Engineering Failure Analysis 47 (2015) 241–251

Contents lists available at ScienceDirect

Engineering Failure Analysis

journal homepage: www.elsevier.com/locate/engfailanal

Phase field modelling of stress corrosion

Per Ståhle a,⇑, Eskil Hansen b
a
Div. of Solid Mechanics, Lund University, Lund, Sweden
b
Centre for Mathematical Sciences, Lund University, Lund, Sweden

a r t i c l e i n f o a b s t r a c t

Article history: The evolution of surfaces exposed to an aggressive environment and mechanical load is
Received 8 September 2013 studied. This is a process of stress corrosion that leads to pitting, crack initiation and grow-
Received in revised form 30 July 2014 ing cracks. In conventional fracture analyses a known or a postulated crack is required. A
Accepted 31 July 2014
serious complication is that a large part of the lifetime of a crack or a surface flaw is spent
Available online 7 September 2014
during the initiation of the crack. The knowledge of the mechanisms leading from a pit,
flaw, scratch, etc. to a crack is very limited. The motivation for the present study is to pro-
Keywords:
vide a model that will increase the understanding of the transition from stress induced sur-
Stress corrosion
Crack initiation
face roughening and pitting to growing cracks.
Surface morphology The evolution of the originally flat surface involves free strain energy, chemical energy
Phase field model and gradient energy. A phase field model is used to capture the driving forces that the free
Anti-plane strain energy causes. The flat surface is unstable and develop a waviness. Initially while the waves
are shallow a spectrum of favoured spatial frequencies are found to be in accordance with
the Asaro–Tiller–Grinfeld theory. Later the surface curvature becomes larger at the depres-
sions than at the higher parts of the surface. This increases the growth rate of formed pits.
The pits finally develop into cracks. Also massive branching of pits and cracks is observed.
Ó 2014 Elsevier Ltd. All rights reserved.

1. Introduction

A strained body exposed to an aggressive environment may suffer from continuous material dissolution during corrosion.
The result is a progressive roughening of its surfaces. After roughening pitting occurs and eventually cracks will form and
grow into the body. Apart from an aggressive environment and a mechanical stress the material needs to have an inherent
sensitivity or to be sensitised to the environment. This can be due to local heating, plastic deformation, fatigue damage, etc.
The aggressive environment can be a bulk aqueous environment surrounding the body or a micro environment, such as
moist in pits, crevices, under deposits and not seldom very local environments created by microbial colonies established
on the body surface.
The corrosion process produces a thin film of metal oxides or hydroxides or compounds thereof. Even though the thick-
ness of this film is typically not more than a few nanometres, it reduces the rate of dissolution by several orders of magni-
tude. If kept intact, this protective film would increase the service life of an exposed structure tremendously. However,
occasionally or repeatedly the film is damaged as a result of variation in load, electro-chemical environment or, as was
recently discovered, by microbiological activity. Some microbes may even have the oxide film or the substrate as an essential
component of their metabolism (cf. [1,2]).

⇑ Corresponding author.
E-mail addresses: per.stahle@solid.lth.se (P. Ståhle), eskil@maths.lth.se (E. Hansen).

http://dx.doi.org/10.1016/j.engfailanal.2014.07.025
1350-6307/Ó 2014 Elsevier Ltd. All rights reserved.
242 P. Ståhle, E. Hansen / Engineering Failure Analysis 47 (2015) 241–251

The focus here is on the evolving surface morphology and initiation of cracks caused by the corrosion. The mechanism is
the mass transportation resulting after dissolution and diffusion of matter into the environment or surface diffusion (see
Fig. 1). The strain energy and the chemical energy provide driving forces for dissolution and transport of molecules. The
observation is an evolving surface waviness that has been explained theoretically by [3–5]. The spectrum of the waves
depends on the strain energy of the body surface and the surface energy. The elastic strain energy tends to increase the wav-
iness of the surface whereas the surface energy tends to decrease the waviness. Waves with wavelengths longer than a stress
dependent critical wavelength grow, while waves with shorter wavelengths decay. Experimental results by Kim et al. [5]
show that the wavelength in, e.g., aluminium is on the scale of a few hundred nanometres when the stress is large and com-
parable to the yield stress.
Several non-linear analyses show that the indents grow markedly faster and the peaks of the surface grow slower. This
sharpens the indents and leads to a formation of pits. The prediction is that the indents eventually approach a singular state,
where the deepest part of the indents locally assumes the geometry of a sharp crack tip or a cusp. In earlier studies [6,7] the
dissolution was assumed to be a function only of the stretching of the surface. A threshold strain was invoked to comply with
the limited elastic properties of the oxide film. It was then predicted that the indent sharpens and become crack like. There-
after, growth continues in the form of a blunted crack or, in a strict sense a deep notch with a relatively sharp tip. Fig. 1(b)
shows a real stress corrosion crack in a nuclear pressure vessel. The crack tip always preserves a finite radius that may
depend on the crack tip driving force and the thermodynamic chemical and mechanical properties of the surface. Further,
the crack frequently branches and the crack width becomes slightly larger as the crack gets longer.

