PHLEXcrack: A Meshless Code for Dynamic Fracture Propagation
Introduction
The Finite Element Method applied to fracture
mechanics to date has been limited to the following approaches:
-
propagating cracks along element boundaries,
-
removing elements along the crack path, and/or
-
remeshing, at least in the vicinity of the
crack.
All of these approaches are very computationally
expensive whenever a reasonably accurate prediction of the crack path is
required. The existence of the finite element mesh is the main obstacle
to obtaining efficient solution algorithms. Recently, a new class of
analysis techniques, based on so called meshless
methods [8], has been developed. Besides the ease of domain
discretization, the ability to introduce cracks arbitrarily in the domain
is the main advantage of these methods when applied to fracture mechanics
problems.
As expected, freedom from meshing constraints
comes at a price, and meshless methods are usually significantly more
computationally expensive than traditional finite element algorithms.
To address these difficult issues, a comprehensive research program was
initiated at Computational Mechanics Co, Inc. (Currently Altair Engineering,
Inc.). The objective of the project, funded partially by the Office of
Naval Research (years 1996-2000), was to develop robust numerical methods
that avoid both the mesh related shortcomings of the FEM and the performance
disadvantages of traditional meshless methods. The outcome of this effort
is a new variant of meshless methods, called the Element Partition Method
(EPM). This method is particularly suitable for fracture mechanics,
and is computationally at least as efficient (and often surpassing)
traditional finite element algorithms. A discontinuous version of EPM,
specifically developed for fracture mechanics applications, produces accurate
solutions of the stress field (static or dynamic) while the crack arbitrarily
passes through the mesh cells.
The EPM technology has been implemented
into a specialized crack propagation program, called PHLEXcrack.
The current version of the code can be used to solve three-dimensional
problems with linear physics (small deformation, small strain theory, static
or dynamic problems), and with arbitrary (user specified) fracture equations
to predict the crack direction and speed. The code is designed to be extensible,
allowing developers/users to change the physical equations of the material
as well as use different crack prediction theories.
Overview
of the PHLEXcrack code
PHLEXcrack is currently released, and
is available for the researchers and commercial applications, although
the fracture mechanics part is limited to linear elasticity model only
with or without thermal stresses. The EPM implementation in PHLEXcrack
allows for the solution of three-dimensional problems using tetrahedral
and hexahedral cells, with arbitrary p-enrichment controlled manually or
by automatic adaptation based on a residual error estimator. Linear, static
and dynamic solution sequences have been implemented, and the current crack
propagation algorithm uses a Freund model [2], combined
with stress intensity coefficients evaluated using linear fracture theory.
Linear EPM volume elements can be connected to traditional Finite Element
entinties (beams, shells, etc.), esp. for modeling external forces and/or
the structure outside of the crack area. A typical hybrid application would
be the analysis of a ship hull or aircraft structure, where only the vicinity
of the crack is modeled using EPM, and the remainder of the model uses
traditional shell elements.
The PHLEXcrack analysis package
is based on the proprietary ProPHLEX[7] finite element/meshless
development environment. This development environment was designed at Altair
specifically for handling computer simulations in an hp-adaptive context.
The current basic capabilities of ProPHLEX include:
-
linear static, dynamic, and eigenvalue
solution sequences
-
nonlinear static and dynamic analysis with
nonlinearities resulting from:
-
material nonlinearities
-
geometry nonlinearities (arbitrary large deformations)
-
rigid contact surfaces
-
a full set of basic (Nastran-like) elements
including:
-
QUAD/TRI (membrane/plate/shell), HEX/TET/PENTA
-
BEAM, ROD, RBE, MPC, SPC, FORCE \
-
hp-adaptive volume elements with robust error
estimation and accuracy control
-
a built-in Graphical User Interface and interactive
postprocessor
-
integration with the Altair's
HyperMesh pre/postprocessor
-
a state-of-the-art linear sparse solver for
the solution of large problems
-
an Object Oriented database with save/restart
capabilities allowing data transfer between different machines (e.g. solution
on CRAY supercomputer and postprocessing on SGI)
-
implementation on various Unix platforms and
a Windows NT version.
Most of these capabilities are available in
PHLEXcrack, although not all are available in the crack propagation solution
sequence. In particular, the nonlinear solid mechanics module has not been
fully tested in connection with EPM and is currently available only in
an alpha testing mode. PHLEXcrack includes preliminary integration
with the
Underwater Shock Analysis [3] ``USA''
code, for accurate simulation of boundary loads resulting from for instance
underwater explosions near immersed or surface structures.
Theoretical
formulation of EPM
The Element Partition Method is a variant
of the Partition of Unity (PoU) Method [1,5], which
serves as a base for several implementations of modern meshless methods.
