Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
The Element-Free Galerkin Method applied on Polymeric Foams
Author: Guilherme da Costa MachadoAdvisor: Marcelo Krajnc Alves, Ph.D.
Co-Advisor: Rodrigo Rossi, Dr.Eng.
Florianópolis - Brazil
September, 2006
Grupo de Mecânica Aplicada e Computacional
Federal University of Santa Catarina
The Element-Free Galerkin Method Moved List Square
Approximation Weight Function definition Imposition of Essential
Boundary Conditions
Celular Solids Introduction Characteristics Mechanical Properties Models examples Methodology
Overview
Finite Strain Elastoplastic formulation Kinematics of deformation The concept of conjugated pairs The Elastoplastic model Strong, Weak and Incremental
formulations Solution Procedure Examples (FEM)
Unilateral Contact and Friction Formulation Problem definition Imposition of the Contact and
Friction terms General Algorithm Examples (EFG)
Conclusions
2
Cellular solids
Introduction
Typically used in energy absorption structures.
Applications areas:• Automotive/Transport industry;• Aerospace industry;• Packing industry;• Construction industry.
Mechanical X low density.
3
Cellular solids
4
Cellular solids
Relative density
Very low = 0,001; Conventional = 0,05 to 0,20; Transition value 0,3 treated as a solid with isolated pores.
Types Polymeric Metallic (aluminum, cooper, nickel, titanium and zinc) Ceramic (carbon) Natural (wood, cork and coral structures)
∗ /m
5
Sólidos celulares
Mec
han
ical
Pro
per
ties
G
IBSO
N &
ASH
BY(
1997
)
6
M
ech
anic
al P
rop
erti
es
GIB
SO
N &
ASH
BY(
1997
)
Cellular solids
Strain
Str
ess
Cellular walls/strutsbuckling plateau
Linear elastic flexion
Densification process
Elastoplastic foam behaviourunder compression
1lim
7
Cellular solids
Modeling Approaches
Periodical Models (GIBSON & ASHBY)• Dependence of the geometrical idealization; • Considers the local cell characteristics;• Limited applications;• Isotropy and cellular interaction, in the majority of the
cases, are not periodic phenomenon.
Random models (ROBERTS,GARBOCZIA e BRYDON)• Voronoi tessellation and Gaussian random field;• Problems to define the RVE and the coordinate number;• Problems at the micro tomography correlation;• Border effects;• Self-contact densification process involves a huge
computational cost;• Explicit solver.
8
Cellular solids
Elastoplastic model for polymeric foams
Methodology
Finite Stains Algorithm; Total Lagrange Description; Rotated Kirchoff stress; Hencky Logarithmic strain measure; Volumetric hardening Law;
Finite Element Method - FEM; Element Free Galerkin Method - EFG;
Contact formulation (Signorini hypothesis); Friction formulation (regularized Coulomb’s law);
9
Required formulations
