Out

MULTISCALE METHODS FOR
MATERIAL FAILURE SIMULATIONS

by
Rong Fan
A Thesis Submitted to the Graduate
Faculty of Rensselaer Polytechnic Institute
in Partial Fulfillment of the
Requirements for the degree of
DOCTOR OF PHILOSOPHY
Major Subject: Civil Engineering

Approved by the
Examining Committee:

_________________________________________
Jacob Fish, Thesis Adviser

_________________________________________
Mark S. Shephard, Member

_________________________________________
Catalin R. Picu, Member

_________________________________________
Lucy T. Zhang, Member

Rensselaer Polytechnic Institute
Troy, New York
February, 2009
(For Graduation May 2009)

UMI Number: 3371297

INFORMATION TO USERS

The quality of this reproduction is dependent upon the quality of the copy
submitted. Broken or indistinct print, colored or poor quality illustrations and
photographs, print bleed-through, substandard margins, and improper
alignment can adversely affect reproduction.
In the unlikely event that the author did not send a complete manuscript
and there are missing pages, these will be noted. Also, if unauthorized
copyright material had to be removed, a note will indicate the deletion.

______________________________________________________________

UMI Microform 3371297
Copyright 2009 by ProQuest LLC
All rights reserved. This microform edition is protected against
unauthorized copying under Title 17, United States Code.
_______________________________________________________________

ProQuest LLC
789 East Eisenhower Parkway
P.O. Box 1346
Ann Arbor, MI 48106-1346

ii

Copyright 2009
by
Rong Fan
All Rights Reserved

iii
CONTENTS
MULTISCALE METHODS FOR MATERIAL FAILURE SIMULATIONS .................. i
CONTENTS ..................................................................................................................... iii
LIST OF FIGURES........................................................................................................... v
ACKNOWLEDGMENT ................................................................................................ viii
ABSTRACT ..................................................................................................................... xi
1. Introduction.................................................................................................................. 1
1.1 Overview............................................................................................................ 1
1.2 Outline of the Thesis .......................................................................................... 3
References.................................................................................................................... 4
2. Mathematical Homogenization of Nonperiodic Heterogeneous Media Subjected to
Large Deformation Transient Loading ........................................................................ 8
2.1 Introduction........................................................................................................ 8
2.2 Mathematical homogenization for geometrically nonlinear problems .............. 9
2.2.1 Asymptotic expansion for geometrically nonlinear problems ............. 10
2.2.2 Boundary conditions for the unit cell problem.................................... 14
2.2.3 The two-scale problem......................................................................... 17
2.2.4 Finite element discretization ................................................................ 18
2.3 Numerical Examples ........................................................................................ 22
2.3.1 Perforated plate .................................................................................... 22
2.3.2 Plate with a center hole ........................................................................ 25
2.3.3 Sandwich shell ..................................................................................... 27
References.................................................................................................................. 31
3. Adaptive Two-Scale Nonlinear Homogenization........................................................ 35
3.1 Introduction...................................................................................................... 35
3.2 Eigendeformation-based reduced order homogenization................................. 36
3.3 Adaptive homogenization model selection...................................................... 39

iv
3.4 Energy-based element erosion criterion........................................................... 41
3.5.1 Composites plate with a center crack................................................... 44
3.5.2 Sandwich shell impact ......................................................................... 48
3.5.3 Crush simulation of a composites circular tube ................................... 52
References.................................................................................................................. 53
4. The rs-method for material failure simulations........................................................... 58
4.1 Introduction...................................................................................................... 58
4.2 The Basic Idea.................................................................................................. 61
4.3 The Formulation............................................................................................... 64
4.4 A simple energy-based element erosion criterion............................................ 68
4.5.1 Plate with a centered crack................................................................... 71
4.5.2 Crack propagation in a concrete beam ................................................ 73
4.5.3 Impact simulation................................................................................. 78
References.................................................................................................................. 79
5. Conclusions and discussions...................................................................................... 83
References.................................................................................................................. 84

v
LIST OF FIGURES
Figure 1.1 Two-scale homogenization method ................................................................. 2
Figure 2.1 Macro and micro coordinate systems............................................................. 11
Figure 2.2 Definition of periodic boundary conditions: (a) initial unit cell (b) deformed
unit cell .................................................................................................................... 15
Figure 2.3 Definition of (a) essential, (b) natural and (c) mixed boundary conditions ... 16
Figure 2.4 Information transfer between the macro and micro problems ....................... 17
Figure 2.5 Unit cell configurations: (a) initial; (b) macro-deformed (intermediate); and (c)
micro-deformed (final) ............................................................................................ 21
Figure 2.6 Geometry, boundary conditions and unit cell discretization: (a) macro-
geometry and boundary conditions and (b) mesh of the unit cell............................ 23
Figure 2.7 Reaction force in x (top) and y (bottom) directions ....................................... 24
Figure 2.8 Unit cell deformation and von Mises stress obtained from the (a) multiscale
and (b) reference solutions....................................................................................... 25
Figure 2.9 Plate with a hole in the center ........................................................................ 25
Figure 2.10 Comparison of the unit cell deformation and von Mises stress at 0.3ms as
obtained with (a) reference solution by direct finite element method; (b) multiscale
solution with essential boundary conditions; (c) multiscale solution with natural
boundary conditions; (d) multiscale solution with mixed boundary conditions...... 26
Figure 2.11 Geometry, layup and impact loading of the sandwich shell ........................ 27
Figure 2.12 Initial configuration of the (a) undeformed coarse-scale (macro) and (b)
undeformed unit cell models ................................................................................... 28
Figure 2.13 Residual velocity of the impactor................................................................. 29
Figure 2.14 von Mises stress at point C (in Figure 2.12(a)) ............................................ 29
Figure 2.15 Deformation of the sandwich plate obtained with the (a) reference and (b)
multiscale solutions at t=15ms................................................................................. 30
Figure 2.16 Local deformation in (a) region a and (b) region b (see Figure 2.15(b)) at
t=15ms as obtained with reference (left) and multiscale (right) solutions .............. 30
Figure 3.1 Effective stress-strain curve ........................................................................... 41

vi
Figure 3.2 Crack closure tractions and displacements in: (a) the analytical model, (b) the
direct finite element mesh simulated by element erosion, and (c) the macro-mesh of
the two-scale problem.............................................................................................. 42
Figure 3.3 (a) plate with a center crack and (b) mesh of the unit cell ............................. 45
Figure 3.4 Comparison of displacement contour in z direction at t=1.6s (Mode II): (a)
reference solution; (b) energy based solution; and (c) stress based solution........... 46
Figure 3.5 The center point O displacement in the loading (z) direction (Mode II) ....... 46
Figure 3.6 Comparison of displacement contour in z direction at t=1.04s (Mode I): (a)
reference solution; (b) energy based solution; and (c) stress based solution........... 47
Figure 3.7 The center point O displacement in the loading (z) direction (Mode I)......... 47
Figure 3.8 Geometry, layup and impact loading of the sandwich shell .......................... 48
Figure 3.9 Initial configuration of the macro and unit cell models: (a) undeformed coarse
scale model; and (b) undeformed unit cell model ................................................... 48
Figure 3.10 Residual velocity of the impactor................................................................. 50
Figure 3.11 von-Mises stress at point C (in Figure 3.9(a)).............................................. 50
Figure 3.12 Deformation of the sandwich plate obtained with the (a) reference and (b)
multiscale solutions at t=12ms................................................................................. 51
Figure 3.13 Local deformation in regions a and b (see Figure 3.12(b)) at t=12ms as
obtained with reference (left) and multiscale (right) solutions................................ 51
Figure 3.14 Finite element model of a circular tube........................................................ 52
Figure 3.15 (a) mesh of the unit cell and (b) phase partitions of unit cell....................... 52
Figure 3.16 Comparison of load-displacement history.................................................... 53
Figure 4.1 A two-level mesh superposition..................................................................... 61
Figure 4.2 The model problem........................................................................................ 63
Figure 4.3 (a) displacement field in the 0 and 1 level meshes and (b) corresponding
shape functions in the two meshes........................................................................... 63
Figure 4.4 (a) Lowest frequency eigenmode
1
1
and (b) corresponding enrichment
functions
0 1
1
( ) ( )
I
N x x ............................................................................................. 64
Figure 4.5 Definition of a local patch
1
( ) t at time t ...................................................... 65

vii
Figure 4.6 Suppressing linear dependency between underlying and superimposed meshes.
Node I in the superimposed mesh overlaps node J in the underlying mesh............ 67
Figure 4.7 Crack closure tractions and displacements: (a) in the analytical model and (b)
in a mesh simulated by element erosion.................................................................. 68
Figure 4.8 (a) Plate with center crack and (b) A superimposed mesh for a plate with a
centered crack.......................................................................................................... 71
Figure 4.9 Error in the stress intensity factor versus number of enrichment modes ....... 72
Figure 4.10 Geometry of notched concrete beam (dimensions in mm) .......................... 73
Figure 4.11 The reference finite element mesh (28x24 elements in the refined region at
the center of the beam) ............................................................................................ 73
Figure 4.12 Height of the crack versus time.................................................................... 74
Figure 4.13 Reaction at support D versus time................................................................ 74
Figure 4.14 A completely cracked beam as obtained with the energy based erosion
criterion in the four meshes ..................................................................................... 75
Figure 4.15 rs-method: (a) Underlying mesh, and (b) superimposed mesh .................... 76
Figure 4.16 Reaction at support D versus time as obtained with the rs-method (six and
nine modes) and the reference solution ................................................................... 77
Figure 4.17 Height of the crack versus time as obtained with the rs-method (six and nine
modes) and the reference solution........................................................................... 77
Figure 4.18 Model problem for impact simulation.......................................................... 78
Figure 4.19 Comparison of the velocity of impactor versus time ................................... 78

viii
ACKNOWLEDGMENT

I would like to take this opportunity to express my sincere thanks and gratitude to
my advisor, Dr. Fish, for his guidance, his support, and all the useful suggestions. Over
the past five years, he guided me to explore and appreciate the multiscale and multigrid
methods. Without his help, this thesis would not have been possible.

I would also like to thank my doctoral committee members: Dr. Shephard, Dr. Picu
and Dr. Zhang for reviewing my thesis and giving valuable comments.

During my thesis work, I got a lot of help from my colleagues and friends. I would
like to thank Dr. Zheng Yuan, Dr. Caglar Oskay, Dr. Aiqin Li, Wei Wu, for all the
helpful discussions, support, encouragement, concerns and friendships. I would like to
give my special thanks to Dr. Zheng Yuan for sharing his knowledge and codes with me.
For that I will always be grateful.

In addition, I would like to thank Janet Pertierra in Civil and Environmental
Engineering Department, Marge Verville and Adam Todorski in Scientific Computation
Research Center, Karen Georgeadis in Multiscale Science and Engineering Center, for
their kindly help.

Finally, I wish to express my deepest appreciation to my parents, brother and sister-
in-law. During the long and stressful journey of education, they have always been there
with encouragement and endless support.

ix
If.
by Rudyard Kipling
If you can keep your head when all about you
Are losing theirs and blaming it on you;
If you can trust yourself when all men doubt you,
But make allowance for their doubting too;
If you can wait and not be tired by waiting,
Or, being lied about, don't deal in lies,
Or, being hated, don't give way to hating,
And yet don't look too good, nor talk too wise;

If you can dream - and not make dreams your master;
If you can think - and not make thoughts your aim;
If you can meet with Triumph and Disaster
And treat those two impostors just the same;
If you can bear to hear the truth you've spoken
Twisted by knaves to make a trap for fools,
Or watch the things you gave your life to, broken,
And stoop and build 'em up with worn-out tools;

If you can make one heap of all your winnings
And risk it all on one turn of pitch-and-toss,
And lose, and start again at your beginnings
And never breath a word about your loss;
If you can force your heart and nerve and sinew
To serve your turn long after they are gone,
And so hold on when there is nothing in you
Except the Will which says to them: "Hold on!"

If you can talk with crowds and keep your virtue,
Or walk with kings--nor lose the common touch;
If neither foes nor loving friends can hurt you;
If all men count with you, but none too much;
If you can fill the unforgiving minute
With sixty seconds' worth of distance run -
Yours is the Earth and everything that's in it,
And - which is more - you'll be a Man, my son!

x

To My Grandfather

xi
ABSTRACT

The objective of this thesis is to develop multiscale methods for material failure
simulations. Three major topics are covered in this thesis: (i) mathematical
homogenization method for geometrically nonlinear problems; (ii) adaptive two-scale
nonlinear homogenization method; and (iii) reduced order superposition finite element
method.
In the first topic, the classical mathematical homogenization theory is generalized to
account for finite unit cell distortions of heterogeneous medium subjected to transient
loading. An auxiliary macro-deformed configuration, where the overall Cauchy stress is
defined, and mixed unit cell boundary conditions for modeling nonperiodic boundary
deformation are introduced and verified. The method is implemented in ABAQUS and
verified against direct finite element simulations.
To reduce the computational cost of direct homogenization method, an adaptive
two-scale nonlinear homogenization method is developed in the second topic. The
adaptive two-scale homogenization procedure combines a reduced order homogenization
method and a direct mathematical homogenization method. A simple strain-based
indicator is used to guide the construction of a nearly optimal model consisting of either
a direct or a reduced order unit cell model. Consistent interscale transfer operators are
developed to map the information between the two homogenization models. The method
has been verified and validated.
Finally, a reduced order superposition finite element method is developed based on
the partition of unit method and the s-version of the finite element method. The degrees
of freedom of the superposition method are reduced by projecting the lowest eigenmodes
(excluding rigid body) of the superimposed fine mesh to the underlying coarse mesh. A
simplified energy-based element erosion criterion is introduced to alleviate mesh size-
dependency for material failure simulations. The method has been integrated in
ABAQUS, and the failure criterion has been verified numerically.

1
1. Introduction

1.1 Overview
Most of the materials are heterogeneous of composite at one or more scales. These
composites may be natural, such as wood, bones, and teeth or they may be man-made,
such as fiber reinforced composites, particle reinforced composites, metal matrix
composites, ceramic matrix composites and polymer matrix composites. Although these
composites have different macroscopic mechanical behavior, they have one common
feature: they possess a heterogeneous structure at one or more scales. It is the composites
at the macroscale, which can be considered as homogenous.
In principle, direct numerical simulation (DNS) can be used to analyze the
mechanical behavior of structures made of composites. However, for composites having
complex microstructure, when it is computationally resolved at the macroscale, the
computational cost is extremely high. This makes DNS for composites impractical.
To overcome this problem, over the past fifty years, several methods have been
developed to establish the relationship between macroscopic (overall or effective)
properties and the microstructure. The early contribution was the effective field method
(see Eshelby [1]), which is based on the closed form of the elliptic inclusion embedded
in the infinite elastic medium subjected to uniform strain. Using the Eshelby solution for
an ellipsoidal inclusion, Hershey [ 2] first established the self-consistent method to
determine the overall elastic moduli of cubic crystals. The self-consistent method was
further developed by Budiansky [3], Hill [4], Christensen and Lo [5] for composites.
The self-consistent method is similar to the Mori-Tanaka method [6], which also utilizes
the Eshelby solution. Based on the variational principles, Hashin [ 7], Hashin and
Shtrikman [8] obtained the upper and lower bounds for the elastic moduli of composite
materials in terms of phase moduli, shape and volume fraction.
Homogenization method based on the mathematical homogenization theory
(asymptotic homogenization theory) was pioneered by Babuska [9], Benssousan [10],
Sanchez-Palencia [11], Bakhvalov and Panasenko [12]. The homogenization method
had a wide application in science and engineering problems, such as continuum

2
mechanics, heat transfer and hydrodynamics. The main advantages of this method are: (a)
it provides both global and local field variables, such stress and strain; (b) once the
geometrical configuration and microscale fields are known, the macroscopic constituents
can be determined without any assumption; (c) it allows to assess the accuracy of the
microscopic response of interest [13].