The Partition of Unity is a set of linearly independent approximation functions,
which only need to be able to approximate a constant function, and are
thus quite easy to build. In order to obtain solution approximations of
a required accuracy, PoU base functions are ``enriched'' typically using
a set of polynomials, but any set of functions can be used for the enrichment.
In particular, asymptotic solutions of fracture mechanics problems have
proven to be very effective in improving the local accuracy of the solution.
In EPM, the base partition of unity comes
from the traditional finite element approximation. This approximation can
subsequently be arbitrarily enriched using higher order polynomials (p-enrichment)
or specifically tailored functions based on the locally asymptotic
solution near crack tip. The combination of these two enrichments produces
very effective and accurate methods for solving fracture mechanics problems.
This version of EPM has many similiarities
to the Finite Element Method (FEM): the PoU approximation satisfies in
general the same requirements as FEM approximations, and they both use
the Galerkin method to generate discretized equations. As a result, most
of the theoretical foundations of FEM, including convergence and error
estimates, apply directly or can easily be extended to EPM. Virtually all
problems which can be solved using FEM, can be also analysed using EPM.
In particular, multiple materials, dynamics, nonlinear solid mechanics
and other similiar problems can be accurately modeled, with efficiency
comparable to or better than FEM. In addition, EPM provides for an easier
implementation of adaptivity, and its accuracy is much less affected by
the quality of the meshing. In our testing we have successfully used meshes
with very badly conditioned tetrahedral cells, while for a similiar mesh
the FEM was locking even after enrichment to order p=3 .
EPM
for domains with cracks
For the cells of the mesh (``elements'') which
are cut by the crack surface a different version of the partition of unity
is used. It is formulated using a Shepard Partition of
Unity [6] within each ``cracked'' cell, effectively producing independent
displacement fields on both sides of the crack without mesh refinement
of remeshing. Still, the resulting approximation is mathematically correct,
preserving the required continuity along inter--element boundaries, and
satisfying rigid body motions and patch test criteria. To illustrate the
idea, lets consider part of an EMP domain with a single crack
Example of element partition discretization
cut by a crack
Cell T1 is not intersected by the
crack, so the approximation at point x will use PoU based on standard
shape functions. For each integration point (Gauss point) within the cells
cut by the crack surface (T2 and T3), the segments between
the point and the cell vertices are tested for the intersection with the
crack surface (e.g. segments 6-y, 7-y, 10-y, and 11-y
in the cell T2). Subsequently the solution at point y is
fully defined by the nodal DOFs of ``visible'' nodes only, here nodes 6
and 7, and is independent of the solution at nodes 10 and 11, which are
located on the other side of the crack surface. The solution along the
path between points r and y is fully C0 continuous,
while an arbitrary jump in the displacement field can appear between points
y and z, or r and s.
Crack
handling in PHLEXcrack
PHLEXcrack offers unprecedented flexibility
in modeling the crack geometry. This is because crack surface definitions
are handled in PHLEXcrack using a powerful geometry engine, a customized
version of the one used in the popular FEM pre/postprocessor Altair's
HyperMesh. Crack data inside the geometry engine is maintained as STL
surfaces (a collection of triangular facets). The computation of geometrical
quantities at the crack front (normal direction, curvature), visibility
calculations (intersection of segments with the crack), and tolerance control
for the accuracy of triangular representation, are handled internally within
the geometry engine. In addition, the geometry engine automatically detects
intersections of multiple crack surfaces and intersections with the boundary
of the domain. Thus, the crack can be completely enclosed within the body,
or can start and propagate along the boundary surface of the domain.
Flowchart of PHLEXcrack
In the dynamic / crack propagation solution
sequence, PHLEXcrack switches between two phases of the solution:
-
solving of the standard dynamic analysis problem,
followed by
-
crack advancement and modification of the
database.
The crack advancement phase consists of the
following steps:
-
calculation of geometrical data at each point
along the active crack front (tangent and normal directions, crack front
curvature, etc.),
-
calculating stress intensity factors and other
relevant quantities at each point along the active crack front,
-
local prediction of crack advancement,
-
modification to the geometry database (crack
advancement, adding / deleting vertices along the crack front to maintain
required accuracy, possible reduction of features of the crack surface(s)
behind the crack front),
-
modification of the EPM database: marking
newly cut cells, changes to the enrichment pattern due to the crack advancement,
and
-
recalculation of the stiffness and mass matrices
for cells which were modified due to the crack movements.