Finite Strain Elastoplastic Formulation
Stress/Strain elastic relation – an elastic law;
Yield function - indicating the stress level to start the plastic flow;
Stress/Strain plastic relation – material hardening/softening law;
10
Kinematics of deformation Movement, strain gradient and the
multiplicative decomposition
Polar decomposition, Cauchy-Greentensor and the log strain measure
( , )
( , )X
e p
X t x X u
xX t
X
j
j
= = +
¶= =
¶=
F
F F F
( )
F R U F R U
U C
C F F
E Uln
T
p p p e e e
e e
e e e
e e
= =
=
=
=
Finite Strain Elastoplastic Formulation
11
Hill’s Principle: work rate invariability
1 1 1 1 1,
2e
o o o or r r r r= ⋅ = ⋅ = ⋅ = ⋅ = ⋅ D D P F S C Et t s
3
1
3
1
3
1
( )
( )
1ln( )( )
2
Te e e
ei i i
i
ei i i
i
ei i i
i
l l
l l
l l
l
l
l
=
=
=
=
= Ä
= Ä
= Ä
å
å
å
C F F
C
U
E
Spectral decomposition
Finite Strain Elastoplastic Formulation
12
Kirchoff rotated stress
e T e= ( ) ( ) = det( )R R Ft t t swhere
Hyperelastic Hencky’s model
( )
[ ]
( )( )
22
3det
12
e
o
ijkl ik jl il jk
ij klijkl
K
r
r m r r m r
r r
d d d d
d d
*
* * * *
* *
= ( )
æ ö÷ç( ) = ( ) + ( )- ( ) Ä÷ç ÷çè ø
=
= +
Ä =
t
D I
I
E
I I
F
I I ( ) ( ) ME c Egr r* *=
Finite Strain Elastoplastic Formulation
GIBSON & ASHBY(1997)
13
22 2( , ) 0
2 2
o oc t c tp p p p
q p q pa aæ öé ù é ù- +÷ç= + - - £ê ú ê ú÷ç ÷÷ç ê ú ê úè øë û ë û
22 2( )( , )
pq p q pnb= +
( )pp vH e
( ),p pv aa e e
t cp pV=
( ) ( ) lno p p pc c p v vp p H Je e= + =-( ) ( ) ln
pp o p p
y a y a ao
LH
Lt e t e e
öæ ÷ç ÷= + =- ç ÷ç ÷çè ø
14Y
ield
Fu
nct
ion
an
d t
he
Pla
stic
Flo
w P
oten
tial
( )
( )
( )
1 1
1 1
1 1
1 1
2
1 1
1 1
,
3,
10
1 0
0, ,
in terms of
where 0
testen n
testen n
n n
teste testen n
n n
n n
p q
p q
p p
q q
q p
p q
kb l
m l
l
l
l
+ +
+ +
+ +
+ +
+ +
+ +
D
D
ìé ùïïê ú+ -ïê úïë û é ùïï ê úï é ùï ê úï ê ú+ - = ê úí ê úï ê úë ûï ê úïï ê úë ûï Dïïïïî
D
D >
G
G
Elastic Prediction
p =F 0
1
testep pn n+ =F F
( )
[ ]
( )( )
1
1 1
11
1 1 1
1 1
1 1
1 1 1
1 1 1
det
1ln
2
( )
32
teste
teste teste teste
teste teste
teste
teste
teste teste
e pn n n
on
n
Te e en n n
e en n
en n
teste en n n
teste D Dn n n
p K tr
q
rr
r
r
-+ +
**+
+
+ + +
+ +
*+ +
*+ + +
+ + +
=
=
=
=
= ( )
é ù= - ê úë û
= ⋅
t
t t
F F F
F
C F F
E C
E
E
D
Plastic Corrector
( ) ( )1 1teste
n n+ +⋅ = ⋅
yes
no
11
expp pn n
n
l++
æ ö¶ ÷ç ÷= Dç ÷ç ÷ç ¶è øtF FG
End
11 1( , , ) 0p teste
n
teste testen n vq p e
++ + £
Return Mapping Algorithm
15
The global problem value
Strong formulation: reference configuration
• For each , determine that it is solution of
div ( , ) ( ) ( , ) 0
( , ) ( , ) ( , )
( , ) ( )
o o
to
uo
X t X b X t em
X t N X t t X t em
u X t u X em
r- = W
= G
= G
P
P
( , )u X t
t ∈ t o , t f
Finite Strain Elastoplastic Formulation
16
The global problem value
Weak formulation: reference configuration
