metals

Article
Finite Element Based Physical Chemical Modeling of
Corrosion in Magnesium Alloys
Venkatesh Vijayaraghavan 1, *, Akhil Garg 2 , Liang Gao 3 and Rangarajan Vijayaraghavan 4
1 School of Engineering, Monash University Malaysia, 47500 Bandar Sunway, Selangor Darul Ehsan, Malaysia
2 Department of Mechatronics Engineering, Shantou University, Shantou 515063, China; akhil@stu.edu.cn
3 The State Key Laboratory of Digital Manufacturing Equipment and Technology, Huazhong University of
Science and Technology, Wuhan 430074, China; gaoliang@mail.hust.edu.cn
4 Materials Engineering Department, KOP Surface Products Singapore Pte Ltd, 77 Science Park Drive 118256,
Singapore; rangarajanv@gmail.com
* Correspondence: ve0005an@e.ntu.edu.sg; Tel.: +60-355-159-711

Academic Editor: Hugo F. Lopez

Received: 18 January 2017; Accepted: 28 February 2017; Published: 7 March 2017

Abstract: Magnesium alloys have found widespread applications in diverse fields such as aerospace,
automotive, bio-medical and electronics industries due to its relatively high strength-to-weight ratio.
However, stress corrosion cracking of these alloys severely restricts their applications in several
novel technologies. Hence, it will be useful to identify the corrosion mechanics of magnesium
alloys under external stresses as it can provide further insights on design of these alloys for critical
applications. In the present study, the corrosion mechanics of a commonly used magnesium alloy,
AZ31, is studied using finite element simulation with a modified constitutive material damage model.
The data obtained from the finite element modeling were further used to formulate a mathematical
model using computational intelligence algorithm. Sensitivity and parametric analysis of the derived
model further corroborated the mechanical response of the alloy in line with the corrosion physics.
The proposed approach is anticipated to be useful for materials engineers for optimizing the design
criteria for magnesium alloys catered for high temperature applications.

Keywords: finite element analysis; corrosion mechanics; AZ31 alloy; computational intelligence

1. Introduction
Research in magnesium (Mg) alloys has found widespread interest in recent years due to their
attractive strength-to-weight ratio, lightweight and high stiffness. These extraordinary properties
make Mg alloys particularly useful for applications requiring high performance and energy savings
such as structural engineering, automotive and biomedical applications [1–3]. It should also be noted
that the weak corrosion resistance of Mg alloys restricts their applications in several novel fields [4–7].
Most of the existing studies in Mg alloys are focused on investigating the mechanical strength based
on specific applications such as structural, mechanical or bio-medical engineering. Hence, there is a
need to study the mechanics of Mg alloys under corrosive conditions. It is also well known that Mg
alloys are more susceptible to stress corrosion cracking (SCC). Hence, an effective deployment of Mg
alloys in new and emerging fields will require an understanding of the behavior of corroded Mg alloys
under mechanical and thermal loading conditions.
Most of the literature studies on Mg alloys are based on their mechanical properties for specific
applications. Argade et al. [8] studied the mechanism of AZ31 Mg alloy under slow strain rate testing
technique. They found that the corrosion of AZ31 is predominantly influenced by its microstructure
which results in a favorable texture for increased adsorption of hydrogen. Su et al. [9] deployed a
controlled phosphate deposition technique for improving the surface biocompatibility of AZ60 Mg

Metals 2017, 7, 83; doi:10.3390/met7030083 www.mdpi.com/journal/metals

Metals 2017, 7, 83 2 of 13

alloy. They found that the coating process is influenced by temperature and pH, which affects the
corrosion resistance of AZ60 alloy. Choudary et al. [10] investigated the SCC mechanism of various
aluminum free Mg alloys in a simulated human fluid environment. It was found that the SCC of
Mg alloys in the simulated condition were attributed to their electrochemical and microstructural
characteristics. Gaur et al. [11] developed a silane based biodegradable coating for improving corrosion
resistance of Mg alloys in physiological environment. They found that formation of a stable and
uniform hydroxide layer on alloys surface facilitated adhesion of silane coating. Wei et al. [12]
investigated the corrosion behavior of plasma formed coatings on Mg alloys with respect to the
aluminum composition in the alloy mix. It was found that the difference in porosity level of coatings is
caused by different discharging behaviors in Mg alloys.
In addition to experimental studies, some simulation studies have also been conducted to
investigate the corrosion mechanics of Mg alloys. Grogan et al. [13] developed a physical corrosion
model based on finite element (FE) technique for predicting mechanical strength of corroded metal
stents. The developed model was able to capture the change in surface due to corroding environment.
Duddu et al. [14] developed an extended FE model for modeling galvanic corrosion in magnesium
alloys. It was demonstrated that the proposed approach can be used to model crevice geometry
without remeshing. Shang et al. [15] deployed molecular dynamic simulations to study the adsorption
behavior of composite coatings on Mg alloys. The presented approach provided an effective application
based surface modification of Mg alloys. In addition to FE modeling, analytical models based
on computational intelligence techniques have also been widely used for predicting corrosion.
For instance, Narimani et al. [16] deployed artificial neural network for predicting the effects of
chemical composition and corrosion characteristics in magnesium alloy pipelines. Similarly, Zadeh
Shirazi and Mohammadi [17] used a hybrid neural network model for predicting corrosion rates in
underwater pipelines.
It is clear from the literature studies that experiments and simulation have been deployed to study
corrosive characteristics in Mg alloys, with simulation being more preferred mode of investigation.
This is because simulation techniques provide a viable alternative in saving cost related to time and
materials. The current state-of-the-art in scientific modeling of corrosion is directed at two fronts,
viz. mechanical and chemical model (Figure 1). These models are capable of generating accurate
solutions in predicting corrosion characteristics with minimal cost and high rapidity. In addition to
these science based models, process modeling based on computational intelligence technique has
also been conducted, as can be seen from the previous literature studies. The advantages of process
modeling include formation of an explicit mathematical relationship which can help in understanding
hidden relationship between various input parameters in corrosion process. It can be seen that the
science based models and process models have their own unique advantages. As a result, it will be
useful to integrate the models which can combine the advantages of accuracy of FE models with model
formation ability of process modeling.
Metals 2017, 7, 83 3 of 13
Metals 2017, 7, 83 3 of 13

Figure1.
Figure 1. State-of-art
State-of-art research
researchinincorrosion
corrosionmechanics
mechanicsof magnesium alloys.
of magnesium alloys.

