Modelling crack growth by level sets

M. Stolarska 1 , D.L. Chopp 1 , N. Moës 2 , and T. Beltyschko 3

1
Department of Engineering Science and Applied Mathematics, Northwestern
University, Evanston, IL 60208, U.S.A.
2
Laboratoire de Mécanique et Matériaux, Ecole Centrale de Nantes, Nantes,
France
3
Department of Mechanical Engineering, Northwestern University, Evanston, IL
60208, U.S.A

ABSTRACT: An algorithm which couples the level set method with the extended finite element
method to model crack growth is described. The level set method is used to represent the crack
location, including the location of crack tips. The extended finite element method is used to
compute the stress and displacement fields necessary for determining the rate of crack growth.
This combined method requires no remeshing as the crack progresses, making the algorithm very
efficient. The combination of these methods has a tremendous potential for a wide range of
applications. A numerical example is presented to demonstrate the accuracy of the combined
methods. In addition, a level set algorithm for modelling crack growth in three dimensions is
described.

INTRODUCTION
In this paper, which is a summary of the work presented in [1], we describe an
algorithm where the level set method (LSM) is coupled with the extended finite
element method (X-FEM) to model crack growth. The LSM is a numerical
scheme developed by Osher and Sethian [2] to model the motion of interfaces. In
the LSM the interface is represented as the zero level set of a function of one
higher dimension. The current formulation of the LSM has no provisions for
modelling free moving endpoints on curves. Here, we present an extension of the
LSM for modelling the evolution of an open curve segment and use this extension
to model fatigue crack growth. We also present an extension of the level set
formulation for crack growth in three dimensions.
The X-FEM [3] algorithm enables the modelling of crack growth without
remeshing. Rather than adapting the mesh so that it coincides with the
discontinuity of the crack, X-FEM allows for the crack to pass arbitrarily through
elements by incorporating enrichment functions to handle the field discontinuities.
In this manner the mesh can remain fixed throughout the evolution of the crack.
The LSM and X-FEM work well, offering complimentary capabilities. The
level set representation of the crack simplifies the selection of the enriched nodes,
as well as the definition of the enrichment functions. In addition to modelling the
crack growth problem, the combined methods were also used to model holes and
material inclusions in [4] and three-dimensional planar crack growth in [5]. The
LSM and X-FEM, as described in this paper, provide a simple and efficient
algorithm for modelling two-dimensional crack growth. Moreover, the LSM
provides a simple and natural method for extending the crack growth model into
three dimensions.

GOVERNING EQUATIONS
In this section we review the governing equations for the displacement field in an
elasto-static analysis. The domain of the problem is Ω with boundary Γ. The
boundary Γ is subdivided into two parts, Γu, where the displacement is prescribed,
and Γt, where the traction is prescribed. In addition to the external boundary, the
crack surface presents an additional boundary, Γc, inside Ω. The coincident crack
surfaces are denoted by Γc+ and Γc- and are traction free.
The strong form of the equilibrium equations and boundary conditions is

∇ ⋅ σ + b = 0 in Ω (1)
σ ⋅ n = T on Γt (2)
σ ⋅ n = 0 on Γc+ and Γc - (3)
u = U on Γ u (4)

where σ is the Cauchy stress tensor, u is the displacement, b is the body force per
unit volume, and n is the unit outward normal. The prescribed traction and
displacement are respectively T and U.
The weak form of the equilibrium equation is

∫ ε (u) : C : ε (v)dΩ = ∫ b ⋅ v dΩ + ∫ T ⋅v dΓ ∀ v ∈U0 (5)

Ω Ω Γt
where ε is the strain tensor, C is the Hooke tensor, and v is a test function in the
space of test functions U0 . For details of the transition from strong form to weak
form see [6].

THE LEVEL SET METHOD

The LSM is a numerical technique for tracking the motion of interfaces. In this
method, the interface of interest is represented as the zero level set of a function
φ(x(t),t). This function is one dimension higher than the dimension of the
interface. The evolution equation for the interface can then be expressed as an
equation for the evolution of φ. For our purposes, cracks will be considered as
one-dimensional manifolds in two-dimensional space.
In general, an (n -1)-dimensional interface γ(t) ∈ Rn can be formulated as the
level set curve of a function φ(x,t): Rn ×R→R, where γ (t)= {x ∈Rn :φ(x,t)=0 }.
The motion of γ(t) is translated into an evolution equation for φ. An important
property of the function φ is that it is defined to be the signed-distance to the
interface, i.e. φ ( x, t ) = ±min x γ ∈γ ( t ) x − xγ . Therefore, at a given point in the
domain we know where we are in relation to the interface. We will make use of
this feature in modelling crack growth.
The LSM has typically been used to track interfaces which are either closed
curves or curves that extend to the boundary of the domain. To represent
interfaces that are open curves the level set method needs to be extended. A crack
is represented as the zero level set of a function ψ(x,t), and a crack tip is
represented as the intersection of the zero level sets of ψ(x,t) and φi(x,t). Here, in
the case of more than one crack tip, the subscript i corresponds to the ith tip.

