Computational modeling

of material failure
Nguyễn Vĩnh Phú*

in collaboration with

Stéphane Bordas, Oriol Lloberas-Valls,

Amin Karamnejad, Erik Lingen, Martijn Stroeven, Bert Sluys

*School of Civil, Environmental & Mining Engineering

University of Adelaide
 Outline
• Computational models for fracture

- Continuum mechanics: LEFM, Cohesive zone models

- Peridynamics

- Continuous/discontinuous description of failure

(Damage models, XFEM, interface elements)
• Multiscale modeling of fracture

- Hierarchical, semi-concurrent and concurrent methods

- Computational homogenization models for fracture
• Image-based modeling

- Conforming mesh methods

- Level Set/XFEM, Finite Cell Method (non-conforming)

- Voxel based methods

2
 Continuum mechanics theories
Cauchy, Euler, Lagrange… S. Silling 2000
Peridynamics
ij,j + ⇢bi = ⇢üi Peridynamics is a formulation of continuum mechanics
that is oriented toward deformations with discontinuities,
= Cijkl ✏kl especially fractures.
ij
1 Integral equation
✏ij = (ui,j + uj,i )
2
+
Hx
Fracture Mechanics 0
Damage Mechanics x

PDE
spatial derivatives of displacements:
do not exist at discontinuities (cracks) 3
No spatial derivatives of displacements
Continuous/discontinuous
description of fracture
weak discontinuity strong discontinuity

Damage Mechanics Fracture Mechanics

Isotropic damage models LEFM, EPFM, CZM
Softening plasticity models
Damage-plastic models 4
 Fracture mechanics models
crack tip Elastic Plastic Fracture Mechanics
(EPFM):
- ductile materials
- an existing crack is required

R⌧D
Cohesive Zone Models (CZMs):
Linear Elastic Fracture - quasi-brittle materials (concrete)
Mechanics (LEFM): - ductile materials
- brittle materials - no existing crack is needed
- ductile materials under
Small Scale Yielding (SSY) condition
- an existing crack is required
5
 Linear Elastic Fracture
Mechanics (LEFM)
SIF Remeshing is a key point.
K
ij =p fij (✓) + H.O.T
2⇡r
crack tip
crack edge

double nodes

1
p
r
Very useful for fatigue life
crack must locate on element edges estimation da = C( K)m
Barsoum element [1970s] dN 6
 Cohesive Zone Models
Barrenblatt 1962
Dugdale 1960
(CZMs) f tensile strength t
t
Hilleborg, 1976 GIc fracture energy
ft

GIc

0
[Extrinsic] Cohesive law
[Initially rigid] TSL
Constitutive equations (Traction Separation Law)
deformation
crack initiation crack direction
separation max
1 ft criterion
7
 Cohesive crack model
Governing equations
(strong form)

Constitutive equations

deformation
separation

8
 Cohesive crack model
Weak form

new term
where

different techniques
9
Crack discretization techniques

Zero-thickness interface elements, 1968 PUM FEM, 1999

Embedded strong discontinuity, 1987

Meshless/Meshfree methods,
1994
 Interface elements
composite delamination
intrinsic
cohesive law

+
u ,u [[u]] ⌧

(+) easy to implement 2D/3D

(+) available in ABAQUS, LD-DYNA

- preprocessing: GMSH
- solver: jem/jive (C++) 11
 Interface elements
inter-granular fracture of polycrystalline material

failure of a fiber reinforced composite

12
Interface elements with discontinuous Galerkin

13
 Partition of Unity Methods
Melenk and
Approximation of the displacement field
Babuska 1996 X
Sum of shape functions is equal to one uh (x) = NI (x)uI
I2S
X X
NJ (x) = 1 (PUM) + NJ (x) (x)aJ
J J2S c
X
NJ (x) (x) = (x)
J

r ✓ J

r
K r
ui = fij (✓)
2µ 2⇡
14
 Extended FEM (XFEM)
Belytschko et al., 1999 nothing but an instance of
PUM for crack problems
X
uh (x) = NI (x)uI
I2S
X
+ NJ (x)H(x)aJ
for LEFM S c J2S c
4
!
X X
t + NK (x) B↵ b↵
K
S
Enrichment functions K2S t ↵=1
⇢
+1 if (x x⇤ ) · n 0
H(x) =
1 otherwise

p ✓ p ✓ p ✓ p ✓
[B↵ ] = r sin , r cos , r sin sin ✓, r cos sin ✓
2 2 15 2 2
Sub-triangulation for numerical integration

Interfacial two cracks

LEFM

Homogeneous LEFM

16 Matlab code
 XFEM for cohesive cracks
X X
h
u (x) = NI (x)uI + NJ (x)H(x)aJ
I2S J2S c

Wells, Sluys, 2001

No good crack tip solution is known, no
tip enrichment!!!

c
S not enriched to ensure zero
˙
ṫ = T[[u]] crack tip opening!!!
˙ = D✏˙ ⇢
+1 if (x x⇤ ) · n 0
H(x) =
1 otherwise

numerical integration 17
XFEM/Cohesive zones

Size effect

18
… convincing examples

Northwestern Univ.