Hence, the main objective of the present research proposes an integrated science based model
andHence, themodel
process main objective of thecorrosion
for predicting present research proposes
characteristics an integrated
of Mg alloy (Figurescience based
2). The Mgmodel
alloy and
process model for predicting corrosion characteristics of Mg alloy (Figure
considered in this study is the AZ31 alloy used in aircraft components, biodegradable metallic2). The Mg alloy considered
in implants,
this studycell
is the AZ31
phone andalloy used
laptop in aircraft
cases, etc. In components,
the present study,biodegradable
a corrodedmetallic implants,
AZ31 alloy cell phone
with cracks
andandlaptop
voids cases, etc.resulting
typically In the present study, a corroded
from an environment AZ31
that leads alloy(e.g.,
to SCC withhydrogen
cracks and voids typically
embrittlement)
resulting from
is modeled an environment
using FE simulation.that Theleads to SCC
corrosive film(e.g., hydrogen
factor is includedembrittlement)
in the study byisconsidering
modeled using
the FE
Pilling–Bedworth ratio (P–B ratio), which is defined as the ratio of the volume
simulation. The corrosive film factor is included in the study by considering the Pilling–Bedworth of corrosive metal
oxide
ratio to the
(P–B volume
ratio), which of is
thedefined
metal. As can ratio
as the be seenof in
theFigure
volume 2, the
of study
corrosivefocuses
metalon oxide
understanding the of
to the volume
themechanical
metal. Asstrength of AZ31
can be seen alloy with
in Figure respect
2, the studytofocuses
the input
on parameters
understanding viz. the
theP–B ratio, strain
mechanical rate of
strength
and alloy
AZ31 temperature. As can
with respect tobe seen
the in Figure
input 2, the FE
parameters viz.model
the P–Bis used tostrain
ratio, predict theand
rate mechanical strength
temperature. As can
of AZ31 alloy in corrosive environment with respect to the input parameters, viz. the P–B ratio,
be seen in Figure 2, the FE model is used to predict the mechanical strength of AZ31 alloy in corrosive
strain rate and temperature. The process modeling based on genetic programming (GP) is the used
environment with respect to the input parameters, viz. the P–B ratio, strain rate and temperature.
to obtain the mathematical relationship between the mechanical strength based on the input
The process modeling based on genetic programming (GP) is the used to obtain the mathematical
parameters. Scientific analysis of the formulated model is also undertaken based on actual corrosion
relationship between the mechanical strength based on the input parameters. Scientific analysis of
science which will provide vital design inputs for maximizing the performance of AZ31 alloys in
thecorrosive
formulated model is also
environment. Theundertaken based on actual
various approaches corrosion
deployed science whichthe
in investigating willcorrosion
provide vital
design
characteristics in AZ31 alloys and the proposed research objective is depicted in Figures 1The
inputs for maximizing the performance of AZ31 alloys in corrosive environment. andvarious
2,
approaches deployed
respectively. in investigating the corrosion characteristics in AZ31 alloys and the proposed
research objective is depicted in Figures 1 and 2, respectively.
Metals 2017,
Metals 7, 83
2017, 7, 83 4 of413
of 13

Finite Element Modeling of AZ31 Alloy

23
0.8
.2 0.7
0.6
5 0.5
4 0.4
3
2
1 True stress
Input data (P-B ratio,
Temperature and Strain
rate)

Genetic Programming

Settings of GP such as population size,

generations, depth of gene, etc.

-
True stress +
model
+
5 Strain rate
P-B ratio
Temperature

Figure 2. FEM-GP
Figure approach
2. FEM-GP in modeling
approach corrosion
in modeling mechanics
corrosion of AZ31 alloy.
mechanics of AZ31FEM-GP:
alloy. Finite Element
FEM-GP: Finite
Modeling-Genetic Programming;
Element Modeling-Genetic P–B ratio: Pilling–Bedworth
Programming; ratio.
P–B ratio: Pilling–Bedworth ratio.

2. Finite
2. Finite Element
Element Modeling
Modeling of of Corroded
Corroded AZ31
AZ31 Alloy
Alloy

2.1.2.1.
Geometry Description
Geometry andand
Description Boundary Conditions
Boundary Conditions
The FEFE
The modeling of AZ31
modeling of AZ31alloy is performed
alloy is performed using ABAQUS/Explicit
using ABAQUS/Explicit Version
Version 6.14 software
6.14 software
(Dassault
(DassaultSystemes,
Systemes,Vélizy-Villacoublay,
Vélizy-Villacoublay, France,
France, 2014)
2014)with fully
with coupled
fully coupledthermal
thermal stress
stressanalysis.
analysis.
TheThethermal
thermal analysis enables
analysis modeling
enables modeling thethe
strength
strength characteristics
characteristicsof of
AZ31
AZ31 alloy
alloyat at
elevated
elevated
temperatures. The model geometry constructed in ABAQUS consists of AZ31
temperatures. The model geometry constructed in ABAQUS consists of AZ31 alloy of dimensions alloy of dimensions
450450
mmmm × 20× mm × 10 ×mm.
20 mm 10 The
mm.most
Thepredominant byproductbyproduct
most predominant of corrosionofincorrosion
Mg alloys in is magnesium
Mg alloys is
oxide (MgO). This
magnesium corrosive
oxide (MgO).product is included
This corrosive in the
product is geometry
included by adding
in the a film by
geometry of MgO
adding on atop of of
film
AZ31 alloy. In the ABAQUS/CAE model, one end of the model geometry is fixed
MgO on top of AZ31 alloy. In the ABAQUS/CAE model, one end of the model geometry is fixed with an ENCASTRE
boundary
with ancondition
ENCASTRE thatboundary
arrests allcondition
rotationalthat
and arrests
translational degrees and
all rotational of freedom. Translational
translational degrees of
degree of freedom along one axis is provided to the other end of the model
freedom. Translational degree of freedom along one axis is provided to the other end of thegeometry for tensile
model
geometry for tensile loading condition. The description of AZ31 model geometry with corrosive film
along with the applied boundary condition is depicted in Figure 3. The material properties of AZ31
Metals 2017, 7, 83 5 of 13