2. Phase field modelling

For a virtually sharp surface, treated as a discontinuity, specific chemical and mechanical properties may be ascribed to
the surface. This simplifies the analysis and has normally no significant influence on the thermodynamical behaviour of the
body. However, when the distance between structural inhomogeneities or other characteristic distances are of the same
order of magnitude as the thickness of the surface, e.g., as at a rough surface, at the tip of a crack or during emission of dis-
locations, etc., the thickness of the surface may play a role and the negligence thereof may lead to unrealistic predictions. As
opposed to this, a diffuse surface model assumes a continuous variation of composition, structure and other properties
within the modelled region (cf. Landau and Lifshitz [8]). This includes the body, the diffuse surface and the surrounding envi-
ronment. Here, the total free energy of the continuous body is formulated as a function of the material composition. The
width of the surface layer is not assumed to be fixed, but instead the model predicts surface layers with a finite width
and the associated surface energy is a result of the thermodynamical state of the material composition of the surface. A dif-
fuse interface model was first applied by Cahn and Hilliard [9], to study the thermodynamics of a coherent interface between
two phases. Their predictions include the width of the interface and the corresponding interfacial energy.
For a general case two types of so-called order parameters are introduced to model the compositional and structural
non-uniformities. An order parameter is defined as a continuous field that is referred to as a phase field. The total free energy

Fig. 1. (a) Corrosion crack penetrating a bimaterial interface between austenitic steel and pressure vessel steel of type SA533C11. (b) The tip of one of the
crack branches. Note that the widest part of the crack is at the crack tip. Crack length is 7 mm and notch width is around 10 lm. Reproduced with
permission from Vattenfall AB, Sweden. Typical wavelength in, e.g., aluminium is on the scale of a few hundred nanometres when the stress is large and
comparable to the yield stress.
 P. Ståhle, E. Hansen / Engineering Failure Analysis 47 (2015) 241–251 243

of the system is formulated as a function of the phase fields. The variation of the free energy with respect to the fields act as
driving forces for the evolution over time, following the Cahn–Hilliard equation for the conserved order parameters and the
Ginzburg–Landau equation for the non-conserved order parameters. In the proposed study an order parameter, w, is used to
distinguish empty space from the solid and to model the properties of the surface of the body.
The goal of this paper is to use relevant physical properties of the surface layer so that the evolution of the corroding body
can be captured. The modelling results in the formation of pits, initiation and growth of cracks and crack branching despite
that no criteria is applied for any of the mentioned events. The tendency to unstable surface behaviour is inherent in the
model and thus no criteria are needed. The body is assumed to be isotropic with a surface energy defined for a planar surface
in the selected environment. Further, the body is assumed to remain in mechanical equilibrium. The tendency of the surface
to change the shape of its reference configuration is governed by a chemical potential. As the surface configuration is varied,
the positive surface energy tends to flatten the surface while the positive strain energy is inclined to roughen the surface.
The fundamental problem is for a semi-infinite body with a planar surface representing the initial structure. A Cartesian
coordinate system x1 ; x2 and x3 is giving a two dimensional representation of the body, in the plane x3 ¼ 0. The initial plane
surface coincide with x2 ¼ 0, dividing the space into the half plane, x2 6 0, initially covered by the body and the halfplane,
x2 < 0, that is empty. The transition region between the empty space and the solid body, does not have to be small compared
to the extent of the considered geometry. However, here the material parameters are selected so that region become small.
This is to make a comparison between the phase-field and known results for a distinct surface, e.g. [3,5,6] possible. The evo-
lution of the body is obtained by minimising the free energy. This is achieved by formulating the total energy as a functional
of the phase field. From this the kinetics is derived along the steepest descent path of the total energy, using the time-
dependent Ginzburg–Landau equation. In the present study the total energy of the structure consists of an elastic strain
energy, a chemical energy and a gradient energy. The variations of these energies are assumed to be the only driving forces
evolving the surface morphology.
The solid material is assumed to be linear elastic with the elastic modulus Eo and Poisson’s ratio m. The energy of the sys-
tem is composed of the Landau chemical potential energy, the gradient energy and the elastic energy. The respective energy
densities are, F ch ; F gr and F el . The total energy density is written
F ¼ F ch þ F gr þ F el : ð1Þ
Here the Landau chemical potential energy density is defined by
F ch ¼ UðwÞ; ð2Þ
where UðwÞ is obtained from a phase diagram. The compositional parameter w is defined in the interval jwj  1, where w ¼ 1
denotes the massive body and w ¼ 1 denotes empty space. Therefore a double-well Landau potential with the require-
ments that U 0 ð1Þ ¼ 0 and U 00 ð1Þ > 0 is selected. The simplest polynominal would be the symmetric double-well potential
 