F.P. van der Mer, TU Delft

M. Duflot

19
 XFEM for material interfaces
Sukumar et al. 2002

rc
xc

across interface, strain field is discontinuous level set representation

abs-enrichment function signed distance function
= ||x xc || rc
(x) = NI (x) I
X X
x uh (x) = NI (x)uI + NJ (x) (x)aJ
I2S J2S c
c <0
,x 20 S min max
trabecular bone, PhD thesis, Tran, NUS
holes=extremely soft inclusions

hole

21
Continuum damage mechanics
Kachanov, 1958, Rabotnov 1969, Hult 1979
CDM is a constitutive theory that describes the progressive loss of
material integrity due to the initiation, coalescence and propagation
of microcracks, microvoids etc. These changes in the microstructure lead
to the degradation of the material stiffness at the macroscale.
nominal stress A
A = ¯ Ā
¯ effective stress Ā 0!1
Ā Ā
=
= (1 ¯ =!)E"
(1 !)¯ , !=1
A A
damage variable
damage variable
Hook’s law: ¯ = E"
= (1 !)E"
[M. Jirasek] 22
 Local damage model
Isotropic damage model C : elasticity tensor
✏eq : equivalent strain [-]
= (1 !)C✏ v
u 3
uX
! = f (✏eq ) Tensile failure ✏eq = t h✏i i2

✏eq = g(✏) [Mazars] i=1

linear softening
Damage evolution law
8
>
<0 if  < i
i c 
E
!= 1  c i if i    c
>
: (1 !)E
1 if  > c ✏
Irreversibility of failure i c
 = max ✏eq stress update: explicit and simple
 Local damage model
In the early 1980s it was found that FE solutions of softening
damage do not converge upon mesh refinement, Z. Bazant, 1984.

Isotropic damage model

= (1 !)C✏
! = f (✏eq )
No energy dissipation !!! ✏eq = g(✏)

Softening plastic models: also suffer from mesh sensitivity.

 Nonlocal damage model
= (1 !)C✏ Cabot and Bazant, 1987

! = f (¯✏eq ) 0!1
Z
✏¯eq (x) = ↵(x ⇠)✏eq (⇠)d⇠ nonlocal l
⌦✓ ◆ eqv. strain
r 2 ?
↵(r) = exp l is the length scale
2l2

SEN beam
25
 Gradient damage model
= (1 !)C✏ = (1 !)C✏
! = f (¯✏eq ) 0!1 ! = f (¯
✏eq )
Z
2
✏¯eq (x) = ↵(x ⇠)✏eq (⇠)d⇠ ✏¯eq cr ✏¯eq = ✏eq
⌦✓ ◆
r 2 l2
↵(r) = exp c=
2l2 secant matrix 2

˙ = (1 !)C✏˙ C✏!˙ Microplane Implicit GD model

Damage Models
Peerlings et al., 1996
(Z. Bazant)
nodal unknowns: displacements
and nonlocal equivalent strain
Continuous vs. discontinuous description

+ easy to implement (2D/3D)

- hard to implement (3D)
+ one single constitutive law
- two separate constitutive laws
+ standard elements
- enriched elements
- incorrect final stage of failure
+ correct final stage of failure
- evolving length scale

bests of both worlds: combined

continuous-discontinuous approaches
(Dr. Nguyen Dinh Giang, Univ. Sydney) 27
 Solution strategies
For a quasi-static analysis of softening solids,
one encounters cases...
F ?: divergence occurs
snap-through

F load control

u
snap-back F

disp control ?: divergence

u
Incremental-iterative procedure u
28
 Path-following methods
Riks 1972 ext
load factor
f = g reference load vector

Newton-Raphson (u, ) arc-length/constraint function

where

1 1
uI = K r, uII = K g
correction
29
 Energy control
Z
✏ = Ba f int
= B T

Gutierrez 2004 ⌦ equilibrium

Z Z
1 T 1 T T 1 T int 1 T
V = ✏ = a B = a f = a g
2 ⌦ 2 ⌦ 2 2

Energy release rate V̇ T

ȧ g

G>0
predefined amount of energy
Arc-length function to be released [Nm]

30
forward Euler
Energy based arc-length control
50

40
reaction [N]

30

20

10

0
0 5 10 15 20 25 30 35
2u [mm]

31
 … and here we are

complex failure mechanism in composites o

[±45 ]