Figure 1.1 Two-scale homogenization method

The basic idea of the two-scale homogenization method for fiber reinforced
composites is illustrated in Figure 1.1. The microstructure is modeled by a
Representative Volume Element (RVE), also called as Unit Cell. The RVE should be
small enough to capture the microstructure of the composites. For nonlinear problems,
the overall stress-strain relationship is obtained numerically. For a given strain (or strain
increment) at any macroscopic material point, a boundary value problem is defined for
the corresponding RVE. The boundary conditions of the RVE are typically assumed to
be periodic. Once the RVE boundary problem is solved the overall stress (macro stress)
is obtained by averaging the micro stress over the RVE domain.
Material point
Initial RVE
and its FEM
mesh
Macroscopic domain
(homogeneous)
Deformed RVE
Macro strain Macro stress
Micro scale problem
Macro scale problem
Fiber
Matrix

3
Recent development of the homogenization method for material damage and fatigue
has been presented in [14, 15, 16, 17]. Terada et al. [18] studied the convergence of
homogenization method. Ghosh et al. [19, 20] developed a multiscale finite element
method by combining the homogenization theory with Voronoi cell finite element
method for microstructural modeling. Smit et al.[21], Miehe et al.[22], Kouznetsova et
al.[ 23 ], Feyel et al.[ 24 ], and Michel et al.[ 25 ] made contributions to improve
computational cost and accuracy of homogenization method.
However, computational cost of homogenization methods remains substantial for
nonlinear problems with complex microstructures. To overcome this difficulty, an
adaptive multiscale method was developed by Ghosh et al.[26]. Fish and Oskay [27]
developed reduced order homogenization method based on the transformation field
analysis pioneered by Dvorak et al. [28, 29].

1.2 Outline of the Thesis
Chapter 2 presents a generalization of the classical mathematical homogenization
theory aimed at accounting for finite unit cell distortions, which gives rise to a
nonperiodic asymptotic expansion. We introduce an auxiliary macro-deformed
configuration where the overall Cauchy stress is defined and nonperiodic boundary
conditions. Verification studies against a direct numerical simulation demonstrate the
versatility of the proposed method.
To reduce the computational cost of direct homogenization method, in Chapter 3
adaptive two-scale homogenization procedure combining a reduced order and a direct
mathematical homogenization method is developed. A simple strain-based indicator is
used to guide the construction of a nearly optimal model consisting of either a direct or a
reduced order unit cell model. Consistent interscale transfer operators are developed to
map the information between the two homogenization models. Verification studies are
conducted on a composites plate with a center crack, impact of a metal-polyurea
sandwich plate and crush of a fabric composites circular tube.
In Chapter 4 a reduced order superposition finite element method is developed
based on the partition of unit method and the s-version of the finite element method. The

4
degrees of freedom of the superposition method are reduced by projecting the lowest
eigenmodes (excluding rigid body) of the superimposed fine mesh to the underlying
coarse mesh. A simplified energy-based element erosion criterion is introduced to
alleviate mesh size-dependency for material failure simulations. The method, coined as
the rs-version of the finite element method (or reduced order s-method), has been
integrated in ABAQUS, and the failure criterion has been verified numerically.
A brief summary and recommendations on future research directions in Chapter 5
conclude the thesis.

References

1 J.D. Eshelby, The Determination of the Field of an Ellipsoidal Inclusion and Related
Problems, Proceedings of the Royal Society of London. Series A, vol.241, pp.376-396,
1957.
2 A.V. Hershey, The Elasticity of an Isotropic Aggregate of Anisotropic Cubic Crystals,
Journal of Applied Mechanics, vol.21, pp.236-241, 1954.
3 B. Budiansky, On the Elastic Moduli of Some Heterogeneous Materials, Journal of the
Mechanics and Physics of Solids, vol.13, pp.223227, 1965.
4 R. Hill, A Self-consistent Mechanics of Composite Materials, Journal of the
Mechanics and Physics of Solids, vol.13, pp.213222, 1965.
5 R. M. Christensen and K.H. Lo, Solutions for Effective Shear Properties in Three
Phase Sphere and Cylinder Models, Journal of the Mechanics and Physics of Solids,
vol.27, pp.315330, 1979.
6 T. Mori and K. Tanaka, Average Stress in Matrix and Average Elastic Energy of
Materials with Misfitting Inclusions, Acta Metallurgica et Materialia, vol.21, pp.571
574, 1973.
7 Z. Hashin, The Elastic Moduli of Heterogeneous Materials, Journal of Applied
Mechanics, vol.29, pp.143150, 1962.

5

8 Z. Hashin and S. Shtrikman, A Variational Approach to the Theory of the Elastic
Behaviour of Multiphase Materials, Journal of the Mechanics and Physics of Solids,
vol.11, pp.127140, 1963.
9 I. Babuska, Homogenization and Application, Mathematical and Computational
Problems, In Numerical Solution of Partial Differential Equations III, SYNSPADE, ed.
B Hubbard, Academic Press, 1975.
10 A. Benssousan, J.L. Lions and G. Papanicoulau, Asymptotic Analysis for Periodic
Structures, North-Holland, 1978.
11 E. Sanchez-Palencia, Non-homogeneous Media and Vibration Theory, Lecture notes
in physics, vol. 127, Springer-Verlag, Berlin, 1980.
12 N. Bakhvalov and G. Panasenko, Homogenization: Averaging Processes in Periodic
Media, Kluwer Academic Publishers, 1989.
13 J. Fish, P. Nayak, and M. H. Holmes, Microscale Reduction Error Indicators and
Estimators for a Periodic Heterogeneous Medium, Computational Mechanics: The
International Journal, vol. 14, pp. 1-16, 1994.
14 J. Fish, Q. Yu and K. L. Shek, Computational Damage Mechanics for Composite
Materials Based on Mathematical Homogenization, International Journal for Numerical
Methods in Engineering, vol. 45, pp. 1657-1679, 1999.
15 J. Fish and Q.Yu, Multiscale Damage Modeling for Composite Materials: Theory and
Computational Framework, International Journal for Numerical Methods in Engineering,
vol. 52, pp. 161-192, 2001.
16 J. Fish and Q.Yu, Computational Mechanics of Fatigue and Life Predictions for
Composite Materials and Structures, Computer Methods in Applied Mechanics and
Engineering, vol. 191, pp. 4827-4849, 2002.
17 Z. Yuan and J. Fish, Towards Realization of Computational Homogenization in
Practice, International Journal for Numerical Methods in Engineering, vol. 73, pp. 361-
380, 2008.
18 K. Terada, M. Hori, T. Kyoya, and N.Kikuchi, Simulation of the Multi-scale
Convergence in Computational Homogenization Approaches, International Journal of
Solids and Structures, vol. 37, pp. 2285-2311, 2000.

6

19 S. Ghosh, K. Lee, and S. Moorthy, Multiple Scale Analysis of Heterogeneous Elastic
Structures Using Homogenization Theory and Voronoi Cell Finite Element Method,
International Journal of Solids and Structures, vol. 32, pp. 2762, 1995.
20 S. Ghosh, K. Lee, and S. Moorthy, Two Scale Analysis of Heterogeneous Elastic-
plastic Materials with Asymptotic Homogenisation and Voronoi Cell Finite Element
Model, Computer Methods in Applied Mechanics and Engineering, vol. 132, pp. 63116,
1996.
21 R. J. M. Smit, W. A. M. Brekelmans, and H. E. H. Meijer, Prediction of the
Mechanical Behavior of Nonlinear Heterogeneous Systems by Multi-level Finite
Element Modeling, Computer Methods in Applied Mechanics and Engineering, vol. 155,
pp. 181-192, 1998.
22 C. Miehe, and A. Koch, Computational Micro-to-macro Transition of Discretized
Microstructures Undergoing Small Strain, Archives of Applied Mechanics, vol. 72, pp.
300-317, 2002.
23 V. Kouznetsova, W.A.M. Brekelmans, and F.P.T. Baaijens, An Approach to Micro-
macro Modeling of Heterogeneous Materials, Computational Mechanics, vol. 27, pp. 37-
48, 2001.
24 F. Feyel, and J.-L. Chaboche, FE2 Multiscale Approach for Modeling the Elastovis-
coplastic Behavior of Long Fiber SiC/Ti Composite Materials, Computer Methods in
Applied Mechanics and Engineering, vol.183, pp. 309-330, 2000.
25 J.C. Michel, H. Moulinec, and P. Suquet, Effective Properties of Composite Materials
with Periodic Microstructure: a Computational Approach, Computer Methods in Applied
Mechanics and Engineering, vol.172, pp. 109-143, 1999.
26 S. Ghosh, K. Lee, and P. Raghavan, A Multi-level Computational Model for Multi-
scale Damage Analysis in Composite and Porous Materials, International Journal of
Solids and Structures, vol. 38, pp. 23352385, 2001.
27 C. Oskay and J. Fish, Eigendeformation-Baed Reduced Order Homogenization for
Failure Analysis of Heterogeneous Materials, Computer Methods in Applied Mechanics
and Engineering, vol. 196, pp. 1216-1243, 2007.

7

28 G. J. Dvorak, Transformation Field Analysis of Inelastic Composite Materials,
Proceedings: Mathematical and Physical Sciences, vol. 437, pp. 311-327, 1992.
29 G.J. Dvorak, Y.A. Bahei-El-Din, and A.M. Wafa, Implementation of the
Transformation Field Analysis for Inelastic Composite Materials, Computational
Mechanics, vol. 14, pp. 201-228, 1994.

8
2. Mathematical Homogenization of Nonperiodic Heterogeneous
Media Subjected to Large Deformation Transient Loading

2.1 Introduction
The chapter presents a mathematical homogenization framework for large
deformation of nonperiodic heterogeneous medium subjected to transient loading. The
subject of mathematical homogenization dates back to the pioneering works of Babuska
[ 1 ], Benssousan [ 2 ], Sanchez-Palencia [ 3 ] and Bakhvalov [ 4 ]. Engineering
homogenization approaches date back to Hill [5]. After more than 30 years and over
1,450,000 Google hits on the word homogenization the present article contribution is
on the following two aspects:
i. Theory: There is a common belief that methods based on the multiple scale
asymptotic expansion [1-4] are limited to small deformation problems and this leads to
so-called computational homogenization approaches, which are based on the Hill-
Mandel relation [ 6 ], in combination with numerical methods (see
[7, 8,9, 10, 11,12, 13, 14,15, 16, 17,18, 19] and for closely related homogenization-like
methods [ 20, 21, 22, 23, 24]). Although mathematical homogenization in combination
with finite element methods have been successfully employed to capture material
nonlinearities ([25,26,27,28] including large unit cell rotations [26,29]), the inherent
assumption of asymptotic methods that macroscopic fields remain constant over a unit
cell domain is questionable at best for large cells undergoing significant distortion. The
present article presents a generalization of the mathematical homogenization theory that
permits consideration of large unit cells undergoing significant nonperiodic distortion.
ii. Practice: Wave propagation in a medium with embedded local weak and/or
strong discontinuities is one of the most attractive applications of the method. Provided
that the usual assumption about the smallness of the local problem holds, the original
initial-boundary value problem can be decomposed into macro and micro problems with
the micro problem being a steady state (or static) problem. This avoids the use of small
finite elements in the macro problem, which govern the time step of explicit calculations.

9
This chapter is organized as follows. Generalization of the mathematical
homogenization theory to account for large nonperiodic unit cell distortions and its
computational implementation are described in Section 2.2. Verification studies against
a direct numerical simulation where local details are computationally resolved are
conducted in Section 2.3.

2.2 Mathematical homogenization for geometrically nonlinear
problems
In deriving the coupled two-scale problem, the governing equations at the smallest
scale of interest (microscale) will be stated in terms of the deformation gradient and the
first Piola-Kirchhoff stress. This pair of work conjugate variables has been originally
advocated by Hill [6] and later adopted in [9, 17, 19]. The choice of the deformation
gradient (and its conjugate stress measure) is convenient in defining the essential
boundary conditions (BCs) on the unit cell. In the discretization phase, the finite element
equations will be reformulated in terms of Cauchy stress for problems where constitutive
equations at the microscale are defined in terms of Cauchy stress.
The strong form of the initial-boundary value problem on domain
X
with
boundary
X
is given as
( )
2
2
0 on
ij
i
Xi X X
j
P
u
b
X t

+ =

F
(2.1)

i
ik ik
k
u
F
X

= +
(2.2)
with BCs:
t
ij Xj Xi X
P N t on
= (2.3)

u
i i X
u u on
= (2.4)
0
t u t u
X X X X X
and = = (2.5)
and initial conditions:
0
0
i i
u u at t
= = (2.6)

10
0
0
i
i
u
v at t
t
= =
(2.7)
where
P denotes the first Piola-Kirchhoff stress tensor;
F the deformation gradient;

X
b the body force vector,
X
N the unit normal to the boundary
X
and
X
the initial
mass density. Lower case subscripts i and j denote spatial dimensions. Subscripts X and
x refer to the initial and deformed configurations, respectively. The superscript
denotes dependence of a variable on the fine (micro) scale features. X is a material
coordinate in the initial macroscopic domain
X
and x is the corresponding deformed
(spatial) coordinate. Summation convention over repeated subscripts is employed except
for the subscripts X and x.

2.2.1 Asymptotic expansion for geometrically nonlinear problems
In the classical mathematical homogenization theory, displacements are expanded as
0 1 2 2 3
( , ) ( , ) ( , , ) ( , , ) ( )
i i i i
u t u t u t u t O
= + + + X X X Y X Y (2.8)
where Y represents material coordinates in the initial microscopic (unit cell)
Y
domain
and y is the deformed (spatial) coordinate in the unit cell domain
y
. The two scale
coordinates are related by
X
Y with 0 < 1 < .
In the classical theory, it is assumed that the size of the unit cell is of order , i.e.
infinitesimally small and therefore the macroscopic displacement
0
( , )
i
u t X is considered
to be constant over the unit cell domain. For large unit cell distortions, the macro
displacement
0
( , )
i
u t X is no longer constant over the unit cell domain and therefore
equation (2.8) has to be modified as described below.
We start by expanding
0
( , )
i
u t X in the Taylors series around an arbitrary reference
point
= X X in the unit cell (see Figure 2.1)

0 2 0
0 0

1

( , ) ( , ) ( ) ( )( )
2
i i
i i j j j j k k
j j k
u u
u t u t X X X X X X
X X X

= + + +

X X
X X (2.9)

11
As
j j j
X X Y = , we can write
0 2 0
0 0 2 3

1
( , ) ( , ) ( )
2
i i
i i j j k
j j k
u u
u t u t Y Y Y O
X X X

= + + +

X X
X X (2.10)
Similarly, the higher order terms ( , , ), for 1
n
i
u t n X Y in (2.8) are expanded around
= X X as
2

2
2 3

1

( , , ) ( , , ) ( ) ( )( )
2
1
( , , ) ( )
2
n n
n n i i
i i j j j j k k
j j k
n n
n i i
i j j k
j j k
u u
u t u t X X X X X X
X X X
u u
u t Y Y Y O
X X X

= + + +

= + + +

X X
X X
X Y X Y
X Y
(2.11)

Figure 2.1 Macro and micro coordinate systems

Substituting (2.10) and (2.11) into (2.8) gives

0 1 2 2 3

( , ) ( , , ) ( , ) ( , , ) ( , , ) ( )
i i i i i
u t u t u t u t u t O
= = + + + X X Y X X Y X Y (2.12)
where
0 0

( , ) ( , )
i i
u t u t = X X (2.13)
0
1 1

( , , ) ( , , )
i
i i j
j
u
u t u t Y
X
= +
X
X Y X Y (2.14)
y
1

Initial
Unit

Y
1

Y
2

y
2

1 1
X , x
2 2
X , x
O
O
O
Deformed
Unit Cell
= X X
= X x

12
1 2 0
2 2

1

( , , ) ( , , )
2
i i
i i j j k
j j k
u u
u t u t Y Y Y
X X X

= + +

X X
X Y X Y (2.15)
In classical homogenization, the spatial derivative is defined as
( , , ) 1 ( , , )
i i i
f f t f t
X X Y

= +

X Y X Y
(2.16)
In the case
( , , ) f t X Y rather than ( , , ) f t X Y , the spatial derivative takes the form
( , ) 1 ( , , )
i i
f t f t
X Y

=

X X Y
(2.17)
The displacement gradient components may be written as
1 2
2

( , , ) ( , , ) ( , , ) 1
( )
i i i i
k k k k
u u t u t u t
O
X Y Y Y

= = + +

X Y X Y X Y
(2.18)
where
1 1 0

( , , ) ( , , ) ( , )
i i i
k k k
u t u t u t
Y Y X

= +

X
X Y X Y X
(2.19)
2 2 1 2 1 2 0

( , , ) ( , , ) ( , , ) ( , , )
i i i i i
j j
k k k j k j k
u t u t u t u t u
Y Y
Y Y X X Y X X

= + + +

X
X X
X Y X Y X Y X Y
(2.20)
The deformation gradient can be expressed as
0 1 2

( , , ) ( , , ) ( )
i
ik ik ik ik
k
u
F F t F t O
X
= + = + +
X Y X Y (2.21)
where
ij
is the Kronecker delta and

1
0
( , , )

( , , ) ( , , )
i
ik ik ik
k
u t
F t F t
Y

= +
X Y
X Y X Y (2.22)

2
1
( , , )
( , , )
i
ik
k
u t
F t
Y
X Y
X Y (2.23)
The overall (macro) deformation gradient is obtained by integrating the leading order
term
( , , )
ik
F t X Y in (2.22) over the unit cell domain, which yields
0
( , ) 1

