consequently, these
matrices have been pre-calculated during the method's
initialization.
The basic principles of the Latin method will be re-
called ®rst; a full description of the method will be
provided within a dynamic 3-D framework introducing
a speci®c strategy for solving the global problem, on the
time-space domain, that we get at each interation.This
approach has been introduced into the ®nite element
code Castem 2000. A few simple examples will subse-
quently be used to highlight the feasibility of this
method, and its capability of describing shock and
frictional unilateral contact problems with good con-
vergence results. A full comparison with industrial codes
will be developed afterwards.
2. Principles of the approach
Two major components constitute this approach
which is the association of a mixed domain decompo-
sition method and a non-incremental resolution tech-
nique.
2.1. Structural decomposition
The studied assembly X is decomposed into two
mechanical quantities: the substructures XE and the in-
terfaces cEE0 . A substructure can only have dialogue with
adjacent interfaces. This exchange is described in Fig. 1.
This ®gure shows the two quantities _WE
and F E which
are the velocity and surface traction ®elds, respectively.
This velocity±force duality, associated with the power
quantity, is very well suited to the dynamic framework.
2.2. Mechanical problem
A linear behaviour of the material within the small
perturbation hypothesis has been considered herein. The
external loadings are time dependent.
Each entity (substructure and interface) has its own
set of equations. The problem can be written as follows:
For a substructure XE
Find a velocity and a stress ®eld (V EM ; t and rEM ; t),de®ned on XE 0; T , that satisfy
· kinematic admissibility
V E joXE _WE; V E 2 V0;T ; 1
· initial conditions
U Et0 U E
0 ; V Et0 V E
0 ; 2· equilibrium equations
divrEM ; t fd q _V EM ; t; 3
rEoXE M ; tn F E; 4
· constitutive law
rEM ; t KU E; 5
where oXE is the boundary of the substructure XE, fd is
the prescribed body force and U E is the displacement
®eld on XE.
For an interface cEE0
Find _WE; _W
E0; F E; F E0
, de®ned on cEE0 0; T , that
satis®es
· constitutive law
R _WE; _W
E0; F E; F E0
0: 6
The interface equations depend on the interface's be-
haviour. In the case of perfectly connected interfaces,
these relations become
8t 2 0; T ; _WE _W
E0; F E F E0 0: 7
2.3. Solution process
In order to solve the problem associated with the
above decomposition, a non-incremental approach,
called the Latin method was proposed by P. Ladeveze in
1985 (more details can be found in Ref. [5]). This
method has yielded some excellent results for quasi-
static loadings [6]. Previous works have shown compu-
tation times divided by 50 for 3-D connections problems
with many contact surfaces [1]. The present study con-
sists of its development in dynamics.
The Latin method is based on three principles. The
®rst principle is to separate diculties in order to avoid
the simultaneity of global and non-linear problems. By
taking into account the mechanical properties of the
equations, two groups can be distinguished: the local in-
space variable and eventually non-linear equations on
the one hand (the associated space will be called C), and
the linear and global in-space variable equations on the
other hand (the associated space will be called Ad).
The chosen separation of the equation de®nes Ad and
C as
WE
γEE'
ΩE
WE'
FE FE'
Fig. 1. Exchange between substructures and interfaces.
46 P. Ladeveze et al. / Computers and Structures 78 (2000) 45±51
The second principle of the method is a two-stage
iteration scheme which alternatively solves each set of
equations. The local stage solves the problem associated
with C and leads to a non-unique solution; it is therefore
necessary to add other equations, called search direction
equations, to be written E. The linear global stage
solves the problem associated with Ad; it becomes nec-
essary once again to add search direction equations to be
written Eÿ.The third principle of the method lies in the resolu-
tion of the global problem. One characteristic of the
method is that the global operators involved in this stage
are constant with respect to the iterations. Thus, they are
all treated during the method's initialization. More de-
tails about the Latin method can be found in Ref. [5].