loading condition. The description of AZ31 model geometry with corrosive film along with the applied
boundaryMetals
condition
2017, 7, 83
is depicted in Figure 3. The material properties of AZ31 alloy and the corrosive
5 of 13
oxide film used in the FE modeling is illustrated in Table 1. The defined model geometry is then
alloydiscrete
meshed into and the corrosive oxide(CPE8RT)
eight-node film used intype
the FE modeling following
elements, is illustratedwhich
in Table 1. The defined
boundary conditions are
model geometry is then meshed into discrete eight-node (CPE8RT) type elements, following which
applied to create mechanical loading of the AZ31 part.
boundary conditions are applied to create mechanical loading of the AZ31 part.

Figure 3. Model geometry of AZ31 alloy and corrosive film considered in study.
Figure 3. Model geometry of AZ31 alloy and corrosive film considered in study.
Table 1. Properties of alloy and film used in FEM simulation [18,19]. FEM: Finite Element Modelling.
Table 1. Properties of alloy and film used in FEM simulation
Parameter
[18,19]. MgO
AZ31 Alloy
FEM:film
Finite Element Modelling.
Thermal Conductivity (k) 96 W/m-K 42 W/m-K
Parameter
Density (ρ) AZ31
1770Alloy
kg/m3 3580 MgO
kg/m3Film
Thermal Young’s modulus
Conductivity (k) (E) 96 44.8 GPa
W/m-K 249 42
GPaW/m-K
Density
Poisson’s(ρ) ratio (υ) 0.35 3
1770 kg/m 3580 kg/m3
0.18
Young’sSpecific
modulus (E)(Cp)
heat 44.8
1000GPa
J/kg/K 249 GPa
877 J/kg/K
Poisson’s ratio (υ) 0.35 0.18
Specificofheat
2.2. Constitutive Modeling (Cin
Voids p ) AZ31 under Corrosion
1000 J/kg/K 877 J/kg/K

AZ31, similar to most magnesium alloys, is susceptible to SCC. SCC results in the formation of
2.2. Constitutive Modeling
mild cracks of Voids
and voids on the in AZ31ofunder
surface Corrosion
the material. These cracks can nucleate and grow further
under application of external loads which results in catastrophic failure of the material. This
AZ31, similar toinmost
phenomenon whichmagnesium
the nucleation alloys, is susceptible
and growth of cracks and to SCC.
voids thatSCC results
lead to ductilein the formation
failure of of
mild cracks and voids on the surface of the material. These cracks can nucleate
metal is modeled in FE simulation using the Gurson–Tvergaard–Needleman (GTN) model [20]. and grow further under
applicationHence, in this work,
of external loadsthewhich
AZ31 alloy
resultssubjected to mechanical
in catastrophic loading
failure is modeled
of the using
material. Thisthe phenomenon
GTN in
damage model with integrated Yld2000 yield criterion [21,22]. The model is programmed in
which the nucleation and growth of cracks and voids that lead to ductile failure of metal is modeled in
ABAQUS solver using the user-defined subroutine module VUMAT. This approach can provide a
FE simulation using theofGurson–Tvergaard–Needleman
better prediction crack evolution and failure of AZ31 alloys (GTN) modelenvironment.
in corrosive [20]. Hence, Theinyield
this work, the
AZ31 alloy subjected
function of GTN todamage
mechanical loading
model is describedis mathematically
modeled using the GTN damage model with integrated
as [23],
Yld2000 yield criterion [21,22]. The 𝑞 2 model is programmed in ABAQUS solver using the user-defined
−3𝑞2 𝑝
subroutine module VUMAT.ΦThis + 2𝑞1 𝑓 ∗ cosℎcan
= 2 approach ( provide 𝑞3 𝑓 ∗2 )prediction
) − (1a +better =0 (1)
of crack evolution and
σ𝑦 2σ𝑦
failure of where
AZ31σalloys in corrosive environment. The yield function of GTN damage model
y is the equivalent stress; p is hydrostatic stress; q is the effective stress; and q1, q2, and q3 are
is described
mathematically as [23], to improve prediction of void interaction effects in the Gurson damage equation.
fitting parameters
The reduction in mechanical strength of the
 materialdue to void coalescence is defined by the
∗  ∗
q2
damage parameter 𝑓 , expressed as a function
∗ of− p
3q2 volume
void 2 𝑓 = 0, the GTN yield
fraction f. ∗For
Φ= 2 +
condition becomes the normal
2q1 f cos h − 1 +that
𝑓 ∗ y= 1 indicates = 0 has failed. It is
q3 f the material (1)
σy yield condition and 2σ
defined as,
where σy is the equivalent stress; p is hydrostatic 𝑓 stress; 𝑓q ≤ is𝑓the
c effective stress; and q1 , q2 , and q3 are
𝑓 ∗ = {of
fitting parameters to improve prediction 𝑓c + δ(𝑓 −
void 𝑓c ) 𝑓c < 𝑓effects
interaction ≤𝑓F in the Gurson damage (2) equation.
𝑓𝐹̅ 𝑓 ≥ 𝑓F
The reduction in mechanical strength of the material due to void coalescence is defined by the
where,
damage parameter f ∗ , expressed as a function of void volume fraction f. For f ∗ = 0, the GTN yield
∗
condition becomes the normal yield condition𝑓F̅ − 𝑓c and𝑞1f += √𝑞112 indicates
− 𝑞3 that the material has failed. It is
δ= , 𝑓F̅ = (3)
defined as,  𝑓F − 𝑓c 𝑞3

 f f ≤ fc
f∗ = f c + δ( f − f c ) f c < f ≤ f F (2)
f ≥ fF

 fF
where, q
f − fc q1 + q21 − q3
δ= F , f = (3)
fF − fc F q3
Metals 2017, 7, 83 6 of 13