( , ) ( , )
Y
i
ik ik Y ik
Y k
u t
F t F ,t d
X
= = +

X
X
X X Y (2.24)

13
where for the time being (see Section 2.2.2 for nonperiodic case) we assume Y-
periodicity of
1
( , , )
i
u t X Y . A function f, is considered to be periodic if
( , ) ( , ) f k f + = X Y L X Y , where L is a characteristic size of the unit cell, i.e., ( , ) f X Y is
equal on the opposite of a unit cell.
The first Piola-Kirchhoff stress ( ) ( )
,
ij
P t

F X formulation is adopted because of its
conjugacy to the deformation gradient. In classical homogenization, the asymptotic
expansion of ( ) ( )
,
ij
P t

F X may be written as
( ) ( ) ( ) ( ) ( ) ( )
0 1 2
, , , , , ( )
ij ij ij
P t P t P t O

= + + F X F X Y F X Y (2.25)
Consider Taylor series expansion of
0
ij
P and
1
ij
P around
= X X yields

0
0 0

( , , ) ( )
ij
ij ij k k
k
P
P P t X X
X
= + +

X
X Y (2.26)

1
1 1

( , , ) ( )
ij
ij ij k k
k
P
P P t X X
X
= + +

X
X Y (2.27)
As
j j j
X X Y = , substituting (2.26) and (2.27) into (2.25) gives
( ) ( )
0
0 1 2

, ( , , ) ( , , ) ( )
ij
ij ij k ij
k
P
P t P t Y P t O
X

= + + +

X
F X X Y X Y (2.28)
Inserting the asymptotic expansions (2.28) into the equation of motion (2.1) and
exploiting the definition of the spatial derivative (2.17) gives
0 0 1
2 0
1
2

( , ) ( , ) ( , ) ( , )

( , ) ( , ) ( ) 0
ij ij ij
i
Xi X
j j j
P t P t P t
u t
b O
Y X Y t

+ + + + =

X Y, X Y, X Y,
X
X Y X Y
(2.29)
Identifying terms with equal powers of yields the following two leading order
momentum equations

0
1
( , )
( ) : 0
ij
j
P t
O
Y
X Y,
(2.30)
0 1
2 0
0
2

( , ) ( , ) ( , )

( ) : ( , ) ( , ) 0
ij ij
i
Xi X
j j
P t P t
u t
O b
X Y t

+ + =

X Y, X Y,
X
X Y X Y (2.31)

14
Integrating (2.31) over the unit cell domain gives the macroscopic equation of motion
2 0
2
( , ) ( , )

( ) ( ) 0
ij
i
Xi X
j
P t
u t
b
t X

+ =

X
X
X X (2.32)
where
( )
0
1

( , ) , ,
Y
ij ij Y
Y
P t P t d

X X Y (2.33)
1 1

( ) ( , ) ( ) ( , )
Y Y
Xi Xi Y X X Y
Y Y
b b d d

= =

X X Y X X Y (2.34)
Remark 1:
Note that even if the original asymptotic expansion (2.8) were periodic in Y, then its
modified version (2.12) would not be periodic due to its dependence on
j
Y . Equation
(2.24) and (2.32) were the only instances where the periodicity assumption was made. In
the next Section we will explore various possibilities on how to satisfy equation (2.24)
without the periodicity assumption.

2.2.2 Boundary conditions for the unit cell problem
In this Section, we will consider various boundary conditions for the unit cell
problem starting with the classical periodic case.
Consider an asymptotic expansion of the displacement field (2.12)
1
0 1 2
( , , )

( , ) ( , , ) ( , ) ( ( , ) ) ( , , ) ( )
i
i i i ij ij j i
u t
u t u t u t F t Y u t O

= = + + +

_
X Y
X X Y X X X Y (2.35)
where we have accounted for equation (2.24).
The leading order term in (2.35) represents the rigid body translation of the unit cell
0
( , )
i
u t X , which is independent of the unit cell coordinates Y . The unit cell distortion is
captured by two ( ) O terms. The first term,
( ( , ) )
ij ij j
F t Y X , represents a uniform
macroscopic deformation, whereas the second term,
1
( , , )
i
u t X Y , captures the deviation
from the uniform field introduced by material heterogeneity.
Figure 2.2(a) and (b) show the initial and deformed shape of the unit cell,
respectively. The dotted line in Figure 2.2(b) depicts the deformed shape of the unit cell

15
due to
( ( , ) )
ij ij j
F t Y X , whereas the solid line shows the contribution of the two
( ) O terms.
At the unit cell vertices,
vert
Y
, the deviation from the uniform field,
1
( , , )
i
u t X Y , is
assumed to vanish. For the remaining points on the boundary of the unit cell, the
deviation from the average
1
( , , )
i
u t X Y is prescribed to be a periodic function. Figure 2.2
depicts two points M and S on the opposite faces of the unit cell with M and S being
master and slave points, respectively. The displacement of the two points is given as
1 M M 1 M

( , , ) ( ( , ) ) ( , , )
i ij ij j i
u t F t Y u t = + X Y X X Y (2.36)
1 S S 1 S

( , , ) ( ( , ) ) ( , , )
i ij ij j i
u t F t Y u t = + X Y X X Y (2.37)
Subtracting equation (2.37) from (2.36) and accounting for periodicity,
1 M 1 S

( , , ) ( , , )
i i
u t u t = X Y X Y , we get the periodic boundary condition
1 M 1 S M S

( , , ) ( , , ) ( ( , ) )( )
i i ij ij j j
u t u t F t Y Y = X Y X Y X (2.38)
where Y
M
and Y
S
represent the coordinates of the master and slave nodes on the unit cell
boundary, respectively.

Figure 2.2 Definition of periodic boundary conditions: (a) initial unit cell (b)
deformed unit cell

As noted in Remark 1, the only instance where we assumed periodicity was in
deriving Equation (2.24). For a nonperiodic medium, we will now re-examine the
derivation of (2.24). For (2.24) to hold, the following condition must be satisfied:
Y
1

1
y
M
M
S
S
2
C
1
C
3
C
4
C
2
y
Y
2

M S
(a)
1
C
3
C
4
C
2
C
Y

y

(b)

16
1
( , , )
0
Y
i
Y
j
u t
d
Y
X Y

Applying Greens theorem and exploiting relations (2.14) and (2.24) the above reduces
to
( )
1

( , , ) ( ( , ) ) 0
Y
i ik ik k Yj Y
u t F t Y N d
X Y X (2.39)
where
Y
is the boundary of
Y
and
Yj
N are the components of the unit normal to the
boundary
Y
.
Alternatively to Equations (2.38) and (2.39), an essential boundary condition
1

( , , ) ( ( , ) ) 0
i ik ik k
u t F t Y = X Y X (2.40)
is often exercised in practice.
The essential boundary condition (2.40) can be enforced in the weak form as
( )
1

( , , ) ( ( , ) ) 0
Y
i ik ik k i Y
u t F t Y d
X Y X (2.41)
where
i
is a Lagrange multiplier representing unknown tractions on
Y
. If we choose
i ij Yj
P N = with
ij
P being constant over
Y
and require (2.41) to be satisfied for
arbitrary
ij
P , we then obtain Equation (2.39). We will refer to Equation (2.39) as a
natural boundary condition. Equation (2.39) is in the sprit of the boundary condition
proposed by Mesarovic and Padbidri [30] who required the unit cell to satisfy average
small strains.

Figure 2.3 Definition of (a) essential, (b) natural and (c) mixed boundary conditions
(a)
1
C 1
y
2
C
3
C
4
C
2
y
y

1
C 1
y
2
C
3
C
4
C
2
y
y
2
C
1
C
3
C
4
C
1
y
y

2
y
(b) (c)

17

The essential (2.40) and natural (2.39) boundary conditions can be combined in so-
called mixed boundary condition as follows
( )
( )
1
1

( , , ) ( ( , ) ) 0

( , , ) ( ( , ) )
Y
i ik ik k Yj Y
i ik ik k Yj
u t F t Y N d
u t F t Y N
X Y X
X Y X
(2.42)
Note that the mixed boundary condition (2.42) is more restrictive than the natural
boundary condition (2.39), but is more compliant than the essential boundary condition
(2.40), which is enforced up to a tolerance . The three types of unit cell boundary
conditions are schematically illustrated in Figure 2.3. The mixed boundary condition can
be implemented in several ways: perhaps the simplest is by defining double nodes on the
boundary and placing a linear or nonlinear spring between them as illustrated in Figure
2.3(c). We will discuss the definition of the spring stiffness in the numerical examples
section. Note that in the case of a very stiff spring the mixed boundary condition (2.42)
coincides with the essential boundary condition (2.40); for an infinitely compliant spring
it reduces to the natural boundary condition (2.39).
We will examine the four types of boundary conditions for their suitability for large
deformation problems. We will denote the four boundary conditions as
( )
1
0
Y
on = g u (2.43)

2.2.3 The two-scale problem

Figure 2.4 Information transfer between the macro and micro problems

Fine scale problem

Coarse scale problem
0
( , )
( , )
i
ik ik
k
u t
F t
X

= +
X
X
( ) ( )
1

( , ) , ,
Y
ij ij Y
Y
P t P t d

X F X Y

18
The two-scale problem derived in the previous sections is two-way coupled. By
introducing simplified notation
( ) ( ) ( ) ( ) ( )
0 0 0

, , , , , ,
ij ij ij
P t P t P t = = F X Y F X Y X Y , the
link between the two scales is schematically shown in Figure 2.4. The fine scale problem
is driven by the overall (coarse scale) deformation gradient
( , )
ik
F t X . It informs the
coarse scale problem about the overall stress
( , )
ij
P t X . The coupled two-scale problem is
summarized below:
(a) Fine scale problem:
( ) ( )
( )
1
1

( , ), ( , , ) :
, ,
0
0
ik i Y
ij
Y
j
Y
Given F t find u t on such that
P t
on
Y
on
=
X X Y
F X Y
g u
(2.44)
(b) Coarse scale problem:
0
2 0
2
0
0
0 0 0
, ( , )
0; 0
ij i X
ij
i
Xi X X
j
t u
ij Xj Xi X i i X
i
i i i
Given P find u t on such that
P
u
b on
X t
P N t on u u on
u
u u at t v at t
t

+ =

= =
= = = =
X
(2.45)

2.2.4 Finite element discretization
Both the coarse and fine scale problems are solved using finite element methods.
The displacement of the fine scale problem
1
( , , )
i
u t X Y is approximated as
1 1 1

( , , ) ( ) ( , )
i B iB
u t N d t = X Y Y X (2.46)
where subscript B denotes the node number;
1
( )
B
N Y are the unit cell shape functions and
1
iB
d are the nodal displacements.

19
Let
( , )
M
jC
d t X be the master (independent) degrees and express
1
iB
d by a linear
combination of
( , )
M
jC
d t X denoted as
1

( , ) ( , ) ( , )
M
iB iBkC kC
d t T t d t = X X X so that the
constraint equation (2.43) in the discrete form is satisfied
( )
1
( ) ( , ) 0
M
B iBjC jC Y
N T d t on = g Y X (2.47)
Then writing the Galerkin weak form of (2.44) and discretizing it using (2.46) yields the
discrete residual equation
1
1 1 1 1
1 1
( ) 0
Y
m m B
kC n iBkC n ji Y
j
N
r T P d
Y
+ +
+ +
d (2.48)
where the left subscript and superscript denote the load increment and the iteration count
(for implicit method) at the coarse scale, respectively;
1 1
1
m
n iB
r
+
+
and
1 1
1
m
n iB
d
+
+
are the residual
and displacement increment in the
th
( 1) m+ iteration of the
th
( 1) n + load increment,
respectively. If the constitutive equations are defined in terms of the Cauchy stress, it is
convenient to restate the unit cell problem as follows:
1 1 1
1 1
1
1 1 1 1
1 1
1
Given and , find such that:
( ) 0

( , ) ( , ) ( , )
y
m m
n ij n ij n iB
m m B
kC n iBkC n ji y
j
M
iB iBkC kC
F d
N
r T d
y
d t T t d t
+ +
+ +
+ +
+ +
d
X X X
(2.49)
where we have exploited the relation between the first Piola-Kirchhoff stress
ij
P and the
Cauchy stress
ji

ji jk ik
J F P = (2.50)
and J in (2.50) is the determinant of
jk
F .
Similarly, the coarse-scale displacements
0
( , )
i
u t X is discretized as
0 0 0

( , ) ( ) ( )
i A iA
u t N d t = X X (2.51)
where
0
( )
A
N X and
0
iA
d are the coarse scale shape functions and nodal displacements,
respectively. Writing the weak form of (2.45) and using discretization (2.51) the discrete
coarse scale equations can be written as

20
0
1 1 1
0 0 0
1 1 1 1 1
0
1 1
Given and , find such that:
( ) 0

1, Go to the next load increment
n i n i n iA
int ext
n iA n ijAB n jB n iA n iA
u
n iA n iA X
b t d
r M d f f
d u on
n n
+ + +
+ + + + +
+ +
+ =
=
+
``
d
(2.52)
where
0
1 n iA
r
+
and
0
1 n iA
d
+
are the coarse scale residual and displacement increments in the
th
( 1) n + load increment, respectively, and
0 0
X
ijAB ij X A B X
M N N d
(2.53)
0
X
int A
iA ji X
j
N
f P d
X
(2.54)
0 0
t
X X
ext
iA A Xi X A Xi X
f N b d N t d

= +

(2.55)
where
ijAB
M is the mass,
int
iA
f and
ext
iA
f are the internal and external forces, respectively.
It is again convenient to express the coarse scale internal and external forces in the
deformed configuration as
0
x
int A
iA ji x
j
N
f d
x

(2.56)
0 0
t
x x
ext
iA A xi x A xi x
f N b d N t d

= +

(2.57)
where
/ , / , .
x X xi Xi xi x Xi X
J b b J t d t d = = = (2.58)
/
ji jk ik
F P J = (2.59)
and J is the determinant of
ij
F .
We now focus on deriving a closed form expression for the overall Cauchy stress
ji
. Substituting (2.50) into (2.33) and recalling
y Y
d Jd = , we have

1
1
( , )
y
ik km mi y
Y
P t F d

X (2.60)
Inserting (2.60) into (2.59) and denoting the volume of the macro-deformed
configuration as
*
y Y
J = , the overall Cauchy stress can be expressed as

21
1
*
1
y
ji jm mi y
y
F d

(2.61)
where
1 1
jm jk km
F F F

= maps the micro-deformed configuration
y
into the macro-
deformed configuration
*
y
as illustrated in Figure 2.5.
The macro problem may be solved using either explicit or implicit time integration
[31]. For the implicit method, consistent tangent stiffness at every material point can be
computed numerically (see [29, 31, 32] for details).

Figure 2.5 Unit cell configurations: (a) initial; (b) macro-deformed (intermediate);
and (c) micro-deformed (final)

Y
1

Y
2

(a)

1
C
3
C 4
C
2
C
1
C
1
y
2
C
1
C
3
C 4
C
*
1
y
*
2
y
(b)
2
C
3
C
4
C
2
y
(c)
F
F
F
Y
*
y

y

22
2.3 Numerical Examples
To verify the proposed formulation we consider three test problems: a perforated
plate, a plate with a centered hole and a delamination of a sandwich shell subjected to
impact loading.

2.3.1 Perforated plate
The geometric configuration of the perforated plate is shown in Figure 2.6(a). The
width W and the length L of the plate are 60.0mm. Circular holes of radius 1mm are
uniformly distributed in a rectangular arrangement. The right side of the plate is fixed;
the top and bottom side are free, and a constant velocity V = 6m/s is applied at the left
side for the duration of 3 mille seconds. Plane strain condition is assumed. The material
is assumed to be hyperelastic. The strain energy is expressed in neo-Hookean form with
initial shear modulus of 160 MPa and bulk modulus equal to 4000MPa. The mass
density of the material is
3 3
1.14 10 / kg m . For multiscale analysis, the macro scale
mesh contains 25 four-node reduced integration quadrilateral elements; all elements are
12mm 12mm. The size of the unit cell is 4 4 mm as shown in Figure 2.6(b). The
reference solution is obtained by direct finite element discretization of the plate with a
sufficient number of elements.