Let s be the entire set of unknowns, associated with
the functional space S. s is de®ned on XE [ cEE0 0; T as
s ; r; _W ; F [E sEn E; rE; _W
E; F E
o: 8
sn is the solution to a linear global stage and sn1=2 is
the solution to the local stage at iteration n. This itera-
tive scheme is described in Fig. 2.
2.4. Global linear stage at iteration n
This stage begins with a known element sn1=2
_WE
n1=2;_W
E0
n1=2; FE
n1=2; FE0
n1=2
that belongs to C and
is de®ned on cEE0 0; T .One must then ®nd an element sn1 (V E
n1; rEn1, de-
®ned on XE 0; T , that satis®es
· motion equation
divrEn1 fd q _V E
n1; 9
rEn1oXE n F E; 10
· constitutive law
rEn1 KU E
n1; 11· search direction (SD)
F En1 ÿ F
E
n1=2 kÿ _WE
n1
ÿ _W
E
n1=2
; 12
F E0n1 ÿ F
E0
n1=2 kÿ _WE0
n1
ÿ _W
E0
n1=2
; 13
where kÿ is a positive constant. The used value is
GTc=Lc, where G is the shear modulus, Lc a characteristic
length and Tc a characteristic time.
The linear global stage leads to a global problem on
each substructure, which serves to considerably reduce
the size of the problem. Moreover, these small problems
are independent and as such can be parallelized. All of
these problems are non-classical in nature. Over the
entire boundary oXE, both density of surface traction
and velocity are applied. The problem associated with
the substructure XE can then be written asZXE
KU En1V dXE
ZoXE
kÿV En1V dSE
Z
XE
q _VE
n1V dXE
Z
oXE
FE
n1=2
kÿ _W
E
n1=2
V dSE 8V 2V
0;T 0 :
14This is a classical formulation of a problem in which
the density of the surface traction is applied: ~F F
E
n1=2 kÿ _WE
n1=2. It has a unique solution which can be
obtained using various techniques [4,8±11].
The space domain is discretized using ®nite elements.
The resolution of the dierential system is performed
with an implicit Newmark scheme [7] over the complete
time interval 0; T .
2.5. Local stage at iteration n
This stage begins with a known element sn (V En ; r
En
that belongs to Ad and is de®ned on XE 0; T .
Γ
Ads ex
s
s
s0s1 s2
( )-
directionsearch of global stage
of local stage
E
( + )E
1 2
3 2
directionsearch
Fig. 2. Presentation of the iterative process.
Ad
kinematic admissibility;initial conditions;equilibrium equations;
8<: C constitutive law on cEE0 :n
P. Ladeveze et al. / Computers and Structures 78 (2000) 45±51 47
One then must ®nd an element sn1=2
_WE
n1=2;_W
E0
n1=2; FE
n1=2; FE0
n1=2
, de®ned on cEE0
0; T , that satis®es
· interface behaviour
R _WE; _W
E0; F E; F E0
0; 15
· search direction (SD)
FE
n1=2 ÿ F En k _W
E
n1=2
ÿ _W
E
n
; 16
FE0
n1=2 ÿ F E0n k _W
E0
n1=2
ÿ _W
E0
n
; 17
where k is a positive constant, equal to kÿ.
2.5.1. Interface with unilateral friction contact
For a unilateral friction contact between XE and XE0 ,
the largest contact zone is supposed to be known.
The velocity ®eld _WE
n1=2 and the surface traction
density FE
n1=2 are written as follows:
_WE
n1=2 NEE0 : _WE
n1=2
N EE0 p _W
E
n1=2; 18
FE
n1=2 N EE0 :FE
n1=2
N EE0 pF
E
n1=2; 19
where N EE0 is the exterior unit vector, oriented from E to
E0, and p denotes the tangential coecient operator
(Fig. 3). Contact conditions are
if WE
n1=2
ÿ W
E0
n1=2
NEE0 > 0 ) separation
) FE
n1=2 FE0
n1=2 0;
Two solutions are possible in the contact case:
if pFE
n1=2
< l FE
n1=2 NEE0
) sticking ) p _W
E
n1=2 p _WE0
n1=2;
if pFE
n1=2
l FE
n1=2 NEE0
) sliding ) 9 k P 0;
such as p _WE
n1=2
ÿ _W
E0
n1=2
ÿkF
E
n1=2:
The resolution of the local stage is explicit.