THE EXTENDED FINITE ELEMENT METHOD

Modelling crack growth in a traditional finite element framework is cumbersome
due to the need for the mesh to match the geometry of the discontinuity. Many
methods require remeshing of the domain at each time step. In X-FEM the need
for remeshing is eliminated. The mesh does not change as the crack grows and is
completely independent of the location and geometry of the crack. The
discontinuities across the crack are modelled by enrichment functions.
To illustrate this, consider the X-FEM displacement approximation for a
vector valued function u(x): R2→R2 given by
  4 
u h ( x, t ) =∑ u i (t ) N i ( x)+∑ b j (t ) N j ( x) H (ψ ( x, t ))+ ∑ N k ( x) ∑ a kl (t ) Bl (r ,θ )  (6)
i∈I j∈J k∈K  l =1 
where Ni (x) is the shape function associated with node i and t is the time. In Eq.
6 J is the set of all nodes whose support is bisected by the crack. The set K
contains all the nodes of the elements containing the crack tip. The nodal degrees
of freedom corresponding to the displacement are ui , bj, and ak .
The second important and distinguishing factor to note in Eq. 6 is the
enrichment functions H(ψ(x,t)) and Bl(r,θ).The function H(y) is defined as
 1 for y > 0
H(y) =  (7)
-1 for y < 0

where y=0 is defined to be along the crack. This implies that the discontinuity
occurs at the location of the crack. The branch function Bl is defined by
 θ θ θ θ 
Bl (r,θ ) =  r sin , r cos , r sin sinθ , r cos sinθ  (8)
 2 2 2 2 

where (r,θ) is a polar coordinate system with its origin at the crack tip and θ=0
tangent to the crack at its tip. The above functions span the asymptotic crack tip
solution of elasto-statics, and r sin θ 2 takes into account the discontinuity across
the crack face.
The introduction of the discrete approximation in Eq. 6 into the principle of
virtual work given by Eq. 5 leads to a system of linear equations. The stress
intensity factors are computed using the domain form of the J –integral as
described in [7]. The direction in which the crack will propagate from its current
tip, θc, is obtained using the maximum hoop stress criteria [3].

LEVEL SET ALGORITHM FOR MODELLING CRACK GROWTH IN 2D

We model one-dimensional crack growth in a level set framework by representing
the crack as the zero level set of a function ψ(x,t). An endpoint of the crack is
represented as the intersection of the zero level set of ψ with an orthogonal zero
level set of the function φi(x,t),where i is the number of tips on a given crack. The
values of the level set functions are stored only at the nodes. The functions are
interpolated over the mesh by the same shape functions as the displacement.
The level set function representing the initial crack is constructed by
computing the signed-distance function for the crack. A difficulty in doing this
arises from the fact that, although the crack tip lies within the domain, the level
set function representing the crack must initially be constructed on the entire
domain. To circumvent this problem, the initial crack is extended tangentially
from its tip and the signed-distance function is constructed from this extended
crack. The level set functions that represent the crack tip are initially defined by
φi (x,0) = (x − x i ) ⋅t ,where t is a unit vector tangent to the crack at its tip and xi is
the location of the ith crack tip. The function φi is planar and has a zero level set
which is orthogonal to ψ at the crack tip. The initial level set functions, ψ and φi,
and the representation of the crack are shown in Figure 1.

φ1n = 0

φ=0
ψ>0

φn+1 = 0
φ2 > 0 1

φ1 < 0 φ2 < 0
F
φ1 > 0

ψ<0

φ1 = 0

φ2 = 0

extension
crack

Figure 1: Construction of initial level set Figure 2: Level set function update.
ψ is functions. recomputed only in white region.