1 1
UðwÞ ¼ p w4  w2 ; ð3Þ
4 2
where p is a temperature dependent material parameter. In the present study the temperature is assumed to be constant.
The parameter p tend to decrease thickness of the surface. Here a simple selection is needed for numerical reasons. Further,
the choice is attractive because there is an analytical solution for a straight edge by Ginzburg and Landau [10] using (3). In an
analysis with the objective to study the detailed physics of the surface region the chosen polynominal could be insufficient.
The gradient energy density is given as follows
gb
F gr ¼ ðrwÞ2 : ð4Þ
2
The material parameter, g b , tend to increase the thickness of the surface. The term F gr as defined by (4) provides a driving
force for diffusion against the gradient of the concentration, rw. It might be of interest to note that the normalised concen-
tration or the density is proportional to ðw þ 1Þ=2.
The elastic modulus is a function of the composition, i.e., E ¼ EðwÞ in the entire modelled region. It should approach
the elastic modulus of the solid, Eo , as w ! 1 and vanish as w ! 1 and the thermodynamic driving force should vanish
as jwj ! 1 which leads to the requirement E0 ðwÞ ! 0 as jwj ! 1. The simplest polynominal approach that fulfil the
requirements is
1
EðwÞ ¼  ðw3  3w  2ÞEo : ð5Þ
4
Poisson’s ratio m is assumed to be independent of w. The elastic deformation of the virtually empty space occurs at insignif-
icant stress. To simplify the analysis anti-plane deformation is assumed. Strains are assumed to be small and rotation is
assumed only to give an insignificant contribution to the free energy. Thus, the only remaining displacement component
is u3 and the elastic strain energy density is given by

EðwÞ
F el ¼ ðru3 Þ2 ; ð6Þ
4ð1 þ mÞ
244 P. Ståhle, E. Hansen / Engineering Failure Analysis 47 (2015) 241–251

where the non-zero strains are given by the displacement gradients as

 2  2
@u3 @u3
ðru3 Þ2 ¼ þ ¼ c213 þ c223 ¼ c2 :
@x1 @x2

The Ginzburg–Landau equation, applicable for non-conserved quantities, is used for matter in this case. This is because the
corrosion is assumed to be a consuming process at which matter disappear. Thus, the evolution of the composition param-
eter w is given as the rate @w=@t proportional to the variation of the free energy density, as follows:

@w dF
¼ Lw ; ð7Þ
@t dw

where Lw is a phase mobility parameter (cf. Cahn and Hilliard [9]). According to Lagranian formalism the variation is given by,

dF @F @F
¼ r : ð8Þ
dw @w @ðrwÞ

For the order parameter w and application of (7), insertion of (3), (2) and (4) into (8) gives the governing equation for the
evolution of w,
  
@w 3Eo
¼ Lw ðru3 Þ2  pw ð1  w2 Þ  g b r2 w : ð9Þ
@t 16ð1 þ mÞ

The right hand side of (9) represents three different phenomena, i.e. the strain energy that enhance the dissolution of mate-
rial, the chemical energy, so-called up-hill diffusion that is the manifestation of the surface energy, and finally the gradient
energy that represents diffusion according to Fick’s law. The material parameters Eo ; p and g b give the magnitude of the
respective phenomenon. The product Lw g b corresponds to the diffusion coefficient of Fick’s second law.
Applying (8) for the displacements gives the following governing equation,

@u3 Lu
¼ r  ½EðwÞru3 ; ð10Þ
@t 2ð1 þ mÞ

where Lu is a material mobility parameter. Eq. (10) describes a diffusive process that drives the mechanical system towards
static equilibrium. Putting @u3 =@t ¼ 0 gives