• Find such that
where( )1; 0nu u ud d+ = " Î
F
( ) ( )
( ){ }( ){ }
1 1 1 1
1
1
;
,
, 0
to o o
n n o o n o n oX
ui p o
ui p o
u u u u d b u d t u dA
u u W u u em
u u W u em
d d r d d
d d d
+ + + +W W G
= ⋅ W - ⋅ W - ⋅
= Î W = G
= Î W = G
ò ò ò
F P
1nu + Î
Finite Strain Elastoplastic Formulation
17
Local Linearization (Newton’s Method)
• Considers as being enough regular
and expanding by Taylor in terms of , we obtain
( ),⋅ ⋅F
( ) ( )11 1 1; ; 0k k k
n n nu u u u u ud d d++ + += +D = " Î
F F
( ) ( ) ( )( ) ( ) ( )
( )1
1 1 1 1 1
1 1 1 1
1 1 11
; ; ;
;o
kn
k k k k kn n n n n
k k k kn n n n oX X
ij ipkn jp ip jk lpijkl
kl klu
u u u u u D u u u
D u u u u u u d
Pu F F F
F F
d d d
d d
tt
+
+ + + + +
+ + + +W
- - -+
é ù+ D + Dë û
é ùD = ⋅ D ⋅ Wë û
¶ ¶é ù = = -ê úë û ¶ ¶
ò
F F F
F
1knu +
Finite Strain Elastoplastic Formulation
18
Global Equilibrium
Find , that satisfies the residual equilibrium equation1hnu +
( ) ( )int1 1 1 0h h ext
n n nr u f u f+ + += - =
( ) ( )( ) ( ) ( )
1
1 1 1
1 1 1 1 1
0k k k
k k k k k
h h hn n n
h h h h hn n n n n
r u r u u
r u u r u Dr u u
+
+ + +
+ + + + +
= +D =
é ù+ D + Dê úë û
( ) ( ) ( ) ( ) ( )11 1 1 t n
o o o
T T Tint h g h ext g gn n o n o o o o of u P u d f b d t dAr
++ + +W W G
= W = W +ò ò ò G
Global Linearization (Newton’s Method)
[ ] ( )K 1 1
1for
k k
k
g hn n
gn
u r u
u
+ +
+
D =-
D
( )1 1 1
kk k gn n nDr u u u+ + +
é ùD = Dë û K
( )o
Tg god
W= WòK G AG
Finite Strain Elastoplastic Formulation
19
While
End
Global Equilibrium AlgorithmStart
( )
( )( )
( ) ( )( )
( )
F I
F F F
C F F
C
U C
R F U
E U
1122
1 1
1
1 1
1 1 1
1
1 1
1
1 1 1
1 1ln
teste
teste teste teste
teste
teste teste
teste teste teste
teste teste
n nX
e pn n n
Te e en n n
en i i i
e en n i i i
e e en n n
e en n
u
l l
l l
l
l
+ +
-+ +
+ + +
+
+ +
-
+ + +
+ +
= +
=
=
= Ä
= = Ä
=
=
å
å
( ) ( )0 00 int1 1 1
0
h h extn n nr r u f u f
error r
+ + += = -
=
max
kr tol
k k
>
<
( )1
khnu +=K K
1
1 1 1
k k kg g gn n nu u u
+
+ + += +D
[ ] ( )1 1
k kg hn nu r u+ +D = - K
01 1
0
n n
k
u u+ +
¬
=
1kerror r +=
1
1
k kr r
k k
+¬
¬ +
Computing at the integration points
yes
no
20
Examples Axisymmetric Uniaxial Compression
0
0
30
0, 082034
0,040470
0,10
928,092
0,049 /
0,00
0,25
1,54
0,30
y
c
m
p
Mpa
p MPa
E MPa
Kg m
c
t
V
r
n
n
g
*
=
=
=
=
=
=
=
=
=
30u mm= -
Finite Strain Elastoplastic Formulation
21TRI6
Uniaxial Compression (Axisymmetric)
Experimental data ZHANG, J. at al (1998)
Finite Strain Elastoplastic Formulation
22
Multiaxial Compression – Axisymmetric Cone Slice
3
0, 082
0, 040
0, 01
928, 092
0, 049 /
0, 00
0,25
1,54
0, 30
oyoc
m
o
p
Mpa
p MPa
E MPa
Kg m
c
t
V
rnng
*
=
===
=====
80u mm=
Finite Strain Elastoplastic Formulation
23
TRI6
Note: Calculated with TRI6 visualized as TRI3 element!