23

Figure 2.6 Geometry, boundary conditions and unit cell discretization: (a) macro-
geometry and boundary conditions and (b) mesh of the unit cell

The reaction force histories obtained with the multi-scale method and the reference
solution in X and Y directions are shown in Figure 2.7. It can be seen that the multiscale
method is in good agreement with the reference solution. Figure 2.8(a) depicts the von
Mises stress in the region around the center hole at t = 3ms as obtained with the
multiscale solution. Figure 2.8(b) compares the multiscale method with a reference
solution in the same subdomain and time instance. Figure 2.8 also shows the lengths
between node A and B, C and D after deformation. In this example, the time step for the
multiscale analysis is 30 times larger than for the reference solution.
V
X
Y
(a)
(b)

24
0
100
200
300
400
500
0 0.0005 0.001 0.0015 0.002 0.0025 0.003
Time(Second)
R
e
a
c
t
i
o
n

f
o
r
c
e
(
N
)
Reference
Multiscale

-1600
-1200
-800
-400
0
0 0.0005 0.001 0.0015 0.002 0.0025 0.003
Time(Second)
R
e
a
c
t
i
o
n

f
o
r
c
e
(
N
)
Reference
Multiscale

Figure 2.7 Reaction force in x (top) and y (bottom) directions

25

Figure 2.8 Unit cell deformation and von Mises stress obtained from the (a)
multiscale and (b) reference solutions

2.3.2 Plate with a center hole

Figure 2.9 Plate with a hole in the center

Geometry and boundary conditions of the plate with a center hole are schematically
shown in Figure 2.9. The width W and the length L of the plate are 100.0mm. A circular
hole of radius r=1.0mm is placed at the center. Plane strain condition is assumed and the
material properties are the same as in the first example. A constant velocity V = 50m/s is
applied at the left side for a period of 0.3 ms. The right side of the plate is constrained in
V
(b) (a)
AB=4.19mm
CD=3.99mm
AB=4.20mm
CD=3.99mm
A
A
B B
C C
D
D

26
the horizontal direction. For the multiscale analysis, the size of the unit cell is
4 4 mm and the corresponding coarse mesh element is located at the center of the plate
as shown by the dashed lines in Figure 2.9. The unit cell mesh is shown in Figure 2.6(b).
Figure 2.10 depicts the deformed configuration and von Mises stress in the unit cell at
0.3ms as obtained by different methods: (a) reference solution by direct finite element
method, (b) multiscale solution with essential boundary conditions (2.40), (c) multiscale
solution with natural boundary conditions (2.39), (d) multiscale solution with mixed
boundary conditions (2.42). The lengths between node A and B, C and D after
deformation are also compared in Figure 2.10.
For the mixed boundary condition, the linear spring stiffness (see Figure 2.3(c)) has
been a priori determined as follows. At every node (one at a time) in the neighboring
subdomain domain to the unit cell (without a hole) we prescribe a small displacement in
the direction normal to the boundary and calculate the reaction force at that node.
Assuming linear spring stiffness the reaction divided by the prescribed displacement is
equal to the spring stiffness at that node. Similarly, a nonlinear spring stiffness (not
considered in this article) can be a priori defined.

Figure 2.10 Comparison of the unit cell deformation and von Mises stress at 0.3ms
as obtained with (a) reference solution by direct finite element method; (b)
multiscale solution with essential boundary conditions; (c) multiscale solution with
natural boundary conditions; (d) multiscale solution with mixed boundary
conditions
(a) (b) (d)
A B A
A A B B B
C
C C C
D
(c)
D D D
AB=3.25mm
CD=4.63mm
AB=2.99mm
CD=4.77mm
AB=2.66mm
CD=5.16mm
AB=3.03mm
CD=4.72mm

27
It can be seen that the essential BC is not equipped with the necessary kinematics
to provide the correct deformation on the unit cell boundary. The natural boundary
condition has the right kinematics but is too flexible. Finally, the mixed boundary
condition seems to be able to capture the overall deformation correctly. The periodic
boundary condition was not considered in this case since the deformation (see Figure
2.10(a)) is clearly not periodic.

2.3.3 Sandwich shell

Figure 2.11 Geometry, layup and impact loading of the sandwich shell

(a)
Polyurea
DH36
RVE
C
V=275m/s
DH36 steel plate
Thickness=5.0mm
Epoxy interface
Polyurea plate
Thickness=5.8mm

28

(b)
Figure 2.12 Initial configuration of the (a) undeformed coarse-scale (macro)
and (b) undeformed unit cell models

The geometry, layup and impact loading conditions are shown in Figure 2.11. The
impactor is modeled as a rigid body with initial velocity of 275m/s and mass equal to
145gm. The circular plate has a diameter of 154mm. It consists of two layers, DH36
steel and polyurea, glued by epoxy. The plate is discretized with 8-node reduced
integration and hourglass control [33, 34] brick elements. The DH-36 structural steel is
modeled as a thermoplastic material with von Mises yield surface proposed by Nemat-
Nasser and Guo [35]. Since the impact process involves large plastic deformation in less
than 1 ms, temperature increases due to plastic deformation. Adiabatic and isothermal
conditions are assumed for the DH36 steel and polyurea, respectively. The polyurea is
modeled using the pressure- and temperature-dependent hyperelastic-viscoelastic
constitutive law [ 36] where hyperelastic effects are modeled using Mooney-Rivlin
model, whereas viscous effects are represented by Prony series. The epoxy is modeled
using cohesive elements proposed by Liechti and Wu [37].
Figure 2.12 displays the geometrical configuration for the multiscale analysis. The
interface and a thin layer of steel and polyurea above and below it define a unit cell (or
representative volume element) as shown in Figure 2.12. The residual velocity of the
impactor obtained with the multi-scale method and direct finite element simulation
(reference solution) is compared in Figure 2.13. The von Mises stress history at point C
(in Figure 2.12(a)) is shown in Figure 2.14 for both the reference and multiscale
solutions. Figure 2.15 compares the overall deformation of the sandwich plate at t =15ms
as obtained with the reference and multiscale solutions. Figure 2.16 compares the unit
cell deformation in the two locations around the interface marked in Figure 2.15(b) as
obtained with the reference and multiscale solutions. The location of points A and B
DH36, thickness=0.2mm
Polyurea, thickness=0.18mm
Epoxy, thickness=0.02mm
A
B

29
prior to the deformation is shown in Figure 2.12(b); the position of the two point at
t =15ms is depicted in Figure 2.16 at the two macroscopic positions denoted as region a
and region b in Figure 2.15(b).

0
50
100
150
200
250
300
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
Time(ms)
R
e
s
i
d
u
a
l

v
e
l
o
c
i
t
y
(
m
/
s
)
Reference
Multiscale

Figure 2.13 Residual velocity of the impactor

0
100
200
300
400
500
600
700
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
Time(ms)
M
i
s
e
s

S
t
r
e
s
s
(
M
P
a
)
Reference
Multiscale

Figure 2.14 von Mises stress at point C (in Figure 2.12(a))

30

(a)

(b)
Figure 2.15 Deformation of the sandwich plate obtained with the (a) reference and
(b) multiscale solutions at t=15ms

(a)

(b)
Figure 2.16 Local deformation in (a) region a and (b) region b (see Figure 2.15(b))
at t=15ms as obtained with reference (left) and multiscale (right) solutions
Reference solution
Multiscale solution
1.11mm
1.92mm
A
B
1.76mm
0.98mm
A
B
Reference solution
Multiscale solution
2.65mm
2.81mm
2.59mm
2.48mm
A
B
B
Debonding
Region a
Region b
A

31

References

1 I. Babuska, Homogenization and Application, Mathematical and Computational
Problems, In Numerical Solution of Partial Differential Equations III, SYNSPADE, ed.
B Hubbard, Academic Press, 1975.
2 A. Benssousan, J.L. Lions and G. Papanicoulau, Asymptotic Analysis for Periodic
Structures, North-Holland, 1978.
3 E. Sanchez-Palencia, Non-homogeneous Media and Vibration Theory, Lecture notes in
physics, vol. 127, Springer-Verlag, Berlin, 1980.
4 N. Bakhvalov and G. Panasenko, Homogenization: Averaging Processes in Periodic
Media, Kluwer Academic Publishers, 1989.
5 R. Hill, A Theory of the Yielding and Plastic Flow of Anisotropic Metals, Proceedings
of the Royal Society of London. Series A, vol.193, pp. 281-297, 1948.
6 R. Hill, Elastic Properties of Reinforced Solids: Some Theoretical Principles, Journal
of the Mechanics and Physics of Solids, vol.11, pp. 357372, 1963.
7 K. Terada, and N. Kikuchi, Nonlinear Homogenization Method for Practical
Applications, Computational Methods in Micromechanics, S. Ghosh and M. Ostoja-
Starzewski (eds.), ASME, New York, vol. AMD-212/MD-62, pp. 1-16, 1995.
8 K. Terada, and N. Kikuchi, A Class of General Algorithms for Multi-scale Analysis of
Heterogeneous Media, Computer Methods in Applied Mechanics and Engineering,
vol.190, pp.5427-5464, 2001.
9 K. Matsui, K. Terada and K. Yuge, Two-scale Finite Element Analysis of
Heterogeneous Solids with Periodic Microstructures, Computers and Structures, vol. 82 ,
pp. 593-606, 2004.
10 R. J. M. Smit, W. A. M. Brekelmans, and H. E. H. Meijer, Prediction of the
Mechanical Behavior of Nonlinear Heterogeneous Systems by Multi-level Finite
Element Modeling, Computer Methods in Applied Mechanics and Engineering, vol. 155,
pp. 181-192, 1998.

32

11 C. Miehe, and A. Koch, Computational Micro-to-macro Transition of Discretized
Microstructures Undergoing Small Strain, Archives of Applied Mechanics, vol. 72, pp.
300-317, 2002.
12 V. Kouznetsova, W.A.M. Brekelmans, and F.P.T. Baaijens, An Approach to Micro-
macro Modeling of Heterogeneous Materials, Computational Mechanics, vol. 27, pp. 37-
48, 2001.
13 F. Feyel, and J.-L. Chaboche, FE2 Multiscale Approach for Modeling the Elastovis-
coplastic Behavior of Long Fiber SiC/Ti Composite Materials, Computer Methods in
Applied Mechanics and Engineering, vol.183, pp. 309-330, 2000.
14 S. Ghosh, K. Lee, and S. Moorthy, Multiple Scale Analysis of Heterogeneous Elastic
International Journal of Solids and Structures, vol. 32, pp. 27-62, 1995.
15 S. Ghosh, K. Lee, and S. Moorthy, Two Scale Analysis of Heterogeneous
Elasticplastic Materials with Asymptotic Homogenization and Voronoi Cell Finite
Element Model, Computer Methods in Applied Mechanics and Engineering, vol. 132,
pp.63-116, 1996.
16 J.C. Michel, H. Moulinec, and P. Suquet, Effective Properties of Composite Materials
with Periodic Microstructure: a Computational Approach, Computer Methods in Applied
Mechanics and Engineering, vol.172, pp. 109-143, 1999.
17 M.G.D. Geers, V. Kouznetsova and W.A.M. Brekelmans, Gradient-enhanced
Computational Homogenization for the Micro-macro Scale Transition, Journal of
Physics IV, vol. 11, pp. 145-152, 2001.
18 C. McVeigh, F. Vernerey, W. K. Liu, and L. C. Brinson, Multiresolution Analysis for
Material Design, Computer Methods in Applied Mechanics and Engineering, vol. 195,
pp. 50535076, 2006.
19 C. Miehe, Strain-driven Homogenization of Inelastic Microstructures and Composites
Based on an Incremental Variational Formulation, International Journal for Numerical
Methods in Engineering, vol. 55, pp. 12851322, 2002.
20 T.I. Zohdi, J.T. Oden, and G.J. Rodin, Hierarchical modeling of heterogeneous bodies,
Computer Methods in Applied Mechanics and Engineering, vol. 138, pp. 273298, 1996.

33

21 Weinan E, B. Engquist, and Zhongyi Huang, Heterogeneous Multiscale Method: A
General Methodology for Multiscale Modeling, Physical Review B, vol. 67, pp. 1-4,
2003.
22 T.Y. Hou, and X.-H. Wu, A multiscale Finite Element Method for Elliptic Problems
in Composite Materials and Porous Media, Journal of Computational Physics, vol. 134,
pp. 169-189, 1997.
23 S. Loehnert, and T. Belytschko, A Multiscale Projection Method for
Macro/Microcrack Simulations, International Journal for Numerical Methods in
Engineering, vol. 71, pp. 14661482, 2007.
24 T. Belytschko, S. Loehnert, and J. Song, Multiscale Aggregating Discontinuities: A
Method for Circumventing Loss of Material Stability, International Journal for
Numerical Methods in Engineering, vol73, pp. 869-894, 2008.
25 J. Fish, K. Shek, M. Pandheeradi, and M.S. Shephard, Computational Plasticity for
Composite Structures Based on Mathematical Homogenization: Theory and Practice,
Computer Methods in Applied Mechanics and Engineering, vol. 148, pp. 53-73, 1997.
26 J. Fish and K. L. Shek, Finite Deformation Plasticity of Composite Structures:
Computational Models and Adaptive Strategies, Computer Methods in Applied
Mechanics and Engineering, vol. 172, pp. 145-174, 1999.
27 J. Fish and Q. Yu, Multiscale Damage Modeling for Composite Materials: Theory
and Computational Framework, International Journal for Numerical Methods in
28 J. D. Clayton and P. W. Chung, An Atomistic-to-Continuum Framework for
Nonlinear Crystal Mechanics Based on Asymptotic Homogenization, Journal of the
Mechanics and Physics of Solids, vol.54, pp. 16041639, 2006.
29 Z. Yuan and J. Fish, Towards Realization of Computational Homogenization in
Practice, International Journal for Numerical Methods in Engineering, vol. 73, pp. 361-
380, 2007.
30 S. D. Mesarovic and J. Padbidri, Minimal Kinematic Boundary Conditions for
Simulations of Disordered Microstructures, Philosophical Magazine, Vol. 85, No. 1, pp.
65-78, 2005.

34

31 T. Belytschko, W.K. Liu, and B. Moran, Nonlinear Finite Elements for Continua and
Structures, J. Wiley & Sons, New York, 2000.
32 C. Miehe, Computational Micro-to-Macro Transitions for Discretized Micro-
structures of Heterogeneous Materials at Finite Strains Based on the Minimization of
Averaged Incremental Energy, Computer Methods in Applied Mechanics and
33 T. Belytschko and L. P. Bindeman, Assumed Strain Stabilization of the Eight Node
Hexahedral Element, Computer Methods in Applied Mechanics and Engineering, vol.
105, pp.225260, 1993.
34 T. Belytschko, J. I. Lin, and T. Chen-Shyh, Explicit Algorithms for the Nonlinear
Dynamics of Shells, Computer Methods in Applied Mechanics and Engineering, vol. 42,
pp. 225251, 1984.
35 S. Nemat-Nasser and W.G. Guo, Thermomechanical Response of DH-36 Structural
Steel over a Wide Range of Strain Rates and Temperatures, Mechanics of Materials, vol.
35, pp. 1023-1047, 2003.
36 S. Nemat-Nasser, Experimental Characterization of Polyurea with Constitutive
Modeling and Simulations, Presentation at ERC ACTD Workshop, Massachusetts
Institute of Technology, 2004.
37 K. M. Liechti, and J.-D. Wu, Mixed-mode, Time-dependent Rubber/Metal
Debonding, Journal of the Mechanics and Physics of Solid, vol. 49, pp. 1039-1072, 2001.

35
3. Adaptive Two-Scale Nonlinear Homogenization

3.1 Introduction
Hierarchical multiscale models are a sequence of mathematical models, which
include increasingly more sophisticated effects, represented by finer scales. The most-
comprehensive member of the sequence depends on the scale where the governing
equations are well understood and the constitutive equations can be effectively
characterized. The modeling error associated with any other member of the sequence is
assessed by comparing it to the most-comprehensive member of the hierarchical
sequence. Loosely speaking, a member of the sequence is considered to be admissibly
accurate if the modeling error in the data of interest is sufficiently small. The goal is to
identify an optimal member of the sequence that is admissibly accurate and most
efficient to solve.
A deterministic error estimate of a particular model in a hierarchy can be assessed
using some physical criterion, such as a magnitude of energy, stress, dissipation energy
[1], etc., or by estimating the validity of the mathematical formulation (for example,
errors introduced by multiscale asymptotic methods can be estimated by comparing the
magnitude of the terms neglected in the asymptotic sequence to those taken into account
[2]), or by quantifying modeling errors in so-called quantities of interest [3]. The error
due to uncertainty can be estimated by the statistical sampling of data and numerical
experiments [4].
Adaptive concurrent modeling guided by error estimators is one of the most
commonly used hierarchical strategies. By this approach various models in the
hierarchical sequence are mapped onto the appropriate space-time window. For example,
in [1] the problem domain was subdivided into regions where phenomenological model,
homogenization model and direct numerical simulation with element size comparable to
the size of heterogeneity were used. Adaptive strategies involving two or more
hierarchical models have been used in [5,6,7,8]. Methods of this type have been very
popular in concurrent linking of discrete and continuum scales [9,10,11].