@E
rw  ru3 þ EðwÞr2 u3 ¼ 0: ð11Þ
@w

The equation is recognised as the equation of static equilibrium for an inhomogeneous material at anti-plane deformation.
An analytical solution is obtained for this system for a geometry representing a straight edge. The analytical solution satisfies
the Eqs. (9) and (11) that are solved for given initial conditions for w and boundary conditions for w and u3 . For a wavy surface
one has to rely on a numerical solution. To make the numerical analysis more efficient (11) is replaced with (10) and a suf-
ficiently large mobility parameter Lu is chosen. The initial conditions of w define the initial shape of the body.

3. The model

Consider a large body with a traction free wavy edge, with a wave amplitude x and the wavelength k (see Fig. 2). The
body initially occupies the region x2 6 x sinð2px1 =kÞ and the remaining part is defined as empty space. Strictly, this would

Fig. 2. Wavy initial body surface with wavelength k and wave amplitude x ¼ 0:1k.
 P. Ståhle, E. Hansen / Engineering Failure Analysis 47 (2015) 241–251 245

imply that w ¼ 1 in the region covered by the body and w ¼ 1 in the remaining body. However, this would not meet the
requirement that w is continuous. To avoid the inconsistency, the initial field w is chosen to be
 
x2 þ x sinð2px1 =kÞ
w ¼  tanh ; ð12Þ

where the parameter  determines the thickness of the layer where w shifts from near 1 to near 1. A remote shear @w=@x1 is
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
prescribed as x21 þ x22 ! 1. Because of the periodicity, the problem is reduced to one for an infinite strip with the width
jx1 j  2k. The following boundary conditions are applied:
@w k
¼ 0 and u3 ¼ kco =2 at x1 ¼ ; ð13Þ
@x1 2
@w k
¼ 0 and u3 ¼ kco =2 at x1 ¼  ; ð14Þ
@x1 2
w ! 1 and c23 ! 0 as x2 ! 1; ð15Þ
and
w ! 1 and c23 ! 0 as x2 ! 1: ð16Þ
As a result of the vanishing shear modulus in the space outside the body, the body surface becomes traction free. The
initial straining of the body is co which gives the initial elastic energy density
1
F el ¼ F o ð2  wÞð1 þ wÞ2 ; at t ¼ 0; ð17Þ
4
where F o ¼ Eo c2o =½4ð1 þ mÞ.

3.1. Exact solution for a dissolving straight edge

A uniaxial solution for chemical equilibrium in the absence of mechanical load is given by [10]. For x=k ¼ 0 in (9) the
initial surface waves disappear and the problem becomes independent of the x1 coordinate.
x2 ¼ aðx2 þ ctÞ, is used. After introducing
To simplify the solving of (9) a Galilean coordinate transformation, ^
wðx2 ; tÞ ¼ f ð^x2 Þ; ð18Þ
insertion of (18) into (9) results in the following

acf 0 2 00
¼ ð3F o  pf Þðf  1Þ þ g b a2 f : ð19Þ
Lw
pffiffiffiffiffiffiffiffiffiffiffiffiffi
Without loss of generality one may put a ¼ p=2g b which simplifies (19) to
1 00 0 2
f  bf þ ðj  f Þðf  1Þ ¼ 0; ð20Þ
2
pffiffiffiffiffiffiffiffiffiffiffi
where b ¼ c=ðLw 2g b pÞ and j ¼ 3F o =p.
In the absence of external load, i.e., j ¼ 0, a stationary solution is found by putting c = 0 which implies that b ¼ 0. Then
(20) is reduced to
1 00 2
f  f ðf  1Þ ¼ 0; ð21Þ
2
with the known solution
f ¼ f o ¼  tanhðax2 Þ; ð22Þ
(c.f. [10]).
A solution for a non-vanishing load, i.e., for an arbitrary j, is obtained after observing that f 0o ¼ f 2o  1. One obtains
1 00 0 2
f þ ðj  bÞf o  f o ðf o  1Þ ¼ 0: ð23Þ
2 o
For b ¼ j, (23) is solved by
f o ð^x2 Þ ¼  tanhðax2 þ ctÞ: ð24Þ
This implies that the edge of the body is moving steadily in the negative x2 direction with the speed
Lw
c ¼ co ¼ 3 Fo: ð25Þ
a
246 P. Ståhle, E. Hansen / Engineering Failure Analysis 47 (2015) 241–251