An important consideration is that, although the actual crack is embedded

inside the domain, the zero level set of ψ cuts through the entire domain. In the
level set framework, the crack is considered to be the zero level set of ψ where φi
≤ 0 for all φi associated with the given crack. This is consistent with the initial
conditions and will continue to be so as the level set functions are updated.
The evolution of φi and ψ for each crack is determined by the crack growth
direction for the given crack, θc. In each step, the displacement of the crack tip is
given by the prescribed vector F = (Fx,Fy). The magnitude of crack extension
depends on the crack growth law. The current location of the crack tip, xi =
(xi,yi), is also used in the equations of evolution.
Let the current values of φi and ψ at step n be φin and ψ n . The algorithm for
the evolution of the level set functions φi and ψ is as follows.

1. We first rotate φin so that F is orthogonal to the zero level set of φ. φin after
rotation is referred to as φi and given by φi = ( x − xi ) ⋅ F F . This gives the
signed distance to φ=0.

2. The crack is extended by computing new values of ψ n+1 only where φi >0. In
this region the newly computed value of ψ is ψ n +1 = ± ( x − xi ) × F F .
Again, this is the signed-distance to ψ=0. The sign of ψ n+1 is chosen so that
it is consistent with the current sign on a given side of the crack.

3. φin +1 is computed so that it represents the updated location of the crack tip. Eq.
9 is in the form of the general iterative equation of evolution for level set
functions as given in [2].

φin +1 = φi − ∆t F (9)

The rotated level set function φi is calculated exactly. Since φin +1 is calculated
from φi , it is important to note that rather than being updated by an iterative
process, φin +1 is also explicitly recalculated at each step. The recalculation of
φin to φin +1 is illustrated in Figure 2.
The location of the new crack tip i can now be determined by finding the
intersection of the zero level sets of φin +1 and the newly extended ψ n +1 .

COUPLING THE LSM AND X-FEM

The LSM and X-FEM couple naturally to model crack growth. The values of ψ
and φi are stored at nodes. Any information needed for crack growth, such as the
location of the crack tip, can be obtained from these nodal values, making it
unnecessary to store any other information pertaining to the crack. The X-FEM
algorithm is an efficient finite element scheme that solves the elliptical problem
that determines the evolution of a crack on a mesh. The mesh is unchanged
throughout the computation of the evolution of the crack. For these reasons, the
LSM and X-FEM work well together.
Moreover, the level set representation of the crack facilitates the
computation of the enrichment. The Heaviside enrichment function is defined so
that the discontinuity is coincident with the crack, ψ =0. Therefore, to determine
the location of a point relative to the crack one merely has to determine the sign of
ψ at that point. The branch enrichment functions are defined in coordinates local
to the crack tip. Because ψ=0 is orthogonal to φ=0, these two level sets create
this needed coordinate system naturally. Finally, the nodes chosen for enrichment
can be determined by the nodal values of ψ and φ.

EXAMPLE - CRACK GROWTH FROM A FILLET

Initial Simulated
Crack Region
75 ρ = 20

75

375
800

Figure 3: Experimental configuration of crack Figure 4: Crack paths for rigid (upper
crack) and growth from a fillet taken from [8]. and flexible (lower crack) constraint.

This example shows the effect of I-beam thickness of the growth of a crack from
a fillet in a structural member. The configuration of the problem is taken from
experimental work found in [8] and is shown in Figure 3. The computational
domain is outlined by the dashed line. Only the limiting cases of a very thick,
rigid I-beam and a very thin, flexible I-beam are discussed. The effects of the
thickness are incorporated into the problem through the boundary conditions.
 The structure is loaded at the top boundary with a load of P =20 kN. The
initial crack is 5 mm in length. Crack growth was simulated for a total of 12
steps, with each step size of length 5 mm. Figure 4 is a close-up of the mesh in
the vicinity of the fillet. The level set representation of the crack (solid line) is
compared to a piecewise-linear segment representation [9] (shown by X ’s).

LEVEL SET FORMULATION FOR CRACK GROWTH IN 3D

We are currently investigating the three dimensional extension of the level set
algorithm described above. Here we present a formulation that will be coupled
with a boundary element method to model crack growth in three dimensions.
Three-dimensional crack growth has also been modelled by X-FEM and the LSM
in [5] and [10]. The difference between the following algorithm and [5] is that
this allows for crack to grow out of plane, i.e. the cracks are not required to stay
flat. In [10], this out of plane growth is modelled by level set methods. However,
the following scheme will prove to be much more efficient than the iterative
process that is used in [10].
The configuration for the level set functions in three dimensions is shown in
Figure 5. As in the 2D case, the crack face corresponds to ψ (x,t) = 0, and the
crack tip, which is now a one-dimensional manifold, is represented by the
intersection of an additional level set function, φ (x,t) = 0, and ψ (x,t) = 0.
Again, these two function are the signed-distance to their zero level set, φ = 0 ⊥
ψ = 0 , and the actual crack is represented as the set {x :ψ = 0 where φ ≤ 0}.