36
This chapter focuses on a hierarchical sequence consisting of two members: the
reduced order homogenization (ROH) model [ 12 ,26] and a detailed or direct
homogenization (DH) model [13,14,15,16,17,18,19,20,21,22,23,24,25]. The two-scale
direct homogenization methods are presented in Chapter 2. The reduced order
homogenization methods are reviewed in Sections 3.2, respectively. Consistent
interscale transfer operators between the two homogenization models consisting of
prolongation or restriction operators as well as a relaxation operator aimed at
equilibrating unit cell equations are developed in Section 3.3. Localization limiters
aimed at alleviating mesh sensitivity arising due to damage are discussed in Section 3.4.
Verification and validation studies are presented in Section 3.5.

3.2 Eigendeformation-based reduced order homogenization
Here we consider small deformation in the material coordinate system (i.e. small
strains, but large rotations). , x y will denote the material coordinate in macroscopic
domain and unit cell domain , respectively. We specifically consider constitutive
equation and kinematic equations in the form
( ) ( ) ( ) ( )
I
ij ijkl kl kl
I
L

=

x x x x x (3.1)
( )
( )
( ) ( )
, , ,
1
2
ij i j j i i j
u u u

= + x x x (3.2)
The constitutive relation described in Equation (3.1) assumes an additive decomposition
of total strains
ij
into elastic and inelastic components (or more generally stated as

eigenstrains
I
kl
); the left superscript I indicates different types of eigenstrains, such as

inelastic deformation (damage, plasticity, etc.), thermal changes, moisture effects and
phase transformation.
Using the constitutive (3.1) and kinematic (3.2) equations, the equilibrium equation
at the fine scale can be expressed as

37
( ) ( )
( )
( ) ( )
1
,
,
, , 0
l
j
I
ijkl kl kl k y
I
y
L u

+ =

y x x y x y (3.3)
where ( )
kl
x is an overall strain field. Following [12, 26] the displacement field is
decomposed using the elastic influence function
ikl
H , the eigenstrain influence function
ikl
h
and the eigenseparation influence function

in
h
as

( ) ( ) ( ) ( )
( ) ( )
( ) ( )
( ) ( )
1 1e 1 1

, , , ,
, ,
, ,
I
i i i i
I
ikl kl
I
ikl kl
I
in n
S
u u u u
H
h d
h dS

= + +
=
+
+
x y x y x y x y
y x
y y x y
y y x y
(3.4)
where the influence (or Greens) functions
ikl
H ,
ikl
h

in
h
are computed by solving a

sequence of elastic boundary value problems independent of and prior to nonlinear
macro analysis. These functions are chosen to satisfy the equilibrium equation (3.3) for
arbitrary ( )
kl
x , ( )
,
I
kl
x y and ( )
,
n
x y . The subscript n in the eigenseparation
influence function denotes the component in the local Cartesian coordinate system of the
interface.
To reduce the computational complexity of solving the unit cell problem, the
eigenstrains are discretized in terms of piecewise constant shape function
( )
( ) N

y as
( )
( )
( )
( )
( )
1
,
n
I I
ij ij
N

=
=
x y y x (3.5)
where

( )
( )
( )
( )
1
0
N

y
y
y
(3.6)

( )
( )
( )
( )
( )
1
,
I I
ij ij
d

x x y (3.7)

38
in which the total volume of the unit cell is partitioned into n nonoverlapping
subdomains denoted by
( )
. Partitions at various scales are denoted by a superscript
enclosed in parenthesis: ( ) ( ) , denote the phase (volume) partitions; ( ) ( ) , denote
the interface partitions.
The eigenseparation ( )
,
n
x y is discretized in terms of a C
0
continuous interface
partition shape function
( )
( ) N

y as
( )
( )
( )
( )
( )

1
,
m
n n
N

=
=
x y y x (3.8)
where
( )
( ) N

y is a linear combination of piecewise linear finite element shape functions
defined over partition

( )
( )
( )
( )
( )
( )
0
a
a S
N S
N
S
y y
y
y

(3.9)

( )
( )
( )
( )
( )
1
,
n n
S
dS
S

=

x x y (3.10)
in which the total interface in the unit cell is divided into m partitions denoted by
( )
S

,
and ( )
a
N y is a linear shape function associated with a finite element mesh node a
along the interface.
Equations (3.3)-(3.8) together with the constitutive equations for the eigenstrains
and eigenseparations form a reduced system of equations from which
( )
( )
I
ij
x and
( )
( )
n
x are computed at each Gauss point. For more details we refer to [12,26]. In the
present manuscript perfect interfaces are considered in the reduced order model.

39
3.3 Adaptive homogenization model selection
In this section, we outline the selection criteria and information transfer of state
variables, between the direct and reduced order homogenization.
Qualitatively, in the regions of high deformation gradients, a more accurate direct
homogenization method is employed, while elsewhere reduced order homogenization
suffices. We will employ a simple switching criterion between the two homogenization
models proposed in [27]. By this approach, given the macroscopic Lagrangian strain
tensor ( ) / 2
ij ki kj ij
F F = , the second invariant of the Lagrangian strain tensor may be
written as:
2 2 2
12 13 23 11 22 22 33 11 33
( ) = + + + +
When exceeds some critical value,
cr
> , the direct homogenization is triggered
for the unit cell corresponding to the macroscopic material point. If on the other hand,
cr
, the reduced order homogenization is employed.
Once the switch between direct homogenization and reduced order homogenization
has been triggered, it is necessary to transfer the state variables, such as damage
parameter and plastic strain, between the two models.
We first consider the information transfer from the reduced order unit cell model to
the detailed unit cell model. Such an information transfer between models has been
exercised in a variety of other disciplines and applications. An operation, known as
prolongation, is the corner stone of multigrid or multilevel methods. A similar idea has
been advocated by Kevrekidis [28] in the so-called Equation Free method where the so-
called lifting operator is employed to lift the information from the coarse scale to the fine
scale. Finally, in the field of mesh generation (see for instance [29]) there is a need to
transfer information between the finite element meshes due to mesh refinement.
Hereafter, we will refer to such coarse-to-fine transfer operation as interpolation.
Nevertheless, interpolation alone does not suffice, since the governing equations on
the fine scale will not necessarily be satisfied after the information transfer. Thus after
interpolation it is necessary to carry out a so-called relaxation that allows for the fine
scale equation residuals to vanish or to be sufficiently reduced. Similar local
equilibration is carried out in other fields: smoothing or relaxation in multilevel methods;

40
relaxation in the equation free method, in the block-step method [30, 31] and in the
quasicontinuum method for complex lattices [32]. In [29], a Hu-Washizu principle was
used for equilibration.
We now focus on the formulation of the information transfer operator between the
reduced order and direct homogenization models. Note that the finite element meshes of
the unit cell models in the two methods are identical. Let
( )
ij
be the constant
eigenstrain in the subdomain
( )
of the reduced order unit cell. Then the
eigenstrain ( )
ij
Y in the detailed unit cell model is defined as a piecewise constant
function
( ) ( )
( )
ij ij

= Y Y
Similarly, the eigenseparations,
( )
( )
n
x , which are described as

0
C functions in the
reduced order model are interpolated using
0
C shape functions (3.9).
In the relaxation process, assume that (i) the macroscopic deformation gradient
( )
ik
F t is invariant due to information transfer and (ii) ( )
ij
Y and the corresponding
eigenseparation constitute the initial values prior to the unit cell deformation. Then we
subject the detailed unit cell model to the macroscopic deformation gradient ( )
ik
F t and
find its equilibrium using nonlinear finite element analysis. Note that this equilibration
process is local in nature, thus the term relaxation, since the macroscopic deformation
( )
ik
F t is assumed to be unchanged. The global equilibration takes place at the next time
step of the macro problem.
Transfer of information from the detailed to the reduced order unit cell is termed
restriction following the terminology used in the multilevel methods. Since the reduced
order eigenstrains and eigenseparations are constant over partition, they are obtained by
averaging the corresponding fine scale fields using equations (3.7) and (3.10),
respectively.

41
3.4 Energy-based element erosion criterion
In the present manuscript we employ continuum damage mechanics constitutive
behavior for microphases and interface (see for instance [ 33, 34]) with an effective
stress-strain curve depicted in Figure 3.1. Strain softening, as depicted in Figure 3.1,
introduces mesh dependence [35,36]. Here we will employ a variant of the localization
limiter proposed in Chapter 4 in the context of linear elasticity.

Figure 3.1 Effective stress-strain curve

We focus on a two-phase material and initially assume that damage initiates in the
matrix phase. We start from the classical definition of fracture toughness
m
G in the
matrix material as the work required to close the crack as shown in Figure 3.2(a).
In classical fracture mechanics the fracture toughness is defined as

0
0
1 1
lim
2
a
m y y
a
G u dx
a

=

.
.
.
(3.11)
where
y
is stress normal to the crack surface and
y
u is the closure displacement. We
now consider an approximation to the fracture toughness
m
G
obtained by element
erosion. For simplicity, we assume the state of constant element stress evaluated at the
element centroid (one-point integration with stabilization [ 37]). The eroded element
elongation perpendicular to the crack surface is given by
y y m
u h = where
y
is the
strain component normal to the crack surface as shown in Figure 3.2(b). If we further
assume that crack propagates normal to the direction of the maximum principal stress in
tension
1
then Eq. (3.11) reduces to
*
m

42

1 1
1
2
m m
G h
(3.12)
where
1
is the normal strain computed in the principal directions of stress. Eq. (3.12)
can be further simplified for linear isotropic materials and by neglecting the remaining
principal stress components, which gives

2
m
m
m
G E
h
=

(3.13)

Figure 3.2 Crack closure tractions and displacements in: (a) the analytical model,
(b) the direct finite element mesh simulated by element erosion, and (c) the macro-
mesh of the two-scale problem

profile before closure
profile after
closure y

a .
y
u
x
y
y

h
m
(1 )
m y
h +
eroded matrix element
(a)
(b)
x
y
*
m
*
h
* *
(1 )
m
h +
eroded matrix
fiber
*
a .
(c)

43
Consider a two-scale problem with crack tractions and displacements in macro-
elements illustrated in Figure 3.2(c). Similarly to Equation (3.11), the fracture toughness
due to cracking in the matrix material can be written as:

*
* * *
*
0
0
1 1
lim
2
a
m m m
a
G u dx
a

=

.
.
.
(3.14)
where subscript m denotes matrix material. The matrix element elongation perpendicular
to the crack surface is approximately given by
* *
m m
u h = . Further assuming that the
crack propagates normal to the direction of the maximum principal stress in tension
*
1
,
Equation (3.14) reduces to
* * * * * *
1 1 1 1 *
1 1
2 2
m
m m
l
G h h
a
=
.
(3.15)
where
m
l is the crack length in the matrix-material,
m
the matrix volume fraction and
*
1
is the normal strain in the principal directions. Following the assumption in Equation
(3.13) and equating (3.12) and (3.15), yields the critical stress in matrix for the two-scale
scale analysis:

*
*
m
m m
m
h
h

= (3.16)
Note that for a homogeneous material 1
m
= and Equation (3.16) reduces to the
criterion proposed in Chapter 4. Also, since the macro-element size is typically larger
than the unit cell size, yields
*
m m
< .
If we assume that following matrix failure, the fiber material may fail next and
repeat the above derivation, we will arrive at a similar scaling law for the fiber phase

*
*
f
f f
f
h
h

= (3.17)
where
f
and
f
h are the critical stress and characteristic element size for direct
numerical simulation and
f
is the fiber volume fraction.

44
To verify the proposed formulation we consider three test problems: a composites
plate with a center crack, delamination of a sandwich shell subjected to impact loading
and crush of a composites circular tube.

3.5.1 Composites plate with a center crack
The geometric configuration of the composites plate with a center crack is shown in
Figure 3.3(a). The width W and the length L of the plate are 80.0mm, and the thickness is
1.0mm. A crack of length 2a=32.0mm is placed at the center. The plate is subjected to
uniform tension load applied at the top and bottom edge of the plate. The load is applied
at constant velocity of 0.003MPa/s. With the consideration of symmetry, one quarter of
the plate is modeled in the simulation. The micro structure of the composite consists of
two materials: matrix and fiber. The fiber is oriented in the loading direction. The size of
the unit cell is 1 1 1 mm and the mesh is shown in Figure 3.3(b). The matrix is modeled
as elastic (Youngs modulus 1.0MPa and Poisson ratio 0.2) with isotropic damage (see
Figure 3.1). The matrix failure strain
m
is 0.04 and the matrix critical stresses for direct
numerical simulation and multiscale simulation,
m
and
*
m
, are 0.01MPa and
0.0074MPa, respectively. The reference solution is obtained by finite element
discretization of the plate with a sufficient number of elements.
We will consider two cases for the fiber strength. In the first case, the fiber does not
break; it is assumed to obey isotropic elasticity with Youngs modulus 10.0MPa and
Poisson ratio 0.2. In the second case, the fiber similarly to the matrix phase is modeled
as elastic with isotropic damage. The fiber failure stain
f
is 0.0267 and the fiber
critical stresses for direct numerical simulation and multiscale simulation,
f
and
*
f
,
are 0.03MPa and 0.022MPa, respectively. The two cases are expected to yield
substantially different behavior. When fibers do not break, the crack is expected
propagate along the fiber (mode II), whereas in the case of weak fibers (case II), the
crack is expected to propagate in a self similar fashion following the initial crack
orientation (mode I). In both cases, crack propagation is simulated using element erosion.

45
We start with the first case of perfect fibers without damage. Figure 3.4(a), (b) and
(c) depict the displacement contours at t=1.6s obtained with the direct numerical
simulation, using adaptive multiscale simulations with the energy based element erosion
criterion (Eq. (3.16)) and without it, respectively. For the adaptive multiscale solution
the crack front in the coarse scale mesh is identified by tracking the failed matrix
elements in the unit cell. The displacement history of the center of the plate (point O in
Figure 3.3(a)) is given in Figure 3.5. The importance of scaling (Eq. (3.16)) is quite
evident. It can be seen that that the element erosion based on the local critical stress or
strain criterion suffers from mesh size dependency. Using the energy release rate based
failure criterion presented in Section 3.4, the mesh dependency is considerably alleviated.
We now consider the second case of relatively weak fibers. Figure 3.6(a), (b) and (c)
depict the displacement contours at t=1.04s as obtained with the direct numerical
simulation, and the adaptive multiscale simulations with the energy based element
erosion criterion (Eq. (3.16) and (3.17)) and without it, respectively. The displacement
history of the center of the plate is given in Figure 3.7. Again, it can be seen that the
mesh dependency is considerably alleviated when the energy release rate based failure
criterion for both fiber and matrix is used.

Figure 3.3 (a) plate with a center crack and (b) mesh of the unit cell
X
Z
(a)
(b)
Fiber
Matrix
Crack
O

46

Figure 3.4 Comparison of displacement contour in z direction at t=1.6s (Mode II):
(a) reference solution; (b) energy based solution; and (c) stress based solution

0
0.05
0.1
0.15
0.2
0.25
0.3
0 0.5 1 1.5 2 2.5
Time(s)
D
i
s
p
l
a
c
e
m
e
n
t
(
m
m
)
Reference
Stress based
Energy based

Figure 3.5 The center point O displacement in the loading (z) direction (Mode II)

(a)
(b)
(c)
Crack
Crack

47

Figure 3.6 Comparison of displacement contour in z direction at t=1.04s (Mode I):
(a) reference solution; (b) energy based solution; and (c) stress based solution

0
0.05
0.1
0.15
0.2
0.25
0 0.4 0.8 1.2 1.6
Time(s)
D
i
s
p
l
a
c
e
m
e
n
t
(
m
m
)
Reference
Energy based
Stress based

Figure 3.7 The center point O displacement in the loading (z) direction (Mode I)

(a)
(b)
(c)
Crack Crack

48
3.5.2 Sandwich shell impact

Figure 3.8 Geometry, layup and impact loading of the sandwich shell

(a)

(b)
Figure 3.9 Initial configuration of the macro and unit cell models: (a) undeformed
coarse scale model; and (b) undeformed unit cell model

Geometry, layup and impact loading conditions are shown in Figure 3.8. The
impactor is modeled as a rigid body with initial velocity of 275m/s and mass equal to
Polyurea
DH36
RVE
C
DH36, thickness=0.2mm
Polyurea, thickness=0.2mm
Epoxy interface
A
B
V=275m/s
DH36 steel plate
Thickness=5.0mm
Epoxy interface
Polyurea
Thickness=5.8mm

49
145gm. The circular plate has a diameter of 154mm. It consists of two layers, DH36
steel and polyurea, glued by epoxy. The plate is discretized with 8-node reduced
integration and hourglass control brick elements. The DH-36 structural steel is modeled
as a thermoplastic material to take into account large plastic deformation occurring in
less than 1 ms [38]. Adiabatic conditions are assumed for the DH36 steel. The polyurea
is modeled as linear elasticity with Youngs modulus 2.52GPa and Poissons ratio 0.484.
The epoxy is modeled using cohesive elements [39].
Figure 3.9 displays the geometrical configuration for the two-scale analysis. The
interface and a thin layer of steel and polyurea above and below define the unit cell as
shown in Figure 3.9. The residual velocity of the impactor obtained with the multiscale
method and direct numerical simulation (reference solution) is compared in Figure 3.10.
The von-Mises stress history at point C (defined in Figure 3.9(a)) is shown in Figure
3.11 for both the reference and multiscale solutions. Figure 3.12 compares the overall
deformation of the sandwich plate at t=12ms as obtained with the reference and
multiscale solutions. Figure 3.13 compares the unit cell deformation at the two locations
around the interface marked in Figure 3.12(b) as obtained with the reference and
multiscale solutions. The location of points A and B prior to the deformation is shown in
Figure 3.9(b); the position of the two point at t=12ms is depicted in Figure 3.13 at the
two macroscopic positions denoted as region a and region b in Figure 3.12(b).