3.2. Perturbed solution for a dissolving wavy edge

The linear relation ship (25) between dissolution rate and strain energy is expected to be valid also for a wavy surface as
x=k ! 0. Initially wavy surfaces are expected to give dissolution rates that vary along the surface because of the variation of
stress that is caused by the wavy boundary. The stress is concentrated to the depressions in the surface which would give
higher dissolution rates and lower dissolution rates at the peaks of the surface. Counteracting this is the gradient energy that
tend to decrease the peaks and elevate the depressions making the surface flatter. This leads to a competition that is depen-
dent on the wave number k ¼ 2p=k.
It is known that waves with an over-critical wave number decay whereas waves with lower wave number increase in
amplitude (see e.g. [3,5,6]). The evolution of the wave amplitude depends on the surface energy, the elastic energy and
the wave number. The surface energy, /, is a manifestation of the free energy connected to the body surface. It is found
by integration of (22) with the following result
2 pffiffiffiffiffiffiffiffiffiffiffi
/¼ 2g b p: ð26Þ
3
The growth rate is
Lw /
x_ ¼ d ðk  kc Þk; ð27Þ
a2
where the non-dimensional constant d is used for calibration. The critical wave number, kc , is known to be
Fo
kc ¼ ; ð28Þ
2/
cf. [3,5].
The exact solution (27) for a shallow wavy edge is used to qualify the numerical implementation. The solution is asymp-
totically correct as a ! 0 giving a vanishing width of the boundary region versus the amplitude of the surface waves, which
in turn should be vanishing as compared with the wave length.

4. Implementation

The attempt is to solve the quasi-static problem that is given by Eqs. (9) and (11). For numerical reasons an artificially
dynamic problem is formulated using (10) with a very large value for the mobility of displacements as compared with
the mobility of the phase.
The Ginzburg–Landau Eqs. (9) and (10) constitute a system of nonlinear parabolic equations. Approximations of the solu-
tions for given initial conditions and boundary conditions will be obtained by combining a semi-implicit time stepping
scheme with dimension splitting.
In order to apply the numerical scheme, the Eqs. (9) and (10) are rewritten as follows:

@w

¼ cðD2x1 þ D2x2 Þw þ F w; ðDx1 u3 Þ2 þ ðDx2 u3 Þ2 and ð29Þ

@t
@u3    
¼ Dx1 GðwÞDx1 u3 þ Dx2 GðwÞDx2 u3 Þ ; ð30Þ
@t
where Djxi ¼ @ j =@xji . To obtain a bounded computational domain, the spatial domain is truncated at x1 ¼ k=2; x2 ¼ 10, and
rescaled into the unite square ½0; 12 . The boundary conditions along the sides x2 ¼ 0 and 1 are then given as
@w @u3
¼ ¼ 0: ð31Þ
@x2 @x2
2
Next, consider the spatial grid f0; h; 2h; . . . ; ðM  1Þh ¼ 1g and the spatial approximations wi;j ðtÞ  wðih; jh; tÞ and
ui;j ðtÞ  u3 ðih; jh; tÞ. The evolution of the spatial approximations are obtained by replacing the terms with spatial derivatives
in the Ginzburg–Landau system by their standard central difference discretisations, e.g.,
1  1 
D2x1 wðih; jhÞ  2
wiþ1;j  2wi;j þ wi1;j ; Dx1 u3 ðih; jhÞ  uiþ1;j  ui1;j
h 2h
and
  1     
Dx1 GðwÞDx1 u3 ði; jÞ  2 Gðwiþ1;j Þ þ Gðwi;j Þ ðuiþ1;j  ui;j Þ  Gðwi;j Þ þ Gðwi1;j Þ ðui;j  ui1;j Þ : ð32Þ
2h
Collecting the approximations wi;j and ui;j in the two vectors W and U, respectively, yields a system of ordinary differential
equations:
 P. Ståhle, E. Hansen / Engineering Failure Analysis 47 (2015) 241–251 247

@W

¼ cðAx1 þ Ax2 ÞW þ F W; ðBx1 UÞ2 þ ðBx2 UÞ2 and ð33Þ

@t
@U 
¼ C x1 ðWÞ þ C x2 ðWÞ U þ b; ð34Þ
@t
where Axi ; Bxi and C xi ðWÞ are matrices and the constant vector b is due to the inhomogeneous Dirichlet boundary conditions
imposed on u3 .
One can proceed by discretising in time over the interval f0; k; 2k; . . . ; Nk ¼ Tg by employing the semi-implicit Euler
scheme. The full approximations WniþMj  wðih; jh; nkÞ and U niþMj  u3 ðih; jh; nkÞ are then given by the linear recurrence
equations
 