Steps 1 and 4 are handled internally within
the geometry engine. Steps 2 and 3 are specific to the particular theory
of fracture propagation (PHLEXcrack is designed to allow for the
substitution of user supplied modules to accomodate different equations
of the crack motion). The last two steps required non-trivial modification
of the time integration code, and were possible primarily because PHLEXcrack
is built using Altair's proprietary ProPHLEX environment, which was designed
for adaptive simulations.
Examples
During the validation process, we solved several
simple examples of static and dynamic fracture problems, including dynamic
crack propagation problems, and compared the results with solutions available
in the literature. (A detailed report [4] and
published paper [9] are available) We present here
a contrived example showing some of the fracture mechanics related capabilities
of the code: a welded T-section of a beam with a preexisting ``half-penny''
crack:
Welded T-section beam with a crack.
Both the web and the flange are made of
the same linear elastic material with the following properties:
-
Young's modulus = 0.2E12
-
Poisson's ratio = 0.3
-
Mass density = 7833
The Freund propagation model is used to advance
the crack.
-
Dynamic fracture toughness in a pure mode
I, Kid = 75000
-
Crack speed limit = 1212
-
The weld material is assumed to be ten times
stiffer than the bulk material ({ i.e.}, Young's modulus = 2.0E12.
-
A torsional load is applied as two equal and
opposite impact forces at the two corners of the edge cross section, with
a value of 5E6 each.
EPM discretization for welded T-section
beam.
Note than only minimal clustering of the
integration cells has been used in the model. In fact, due to the shape
of the crack, less than 20 elements are initially cut by the crack. Although
the initial crack seems to be aligned with the nodes, PHLEXcrack automatically
detected this situation and distorted all cells in the crack vicinity.
This process is repeated after every crack advancement to avoid cell vertices
being too close to the crack surface.
Following figures show the crack advancement
at two distinct points in time. Note that crack surface was preserved during
the computations (no defeaturing behind the crack front) so the history
of the crack advancement is visible in both figures.
Crack surface of welded T-section example
at time 0.0015
Crack surface of welded T-section example
at time 0.0030
Conclusions
and Suggested Future Work
PHLEXcrack is a very powerful tool to analyze
various fracture mechanics problems, including:
-
stress intensity calculations for various
crack configurations,
-
pseudo--static and dynamic prediction of the
crack growth using linear theory based on the Freund model.
Note that due to the use of the crack definition
independent of the domain discretization, PHLEXcrack can be used to rapidly
analyze various ``what if'' cases with different
crack configurations, without the need to regenerate the model of the structure.
This could significantly reduce the manual effort and timeframe for the
analysis of lifetime prediction for aging parts and structures.
Possible expansions of the code capabilities
may include:
-
implementation of different crack propagation
theories,
-
application to long term fatigue--type crack
propagation,
-
implementation of various fracture models
(including local non--elastic behaviour near the crack front).
PHLEX environment already has most of the
capabilities needed for nonlinear analysis, so these extensions should
be possible with relatively small amount of extra effort.
References
[1]
I. Babuska and J.M. Melenk. The partition of unity finite element method.
Technical Report BN-1185, Inst. for Phys. Sc. and Tech., University of
Maryland, June 1995.
[2]
L.B. Freund, Dynamic Fracture Mechanics, Cambridge University Press,
Cambridge, 1990.
[3]
J.A. DeRuntz, Jr, Underwater Shock Analysis code, Theory Manual. Unique
Software Applications, Colorado Springs, Colorado, 1998.
[4]
T.J. Liszka, C.A.M. Duarte, and O.N. Hamzeh. Hp-meshless cloud method for
dynamic structure in fluid-structure interaction, SBIR-ONR-N00014-96-C-0329.
The Computational Mechanics Company, 1999.
[5]
J.M. Melenk, and I. Babuska, The partition of unity finite element method:
Basic theory and applications, Computer Methods in Applied Mechanics
and Engineering, 139:289-314, 1996.
[6]
D. Shepard, A two-dimensional function for irregularly spaced data. In
ACM National Conference, pp 517-524, 1968.
[7] T.J.Liszka, et al. ProPHLEX - An hp-adaptive finite element kernel
for solving coupled systems of partial differential equations in computational
mechanics, Computer Methods in Applied Mechanics
and Engineering, 150:251-271, 1997.
[8] T.J.Liszka, C.A.M. Duarte, hp-meshless cloud method,
Computer Methods in Applied Mechanics and Engineering, 139:236-288,
1996.
[9] C.A.Duarte, O.N.Hamzeh, T.J.Liszka, W.W.Tworzydlo, A generalized finite
element method for teh simulation of three-dimensional dynamic crack propagation
, Computer Methods in Applied Mechanics and Engineering,
190:2227-2262, 2001.