TRI6
TRI3
24
25
The discretization of the differential equations system uses the weak Galerkin form;
Do not required a finite element mesh;
The discretization is based on a set of nodes (discrete data);
The connection in terms of nodes interactions may be change constantly, and modeling fracture (ability to simulate crack growth), free surfaces, large deformations, etc. is considerably simplified;
Accuracy can be controlled more easily, since in areas where more refinement is needed, nodes can be added easily;
Meshfree discretization can provide accurate representation of geometric object;
A special strategy to impose the essential boundary conditions.
Main characteristics BELYTSCHKO et al. (1994)
Element-Free Galerkin Method (EFG)
26
Element-Free Galerkin Method (EFG)
Comparison of the deformations at different time stages for a block of hyperelastic material under compression by using:
EFG FEM
LI & LIU(2002)27
Element-Free Galerkin Method (EFG)
Introduction
Enables the construction of an approximate function
that fits a discrete set of data
Moving Least Square Approximation (MLSA).LANCASTER & SALKAUSKAS (1981);
( )hu X
{ }, I=1,...,nI Iu u=
28
Simple Least Squares Method Finding the best-fitting curve to a given set of discrete points (or nodes)
( ) { } { }( ){ }
[ ]
[ ] ( ) ( ) ( )
A
A
22
1 1
1
1
( ) ,
argmin
.
n nh
I I II I
m
n
I II
n
I I I II
J a u x u p a u
a J a a
a u b
p x p x b p x
* *
= =
* *
=
=
= - = -
= "
=
é ù= Ä =ë û
å å
å
å
R
where and
ALL values uI , I=1,..n exerts influence to determinate ALL the aj coefficients, for the n discrete points.
Element-Free Galerkin Method (EFG)
{ } ( ) ( ) ( )1
, I=1,...,n ,m
hI I j j
j
u u u x p x a p x a=
= = =å
discrete pointsapproximation function
29
Element-Free Galerkin Method (EFG) ( ) ( ) ( ) ( )
( )[ ] ( ) ( ) ( )
( ) ( )
( )[ ] ( )
( ) ( )[ ]
2
1
1
1
1
1
,n
I I II
n
I I II
I I I
n
I II
n
I II
J a w x x p x a x u
x w x x p x p x
b w x x p x
x a x u b
a x x u b
=
=
=
-
=
é ù= - -ê úë û
é ù= - Äë û
= -
=
ì üï ïï ï= í ýï ïï ïî þ
å
å
å
å
A
A
A
( ) ( ) ( ) ( )[ ] ( )A 1
1 1
( ) , ,n n
hI I I I
I I
u x p x a x u p x x b u x-
= =
= = = Få å
intrinsic base functionsglobal shape
functions
moment matrix
( ) 2 2 11, , , , , ,..., ,... ,T k k k
k
p x x y x xy y x xy y
Co
-é ù= ê úë û
whit consistency order
( ) ( ) ( )[ ]A 1,I Ix p x x b-F =
The Moving Least Square Approximation (MLSA)
30
Influence Domain definition
( )Iw x x-
Weight function – Quartic Spline
( )
max
max
2 3 41 6 8 3 para 1;
0 para 1;
max ,
II
I I
I i I Ii
r r r rw r
r
x xrr
r s r
r x x i L
ì - + - £ïïï= íï >ïïî
-= =
⋅
= - Î
rI max
x1
xI
x2
x3 x4
x5Influence
factor
Associated list
Parameterized radius
Element-Free Galerkin Method (EFG)
31
Stability Conditions LIU et al. (1996)
The choice of the size of the support to assure
The consistency order of the intrinsic base
( ){ } ( )card 0 dimi ix x xé ùF ¹ ³ ë û A
( )1
x-é ùë û
A
( ) 21, ,T
p x x y xé ù= W Î " Î Wê úë û