2 2
I  ckðAx1 þ Ax2 Þ Wnþ1 ¼ Wn þ kF Wn ; ðBx1 U n Þ þ ðBx2 U n Þ and ð35Þ
n h io
I  k C x1 ðWnþ1 Þ þ C x2 ðWnþ1 Þ U nþ1 ¼ U n þ kb: ð36Þ

To speed up the implementation we also employ a dimension splitting strategy. This implies that the above recurrence equa-
tions are replaced by

2 2
ðI  ckAx1 ÞðI  ckAx2 ÞWnþ1 ¼ Wn þ kF Wn ; ðBx1 U n Þ þ ðBx2 U n Þ and ð37Þ

I  kC x2 ðWnþ1 Þ V ¼ U n ; I  kC x1 ðWnþ1 Þ U nþ1 ¼ V þ kb: ð38Þ

Note that the splitting approach only requires the solution of linear equation systems involving block diagonal matrices, i.e.,
the equation systems decouples into several simpler problems which can be solved in parallel.
The desired goal is to fulfil mechanical equilibrium, i.e. @u3 =@t ¼ 0. This is solved by using a large mobility constant Lu so
that close to mechanically static or quasi-static conditions are achieved.

5. Results and discussion

First the problem for a straight edge is studied, i.e. x=k ¼ 0 without mechanical load. The length parameter  is set to 1
which is 10% of height of the strip. This is assumed to be sufficient to make the boundary region, where w essentially switches
from 1 to 1, small so that its size has little influence on the result, e.g., (24), (25) and (28). The inverse length a is put to 200.
The desired stationary state was assumed after around 10 time increments. During the process the width of the boundary
region decreased from 5.3 to 0.0106 if the region is supposed to be limited by jwj < 0:99 and to 0.0152 if the limit is set to
jwj < 0:999. The boundary regions then have a width of 5:3=a and 7:6=a respectively. The width may be compared with
the height and width of the mesh that is 200=a. The resulting variation of the phase field w with respect to the x2 coordinate
is shown in Fig. 3. The result from two different mesh sizes are compared. For the courser mesh the resolution is around 5
nodal points covering the boundary region in which the phase undergo the transition from solid body to virtually empty
space. With the finer mesh, the boundary region is covered by around 10 mesh elements. The markers showing the nodal
results indicate the mesh resolution. Included in the figure is the analytical result (24) by Ginzburg and Landau [10].
As observed the finer resolution gives rather accurate results. About 10 elements covering the boundary region seems to
be sufficient to achieve a good accuracy. Also the courser mesh gives a sufficiently correct result with a shift of the position of
the boundary a distance as little as 1=a or 0.05% of the body height. Even so, the higher resolution is selected for the remain-
ing study.
Second, an applied stress resulting in a dissolution of material at the edge of the body is studied. The dissolution means a
propagation of the edge, i.e. the propagation of the contour w ¼ 0 as given by (25). In Fig. 4 the numerical result is plotted

Fig. 3. The phase w across the boundary layer region. Analytical solution (22) (cf. [10]) and numerical result with course () and fine ( ) resolutions.
Mechanical load is not applied.
248 P. Ståhle, E. Hansen / Engineering Failure Analysis 47 (2015) 241–251

versus the remote strain energy density (as ax2 ! 1). As observed the numerical result follows closely the analytical pre-
diction that the dissolution rate is proportional to the remote strain energy density F o .
For a study of a wavy surface the initial wave amplitude is set to x ¼ 0:1k. The initial value of  in (12) is 0.002 which
gives a boundary width of around a tenth of the amplitude. The wavy surface leads to a variation of the dissolution rate along
the body boundary because of the variation of stress that arise when the non-straight boundary is stretched. The dependence
on the wave number k is given by (27).
Waves with an over-critical wave number are predicted to decay whereas waves with lower wave number increase in
amplitude as is known from analyses of bodies with sharp boundaries [6,5]. As observed in Fig. 5, the phase field model have
the same features. The result shows that for a wave number larger than the critical wave number, kc , the growth rates of the
wave amplitudes are negative and with a maximum growth rate for k  0:6kc . For the analytical result (27) the constant d is
calibrated to match the derivative @ x_ =@k at k ¼ kc . The calibration gives d ¼ 1:28. The result is included in Fig. 5. The agree-
ment with the analytical result is reasonable for wave numbers 0:5kc < k 6 kc . For longer waves with k < 0:5kc , the deviation
is larger and for wave numbers less than 0:3kc the system could not be computed because of numerical problems.
With regard to the Eqs. (9) and (10) the time scale,
1
ts ¼ ; ð39Þ
Lw p
and the length scale
sffiffiffiffiffiffiffiffi
2g b
‘s ¼ ; ð40Þ
p