- matrix cracking
- delamination of plies

F.P. van der Mer

EFM, 2008 32
 Peridynamics
S. Silling 2000

time integration:Verlet integration

continuum version of MD (molecular dynamics)
XFEM can do this as well but would
require genius programmers (2D)
glass See phase field models for similar
capacities
crack branching
Bobaru, 2010

33
 Multiscale methods

multiple length scales

- better constitutive models

Multiscale models:
- design new materials
34
 Classification
After Ted Belytschko ARLEQUIN method

h i = C : h✏i

pile installation
Wriggers,2011

hierarchical methods semi-concurrent concurrent methods

 Arlequin method
[H. Ben Dhia, 1998] - partition of unity for energy in gluing zone
M m
↵ +↵ =1
- Lagrange multipliers to glue two models

a multi-scale/multi-model method

2D/1D FEM

3D/2D FEM
36 Mortar Method
Arlequin FEM

37 work in progress
 Heterogeneous materials
macroscopically homogeneous but microscopically heterogeneous

- size, shape
macroscopic behavior depends on - spatial distribution
phenomenological - volume fraction
= f (✏, ↵) - mechanical properties
constitutive models
two many params 38
of the constituents.
the identification of these parameters is generally difficult
 Direct Numerical
Simulation (DNS)
Unger, Eckard, 2011

Due to the high numerical effort and memory demand of

DNS, it is, in general, not possible to simulate the full
structure on the micro-/meso-scale with the computational
power available nowadays
44 cm

1 100 00039 degrees of freedom

 Homogenization
Homogenization = replace a heterogeneous material with an
equivalent homogeneous material. Voigt, 1910

Reuss 1929

Hill 1965

separation of scales

40 RVE=Representative Volume Element

 Homogenization
h i = C : h✏i
1 FEM

bottom-up approach Z
1
h i= m d⌦
|⌦m | ⌦m
2 Z
1
RVE h✏i =
|⌦m | ⌦m
✏m d⌦

6 independent loads are needed to determine 36 constants

+ simple, efficient C effective properties

- restricted to an assumed macro constitutive model
- linear elasticity, hyperelasticity, simple plasticity 41
 Artificial microstructures
Real microstructures: hard to obtain and not meshable
Statistically equivalent to real microstructures
Easy to discretized into finite elements

Build tailor made materials

F. Fritzen 2010
42
Computational Homogenization
FE2 method
[Renard, 1987, Smit 1998,
F. Feyel, 2000]

\sigma
+ nonlinear, large deformation
- computationally expensive
- 2D problems at laboratory scale
Micro problems are solved in parallel
- not always robust!!! 43
 Troubles with softening RVEs

strain localization

meso-structure of concrete

- RVE does not exist for softening materials

- CH cannot be applied for softening materials
44
Failure zone averaging

45 Nguyen et al, 2010

RVE does exist for softening materials by
using the failure zone averaging technique
46
Discontinuous CH model
MACRO MICRO

localization
band
discrete crack
47 Nguyen et al, 2011
Example

DNS

CH
48 RVE
Dynamic discontinuous
CH model
- macro: implicit dynamics
- micro: quasi-static
Vo

A. Karamnejad,
Nguyen, Sluys, 2012

49
More information

50
Image-based modeling

51
Traditional FE analysis

Geometry
Mesh FE solver
(CAD)
52
There are many cases in which such CAD geometries
are not available. However, image data are so ready:
medicine, material sciences...
(1) Industry

Simpleware

See also the FREE program OOF2, NIST, USA

53
(2) universities microstructure of cement paste
voxel based method
each voxel = one finite element

zig-zag boundary

- incorrect volume fraction

- images with high resolution are required P. Wriggers
- too large problem size!!! 54
 Level set/XFEM

Finite Cell Method (FCM) (Fictitious Domain Methods)

55
 Tools
Matlab is not enough. Consider Fortran, C++, Python.
Move to Ubuntu Linux to make your programming life much easier.
• Preprocessing: GMSH, GID, ANSYS, ABAQUS
•trilinos.sandia.gov
Solvers: 

- FEM: FEAP, OOFEM, libMesh, KRATOS, Code Aster,
/

TRILINOS, PERMIX, OpenSees (earthquake, structures)


- DEM: LAMMPS, KRATOS,YADE… 

OpenMPI
- CFD: OpenFoam, KRATOS… 

ParMETIS
- PD: LAMMPS
• Postprocessing: GMSH, PARAVIEW, MATLAB,TECPLOT
domain decomposition 56
 Prof. S. Bordas

Dr. O. Lloberas Valls

Prof. L.J. Sluys

A. Karamnejad Dr. E. Lingen Dr. M. Stroeven

Habanera develops jem/jive C++ library
Thank you for
your attention