Metals 2017, 7, 83 6 of 13
where f c is critical void volume fraction at which void coalescence first occurs, and f F is the void
volume
wherefraction at which
𝑓c is critical voidthe material
volume fails at
fraction due to crack
which void. propagation.
coalescence first occurs, and 𝑓F is the void
The rate
volume of increase
fraction inthe
at which total void volume
material to crackf propagation.
fails duefraction is due to growth of existing voids as well as
.
The of
nucleation rate of increase
new voids, fin
new total void
. This canvolume fractionas,
be described 𝑓̇ is due to growth of existing voids as well as
nucleation of new voids, 𝑓new ̇ . This can be described as,
. . .
f ̇ = f growth
̇ + f new ̇ (4)
𝑓 = 𝑓growth + 𝑓new (4)
TheThe growth
growthrate
rateofofexisting
existingvoids
voidsis
is proportional
proportional toto the
thehydrostatic
hydrostaticcomponent
component
ofof plastic
plastic strain
strain
.p 𝑝
rate
rate (ε̇),𝑘𝑘given
(εkk by,by,
), given
.p
f growth = (1 − f )ε 𝑝 (5)
𝑓growth = (1 − 𝑓)ε̇kk 𝑘𝑘 (5)
The nucleation rate of new voids is generally defined by strain-controlled nucleation rule which
The nucleation rate of new voids is generally defined by strain-controlled nucleation rule which
assumes that the voids undergo nucleation as second phase particles which is normally distributed for
assumes that the voids undergo nucleation as second phase particles which is normally distributed
the total particle population. This is described as,
for the total particle population. This is described as,
. f𝑓N𝑁 11 ε̅𝑝p − ε𝑁
  
̇ √ − N 𝑝𝑙pl
𝑓new =
f new = exp [− 2 ( S )] ε̅ε𝑚m (6) (6)
S𝑆N𝑁 √2π2π 2 𝑆𝑁N
where
where f N 𝑓represents
𝑁 representsthethe volume
volume fraction
fraction of void-nucleating
of void-nucleating particles,
particles, and and ε𝑁 Sand
ε N and 𝑆𝑁the
N are are the
average
andaverage
standard anddeviation
standard deviation of the
of the strains atstrains
which at which particles
particles nucleatenucleate voids, respectively.
voids, respectively.
TheThe steps
steps involvedininthe
involved themodified
modified GTN
GTN damage
damage model
modelusing
usingVUMAT
VUMATsubroutine
subroutineis is
illustrated
illustrated
in the form of a flowchart [22] in Figure
in the form of a flowchart [22] in Figure 4. 4.

Figure4.4.Constitutive
Figure Constitutive modeling
modeling of AZ31
AZ31 alloy
alloyusing
usingFE
FEapproach
approach[22].
[22].
Metals 2017, 7, 83 7 of 13
Metals 2017, 7, 83 7 of 13

The modified
The modifiedGTN GTNmodelmodel is used
is used to predict
to predict the strength
the strength characteristics
characteristics of AZ31ofalloy
AZ31 byalloy by
varying
varying the P–B ratio (x 1), temperature (x2) and strain rate (x3). The range of tested input parameters
the P–B ratio (x1 ), temperature (x2 ) and strain rate (x3 ). The range of tested input parameters tested are
testedin
given are given
Table 2. in
The Table 2. The computes
simulation simulationthecomputes the engineering
engineering stress and engineering
stress and engineering strain,
strain, from which
from which the true stress and true strain are calculated using classical mechanics
the true stress and true strain are calculated using classical mechanics relations. relations.

Table 2. Input
Table 2. Input parameters
parameters for
for tensile
tensile loading
loading simulation
simulation of
of AZ31
AZ31 alloy.
alloy.

Process Input Variable Values Units

Process Input Variable Values Units
P–B ratio 0.2, 0.3, 0.4, 0.5 No units
P–B ratio
Temperature 0.2,
300, 350,0.3,400,
0.4, 450
0.5 No units
K
Temperature 300, 350, 400, 450 K −1
Strain rate 0.0001, 0.001, 0.01 −1 s
Strain rate 0.0001, 0.001, 0.01 s

2.3. Validation of Finite Element Model

2.3. Validation of Finite Element Model
The stress–strain plot of the AZ31 alloy under uni-axial tensile loading is shown in Figure 5. For
The stress–strain
comparison plotwith
of FE results of the AZ31
actual alloy
data, theunder uni-axialtensile
experimental tensileloading
loadingcharacteristics
is shown in Figureof AZ31 5.
For comparison
obtained from Ref. of [24]
FE results
are alsowith actualindata,
included the experimental
the figure. It can be seentensile loading
from this characteristics
plot that the FE model of
AZ31 obtained from Ref. [24] are also included in the figure. It can be seen from
was able to predict the stress characteristics of AZ31 with an accuracy of 92%. It is noted that the this plot that the FE
model
model was able toapredict
predicted the high
relatively stressyield
characteristics of AZ31towith
stress compared an accuracy
experiments. Thisof 92%.
can be It attributed
is noted that to
the model
several predicted
factors such aasrelatively
the materialhighdefects,
yield stress compared
temperature to experiments.
fluctuations, Thissettings
machine can be attributed
in tensile
to several
tester, etc. factors such asthe
Furthermore, theFEmaterial
geometrydefects,
of AZ31temperature
also factoredfluctuations, machine
the corrosive settings in tensile
film component which
tester, etc. Furthermore, the FE geometry of AZ31 also factored the corrosive
may have caused certain variations in the yield stress prediction. This clearly demonstrates the film component which
may have
ability of thecaused certain variations
FE simulation in the required
for generating yield stressdataprediction.
sets that can Thisbe clearly demonstrates
subsequently utilizedtheby
ability of the FE simulation
optimization algorithms. for generating required data sets that can be subsequently utilized by
optimization
The tensile algorithms.
loading simulation performed by ABAQUS/CAE software is depicted in the form of
The tensile
screen shots in Figureloading6.simulation performed
It can be seen that theby ABAQUS/CAE
stress distribution software
in the AZ31 is depicted
alloy as in theas
well form
the
of screen shots in Figure 6. It can be seen that the stress distribution in the
corrosive film depends markedly on the simulation temperature due to the development of thermal AZ31 alloy as well as the
corrosive
stresses infilmthe depends
material. markedly
A total of on thesteps
5000 simulation
were used temperature
to complete duetheto tensile
the development of thermal
loading simulation of
stresses in the material. A
the AZ31 material in the CAE model.total of 5000 steps were used to complete the tensile loading simulation of
the AZ31 material in the CAE model.