Φ=0
Φ=0

Ψ>0
µ=||r||
1000111
r11 000
00
11
001010111
11 000
F
Ψ=0 Φ>0
Ψ=0
update
Φ<0
Ω
no update
Ω
Ψ<0

Figure 5: Configuration of level set Figure 6: Geometry of update in 3

functions in 3 dimensions. dimensions.
 The update of level sets in three dimensions follows the same general
procedure as for the two dimensional case and is described below. The extensions
arise in the fact that the two dimensional calculations are in essence done on a
series of planes perpendicular to the crack. In addition, because the velocity at the
crack tip depends on position, it must be extended to the entire computational
domain in order for the level set functions to be updated everywhere. Finally, in
order to find the position of a point in the computational domain relative to the
crack tip an additional level set function must be introduced. This function is
defined as the perpendicular distance to the crack tip, and is denoted as µ(x,t).
The update geometry along with µ(x,t) is shown in Figure 6. Assume that the
values of φ, ψ, and µ at time n are φn, ψn, and µn.
1. φ = 0 ∩ ψ = 0, i.e. the crack tip, is found using methods described in [11],
and µn+1 is computed using a Fast Marching Method (FMM), a method based
on level sets the details of which are given in [12]. The position relative to the
crack tip is given by r = µ ∇µ .

2. We then extend the velocity vector given on the crack tip into the entire
computational domain by an extension of the FMM.
3. φn is rotated so that it is orthogonal to the velocity field on the crack tip.
Given a velocity field F, this is done geometrically and given by φ = r ⋅
F/||F||. This and the following step are taken from the 2D case.

4. The crack is extended by computing new values of φn+1 where φ > 0 (Ωupdate).
This is also done geometrically and given by ψn+1 = ±||r× (F/||F||)||.
5. φn+1 is updated using Eq. 9.
A new crack tip velocity is now calculated using a boundary element method, or
any other method which solves the necessary equilibrium equations, and the
process is repeated. Because the level sets are updated geometrically, it is simple
to implement and efficient.
CONCLUSIONS
The LSM and X-FEM couple naturally to solve the elasto-static fatigue crack
problem. The level set formulation is used to model the crack and update the
crack tip at each iteration. The geometry of the crack is easily represented by two
signed-distance functions whose zero level sets are orthogonal to one another at
the crack tip. These two properties facilitate the calculation of the enrichment
functions. We use X-FEM to solve the elasto-static problem and determine the
direction of crack growth. The example shows that the results obtained by the
level set formulation are comparable to those obtained with a piecewise-linear
segment representation of the crack. The advantages of the LSM and X-FEM in
two dimensions are simple and useful. Moreover, the method presented here can
also be extended to three dimensions.

REFERENCES
1. Stolarska, M., Chopp, D.L., Moës, N., Belytschko, T. (2001) International
Journal for Numerical Methods in Engineering 51, 943-960.
2. Osher, S., Sethian, J.A. (1988) Journal of Computational Physics 79(1), 12-
49.
3. Moës, N., Dolbow, J., Belytschko, T. (1999) International Journal for
Numerical Methods in Engineering 46, 131-150.
4. Sukumar, N., Chopp, D.L., Moës, N., Belytschko, T. (2001) Computer
Methods in Applied Mechanics and Engineering 190, 6183-6200.
5. Sukumar, N., Chopp, D.L., Moran, B. (2002) Engineering Fracture
Mechanics. To appear.
6. Belytschko, T., Black, T. (1999) International Journal for Numerical
Methods in Engineering 45, 601-620.
7. Moran, B., Shih, C.F. (1987) Engineering Fracture Mechanics 27, 615-641.
8. Sumi, Y., Yang, C., Wang, Z. (1995) Technical Report, Department of Naval
Architecture and Ocean Engineering, Yokohama University, Japan.
9. Dolbow, J., Moës, N., Belytschko, T. (2000) Finite Elements in Analysis and
Design 36, 235-260.
10. Gravouil, A., Moës, N., Belytschko, T. (2002) International Journal for
Numerical Methods in Engineering 53, 2569-2586.
11. Chopp, D.L. (2001) SIAM Journal of Scientific Computing 23, 230-244.
12. Sethian, J.A. (1999) SIAM Review 41, 199-235.