In the normal direction, a scalar indicator Cn gives
the solution:
Cn 1
2_W
E0
n1=2
ÿ _W
E
n1=2
ÿ 1
2kF
E0
n1=2
ÿ F
E
n1=2
N EE0 ; 20
if Cn > 0 ) separation, and if Cn 6 0 ) contact.
In the tangential direction, a vectorial indicator Gn
gives the solution:
Gn 12kp _W
E0
n1=2
ÿ _W
E
n1=2
ÿ 1
2p F
E0
n1=2
ÿ F
E
n1=2
:
21
The sliding conditions can then be written, using the
sliding limit g l FE
n1=2 NEE0
, as follows:
sticking if kGnk 6 g
)p F
E
n1=2
ÿp F
E0
n1=2
Gn
p _WE
n1=2
and p _W
E0
n1=2
SD;
22
M
ΩE
ΩE'
γEE'
NEE'
Fig. 3. Contact.
if WE
n1=2
ÿ W
E0
n1=2
NEE0 0 ) contact )
_WE
n1=2 NEE0 _WE0
n1=2 NEE0 ;
FE
n1=2 FE0
n1=2 0;
FE
n1=2 N EE0 6 0:
48 P. Ladeveze et al. / Computers and Structures 78 (2000) 45±51
sliding if kGnk > g
)p F
E
n1=2
ÿp F
E0
n1=2
g Gn
kGnk
p _WE
n1=2
and p _W
E0
n1=2
SD:
23
In this manner, contact with friction problems are no
more dicult to solve than are perfect connection
problems. Moreover, the convergence characteristics of
the method are not aected by the presence of contact
interfaces.
2.6. Control of the iterative process
An error estimator can be based upon the degree of
non-satisfaction of the behaviour of the interfaces for an
approximate solution sn, an element of Ad. In order to
stop the algorithm, we use an energy indicator, which is
computed after each local stage as follows:
g Rksn ÿ sn1=2k2E
Rksnk2E ksn1=2k2
E
;
where
ksk2E
ZoXE
F Tkÿ1F dSE Z
oXE
_WTk _W dSE: 24
3. Examples
3.1. Beam in traction
This example demonstrates the capacity of the
method in treating shock-related problems. An example
problem is presented in Fig. 4. Two beams, composed of
the same material, are separated by a gap j and then
shocked by a compression load.
The meso-model employed is composed of three
substructures each of them containing ®fty elements
(eight-node cubes). The time step for the studied interval
consists of 250 discretized time values to represent the
loading history.
The solution is given in Fig. 5 with a plot of the
uniaxial stress along the beams. Once the wave has
moved suciently, with respect to the ®rst beam, the gap
vanishes. This phenomenon creates one wave in each
beam; both of these waves have the same speed and the
same level.
In Fig. 6, the state of the uniaxial stress is repeated
and associated with the time history of this stress for a
single point. This procedure reveals, by means of the
iterations, how the solution is obtained. At the ®rst it-
eration, only one sub-structure can perceive the loading.
Therefore, the entire structure requires several iterations
in order to be in¯uenced by the loading. In any event, at
each iteration, the state of all of the unknowns is ob-
tained throughout the entire structure and over the en-
tire studied time interval.
Fig. 4. Studied problem.
Fig. 5. Problem solution.
Fig. 6. Method behaviour.
P. Ladeveze et al. / Computers and Structures 78 (2000) 45±51 49
At this stage, the equilibrium equations are solved
globally. Thus, the contact conditions can be trans-
gressed and get veri®ed only at convergence. It is im-
portant to note that corrections are done not only in
space but in time as well. Both the level and the moment
of the shock are therefore modi®ed throughout the it-
erations. At convergence (six iterations), the theoretical
solution is correctly obtained.
3.2. Two-specimen crushing
This example demonstrates the capacity of the
method in treating friction problems. Two specimens,
whose mechanical characteristics vary greatly, are in
contact on a wall. The only non-zero sliding coecient is
between the two sub-structures. By virtue of symmetry,
only a quarter of the structure needs to be studied. This
problem and its corresponding meso-model are pre-
sented in Fig. 7.