50
0
50
100
150
200
250
300
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
Time(ms)
R
e
s
i
d
u
a
l

v
e
l
o
c
i
t
y
(
m
/
s
)
Reference
Multiscale

Figure 3.10 Residual velocity of the impactor

0
200
400
600
800
1000
1200
1400
1600
0 0.1 0.2 0.3
Time(ms)
M
i
s
e
s

S
t
r
e
s
s
(
M
P
a
)
Reference
Multiscale

Figure 3.11 von-Mises stress at point C (in Figure 3.9(a))

51

(a)

(b)
Figure 3.12 Deformation of the sandwich plate obtained with the (a) reference and
(b) multiscale solutions at t=12ms

Figure 3.13 Local deformation in regions a and b (see Figure 3.12(b)) at t=12ms as
obtained with reference (left) and multiscale (right) solutions
Reference solution
Multiscale solution
1.15mm
2.36mm
B
2.18mm
1.08mm
A
B
A
(a)
Reference solution
Multiscale solution
2.76mm
3.04mm
A
B 2.91mm
2.64mm
B
A
(b)
Debonding
Region a
Region b

52
3.5.3 Crush simulation of a composites circular tube
The circular tube model is illustrated in Figure 3.14. The inner diameter of the tube
is 5.96cm, and the thickness is 0.19cm. The length of the tube is 18.0cm. The tube
bottom is constrained in the vertical direction. The top of the tube is subjected to
compression by a rigid body with a constant velocity 25cm/s. Only one quarter of the
tube is modeled in the simulation with the consideration of symmetry. Figure 3.15(a)
shows the mesh of a unit cell for micro-scale analysis. The microstructure of the braided
composites without matrix is shown in Figure 3.15(b). The unit cell is partitioned into
four phases: one for matrix, one for axial tow and two for bias tows. The tow is modeled
as elastic material with isotropic damage [40,41]. Figure 3.16 compares the multiscale
simulation and experiment results for load-displacement history. Good agreement with
the experiment result can be observed.

Figure 3.14 Finite element model of a circular tube

Figure 3.15 (a) mesh of the unit cell and (b) phase partitions of unit cell
(a)
(b)

53
0
5
10
15
20
25
30
35
0 20 40 60 80 100
Displacement (mm)
L
o
a
d

(
K
N
)
Multiscale
Experiment

Figure 3.16 Comparison of load-displacement history

References

1. S. Ghosh, J. Bai, and P. Raghavan, Concurrent Multi-level Model for Damage
Evolution in Microstructurally Debonding Composites, Mechanics of Materials, vol. 39,
pp. 241266, 2007.
2. J. Fish, P. Nayak, and M. H. Holmes, Microscale Reduction Error Indicators and
Estimators for a Periodic Heterogeneous Medium, Computational Mechanics, vol. 14, pp.
323-338, 1994.
3. J. T. Oden, and K. S. Vemaganti, Estimation of Local Modeling Error and Goal-
oriented Adaptive Modeling of Heterogeneous Materials I. Error Estimates and Adaptive
Algorithms, Journal of Computational Physics, vol. 164, pp. 22-47, 2000.

54

4. R. Ghanem, and M. Pellissetti, Adaptive Data Refinement in the Spectral Stochastic
Finite Element Method, Communications in Numerical Methods in Engineering, vol. 18,
pp. 141-151, 2002.
5 . J. T. Oden, and T. I. Zohdi, Analysis and Adaptive Modeling of Highly
Heterogeneous Elastic Structures, Computer Methods in Applied Mechanics and
6. J. Fish and A. Wagiman, Multiscale finite element method for a locally nonperiodic
heterogeneous medium, Computational Mechanics, vol. 12, pp. 164-180, 1993.
7. J. Fish and V. Belsky, Multigrid Method for a Periodic Heterogeneous Medium. Part 2:
Multiscale Modeling and Quality Control in Mutidimensional Case, Computer Methods
in Applied Mechanics and Engineering, vol. 126, pp. 17-38, 1995.
8. M.W. Beall, V.Belsky, J.Fish, and M.Shephard, Automated Multiple Scale Fracture
Analysis, Journal of Multiscale Computational Engineering, vol. 1, pp. 29-48, 2002.
9 . T. Belytschko and S. P. Xiao, Coupling Methods for Continuum Model with
Molecular Model, International Journal for Multiscale Computational Engineering, vol.
1, pp. 115-126, 2003.
10. M. A. Nuggehally, M. S. Shephard, C. R. Picu and J. Fish, Adaptive Model Selection
Procedure for Concurrent Multiscale Problems, International Journal for Multiscale
Computational Engineering, vol. 5, pp. 369-386, 2007.
11. J. Fish, M. A. Nuggehally, M. S. Shephard, C. R. Picu, S. Badia, M. L. Parks, and M.
Gunzburger, Concurrent AtC Coupling Based on a Blend of the Continuum Stress and
the Atomistic Force, Computer Methods in Applied Mechanics and Engineering, vol.
196, pp. 4548-4560, 2007.
12. J. Fish and Z. Yuan, N-scale Model Reduction Theory, in J. Fish (eds.), Bridging the
Scales in Science and Engineering, Oxford University Press, 2008
13 . K. Terada and N. Kikuchi, Nonlinear Homogenization Method for Practical
Applications, Computational Methods in Micromechanics, in S. Ghosh and M. Ostoja-
Starzewski (eds.), ASME, New York, vol. AMD-212/MD-62, pp. 1-16, 1995.

55

14. K. Terada and N. Kikuchi, A Class of General Algorithms for Multi-scale Analysis
of Heterogeneous Media, Computer Methods in Applied Mechanics and Engineering,
vol. 190, pp. 5427-5464, 2001.
15 . K. Matsui, K. Terada and K. Yuge, Two-scale Finite Element Analysis of
Heterogeneous Solids with Periodic Microstructures, Computers and Structures, vol. 82,
pp. 593-606, 2004.
16. R.J.M. Smit, W.A.M. Brekelmans, and H.E.H. Meijer, Prediction of the Mechanical
Behavior of Nonlinear Heterogeneous Systems by Multilevel Finite Element Modeling,
Computer Methods in Applied Mechanics and Engineering, vol. 155, pp. 181-192, 1998.
17. C. Miehe and A. Koch, Computational Micro-to-macro Transition of Discretized
Microstructures Undergoing Small Strain, Archive of Applied Mechanics, vol. 72, pp.
300-317, 2002.
18. V. Kouznetsova, W.-A. Brekelmans, and F.P.-T. Baaijens, An Approach to Micro-
macro Modeling of Heterogeneous materials, Computational Mechanics, vol. 27, pp. 37-
48, 2001.
19 . F. Feyel and J.-L. Chaboche, FE2 Multiscale Approach for Modeling the
Elastoviscoplastic Behavior of Long Fiber sic/ti Composite Materials, Computer
Methods in Applied Mechanics and Engineering, vol. 183, pp. 309-330, 2000.
20. S. Ghosh, K. Lee, and S. Moorthy, Multiple Scale Analysis of Heterogeneous Elastic
International Journal of Solids and Structures, vol. 32, pp. 27-62, 1995.
21 . S. Ghosh, K. Lee, and S. Moorthy, Two Scale Analysis of Heterogeneous
Elasticplastic Materials with Asymptotic Homogenization and Voronoi Cell Finite
Element Model. Computer Methods in Applied Mechanics and Engineering, vol. 132, pp.
63-116, 1996.
22 . J.-C. Michel, H. Moulinec, and P. Suquet, Effective Properties of Composite
Materials with Periodic Microstructure: a Computational Approach, Computer Methods
in Applied Mechanics and Engineering, vol. 172, pp. 109-143, 1999.

56

23 . M.G.D. Geers, V. Kouznetsova, and W.A.M. Brekelmans, Gradient-enhanced
Computational Homogenization for the Micro-macro Scale Transition, J. Phys. IV, vol.
11, pp. 145-152, 2001.
24. C. McVeigh, F. Vernerey, W. K. Liu, and L. C. Brinson, Multiresolution Analysis
for Material Design, Computer Methods in Applied Mechanics and Engineering, vol.
195, pp. 50535076, 2006.
25 . C. Miehe, Strain-driven Homogenization of Inelastic Microstructures and
Composites Based on an Incremental Variational Formulation, International Journal for
Numerical Methods in Engineering, vol. 55, pp. 12851322, 2002.
26. C. Oskay and J. Fish, Eigendeformation-Baed Reduced Order Homogenization for
Failure Analysis of Heterogeneous Materials, Computer Methods in Applied Mechanics
and Engineering, vol. 196, pp. 1216-1243, 2007.
27. E.B. Tadmor, M. Ortiz, and R. Phillips, Quasicontinuum Analysis of Defects in
Solids, PHILOS. MAG. A, vol. 73, pp. 1529-1563, 1996.
28 . I.G. Kevrekidis, C.W. Gear, and G. Hummer, Equation-Free: The Computer-
Assisted Analysis of Complex, Multiscale Systems, Am. Inst. Chemical Engineers J., vol.
50, pp. 13461354, 2004.
29 . M. Ortiz and J.J Quigley, Adaptive Mesh Refinement in Strain Localization
Programs, Computer Methods in Applied Mechanics and Engineering, vol. 90, pp. 781-
804, 1991.
30. J. Fish and Q.Yu, Computational Mechanics of Fatigue and Life Predictions for
Composite Materials and Structures, Computer Methods in Applied Mechanics and
31. E. Gal, Z. Yuan, W. Wu, and J. Fish, A Multiscale Design System for Fatigue Life
Prediction, International Journal of Multiscale Computational Engineering, vol. 5, pp.
435-446, 2007.
32 . E. B. Tadmor and R. E. Miller, The Theory and Implementation of the
Quasicontinuum Method, in S. Yip (eds.), Handbook of Materials Modeling, chap. 2.13,
Springer Netherlands, 2005.

57

33. J. Fish, Q. Yu, and K. Shek, Computational Damage Mechanics for Composite
Materials Based on Mathematical Homogenization, International Journal for Numerical
34. J. C. Simo and J. W. Ju, Strain- and Stress-based Continuum Damage Models. I:
Formulation, International Journal of Solids and Structures, vol. 23, pp. 821-840, 1987.
35. J. Ozbolt and Z.P. Bazant, Numerical Smeared Fracture Analysis: Nonlocal Micro-
crack Interaction Approach, International Journal for Numerical Methods in Engineering,
vol. 39, pp. 635-661, 1996.
36. T. Belytschko and D. Lasry, A study of Localization Limiters for Strain Softening in
Statics and Dynamics, Computers and Structures, vol. 33, pp. 707-715, 1989.
37. T. Belytschko and I. Leviathan, Physical Stabilization of the 4-node Shell Element
with One Point Quadrature, Computer Methods in Applied Mechanics and Engineering,
vol. 113, pp. 321-350, 1994.
38. S. Nemat-Nasser and W.G. Guo, Thermomechanical Response of DH-36 Structural
35, pp. 1023-1047, 2003.
39. K. M. Liechti and J.D. Wu, Mixed-mode, Time-dependent Rubber/Metal Debonding,
Journal of the Mechanics and Physics of Solid, vol. 49, pp. 1039-1072, 2001.
40. Z. Yuan, Multiscale Design System. Ph.D. thesis, Rensselaer Polytechnic Institute,
New York, 2008.
41. Z. Yuan and J. Fish, Multiscale Modeling for Crash Prediction of Composite
Structures, ACC Progress Report, New York, 2008.

58
4. The rs-method for material failure simulations

4.1 Introduction
Studies of causes and mitigation of material failure have been and still are one of the
most important subjects in scientific and engineering community. The number of papers,
workshops and conferences focusing on this subject are still on the rise; a recent Google
search of the key-word material failure provided over 59 million hits. Indeed, material
failure comes in different shapes and forms: defect nucleation and growth, coalescence,
micro-cracking, macro-crack growth, shear banding, just to mention a few. The
challenges are in both understanding the mechanisms of material failure and its
computational modeling. In this chapter, we focus on the latter.
From computational point of view, the main difficulty stems from the multiscale
nature of failure. Even if failure mechanisms of bond breaking at a discrete scale, defect
nucleation and growth, shear banding and macro-fracture leading to catastrophic
structural failure were well understood, the so-called formidable tyranny of scales
coined by the NSF Simulation Based Engineering Science (SBES) report [1], poses
tremendous challenges. Limiting the discussion to continuum scales only, our plan is to
first briefly review existing practices and then outline our vision for new opportunities
culminating in the proposed method.
Finite element modeling of fracture propagation has traditionally been performed by
remeshing and by inserting crack surfaces along a crack path allowing the adjacent
elements to separate [2]. While complete remeshing of large models may overshadow
the entire computational cost, mesh modification, which is local remeshing [3], provides
an attractive alternative. In case of dynamic fracture, external time-varying tractions are
often applied along crack surfaces to reduce the effect of waves generated by creation of
new surfaces [4].
The extended finite element method (X-FEM) pioneered by Belytschko and his
associates [5][6] alleviates the need for remeshing of crack surfaces. X-FEM allows the
crack to pass arbitrarily through elements by incorporating enrichment functions through
the notion of local partition of unity. X-FEM is in particular attractive when used in

59
conjunction with the level set method pioneered by Osher and Sethian [7]. The level set
representation of the crack simplifies the selection of the enriched nodes, as well as the
definition of the enrichment functions.
Strouboulis et al. [8] have used the partition of unity framework (PUM [9]) to model
holes and cracks in two-dimensions, whereas Duarte et al. [ 10] have studied the
simulation of three-dimensional dynamic crack propagation. Alternatively, local
enrichment schemes where various failure mechanisms including discontinuity in strains
[11,12,13], curvatures [14] and displacements [15,16] can be embedded at the element
level. Cohesive elements that allow for separation along element boundaries [17,18]
provide an attractive alternative since they do not require enrichment. The nonsmooth
crack growth in this case can be attributed to non-smooth fracture surfaces provided that
the mesh is sufficiently fine.
Another category of methods that alleviates the need for remeshing is often known
as mesh-free formulation. Methods belonging to this category include: Smooth Particle
Hydrodynamics (SPH) originally developed by Gingold [19] and Lucy [20], Element
Free Galerkin Method (EFGM) [21], Reproducing Kernel Particle Method (RKPM) [22]
and various variants of these methods.
Despite significant progress in understanding and developing various methods for
propagating strong and weak discontinuities, commercialization of these technologies
has been rather slow. Notable exceptions are fracture codes based on local mesh
modification (FRANC3D [23,24]) and Smooth Particle Hydrodynamics (EPIC [25]).
Most of the commercial explicit Lagrangian codes, such as LS-DYNA and ABAQUS
employ the erosion element algorithm, where element deletion is controlled by certain
local failure criterion. The attractiveness of this technology stems from its simplicity. In
fact, this technology results in CPU time decrease since fracture often appears in highly
deformed elements. The method is accurate provided that sufficiently small elements
and adequate erosion criterion are used. One of its drawbacks stems from the mass loss -
the effect of which is very severe for large elements, but could be partially circumvented
by associating masses with nodes. The element erosion based on local critical stress or
strain criterion is known to suffer from mesh size dependency, but energy based failure
criterion [26], have shown to yield mesh size independent results [27]. The energy

60
based failure criterion necessitates calculation of path dependent integrals (in dynamics),
which have limited the implementation to two-dimensional settings.
We now outline our vision for the failure simulation code. First, we believe it should
have a hierarchical multilevel structure so that failure modes associated with various
scales can be hierarchically introduced. By hierarchy, we mean that fine scale features
could be introduced without modifying the coarse scale model. Hierarchical structure
allows reuse of coarse model meshes and computations and is consistent with a multigrid
philosophy. Secondly, failure characteristic are often very complex consisting of
interacting weak and strong discontinuities. Therefore, failure modes should not be a
priori defined, but rather computationally resolved possibly in an auxiliary local patch
and then hierarchically introduced into coarse scale. Finally, the technology should be
simple, preferably compatible with commercial software architectures.
We now describe one such candidate that possesses aforementioned characteristics.
As a multilevel scheme we choose the multilevel s-version of the finite element method
[25], which was originally introduced for two levels in [28, 29]. The s-method consists
of overlaying a basic coarse mesh with a hierarchy of local patches [30] engineered of
resolving local features. For instance, the base coarse mesh can be crack free, whereas
the discontinuity can be embedded in the superimposed mesh(es) only [31]. Interesting
variants of this approach have been recently reported in [32] and [33]. The s-version,
however, suffers from two shortcomings: (i) the computational complexity and (ii) the
need for remeshing the superimposed mesh. While sufficiently fine superimposed
meshes are capable of resolving any failure characteristics, they may involve many more
degrees-of-freedom than, for instance, XFEM. The computational complexity of the
superimposed mesh can be reduced in several ways. One possibility is to carry out modal
analysis in the superimposed patch(es) and then to enrich the coarse scale approximation
by critical modes capturing failure characteristics. This obviously has to be repeated with
failure evolution. If failure modes extend over many coarse scale elements, it might be
advantageous to project these modes onto local supports of the coarse mesh (PUM), to
preserve sparsity at the expense of additional degrees of freedom. To remedy the need
for remeshing the superimposed patches and to take advantage of commercial software
architectures we will employ element erosion technology with a simplified energy based

61
element erosion criterion in the superimposed patches only. The resulting method will be
termed as the rs-version of the finite element method, or reduced order s-version with
element erosion.
The outline of this chapter is as follows: The basic idea of the rs-method is outlined
in Section 4.2. Section 4.3 details the formulation, algorithmic details and
implementation in ABAQUS. A simplified variant of the energy release-based failure
criterion is given in Section 4.4. Numerical examples are given in Section 4.5.