Figure 5. Comparison of tensile force of SS304 alloy predicted from FEM simulation with experimental
Figure 5. Comparison of tensile force of SS304 alloy predicted from FEM simulation with
data obtained from Ref. [24].
experimental data obtained from Ref. [24].
Metals 2017, 7, 83 8 of 13
Metals 2017, 7, 83 8 of 13

(a)

(b)

(c)

(d)
Figure
Figure 6.
6. FE
FE modeling of mechanical
modeling of mechanical stress
stressbuildup
buildupofofcorrosive
corrosiveAZ31
AZ31alloy
alloyatattemperatures:
temperatures:
(a)(a)
300300
K;
K; (b) 350 K; (c) 400 K; and (d) 450 K. It can be seen that, regardless of temperature, the stress
(b) 350 K; (c) 400 K; and (d) 450 K. It can be seen that, regardless of temperature, the stress concentration
concentration is highest
is highest at corrosive film at corrosive
near film near
pits and crack pits and
tips which leadscrack tips which
to material leads to material
deterioration.
deterioration.
Metals 2017, 7, 83 9 of 13

3. Analytical Modeling of Stress Corrosion Characteristics of AZ31 Alloy

The FE model generates the strength data of AZ31 alloy by systematically varying the input
parameters related to P–B ratio, temperature and strain rate. A total of 48 data sets are then used to
obtain non-linear analytical model using GP technique. In GP cluster, the training data are used to
train the GP model and the validity of obtained model is tested on testing data. Kennard-and-Stone
algorithm is integrated within the GP algorithm to ensure randomness of data selection for training and
testing sets. The GP cluster is then used to derive mathematical model that describes the relationship
between the true strength of the alloy (y) as a function of input variables, viz. P–B ratio (x1 ), temperature
(x2 ) and strain rate (x3 ). The descriptive statistics of the data sets generated from the FE model is
shown in Table 3.

Table 3. Descriptive statistics of true stress data from FE simulation.

P–B Ratio, x1 Temperature, x2 Strain Rate, x3 True Stress, y1

Parameter
(No Unit) (K) (s−1 ) (MPa)
Mean 0.35 375 0.034 137.43
Median 0.35 375 0.001 121.92
Standard deviation 0.11 56.49 0.047 49.30
Kurtosis −1.38 −1.38 −1.53 −0.49
Skewness −1.842 × 10−15 0 0.730 0.75
Minimum 0.2 300 10−4 69.82
Maximum 0.5 450 0.1 235

The accuracy and effectiveness of the evolutionary algorithm framework depends on parameter
settings, which control the generation of different sizes of mathematical models. A trial and error
approach is used by the authors to determine the optimum parameter settings in the present study.
The optimal settings determined are listed as follows: the population size is 400, the number of
iterations is 20, the number of maximum genesis 6, the probability for crossover is 0.80, and the number
of generations is 200. The authors deployed the GP algorithm from the GPTIPS toolbox in MATLAB
framework developed by Searson et al. [25]. The best GP model (Equation (7)) for the true stress of
AZ31 alloy (YAZ31 ) is selected based on the minimum mean absolute percentage error value among all
the runs.

YAZ31 (MPa) = 68.18 + (1036.91 × x1 ) + [0.71 × tan{( x2 ) × (log(log( x3 )))}]

−[920.58 × {log(cos(log( x2 )))}] − [ 1130.62 × { tan{(tan( x1 ))}][ (7)
+0.53 × { tan(tan(log(log( x3 ))))}] + [0.49 × {x2 × (tan(tan( x1 )))}]

where x1 , x2 , x3 are values of P–B ratio, temperature (K) and strain rate (s−1 ), respectively. YAZ31 is the
value of true stress derived from the model based on the values of x1 , x2 , x3 .

3.1. Performance Evaluation of the Analytical Model

The evaluation of the FEM-GP model performance is conducted by analyzing the statistical
metrics of the model as defined by the root mean square error (RMSE), the coefficient of determination
(R2 ), and the mean absolute percentage error (MAPE). These metrics provides an indicator of model
performance as they measure the deviations of the predicted from actual values. The definition of
these model metrics represented by mathematical equations is available in Ref. [26]. The performance
and validation analysis of the FEM-GP model is shown in Figure 7. It can be seen from the figure that
the FEM-GP model is able to predict the true stress of AZ31 with close fit and higher accuracy. Hence,
it can be concluded that the FEM-GP model can be used as an effective approach for generalizing the
true stress of AZ31 alloy under corrosion.
Metals 2017, 7, 83 10 of 13
Metals 2017, 7, 83 10 of 13