R
Element-Free Galerkin Method (EFG)
rI max
x1
xI
x2
x3 x4
x5
32
Stability Conditions LIU et al. (1996)
( ) 2 2 11, , , , , ,..., ,... ,T k k kp x x y x xy y x xy y-é ù= ê úë û
Element-Free Galerkin Method (EFG)
33
Stability Conditions LIU et al. (1996)
Element-Free Galerkin Method (EFG)
34
Imposition of the Essential Boundary Conditions using The Augmented Lagrange Method
Element-Free Galerkin Method (EFG)
( )I J IJx dF ¹
( )( )( ) ( )
1
1 1
1 1 1
1 1
1
k ikh
n
k i k i k kh g
n n
u h un u n n o
u
h g g h g g gn n u u
Q u u em
u u u u
le
d d l l
+
+ +
+ + +
+ +
é ùê ú= - + - Gê úë û
= = =
N
( )( ) ( ) ( )
1 1 1 1,kk i k i k i
u
uo
u h h u h u g u gn n o n nu u Q u d f u f u
ld d d d+ + + +G
= - ⋅ G = ⋅ + ⋅ò
F
35
New Problem Definition
Find that
1 1 1 1( , ) ( , ) ( , ) ( , ) 0k k u k c kn n n nu u u u u u u u ud d d d d+ + + += + + = " Î F F F F
1knu + Î
( ) ( )
( )
P
1
1
1 1 1 1
1 1 1
1 1 1
;
( , ) ( , , )
, ( , , )
to o o
k
u no
k
c no
n n o o n o n oX
u k u kn n n u u o
c k c kn n n o
u u u u d b u d t u dA
u u Q u u dA
u u Q u u dAn n
d d r d d
d e l d
d e l d
+
+
+ + + +W W G
+ + +G
+ + +G
= ⋅ W - ⋅ W - ⋅
= - ⋅
= - ⋅
ò ò ò
ò
ò
F
F
F
where
Unilateral Contact with Friction
36
( , ) ( , ) ( , )g X u X t Y X t X tn né ù= - + - ⋅ê úë û
[0,1]argmin ( , ) ( )
ll X u X t Y l*
Î= + -
37
Unilateral Contact with Friction
Imposition of the Normal Contact and Friction Terms Normal and Tangential Works
( ) ( )( ) ( )( )1 1 1, , ,c c k i c k in n T nu u u u u und d d+ + += +
F F F
( )( ) ( )( ) ( )
( )( ) ( )( ) ( )
1 1
1 1 1
1 1 1
1 1
, , ,
, , ,
c k
c n no
c c k i
c n n no
c k i k i k i cn n n o
c k i k i cT n T n T T o
u u Q u u d
u u Q u Q e u d
n n n n
n
d e l n d
d e d
+ +
+ + +
+ + +G
+ +G
= - ⋅ G
= - ⋅ G
ò
ò
F
F
38
Unilateral Contact with Friction
Normal Contact
Augmented Lagrange Method
( )( ) ( )( )1 1 11 1
1, ,c k k
n n n
k i k in nQ u g un n n n
n
e l le+ + ++ += +
( )( ) ( ) ( ) ( )
1 11 1,k i k i k ic c
c n no
c h h k i h c gn n ou u Q u d f un n nd n d d
+ ++ +G
= - ⋅ G = ⋅ò
F
39
Unilateral Contact with Friction
Tangential Contact Regularized Coulomb's Law (Penality Method)
Stick condition
Slip condition
( ) ( )( ) ( ) ( )
( ) ( )
( )
( )
1 1 1 1
1
1 1
1
,
1
0
0
k i k i k i k ic c c c
n n n n
k ic
k ic k i n
k in n c
n
T T f
T TT
Q Q Q c Q
QQ u
Q
n n
n
n
ge
g
g
+ + + +
+
+ +
+
¡ = -
æ ö÷ç ÷ç ÷= - +ç ÷ç ÷ç ÷çè ø³
¡ =
0¡ £
0¡ >
40
Unilateral Contact with Friction
QTn1cki cfQ n1
cki QTn1c teste
ki
QTn1c teste
ki
Frictional Algorithm
( ) ( )
1 1
k ic c k i
n n n nT T TQ Q u un + +
( )( )( )
1 1
1k itestec c k i
Tn n n nT T T TQ Q u ue+ +
= - -
( ) ( )
1 1
k ik i testec c
n nT TQ Q+ +
=
SLIP CONDITION
no
yes
STICK CONDITION
( ) ( )
1 1( , ) 0
k iteste k ic c
n nTQ Qn+ +¡ £
Start
End
41
Unilateral Contact with Friction
Examples using EFG Uniaxial Compression (with the same d.o.f of the FEM example)
0,082034
0,040470
0,50
928,09288
0,049
0,00
0,25
1,54
0,30.