Since this problem has no analytical solution, the
reference then becomes our solution based on a large
number of iterations (30 iterations). This solution has
been presented in Fig. 8. The central displacement of the
interface is plotted for both specimen. The solid lines
represent the hard specimen and the dashed lines rep-
resent the soft specimen. The evolution of these curves
throughout the method's iterations is also plotted. Once
again, the method's classical behaviour appears. The
plotted quantities are relevant to the substructures; thus,
they satisfy the contact condition only at convergence.
The solution in terms of tangential displacement is
plotted in Fig. 9. The solid lines represent the hard
specimen and the dashed lines represent the soft speci-
men. The studied point is located at the center of the two
specimen's interface. Four dierent friction coecients lhas been studied.
Fig. 7. Studied problem.
Fig. 8. Evolution in contact during the iterations.
Fig. 9. Evolution in sliding.
Fig. 10. Convergence indicator.
50 P. Ladeveze et al. / Computers and Structures 78 (2000) 45±51
All these cases yield dierent behaviour for the
structure even when the moment of the separation is the
same. No sticking occurs when the sliding coecient is
zero, and no sliding occurs when l is suciently large.
For the other two cases, three phases are encountered:
sticking, sliding and separation.
In Fig. 10, the method's convergence behaviour is
shown. It can be easily observed both that the value of
the sliding coecient does not aect this convergence
behaviour, and that sliding problems can be treated with
the same facility as frictionless problems.
4. Conclusion
A new approach for solving three-dimensional dy-
namic assemblage problems has been presented here. It
is based on two components: a mixed decomposition
of the structure which provides signi®cant modularity
to the problem description, and an iterative solution
scheme that is well-adapted to the problem.
The initial numerical results in the case of elastic
problems have shown the capacity and behaviour of this
approach, in addition to demonstrating its capability in
solving friction and frictionless problems with the same
facility. Moreover, the special treatment of contact con-
ditions and the resolution technique used here involving
constant operators through out the iterations make this
approach suitable for 3-D computations. Other com-
parisons with industrial codes will be conducted subse-
quently using problems with large numbers of d.o.f.
References
[1] Blanze C, Champaney L, Cognard JY, Ladeveze P. A
modular approach to structure assembly computations,
application to contact problems. Int J Comput Aided
Engng Software 1996;13:15±32.
[2] Chabrand P, Dubois F, Raous M. Various numerical
methods for solving unilateral contact problems with
friction. Math Comput Modelling 1998;28:97±108.
[3] Champaney L, Cognard JY, Dureisseix D, Ladeveze P.
Large-scale application on parallel computers of a mixed
decomposition method. Computat Mech 1997;19:
253±63.
[4] Hulbert GM. Time ®nite element methods for structural
dynamics. Int J Numer Meth Engng 1992;33:307±11.
[5] Ladeveze P. Non-linear computational structural mechan-
ics. New york: Springer, 1999 [french version 1996].
[6] Ladeveze P, Cognard JY, Talbot P. A non-incremental
adaptative computational approach in thermo-viscoplas-
ticity. IUTAM, 1999. p. 281±91.
[7] Newmark NM. A method of computation for structural
dynamics. J Engng Mech Div 1959;EM3:67±94.
[8] Park Z, Housner J. Semi-implicite transient analysis
procedures for structural dynamics analysis. Int J Numer
Meth Engng 1982;18:609±22.
[9] Sotelino E. A concurent explicit-implicite algorithm in
structural dynamics. Comput Struct 1992;51:181±90.
[10] Wood WL. A further look at Newmark Houbolt etc time
stepping formulae. Int J Numer Meth Engng 1984;20:
1009±17.
[11] Zienkiewick O, Wood WL, Hine N. A uni®ed set of single
step algorithms part1: general formulation and applica-
tions. Int J Numer Meth Engng 1984;20:1529±52.
P. Ladeveze et al. / Computers and Structures 78 (2000) 45±51 51