4.2 The Basic Idea
As a prelude, we start with a brief overview of a two-level structured s-version. In
the structured s-version, which has found its commercial implementation in COMET-AR
[ 34], element boundaries of the base (underlying) mesh coincide with those of the
superimposed elements as shown in Figure 4.1. This provides considerable
simplification in integration of finite element matrices at the expense of optimal
placement of the superimposed patch typically placed in the critical regions identified by
some error indicators [28].

Figure 4.1 A two-level mesh superposition

The displacements in the superimposed region are approximated as:
Underlying coarse mesh
Superimposed fine mesh
Overlapping nodes
Dangling nodes
int

62

0 0 0 0
0 1 0 0 1 1 1
1
int
( ) ( ) on
( )
( ) ( ) ( ) ( ) on
( ) 0 on
i I Ii
i
i i I Ii J Ji
i
u x N x d
u x
u x u x N x d N x d
u x
=
=

+ = +
=
(4.1)
where the superscripts denote the mesh level; lower case subscripts denote spatial
dimensions and capital subscripts denote node numbers; summation convention is
employed over repeated subscripts (both spatial dimensions and nodes). In Eq. (4.1),
0
is the entire problem domain and
1
the domain of the superimposed patch. At
the interface
int
between the two meshes the displacement in the superimposed mesh
1
( )
i
u x are imposed to vanish to ensure
0
C continuity of the solution. Consequently, at
the discrete level, the nodes on
int
, termed as dangling nodes, are constrained, i.e.
1
0
Ji
d = ;
0
n and
1
n represent the number of mesh nodes in the two meshes, respectively.
Displacements in the underlying and superimposed meshes are discretized in terms of
shape functions denoted as
0
I
N and
1
I
N , respectively. Linear dependency is eliminated
by constraining the so-called overlapping nodes in the superimposed mesh, i.e. the
superimposed mesh nodes, which coincide with the underlying coarse mesh nodes as
shown in Figure 4.1.
For the rs-method, the displacements in the superimposed region are approximated
as:
0 1
0 0 0
0 0 0 1 1 1
1
( ) on
( )
( ) ( ) ( ) on
0 on
i i
I Ii
i
I Ii J Ni JN
u u
JN int
N x d
u x
N x d N x x d
d

=
+
=
_ _
(4.2)
where
1
( )
Ni
x is N
th
-mode displacement component in the spatial direction i . The modes
1
( )
Ni
x are selected to be the lowest eigenmodes (excluding rigid body) computed in the
superimposed mesh. An alternative weak compatibility condition, detailed in Section 4.3,
has been found to provide superior accuracy.
To illustrate the basic idea of the reduced order method (rs-method), we consider a
one-dimensional model problem as shown in Figure 4.2, where discontinuity is modeled
by double nodes in the superimposed mesh. For comparison with the s-method, we use

63
the same underlying and superimposed meshes. The shape functions and the
displacement fields are shown in Figure 4.3. We select a single non rigid-body mode,
which obviously captures the discontinuity as shown in Figure 4.4(a). It can be seen that
for this model problem this mode is identical to the discontinuous piecewise constant
function a priori constructed in XFEM. The resulting enrichment functions,
0 1
1
( ) ( )
I
N x x ,
are shown in Figure 4.4(b).

Figure 4.2 The model problem

Figure 4.3 (a) displacement field in the 0 and 1 level meshes and (b) corresponding
shape functions in the two meshes
(b)
Level 0 Shape function
0
( )
I
N x
Level 1 Shape function
1
( )
J
N x
(a)
x
u
0
u
1
u
Level 0: Coarse mesh
Level 1: Superimposed mesh
x
x

64

Figure 4.4 (a) Lowest frequency eigenmode
1
1
and (b) corresponding enrichment
functions
0 1
1
( ) ( )
I
N x x

4.3 The Formulation
Consider a problem domain with boundary , consisting of the prescribed
displacement boundary
u
and the prescribed traction boundary
t
. The strong form is
given by

0 on
ji
i
j
j ji i t
i i u
b
x
n t on
u u on
+ =
=
=
(4.3)
In local regions where failure takes place at a certain time instant t, patch(es)
1
( ) t
designated to resolve failure characteristics (weak and strong discontinuities) are
superimposed as shown in Figure 4.5. For simplicity, we focus on a two-level scheme.
The weak form is obtained in a usual manner. It seeks for
i
u U such that

,
0
d d d
t
i j ij i i i i i
w wt wb w U

= +

(4.4)
where
(b)
(a)
1
2
0 1
1 1
( ) ( ) N x x
0 1
2 1
( ) ( ) N x x

65
1
1
0
{ | , on }
{ | , 0 on }
i i i i u
i i i u
U u u H u u
U w w H w
= =
= =
(4.5)

Figure 4.5 Definition of a local patch
1
( ) t at time t

The weight functions are decomposed and discretized similarly to the trial functions
given in Eq. (4.2)
0 0 0 0
0 1 0 0 0 1 1 1
0 1
1
( ) on
( )
( ) ( ) ( ) on
0 on
0 on
i I Ii
i
i i I Ii I Ni IN
i i u
i int
w N x a
w x
w w N x a N x x a
w w
w
=
=

+ = +
= =
=
(4.6)
The resulting discrete equations are summarized below

( )
1 1
0
0 0 0
0 1
1 0 1 0 1
d d d 0
d d d 0
t
t
I
Ii ij I i I i
j
I Ni
Ni ij I Nj i I Nj i
j
N
r N t N b
x
N
r N t N b
x
= =

= =

(4.7)
For nonlinear problems, the tangent stiffness matrix is obtained by consistent
linearization of (4.7).
An alternative to the strong compatibility condition (4.2)b is a weak compatibility
given by

int
1
0
i i
u d
(4.8)
0
=

t

u

1
( ) t

66
The function
0
( )
i
x H is approximated as piecewise constant over element edges
int
e

(or faces in three dimensions) positioned at the interface
int int
e
e
=

int
( )
e e
i i
e
x N
=

(4.9)
where

int
1 on
0 elsewhere
e
e
N

=

(4.10)
Substituting (4.9) and (4.2) into (4.8) yields the so-called multi-point constraints (MPC)
equations for
1
int IN
d

mod
int
0 1 1
int
1
0 ,
e
N
e
I Ni IN
N
I
N d i
=
=

(4.11)
The MPC condition (4.11) can be enforced either by using penalty method Lagrange
multipliers method or by requiring the weight functions
i
w in Eq. (4.4) to satisfy
mod
int
0 1 1
int
1
0 ,
e
N
e
I Ni IN
N
I
N a i
=
=

(4.12)
To complete the formulation, the following two issues remain to be addressed: (i)
linear dependency between
0
i
u and
1
i
u and (ii) selection of
1
Nj
. To clarify these issues
consider two cases depicted in Figure 4.6: case A without a crack and case B with a
crack. In both cases, the underlying mesh is crack free. The crack is represented in the
superimposed mesh (case B) only. For the crack-free case, it is possible that a non-rigid
body mode extracted from the superimposed mesh would be linear dependent with a
deformation represented by the underlying mesh. In this cases the overlapping nodes I in
the superimposed mesh have to be constrained to eliminate linear dependence. On the
other hand, if
1
Nj
captures discontinuity (case B), the underlying crack-free mesh will
not represent the same deformation mode. In this case, the overlapping node J is left
free. Thus, the strategy for
1
Nj
extraction is as follows: (i) loop over all the underlying
mesh nodes and constrained those, which fall into the category of nodes described in
case A (see Figure 4.6); (ii) extract the lowest energy (excluding rigid body) modes from
the superimposed mesh and project them onto local supports, i.e. compute
0 1
I Nj
N . In

67
theory, it is possible that
0 1
I Nj
N would show no discontinuity on one or more patches in
which the overlapping node has to be constrained.

Figure 4.6 Suppressing linear dependency between underlying and superimposed
meshes. Node I in the superimposed mesh overlaps node J in the underlying mesh

To this end, we describe the implementation of the rs-method in ABAQUS [35].
Our implementation strategy is guided by the limitations of working with commercial
software architecture. In ABAQUS, we use a user-defined element subroutine UEL to
control the coarse mesh and the superimposed mesh analyses. The basic steps are
summarized below:
1. In the first iteration of each load increment, perform modal analysis in the
superimposed mesh
1 1 1
IiJj Jj Ii
K =
1 1
1
1 1 1
1
d d d
t
I
IiJj ij I i I i
Jj j
N
K N t N b
d x

=

and store the extracted modes
1
Ii
in the external file.
Underlying mesh without crack
Case A: Superimposed
mesh without crack
Case B: Superimposed
mesh with crack

I
I
J
crack

68
2. Read
1
Ni
from the external file, interpolate for the velocities in the
superimposed mesh nodes
( ) ( ) ( )
1 0 1 0 0 1 1 1 1
( )
I I I Ii I I Ni I IN
I
i
u x x d x x N N d + =
`
`
`

and save them to the external file.
3. Update the stresses in the Gauss points of the superimposed mesh and those in
the coarse mesh elements not overlapped by the superimposed mesh.
4. Calculate the residuals based on Eq. (4.7), compute the tangent stiffness matrix
as
0 0
0 1
1 1
0 1
Ii Ii
Jj JM
IN IN
Jj JM
r r
d d
r r
d d

and iterate (Newton or related method) until convergence.
Note that once the enrichment shape functions are defined based on Eq. (4.2) the
last step is carried out automatically by a commercial software of choice.

4.4 A simple energy-based element erosion criterion

Figure 4.7 Crack closure tractions and displacements: (a) in the analytical model
and (b) in a mesh simulated by element erosion
profile before closure
profile after
closure y

a .
y
u
x
y
y
h
(1 )
y
h +
eroded element
(a) (b)

69
We start from a classical definition of the fracture toughness (or critical energy
release rate)
c
G as the work required to close a crack by a small increment as shown in
Figure 4.7(a).
In the classical fracture mechanics the fracture toughness is defined as

0
0
1 1
lim
2
a
c y y
a
G u dx
a

=

.
.
.
(4.13)
where
y
is stress normal to the crack surface and
y
u is a closure displacements.
We now consider an approximation to the fracture toughness
c
G
obtained by
element erosion. For simplicity, we assume the state of constant element stresses
evaluated at the element centroid (one-point integration with stabilization [36]). The
eroded element elongation perpendicular to the crack surface is given by
y y
u h = ,
where
y
is the strain component normal to the crack surface as shown in Figure 4.7(b).
If we further assume that the crack propagates normal to the direction of the maximum
principal stress in tension
1
then Eq. (4.13) reduces to

1 1
1
2
c
G h
(4.14)
where
1
is the normal strain computed in the principal directions of stress. Eq. (4.13)
can be further simplifying for linear isotropic material and by neglecting the remaining
principal stress components, which gives

2
c
c
c
G E
h
=

(4.15)
where
c
h is a characteristic material lengthscale (aggregate size, grain size, etc.) and
c

is a critical stress computed from Eq.(4.15). The maximum principal stress at which an
element of length
c
h h = is eroded is then
1er c
= . On the other hand, for elements of
size
c
h h , the critical stress for element erosion is

1
c
er c
h
h
= (4.16)

70
For instance, if / 2
c
h h = , i.e., we refine the mesh a by a factor of two, the erosion stress
increases by 40% compared to the base mesh (
c
h h = ). The criterion for element erosion
(4.16) is in the spirit of cohesive crack models [37].

Remark:
For elements neighboring the crack tip, Eq. (4.16) has an alternative interpretation. Let
c
r be the distance from the crack tip (assumed to be material property) at which the
critical stress
c
is measured. For different meshes, the distance of the element centroid
to the crack tip is denoted by r and is of order of element size, i.e, r h . Since stresses
at the crack tip are varying as 1/ r , the element erosion stress
1er
is then

1
c
er c
r
r
= (4.17)
Note that Eq. (4.16) is applicable to all the elements in the mesh, whereas Eq. (4.17) is
limited to elements neighboring the crack tip and would require tracking the crack fronts.

Our numerical experimentation agenda includes three test problems. First, we
investigate the accuracy of the method to predict the stress intensity factor. In the
second example, we study quasi-static crack propagation in a concrete beam with energy
release-based and stress-based element erosion criteria. Finally, we consider a time
dependent impact problem without element erosion.

71
4.5.1 Plate with a centered crack

(a)

(b)
Figure 4.8 (a) Plate with center crack and (b) A superimposed mesh for a plate with
a centered crack

The geometry of the plate with a centered crack is shown in Figure 4.8(a) and (b).
The width w of the plate is 10cm and the length L is 10.25cm. A crack of length
2a=1.0cm is placed at the center. Plane stress condition is assumed with the following
material properties: Youngs modulus 2.0 GPa and Poisson ratio 0.2. The plate is
subjected to mode one tension load of 10.0 MPa. The width of the superimposed patch is
1.25cm and the length is 2.0cm. The global domain is discretized with 4041 four-node
Superimposed

72
quadrilateral elements; the superimposed patch is placed over 58 underlying elements.
Each underlying element in the superimposed patch is overlaid by 8 8 four-node
quadrilateral elements.
Figure 4.9 gives the error in the stress intensity factor versus number of modes; both
strong and weak compatibility conditions at the interface have been considered. It can be
seen that the weak compatibility condition provides superior accuracy at low number of
modes, but as the number of modes increases the two methods of enforcing
compatibility yield similar performance.