oy
oc
m
o
p
Mpa
p MPa
E MPa
c
t
V
r
n
n
g
*
=
=
=
=
=
=
=
=
=
61 10tol -=
ū −30mm
6
1,5
10u
s
e -
=
=
100
7
load steps
pintg
42
Unilateral Contact with Friction
Uniaxial compression and the densification process
ZHANG, J. at al (1998)
Unilateral Contact with Friction
43
Axisymmetric Cone Slice (with the same d.o.f of the FEM example)
61
6
10
1,5
10u
tol
s
e
-
-
=
=
=
1000
7
load steps
pintg
80u mm= -
Unilateral Contact with Friction
44
45
BA
46
EFG
FEM
108,74.10-
( )71,28 15%-
0,00
61,22- 47
Note: Calculated with TRI6 visualized as TRI3 element!
TRI6 TRI3
Indentation Test6
1
62
10
10
tol
tol
-
-
=
=
4
3
6
1,5
0,10
10
10
10
f
T
u
s
c
ne
e
e
-
-
-
=
=
=
=
=
1000
7
steps
pintg
15u mm=-
Unilateral Contact with Friction
48
3%
0.5489
0.5659
49
Conclusions The constitutive model
The features a single-surface yield criteria; a non-associated plastic flow law; the relative density dependence ;
Show a good prediction for the responses of rigid polymeric foams under simple and complex monotonic loading conditions.
The yield surface parameters are simple and can be obtained from two independent experiments Uniaxial compression (ISO 844 / ASTM D1621-04a); Hydrostatic compression;
The relative density dependence has as a consequence a correction in the tangent modulus.
r*( )
( )( )=
rr
**
æ ö æ ö¶¶ ¶÷ç ÷ç+÷ç ÷ç÷ ÷çç ÷ è ø¶ ¶ ¶è ø
ee EE
F F F t
50
EFG x FEM
EFG method showed to be:
Able to withstand the analysis of very large deformation processes, without remeshing and breaking up;
More robust to capture high deformation levels and deformation gradients;
More expensive with respect the computational aspect (more integration points);
In another hand, load step size bigger than the FEM. (small number of interactions!)
The choice of the penalties values (in both cases EBC and Contact) implies in changes at the convergence rate;
Difficulties to find the “right” value for each problem.
Conclusions
51
Numerical Aspects
The Returning Map Algorithm do not capture critical points (limit or path bifurcation).
For this reason, strategies was performed to improve the condition number of the local tangent modulus at the plateaus. (matrix is singular if the condition number is infinite)
Compression causes stress bottom up due to foam consolidation.
Problems to insure convergence (Turning causes low convergence rates);
Further, the hardening laws are interpolated by a polynomial fitting. After 60% of the logarithm strain, this fitting don’t represent the experimental data.
For this reason, the divergence between the numerical and experimental data is plausibly.
At overall, the numerical procedure showed to be adequate to describes the rough non-linearities (material and contact with friction).
Conclusions
52