are utilised. After normalisation of the displacements and spatial coordinates with respect to ‘s and the rates of displace-
ments and the phase with respect to t s one is left with the following transformed equivalents to (9) and (10):
^ h i 
@w ^u
¼  k1 ðr ^ 3 Þ2  w ^ 2Þ  1 r
^ ð1  w ^ 2w
^ ð41Þ
@^t 2

Fig. 4. Dimensionless dissolution rate c as a function of the dimensionless remote elastic strain energy density F o close to the surface due to an applied
mechanical load. The rate is taken as the propagation rate of the contour w ¼ 0. The markers show the numerical result. The dashed line shows the
analytical result (25).

_ of the surface waves as a function of their spatial frequency k ¼ 2p=k. The dashed curve is the growth rate
Fig. 5. The growth rate of the amplitude x
according to (27).
 P. Ståhle, E. Hansen / Engineering Failure Analysis 47 (2015) 241–251 249

and

@u^3 h i
^  ðw3  3w  2Þr
¼ k2 r ^u^3 ; ð42Þ
^
@t
where

3Eo Lu Eo 1
k1 ¼ and k2 ¼ pffiffiffiffiffiffiffiffiffiffiffi ð43Þ
16ð1 þ mÞp Lw 8ð1 þ mÞ 2g b p

The latter parameter is chosen to k2 ¼ 142, which is supposed to be large enough to provide quasi-static conditions. This is in
the sense that the displacement evolution is much faster than the evolution of the phase. It also makes the solution reason-
ably independent of k2 .
An attempt to compute the subsequent evolution of the wavy surface a calculation was performed for k1 ¼ 1. The bound-
ary conditions given by co ¼ 2 and the initial wave amplitude is chosen to ‘s =100. They are imposed on x1 =‘s ¼ 0 and 200‘s
respectively
pffiffiffi on the boundaries at x2 =‘s ¼ 0 and 200. The effective stress according to von Mises effective stress is given by
re ¼ 3cEðwÞ=2ð1 þ mÞ. A constant time step is used and is set to 0:001ts . The load is applied instantly in the beginning of the
calculations.
Fig. 6a and b shows the phase field and the corresponding effective stress distribution. As the figure shows, the surface
waves develop into deep pits that tend to grow in perpendicular to the load direction. This is opposed to the normally
observed single pit or crack that take the lead and causes the other pits to stop because of an insufficient driving force. Here
it seems as if a decrease of the driving force results in a thinner pit that would increase the stresses at the pit bottom which in
turn will increase the dissolution rate. This seems to have a stabilizing effect on the growth rates to some extent impeding
any single notch from advancing ahead of the others.
In the analysis, electro-chemical dissolution or general corrosion of materials has been ignored. Here a supposedly con-
stant dissolution rate of the material that occur irrespective of mechanical load is included. To add this to the original phase
rate @w=@t, given by (9), a phase gradient term is introduced according to the following
@wg @w
¼  #Lw F o ðrwÞ2 ; ð44Þ
@t @t
where # is an arbitrary dimensionless constant. As is readily seen # ¼ 0 preserves the former result.
The conditions are the same as for the previous case, i.e. k1 ¼ 1. The initial conditions and the boundary conditions are
also the same. The time step is chosen to 0:005ts . A comparison with the previous calculation that was identical apart from
the five times shorter times step and that it covered only t  2t s shows no significant differences.
Fig. 7a shows that growing pits form cracks when the general corrosion is absent, i.e., for # ¼ 0. In Fig. 7a one observes
that the initial notches increase in width during the growth. As the bottom of the notches become wider another observation
is a tendency to branching. Incipient branching is already observed at t ¼ 2ts and more clearly at t ¼ 4t s and at 20t s . A

Fig. 6. (a) The morphology of the unstable surface after the time t ¼ 2t s . Red shows
pffiffiffi the solid material and blue is empty space. (b) The corresponding
distribution of von Mises effective stress re =rrem , where the remote stress rrem ¼ 3co Eo =2ð1 þ mÞ. At the bottom of the deep notches the effective stresses
are observed to be around 2:7rrem . (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this
article.)
 250 P. Ståhle, E. Hansen / Engineering Failure Analysis 47 (2015) 241–251