Metals 2017, 7, 83 10 of 13

(a) (b)
Figure 7.7. Performance
Performanceof ofFEM-GP
FEM-GPmodel
modeland
andANN
ANNofoftrue stress
true of of
stress AZ31
AZ31for (a) training data and (b)
Figure (a) (b) for (a) training data and
(b) testing data. R : coefficient of determination; MAPE: mean absolute percentageerror;
testing data. R 2: coefficient of determination; MAPE: mean absolute percentage
2 error; RMSE: root
RMSE: root
mean Figure
square 7. error;
Performance
GP: of FEM-GP
Genetic model and ANN
Programming; of Artificial
ANN: true stress Neural
of AZ31 Network.
for (a) training data and (b)
mean square error;2 GP: Genetic Programming; ANN: Artificial Neural Network.
testing data. R : coefficient of determination; MAPE: mean absolute percentage error; RMSE: root
mean square error; GP: Genetic Programming; ANN: Artificial Neural Network.
3.2. Model Analysis Based on Corrosion Physics
3.2. Model Analysis Based on Corrosion Physics
3.2. Model Analysis
Corrosion Based onisCorrosion Physics
Corrosion ofofmaterial
material a complex
is a complex process
process in which
in which severalseveral factorsconsiderable
factors exhibit exhibit considerable
influence
influence Corrosion
on the material
of behavior.
material is a Hence, any
complex analytical
process in model
which describing
several factors the material
exhibit strength of
considerable
on the material behavior. Hence, any analytical model describing the material strength of the
the component
influence onunder corrosion
thecorrosion
material must be ableany
behavior. to capture the actual physics the involved during corrosion.
component under must beHence,able to captureanalytical
the model
actualdescribing
physics involved material strength
during of
corrosion.
For this, parametricunder
the component analysis is conducted
corrosion which
must be able showsthe
to capture the variation
actual physics ininvolved
the trueduringstresscorrosion.
based on the
For this, parametric analysis is conducted which shows the variation in the true stress based on the
For this,input
considered parametric
factors. analysis
Figureis8conducted
shows thewhicheffectshows
of each theofvariation in the true
the considered stress based
individual on the
factors on the
considered input factors. Figure 8 shows the effect of each of the considered individual factors on the
considered input factors. Figure 8 shows the effect of each of the considered
true stress of AZ31 alloy under corrosion. It can be seen that as the P–B ratio increases the individual factors onmaterial
the
true stress of AZ31 alloy
true degrades
stress of AZ31
under
alloyThe
under
corrosion. It can be
corrosion. Itiscan
seenthat
be seen
thatasasthe
theP–B
P–B ratio increases the material
strength rapidly. degradation more pronounced whenratioP–Bincreases
ratio exceeds the material
0.3. Based
strength degrades
strength degradesrapidly. TheThe
rapidly. degradation
degradation is is
more
morepronounced
pronouncedwhen when P–BP–B ratio exceeds0.3.
ratio exceeds 0.3.Based
Based on
on corrosion physics, it can be stated that the corrosive film does not protect the alloy from
corrosion physics, it can be stated that the corrosive film does not protect
on corrosion physics, it can be stated that the corrosive film does not protect the alloy from the alloy from degradation.
degradation. Similar trend can be seen for the case of temperature in which a corrosive AZ31 alloy
Similar trend canSimilar
degradation. be seentrendfor the
can case of temperature
be seen for the case ofin which a corrosive
temperature in which AZ31 alloyAZ31
a corrosive exhibitsalloylower
exhibits lower
exhibits material
lowerat materialstrength at higher working temperatures. This indicates that AZ31 alloy
material strength higherstrength
workingattemperatures.
higher workingThis temperatures.
indicates Thisthat indicates
AZ31 alloy thatunder
AZ31 alloy
corrosive
underunder
corrosive conditions is not suitable for high temperature applications. However, the strain
conditions iscorrosive conditions
not suitable for highis not suitable forapplications.
temperature high temperature applications.
However, the strain However,
rate seems the tostrain
have no
rate seems
rate to have
seems to haveno no
effect onon
effect thethematerial
material strength
strength of AZ31
of AZ31alloyalloywhich
which agrees
agrees well well
with with previous
previous
effect on the material strength of AZ31 alloy which agrees well with previous experimental study [24].
experimental
experimentalstudy [24]. Hence, atatambient temperatures,the theeffect
effectofof loading (i.e.,(i.e., strain
rate)rate)
does does
Hence, at ambientstudy [24].
temperatures,Hence,
the ambient
effect oftemperatures,
loading (i.e., strain rate)loading
does not havestrain
any negative or
not have
not have any any negative
negative or or positiveimpact
positive impacton on the
the strength
strength ofofthe theAZ31
AZ31 alloy.
alloy.This corroborates
This corroborates with with
positive impact on the strength of the AZ31 alloy. This corroborates with the fact that material failure
the that
the fact fact that material
material failure
failure underSCC
under SCCoccurs
occurs not
not because
becauseofofexternal
external load itself,
load but but
itself, due due
to theto the
undercrack
SCCgrowth
occursand notnucleation
because of of external
voids load
that itself,
results frombutresidual
due tostresses,
the crack growth
loading, and
etc. nucleation
Hence, from from of
crack growth and nucleation of voids that results from residual stresses, loading, etc. Hence,
voidsthis
thatanalysis
resultsitfrom residual stresses, loading, etc. Hence, from this analysis
can be concluded that the derived analytical model was able to confirm well with the it can be concluded that
this analysis it can be concluded that the derived analytical model was able to confirm well with the
the derived
observed analytical
physics inmodel
corrosionwasprocess.
able to confirm well with the observed physics in corrosion process.
observed physics in corrosion process.