0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
0 2 4 6 8 10 12 14
Number of Modes
E
r
r
o
r

o
f

K
I

(
%
)
Strong Compatibility
Weak Compatibility

Figure 4.9 Error in the stress intensity factor versus number of enrichment modes

73
4.5.2 Crack propagation in a concrete beam [38]

Figure 4.10 Geometry of notched concrete beam (dimensions in mm)

Figure 4.11 The reference finite element mesh (28x24 elements in the refined region
at the center of the beam)

Consider a beam made of concrete with the following material properties: Youngs
modulus 24800MPa, Poissons ration 0.18, and a critical fracture stress 6.0MPa. The
geometry and dimensions of the beam are shown in Figure 4.10. Plane stress is assumed
with a beam thickness of 156mm. A linearly increasing velocity is applied at point C
starting from 0 to 0.75mm/s with ramping time of 0.38 seconds. The rigid beam AB is
used to transmit the load to the concrete beam. We first study the sensitivity of the
solution (reaction force at point D versus time and crack height versus time) to the mesh
size. Figure 4.11 depicts the reference mesh, which has 28x24 elements in the refined
region at the center. We consider three additional meshes denoted as mesh 1, obtained
by splitting each element in the reference mesh into four elements, mesh 2 obtained by
A
V
C B
D E
397 397 6 6
397
82
Initial crack

74
splitting each element in mesh 1 into four elements, and mesh 3 obtained by splitting
each element in mesh 2 into four elements.
0
50
100
150
200
250
300
350
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
Time(s)
H
e
i
g
h
t

o
f

c
r
a
c
k
(
m
m
)
reference
EE-mesh 1
EE-mesh 2
EE-mesh 3
LSE-mesh 1
LSE-mesh 2
LSE-mesh 3

Figure 4.12 Height of the crack versus time

0
20000
40000
60000
80000
100000
120000
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
Time(s)
L
o
a
d
(
N
)
reference
EE-mesh 1
EE-mesh 2
EE-mesh 3
LSE-mesh 1
LSE-mesh 2
LSE-mesh 3

Figure 4.13 Reaction at support D versus time

75

Figure 4.14 A completely cracked beam as obtained with the energy based erosion
criterion in the four meshes

76
We study two failure criteria: (i) local stress-based erosion (LSE) criterion by which
an element is removed when the maximum principal stress in tension reaches the
fracture stress, and (ii) energy-based erosion (EE) criterion described in Section 4.4.
Figure 4.12 and Figure 4.13 depict the height of the crack and the reaction at point D
versus time as obtained with the two failure criteria. It can be seen that the local stress-
based criterion suffers from significant mesh dependence, whereas energy based
criterion is almost insensitive to the mesh size. The finite element meshes of the fully
cracked beam as obtained with the energy-based erosion criterion are shown in Figure
4.14.
To this end, we study the performance of the rs-method for the notched concrete
beam. The underlying and the superimposed meshes are shown in Figure 4.15. 4x4
elements were superimposed on each underlying element crossing the path of the crack
as shown in Figure 4.15(b). We considered nine and six modes extracted from the
superimposed patch with strong compatibility condition (fewer modes could have been
used with weak compatibility). Figure 4.16 and Figure 4.17 depict the height of the crack
and the reaction at point D versus time as obtained with the rs-method (nine and six
modes) and utilizing the energy based failure criteria.

(a) (b)
Figure 4.15 rs-method: (a) Underlying mesh, and (b) superimposed mesh

77
0
20000
40000
60000
80000
100000
120000
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
Time(s)
L
o
a
d
(
N
)
Reference
9 modes
6 modes

Figure 4.16 Reaction at support D versus time as obtained with the rs-method (six
and nine modes) and the reference solution

0
50
100
150
200
250
300
350
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
Time(s)
H
e
i
g
h
t

o
f

C
r
a
c
k
(
m
m
)
Reference
9 modes
6 modes

Figure 4.17 Height of the crack versus time as obtained with the rs-method (six and
nine modes) and the reference solution

78
4.5.3 Impact simulation
In the last example, we study the rs-method for impact simulation without element
deletion. The length of the beam is 132mm, and the height is 4.83mm. Figure 4.18 gives
the problem setup.

Figure 4.18 Model problem for impact simulation

160
180
200
220
240
260
280
300
0 0.00001 0.00002 0.00003 0.00004
Time(s)

V
e
l
o
c
i
t
y

o
f

i
m
p
a
c
t
o
r

(
m
/
s
)
rs-method
Reference

Figure 4.19 Comparison of the velocity of impactor versus time

The beam is made of DH-36 structural steel with a material model developed by
Nemat-Nasser and Guo [39]. The rigid body has a round nose of radius 9mm. The mass
of the rigid body is 3% of the mass of the shaded area. The initial velocity of rigid body
is 275m/s. Free boundary conditions at the two ends of the beam and plain strain were
assumed. Only half of the beam was modeled due to symmetry. The shaded domain

79
under the impactor in Figure 4.18 shows the placement of the superimposed patch. Three
modes were extracted for enrichment.
Figure 4.19 shows the impactor velocity versus time. An excellent agreement
between the reference (no modal reduction) and the rs-version is observed.

References

1 Simulation - Based Engineering Science, National Science Foundation report sbes0506,
http://www.nsf.gov/pubs/reports/sbes_final_report.pdf , May 26, 2006
2 M.S. Shephard, N.A.B. Yehia, G..S. Burd, and T.J. Weidner, Automatic Crack
Propagation Tracking, Computers & structures, vol. 20, pp.211- 223, 1985.
3 X. Li, M.S. Shephard, and M.W. Beall, 3-D Anisotropic Mesh Adaptation by Mesh
Modifications, Computer Methods in Applied Mechanics and Engineering, vol. 194, pp.
4915-4950, 2005.
4 R.H. Dodds Jr., M. Tang, and T.L. Anderson, Numerical Procedures to Model Ductile
Crack Extension, Engineering Fracture Mechanics, vol. 46, pp. 253-264, 1993.
5 T. Belytschko, and T. Black, Elastic Crack Growth in Finite Elements with
MinimalRemeshing, International Journal for Numerical Methods in Engineering, vol.
45, pp. 601620, 1999.
6 N. Sukumar, N. Mos, B. Moran, and T. Belytschko, Extended Finite Element Method
for Three-dimensional Crack Modeling, International Journal for Numerical Methods in
7 S. Osher and J.A. Sethian, Fronts Propagating with Curvature Dependent Speed:
Algorithms Based on HamiltonJacobi Formulations, Journal of Computational Physics,
vol. 79, pp.1249, 1988.
8 T. Strouboulis, I. Babuska, K. Copps, The Design and Analysis of the Generalized
Finite Element Method, Computer Methods in Applied Mechanics and Engineering, vol.
181, pp.4369, 2000.

80

9 J.M. Melenk, and I. Babuska, The Partition of Unity Finite Element Method: Basic
Theory and Applications, Computer Methods in Applied Mechanics and Engineering,
vol. 139, pp.289314, 1996.
10 C. A. Duarte, O. N. Hamzeh, T. J. Liszka, and W. W. Tworzydlo, A Generalized
Finite Element Method for the Simulation of Three-dimensional Dynamic Crack
Propagation, Computer Methods in Applied Mechanics and Engineering, vol. 190, pp.
22272262, 2001.
11 T. Belytschko, J. Fish and B.E. Engelmann, A Finite Element with Embedded
Localization Zones, Computer Methods in Applied Mechanics and Engineering, vol. 70,
pp. 59 89, 1988.
12 J. Fish and T. Belytschko, Elements with Embedded Localization Zones for Large
Deformation Problems, Computers and Structures, vol. 30, pp. 247-256, 1988.
13 M. Ortiz, Y. Leroy, A. Needleman, A Finite-Element Method for Localized Failure
Analysis, Computer Methods in Applied Mechanics and Engineering, vol. 61, pp. 189-
214, 1987.
14 T. Belytschko and J. Fish, Embedded Hinge Lines for Plate Elements, Computer
15 F. Armero and K.Garikipati, An Analysis of Strong Discontinuities in Multiplicative
Finite Strain Plasticity and Their Relation with the Numerical Simulation of Strain
Localization, International Journal of Solids and Structures, vol. 33, pp. 2863-2885,
1996.
16 J. C. Simo, J. Oliver, and F. Armero, An Analysis of Strong Discontinuities Induced
by Strain-sofetening in Rate-independent Inelastic Solids, Computational Mechanics, vol.
12, pp. 277-296, 1993.
17 G.T. Camacho and M. Ortiz, Computational Modeling of Impact Damage in Brittle
Materials, International Journal of Solids and Structures, vol. 33, pp.2899 2938, 1996.
18 M. Ortiz, and A. Pandolfi, Finite-deformation Irreversible Cohesive Elements for
Three-dimensional Crack-Propagation Analysis, International Journal for Numerical
Methods in Engineering, vol. 44, pp. 12671282, 1999.

81

19 R.A. Gingold and J.J. Monaghan, Smoothed Particle Hydrodynamics: Theory and
Application to Non-spherical Stars, Monthly Notices Royal Astronomical Society, vol.
181, pp. 375-389, 1977.
20 L.B. Lucy, A Numerical Approach to the Testing of Fusion Process, Astronomical
Journal, vol. 88, pp. 1013-1024, 1977.
21 T. Belytschko, Y.Y. Lu, and L. Gu, Element-free Galerkin Methods, International
Journal for Numerical Methods in Engineering, vol. 37, pp. 229-256, 1994.
22 W. K. Liu, S. Jun, S. Li , J. Adee, and T. Belytschko, Reproducing Kernel Particle
Methods for Structural Dynamics, International Journal for Numerical Methods in
23 B.J. Carter, C.-S. Chen, A.R. Ingraea and P.A. Wawrzynek, A Topology Based
System for Modeling 3d Crack Growth in Solid and Shell Structures, in Proceedings of
the 9th International Congress on Fracture, Sydney, Australia, Elsevier Science
Publishers, Amsterdam, 1997.
24 B.J. Carter, P.A. Wawrzynek, and A.R. Ingraffea, Automated 3-d Crack Growth
Simulation, International Journal for Numerical Methods in Engineering, vol. 47, pp.
229253, 2000
25 J.R. Johnson, R.A. Stryk, T.J. Holmquist, and S.R. Beissel, Numerical Algorithms in
a Lagrangian Code, Report, Wright Laboratory, Elgin Air Force Base, 1997.
26 T. Nakamura, C.F. Shih, L.B. Freund, Computational Methods Based on an Energy
Integral in Dynamic Fracture, International Journal of Fracture, vol. 27, pp.229-243,
1985.
27 S.R. Beissel, G.R. Johnsona, C.H. Popelar, An Element-failure Algorithm for
Dynamic Crack Propagation in General Directions, Engineering Fracture Mechanics, vol.
61, pp. 407-425, 1998.
28 J. Fish and S. Markolefas, Adaptive s-method for Linear Elastostatics, Computer
29 J. Fish, The s-version of the Finite Element Method, Computers & Structures, vol. 43,
pp. 539-547, 1992.

82

30 J. Fish and A. Wagiman, Multiscale Finite Element Method for a Locally
Nonperiodic Heterogeneous Medium, Computational Mechanics, vol. 12, pp.164-180,
1993.
31 J. Fish, Hierarchical modeling of discontinuous fields, Communications in Applied
Numerical Methods, vol. 8, pp. 443-453, 1992.
32 S. Lee, J. Song, Y. Yoon, G. Zi, T. Belytschko, Combined Extended and
Superimposed Finite Element Method for Cracks, International Journal for Numerical
33 S. Loehnert, and T. Belytschko, A Multiscale Projection Method for
Macro/Microcrack Simulations, International Journal for Numerical Methods in
34 G. Stanley, I. Levit, B. Stehlin and B. Hurlbut, CSM Task 15 Report, Lockheed, Palo
Alto Research Laboratory, 1991.
35 ABAQUS Analysis Users Manual, 2006.
36 T. Belytschko and I. Leviathan, Physical Stabilization of the 4-node Shell Element
with One Point Quadrature, Computer Methods in Applied Mechanics and Engineering,
vol.113, pp. 321-350, 1994.
37 A. Hillerborg, M. Modeer, P.E. Petersson, Analysis of Crack Formation and Crack
Growth in Concrete by Means of Fracture Mechanics and Finite Elements, Cement and
Concrete Research, vol. 6, pp. 773782, 1976.
38 ABAQUS Benchmarks Manual, 2006.
39 S. Nemat-Nasser, and W.G. Guo, Thermomechanical Response of DH-36 Structural
35, pp. 1023-1047, 2003.

83
5. Conclusions and discussions

In Chapter 2, mathematical homogenization theory has been generalized to
account for finite nonperiodic deformation of heterogeneous medium subjected to
transient loading. An auxiliary macro-deformed configuration, where the overall Cauchy
stress is updated, and mixed unit cell boundary conditions for modeling nonperiodic
deformation were introduced and verified. In all verification problems considered, the
overall deformation as obtained with the multiscale method has been found to be in good
agreement with the reference solution; the error in the localized fields was often as high
as 10% (sandwich plate problem) but the time integration step was reduced by a factor of
30 compared to the reference solution.
An adaptive two-scale homogenization procedure aimed at optimizing
computational resources is developed in Chapter 3. Selection of either detailed or
reduced order unit cell model is guided by the norm of the Green-Lagrange strain tensor.
The importance of scaling the critical phase stress is illustrated. The two-scale adaptive
homogenization scheme has been verified against the direct numerical simulation of
crack propagation in the composites plate with a center crack and impact of the metal-
polyurea sandwich plate. It has been validated against crush experiments of a fabric
composites circular tube.
In Chapter 4, a new superimposed finite element method with reduced order was
developed and verified. The method builds on excellent closely related methods
including: XFEM [1], PUM [2], GFEM [3], and s-method [4]. We have demonstrated
that it is compatible with commercial software architectures by integrating it with
ABAQUS.
In Chapter 4, it has been implicitly assumed that a physical dimension of a
superimposed patch is much smaller than that of the global problem and that a
computational complexity of solving a problem on a single patch is much smaller than
solving the global problem. If this were not the case, then it would be instrumental to
device an alternative method to eigenmode extraction on a patch. This, for instance, can
be accomplished by subjecting a local patch to overall constant and progressively higher

84
order strain fields, similarly to what is done in the homogenization theory, and then to
approximate the eigenmodes by a linear combination of solutions obtained on a patch.

Recommendations for the future work are:
(a) Model reduction to reduce the cost of unit cell calculations for larger
deformation problems (large strain) [5];
(b) Alternative formulation of mixed boundary condition that would not require
a priori calibration. For example, consider a larger domain which contains
the unit cell domain with appropriate boundary conditions, and average the
stress over the unit cell domain only [6];
(c) Extend the two-scale homogenization method for larger deformation
problems to multiple scales with reduced order;
(d) Develop multiscale like multigrid solver (from the multigrid point of view,
multiscale method can be considered as a solver with respect to direct finite
element method);
(e) Generalize the classical mathematical homogenization theory for
geometrically nonlinear problems with higher order terms [7,8];
(f) Nonuniform transformation field analysis [9,10];
(g) Parallel implementation.

References

1 T. Belytschko and T. Black, Elastic Crack Growth in Finite Elements with Minimal
Remeshing, International Journal for Numerical Methods in Engineering, vol. 45, pp.
601 620, 1999.
2 J.M. Melenk, and I. Babuska, The Partition of Unity Finite Element Method: Basic
Theory and Apapplications, Computer Methods in Applied Mechanics and Engineering
vol. 139, pp.289314, 1996.

85

3 T. Strouboulis, I. Babuska, K. Copps, The Design and Analysis of the Generalized
Finite Element Method, Computer Methods in Applied Mechanics and Engineering, vol.
181, pp. 4369, 2000.
4 J. Fish and S. Markolefas, Adaptive s-method for Linear Elastostatics, Computer
5 J.Yvommet, and Q.-C. He, The Reduced Model Multiscale Method (R3M) for the
Non-linear Homogenization of Hyperelastic Media at Finite Strains, Journal of
Computational Physics, vol. 223, pp. 341-368, 2007.
6 M. Dong and S. Schmauder, Transverse mechanical behaviour of fiber reinforced
composites FE modelling with embedded cell models, Computational Materials
Science, vol. 5, pp. 53-66, 1996.
7 V.G. Kouznetsova, M.G.D. Geers, and W.A.M. Brekelmans, Multi-scale Second-order
Computational Homogenization of Multi-phase Materials: a Nested Finite Element
Solution Strategy, Computer Methods in Applied Mechanics and Engineering, vol. 193,
pp. 5525-5550, 2004.
8 O. Lopez-Pamies and P. Ponte Castaeda, On the overall behavior, microstructure
evolution, and macroscopic stability in reinforced rubbers at large deformations: I
Theory, Journal of the Mechanics and Physics of Solids, vol. 54, pp. 807830, 2006.
9 J.C. Michel and P. Suquet, Computational Analysis of Nonlinear Composite Structures
Using the Nonuniform Transformation Field Analysis, Computer Methods in Applied
Mechanics and Engineering, vol. 193, pp. 5477-5502, 2004.
10 J.C. Michel and P. Suquet, Nonuniform Transformation Field Analysis, International
Journal of Solids and Structures, vol. 40, pp. 6937-6955, 2003

Out

Hochgeladen von

Dokumentinformationen

Originalbeschreibung:

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Out

Hochgeladen von

Copyright:

Verfügbare Formate

MULTISCALE METHODS FOR

MATERIAL FAILURE SIMULATIONS

P denotes the first Piola-Kirchhoff stress tensor;

F the deformation gradient;

= X X in the unit cell (see Figure 2.1)

( , , ) f t X Y rather than ( , , ) f t X Y , the spatial derivative takes the form

into elastic and inelastic components (or more generally stated as

); the left superscript I indicates different types of eigenstrains, such as

and the eigenseparation influence function

are computed by solving a

x , which are described as

Das könnte Ihnen auch gefallen