Fig. 7. Simulation of stress driven dissolution. Time is increasing from top to bottom. In column (a) the phase field displays the result of pure stress
corrosion (# ¼ 0, see Eq. (44)) and in (b) the phase field displays the result when general corrosion is added (# ¼ 1).

hypothesis could be that the wider traction free area at the bottom of the notch at some point provide sufficient space for
two surface waves that are in the range of wavelengths that supported to initiate the shape instability that was reported for
flat surfaces (see Fig. 7). Apart from the wider area also the spectrum of waves that initiate surface instabilities cover shorter
wavelengths at the plane surface because of the increased stresses at the bottom of the notch.
Another observation is that the parts of the growing notches and cracks are getting smoother with increasing time. The
effect is not totally surprising because of the expected absence of stress along the crack surfaces. Therefore, in the wake
behind the propagating tip of both notches and cracks only chemical and gradient energies are present. When the strain
energy density is small or absent the critical wave number decrease or vanishes. This means that only long wave perturba-
tions of the surface grow and most likely the short features that occur at e.g. at t ¼ 2ts has become more vague at t ¼ 4t s and
disappeared at t ¼ 20ts .
To explore the effect of general corrosion, that phase field was computed using (44) for an initially slightly wavy surface.
Conditions are identical to the previous calculation apart from the parameter # ¼ 1. The result is displayed in Fig. 7b. For this
choice of # the general contribution to the dissolution rate is comparable in magnitude to the dissolution caused by the strain
energy.
The variation of notch shapes is present for the case without general corrosion but even more obvious in this case. The
observation gives an indication of the sensitiveness to smaller disturbances such as initial surface roughness and environ-
mental and material variations.
One might have expected a slower crack/notch growth rate because of the more blunt bottoms of the respective. The
bluntness is expected to mitigate the stress concentration but seemingly the added aggressiveness caused by the added gen-
eral corrosion. Without general corrosion, judging from the tip widths at, e.g., t ¼ 20t s pthat
ffiffiffi are around three times less than
the case with general corrosion should lead to an increase of local stress that is around 3 times larger (cf. [11]). By virtue of
(25) the growth rate of the notch should then be 3 times larger for the case without general corrosion. The result is the oppo-
site which indicates a competition between the effect of locally decreased load effect of increased environmental
aggressiveness.

6. Conclusions

Stress corrosion can be modelled as a moving boundary problem using a phase field model. The phase field modelling
simplifies the analysis and the tracking of the moving boundary. Formation of cracks and crack growth are captured.
An analytical solution for a steady state dissolving flat surface is derived. The instability of a flat surface subjected to
mechanical load that is in agreement with results of analyses for models with distinct body surfaces.
The governing Eqs. (9) and (11) are deterministic but rather sensitive to initial conditions. The exact shape of the body
surface after some time may be very different but the general character of the body shape is preserved if the initial
conditions are slightly altered.
 P. Ståhle, E. Hansen / Engineering Failure Analysis 47 (2015) 241–251 251

Acknowledgement

Support from The Swedish Research Council under 621-2011-5561 and 621-2011-5588 is acknowledged.

References

[1] ASME, American Society of Mechanical Engineers. Materials handbook, corrosion, vol. 13. Academic Press; 2005.
[2] Beech IB, Sunner J. Curr Opin Biotechnol 2004;15:181–6.
[3] Asaro RJ, Tiller WA. Metall Trans 1972;3:1789.
[4] Grinfeld MA. Sov Phys Dokl 1986;31:831;
Grinfeld MA. J Nonlinear Sci 1993;3:35.
[5] Kim KS, Hurtado JA, Tan H. Phys Rev Lett 1999;83:19.
[6] Stahle P, Bjerken C, Jivkov AP. Int J Solids Struct 2007;44:1880–90.
[7] Jivkov AP, Stahle P. Eng Fract Mech 2002;69(18):2095–111.
[8] Landau L, Lifshitz E. Phys Zeit Sowjetunion 1935;8:153.
[9] Cahn JW, Hilliard JE. J Chem Phys 1958;28:258.
[10] Ginzburg VL, Landau LD. Zh Eksp Teor Fiz 1950;20:1064.
[11] Neuber H. Theory of notch stresses: principles for exact calculation of strength with reference to structural form and material. Berlin: Springer Verlag;
1958.