(a) (b)

(a) Figure 8. Cont. (b)

Metals 2017, 7, 83 11 of 13
Metals 2017, 7, 83 11 of 13

(C)
Figure 8. Parametric analysis of true stress of AZ31 with (a) P–B ratio, (b) Temperature and (c) Strain
Figure 8. Parametric analysis of true stress of AZ31 with (a) P–B ratio, (b) Temperature and (c)
rate. Square markers ( ) in these plots refer to the data samples of true stress and the corresponding
Strain rate. Square markers () in these plots refer to the data samples of true stress and the
inputs.
corresponding inputs.
A sensitivity analysis was also conducted to determine the relative importance of each of the
input
A parameters
sensitivity on the was
analysis considered output. Thetosensitivity
also conducted determine analysis is defined
the relative by partial derivative
importance of each of the
inputofparameters
strength withonrespect to temperature.
the considered output.While
Thethe partial derivative
sensitivity analysisisiscomputed,
defined by thepartial
other inputs
derivative
such as the P–B ratio and strain rate are kept constant at their mean values. The effect of the two
of strength with respect to temperature. While the partial derivative is computed, the other inputs
inputs P–B ratio and strain rate, which are at mean values, is still there on the strength. Therefore,
such as the P–B ratio and strain rate are kept constant at their mean values. The effect of the two
the strength depends on the inputs in a complicated way. The relative importance of the considered
inputs P–B ratio and strain rate, which are at mean values, is still there on the strength. Therefore, the
input parameters on the mechanical strength of AZ31 alloy under corrosion is depicted in Table 4. It
strength
can depends on the
be seen that the inputs in a complicated
most important amongst way.
thoseThe relative
studied importance
under corrosion of the temperature,
is the considered input
parameters
followed by P–B ratio and strain rate. This analysis corroborates with actual corrosion Table
on the mechanical strength of AZ31 alloy under corrosion is depicted in 4. It can be
mechanisms
seen in
that the most alloy
magnesium important
system,amongst those
as they are morestudied under
susceptible to corrosion is thetemperatures.
SCC at elevated temperature,As followed
AZ31 by
is primarily
P–B ratio usedrate.
and strain for high
Thistemperature applications,with
analysis corroborates controlling
actual the system temperature
corrosion mechanismsisin crucial to
magnesium
alloyavoid tensile
system, strength
as they aredegradation under corrosion,
more susceptible to SCC at which can lead
elevated to material failure.
temperatures. As AZ31 is primarily
used for high temperature applications, controlling the system temperature is crucial to avoid tensile
Table 4. Sensitivity analysis of true stress of AZ31 alloy.
strength degradation under corrosion, which can lead to material failure.
Percentage Contribution
Input Parameter
to stress
Table 4. Sensitivity analysis of true True Stress
of AZ31(%)alloy.
P–B ratio 36
Input ParameterTemperature (K) Percentage Contribution62 to True Stress (%)
Strain rate (s−1) 2
P–B ratio 36
Temperature (K) 62
4. ConclusionsStrain rate (s−1 ) 2
The present work introduced an integrated FE based computational intelligence approach for
4. Conclusionscorrosion mechanics of AZ31 alloy under mechanical loading. FE modeling of AZ31 alloy
modeling
with a modified constitutive damage model was used to predict the true stress responses of AZ31
The
alloypresent workcorrosive
with varying introduced an integratedtemperature
film concentration, FE based computational intelligence
and strain rate. The approach
data obtained from for
modeling corrosion
the model mechanics
were used of aAZ31
to derive alloy
genetic under mechanical
programming loading.
model, which couldFE modeling
predict of AZ31
the strength of alloy
with athe alloy with
modified an accuracy
constitutive of 92%model
damage and confirmed
was used the data trend
to predict withstress
the true actualresponses
corrosionofphysics.
AZ31 alloy
with Sensitivity and parametric
varying corrosive analysis of the genetic
film concentration, programming
temperature model
and strain showed
rate. that the
The data mechanical
obtained from the
strength of the AZ31 alloy highly dependent on system temperature and does
model were used to derive a genetic programming model, which could predict the strength of not show much
the alloy
with variation with regards to applied strain rate. The proposed model can be used to determine optimal
an accuracy of 92% and confirmed the data trend with actual corrosion physics. Sensitivity and
design parameters for AZ31 alloy under corrosive environments without the need of experiments.
parametric analysis of the genetic programming model showed that the mechanical strength of the
This can help in reducing costs associated with material fabrication and testing, which can also lead
AZ31toalloy highly dependent on system temperature and does not show much variation with regards
sustainable materials design.
to applied strain rate. The proposed model can be used to determine optimal design parameters
for AZ31 alloy under corrosive environments without the need of experiments. This can help in
reducing costs associated with material fabrication and testing, which can also lead to sustainable
materials design.
Metals 2017, 7, 83 12 of 13

Author Contributions: The work presented here was carried out in collaboration between all authors.
Venkatesh Vijayaraghavan and Akhil Garg defined the research theme. Venkatesh Vijayaraghavan performed
the finite element modeling and wrote the paper. Akhil Garg performed the genetic programming computations.
Liang Gao provided the genetic programming framework and provided guidance on interpreting the trends from
the evolutionary model. Rangarajan Vijayaraghavan designed methods and experiments, analyzed the data, and
interpreted the results. All authors have contributed to, seen and approved the manuscript. The authors hope that
this paper can contribute to the successful application of finite element modeling to determine optimal design
parameters of AZ31 alloy without the need to conduct experiments.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Çelik, İ. Structure and surface properties of Al2 O3 –TiO2 ceramic coated AZ31 magnesium alloy. Ceram. Int.
2016, 42, 13659–13663. [CrossRef]
2. Lentz, M.; Risse, M.; Schaefer, N.; Reimers, W.; Beyerlein, I.J. Strength and ductility with {1011}-{1012} double
twinning in a magnesium alloy. Nat. Commun. 2016, 7, 11068. [CrossRef] [PubMed]
3. Sunil, B.R.; Reddy, G.P.K.; Patle, H.; Dumpala, R. Magnesium based surface metal matrix composites by
friction stir processing. J. Magnes. Alloys 2016, 4, 52–61. [CrossRef]
4. Jafari, S.; Harandi, S.E.; Raman, R.K.S. A review of stress-corrosion cracking and corrosion fatigue of
magnesium alloys for biodegradable implant applications. JOM 2015, 67, 1143–1153. [CrossRef]
5. Jafari, S.; Raman, R.K.S.; Davies, C.H.J. Corrosion fatigue of a magnesium alloy in modified simulated
body fluid. Eng. Fract. Mech. 2015, 137, 2–11. [CrossRef]
6. Raman, R.K.S.; Harandi, S.E. Understanding corrosion-assisted cracking of magnesium alloys for bioimplant
applications. In Magnesium Technology; John Wiley & Sons, Inc.: Hoboken, NJ, USA, 2016.
7. Raman, R.K.S.; Jafari, S.; Harandi, S.E. Corrosion fatigue fracture of magnesium alloys in bioimplant
applications: A review. Eng. Fract. Mech. 2015, 137, 97–108. [CrossRef]
8. Argade, G.R.; Yuan, W.; Kandasamy, K.; Mishra, R.S. Stress corrosion cracking susceptibility of ultrafine
grained AZ31. J. Mater. Sci. 2012, 47, 6812–6822. [CrossRef]
9. Su, Y.; Guo, Y.; Huang, Z.; Zhang, Z.; Li, G.; Lian, J.; Ren, L. Preparation and corrosion behaviors of calcium
phosphate conversion coating on magnesium alloy. Surf. Coat. Technol. 2016, 307, 99–108. [CrossRef]
10. Choudhary, L.; Raman, R.K.S.; Hofstetter, J.; Uggowitzer, P.J. In Vitro characterization of stress corrosion
cracking of aluminium-free magnesium alloys for temporary bio-implant applications. Mater. Sci. Eng. C
2014, 42, 629–636. [CrossRef] [PubMed]
11. Gaur, S.; Raman, R.K.S.; Khanna, A.S. In vitro investigation of biodegradable polymeric coating for corrosion
resistance of Mg-6Zn-Ca alloy in simulated body fluid. Mater. Sci. Eng. C 2014, 42, 91–101. [CrossRef] [PubMed]
12. Wei, F.; Zhang, W.; Zhang, T.; Wang, F. Effect of variations of Al content on microstructure and corrosion
resistance of PEO coatings on Mg-Al alloys. J. Alloys Compd. 2017, 690, 195–205. [CrossRef]
13. Grogan, J.A.; Leen, S.B.; McHugh, P.E. A physical corrosion model for bioabsorbable metal stents.
Acta Biomater. 2014, 10, 2313–2322. [CrossRef] [PubMed]
14. Duddu, R.; Kota, N.; Qidwai, S.M. An Extended Finite Element Method Based Approach for Modeling
Crevice and Pitting Corrosion. J. Appl. Mech. Trans. ASME 2016, 83, 081005. [CrossRef]
15. Shang, W.; Wang, X.; Wen, Y.; He, C.; Wang, Y.; Zhang, L.; Zhang, Z. Corrosion resistance and molecular
dynamics behavior of the MAO/SAM composite coatings on magnesium alloy. Prot. Met. Phys. Chem. Surf.
2016, 52, 847–853. [CrossRef]
16. Narimani, N.; Zarei, B.; Pouraliakbar, H.; Khalaj, G. Predictions of corrosion current density and potential by
using chemical composition and corrosion cell characteristics in microalloyed pipeline steels. Meas. J. Int.
Meas. Confed. 2015, 62, 97–107. [CrossRef]
17. Zadeh Shirazi, A.; Mohammadi, Z. A hybrid intelligent model combining ANN and imperialist competitive
algorithm for prediction of corrosion rate in 3C steel under seawater environment. Neur. Comput. Appl. 2016.
[CrossRef]
18. Zhao, P.J.; Chen, Z.H.; Dong, C.F. Failure Analysis of Warm Stamping of Magnesium Alloy Sheet Based on
an Anisotropic Damage Model. J. Mater. Eng. Perform. 2014, 23, 4032–4041. [CrossRef]
19. Vasilos, T.; Mitchell, J.B.; Spriggs, R.M. Mechanical Properties of Pure, Dense Magnesium Oxide as a Function
of Temperature and Grain Size. J. Am. Ceram. Soc. 1964, 47, 606–610. [CrossRef]
Metals 2017, 7, 83 13 of 13

20. Acharyya, S.; Dhar, S. A complete GTN model for prediction of ductile failure of pipe. J. Mater. Sci. 2008, 43,
1897–1909. [CrossRef]
21. Oh, C.S.; Kim, N.H.; Kim, Y.J.; Baek, J.H.; Kim, Y.P.; Kim, W.S. A finite element ductile failure simulation
method using stress-modified fracture strain model. Eng. Fract. Mech. 2011, 78, 124–137. [CrossRef]
22. Zhao, P.J.; Chen, Z.H.; Dong, C.F. Damage and Failure Analysis of AZ31 Alloy Sheet in Warm Stamping
Processes. J. Mater. Eng. Perform. 2016, 25, 2702–2710. [CrossRef]
23. Needleman, A.; Tvergaard, V. An analysis of dynamic, ductile crack growth in a double edge cracked
specimen. Int. J. Fract. 1991, 49, 41–67. [CrossRef]
24. Lin, Y.C.; Liu, Z.H.; Chen, X.M.; Chen, J. Uniaxial ratcheting and fatigue failure behaviors of hot-rolled
AZ31B magnesium alloy under asymmetrical cyclic stress-controlled loadings. Mater. Sci. Eng. A 2013, 573,
234–244. [CrossRef]
25. Searson, D.P.; Leahy, D.E.; Willis, M.J. GPTIPS: An Open Source Genetic Programming Toolbox for Multigene
Symbolic Regression. In Proceedings of the International MultiConference of Engineers and Computer
Scientists 2010 (IMECS 2010), Hong Kong, China, 17–19 March 2010.
26. Vijayaraghavan, V.; Castagne, S. Computational model for predicting the effect of process parameters on
surface characteristics of mass finished components. Eng. Comput. 2016, 33, 789–805. [CrossRef]

© 2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).