2.

3 Simulation of Grain Growth in the HAZ in 3D

2.3.1 Introduction

Grain growth is an important aspect of welding metallurgy. An important

microstructural feature in the HAZ is the significant grain size gradient, which is caused
by the steep thermal gradients in this region. Therefore, quantitative prediction of grain
growth in the HAZ needs to combine grain growth kinetics with the thermal cycles at
various locations in this region. In the past decade, the Monte Carlo (MC) technique has
been widely used to simulate the evolution of grain structure under isothermal
conditions.25-27 Compared with the analytical method,30 the Monte Carlo simulation of
grain growth can provide not only grain growth kinetics but also detailed information on
temporal grain morphology evolution. More recently, the Monte Carlo technique has
been used to simulate grain growth in the HAZ of welds. 28,29 The simulation of grain
growth in the HAZ by the MC technique is more realistic than by the classical analytical
equation, because the temperature gradient near the fusion line is steep and the atomic
mobility can vary across a single grain. This phenomena can not be taken into account by
the classical analytical equation. However, it can be effectively simulated by the MC
technique, in which the grains are subdivided into many discrete small units and each
unit can have its own atomic mobility depending on the local temperature.

2.3.2 Grain Growth Kinetics

Grain growth is the process by which the mean grain size of an aggregate of
crystals increase. The driving force for this results from the decrease in total grain
boundary energy which accompanies reduction in total grain boundary area. Two types
of grain growth phenomena have been distinguished-normal growth and abnormal
growth. In the normal grain growth, the microstructure exhibits a uniform increase in
grain size and the normalized grain size distribution function is invariant with time. The
abnormal grain growth is characterized by the growth of just few grains to very large
size. These grains grow by consuming the surrounding grains, until the fine grains are
entirely replaced by the few coarse grains. Abnormal grain growth usually occurs when
normal grain growth is impeded by second phase particles. In the present work, only
normal grain growth phenomena is studied.

The normal grain growth phenomena has been extensively investigated because
of the importance of its resulting microstructure in controlling the physical and
mechanical properties of the polycrystalline materials. In a classical paper of the early
1950s, Burke and Turnbull deduced a parabolic relationship for grain growth kinetics
under isothermal conditions. The migration of a grain boundary was modeled as
occurring by atom transport across the boundary under a pressure due to surface
curvature. The surface curvature forces were considered on an isolated section of
boundary. The boundary tends to migrate towards its center of curvature as this reduces
the area of boundary and hence the energy associated with it. Several assumption were
made on his derivation: (1) the radius of curvature, r, is proportional to the mean grain
radius of grains, R; (2) the grain boundary migration velocity is proportional to dR/dt;
(3) the grain boundary free energy,g, is independent of grain size and time; (4) the mean
driving force is proportional to 2g/R. The parabolic grain growth kinetics can be
expressed as:
2 2
D  D 0  Kt

where D is the mean grain size at time t, D 0 is the initial mean grain size, and K is a
kinetic constant. Following this, other different theories were proposed to predict the
time-dependence of average grain size and size distribution by analytical methods. Most
these analytical theories also predict a parabolic growth law. However, the
experimentally determined grain growth exponent values have been reported varying
from 2 to 4 even for pure metals, as shown in Table 2.*. The deviation of the
experimental values from theoretical prediction has been attributed to the presence of
additional factors not considered in the analytical models. These additional factors
include impurity segregation, second particles, and anisotropies etc..
2.3.3 Grain Growth in the HAZ simulated by analytical method

The grain growth behavior in the HAZ of welds is complicated by the steep
temperature gradients and transient thermal cycles in this region. Several studies are
available in the literature to use analytical equation to calculate grain size in the HAZ. 66-88
One popular analytical method was proposed by Ashby et al to simulate grain growth in
the HAZ based on the idea of integrating the parabolic grain growth kinetic equation
over the weld thermal cycles.

Assuming that grain growth is diffusion controlled, driven by grain boundary

energy, and requires no nucleation, then the extent of transformation depends on the
integrated number of diffusive jumps during the weld thermal cycle. The austenite grain
growth in the HAZ can be calculated by a kinetic equation as:

  Q 
2  
g  g 20  k 1   e RT( t ) dt 

0 
 (3.1)

where g is the grain size after time t, g0 is the initial grain size, k1 is a kinetic constant, Q
is the activation energy for grain growth, T(t) is the temperature varying with time, and R
is the gas constant. The term in the bracket of equation (1) is the “kinetic strength” of
the cycle, which has been previously defined in section *.*. Based on the concept of
“kinetic strength”, the grain growth equation can be simplified as:

Q

g 2  g 20  k 1e RTP
(3.2)

where t is a characteristic time constant for the thermal cycle, T p is the peak temperature
of the thermal cycle, a is a variable determined by Q and Tp.16

The kinetic constant k1 in equations (3.1) and (3.2), which contains uncertain
kinetic factors, can be eliminated by using experimental data. Under a specified
experimental condition, the grain size at a point of the HAZ can be expressed as:
 Q

RTp*
g *2  g 20  k 1 *  * e (3.4)
where g* is the measured grain size produced by a thermal cycle characterized by a peak
temperature Tp* and a characteristic time constant t*. Combining equations (3.2) and
(3.4), the grain size for a specified weld thermal cycle can be calculated by the following
equation:
Q 1 1 
g 2  g 20     
R  TP* TP 
 e
g*2  g 20 * * (3.5)
Equation (3.5) establishes a relationship between grain size g and the thermal
cycle characterized by variables Tp and t, which allows grain size in the HAZ to be drawn
in a Tp and t space.

Fig.*.* shows the grain size distribution in the HAZ of …predicted by this
algorithm.

2.3.4 Simulation of Grain Growth by Monte Carlo (MC) Technique

2.3.3.1 Simulation methodologies of Monte Carlo (MC) technique

The simulation methodologies of MC technique have been given in detail in

literature66 and are briefly described here. A computer image of the polycrystalline
microstructure is first created by mapping the continuum grain structure onto a discrete
lattice. This is accomplished by dividing the material into small area (2D) or volume
(3D) elements, and placing the center of these elements onto the lattice points. The lattice
can be triangular or square in two dimensions, and simple cubic in three dimensions.
Each of the grid points will be assigned a random orientation number between 1 and Q,
where Q is the total number of grain orientations. A grain boundary segment is defined to
lie between two sites of unlike orientation. In other word, two adjacent grid points having
the same orientation number are considered to be a part of the same grain; otherwise they
belong to different grains. The grain boundary energy is specified by defining an
interaction between nearest neighbor lattice sites. The local interaction energy is
calculated by the Hamiltonian:
 nn
E   J  ( SiS j  1)
j1 (1)
where J is a positive constant which sets the scale of the grain boundary energy, d is the
Kronecker’s delta function, Si is the orientation at a randomly selected site i, and S j are
the orientations of its nearest neighbors. The sum is taken over all nearest neighbor sites
(nn). The nearest neighbor pairs contribute J to the system energy when they are of
unlike orientation and zero otherwise.

The kinetics of grain boundary migration are simulated by selecting a site

randomly and changing its orientation to one of the nearest neighbor orientations based
on energy change due to the attempted orientation change. The probability of orientation
change is defined as:
p=1 DE £ 0 (2.1)
p = exp(-DE/kBT) DE > 0 (2.2)
where DE is the change of energy due to the change of orientation, k B is the Boltzman
constant, and T is the temperature. Successful transitions at the grain boundaries to
orientations of nearest neighbor grains thus corresponds to boundary migration.

2.3.3.2 Relation between MC simulation step (tMCS) and real time-temperature

One critical issue to apply the MC technique to a real metallurgical phenomena is

establish the relation between the MC simulation step (tMCS) and real time-temperature.
Depending on the nature of the problem, different methods4 have been used to convert
real time to tMCS. These methods can be classified into:66 Atomistic Model, Grain
Boundary Migration (GBM) Model, and Experimental Data Base (EDB) Model.

2.3.3.2.A Atomistic model

In the atomic model, the grid points in the simulated domain are treated as atoms.
The MC simulation step is thus correlated to atomic diffusion across grain boundary and
connected to real time-temperature by the following relationship:
  Q
t mcs    exp  t
 RT 

where tmcs is the MC simulation step, n is the atomic vibration frequency, Q is the
activation energy for grain growth, R is the constant, T is temperature, and t is real time.
To apply equation () for grain growth in the HAZ of welding, the experienced
temperature - time curve at a location of the HAZ is divided into many short intervals.
The temperature in each time interval is assumed to be constant. The MC simulation step
for a transient thermal cycle is thus obtained by summation of the MC simulation step at
each time interval, which can be expressed as:

m  Q 
t mcs     exp    t i
i 1  RTi 

where Ti is the temperature within the time interval Dti, m is the number of the time
intervals.

2.3.3.2.B Grain Boundary Migration (GBM) Model

Based on Burke and Turnbul’s well known parabolic law for isothermal grain
growth30 and the quantitative description of the mobility of grain boundary derived by
Porter et al.,43 Gao et al.28 proposed an analytical equation to calculate the isothermal
grain growth kinetics from the first principles:
 4 AZVm2 S    Q
L L 
2 2
 exp( a )   exp t
0
 Na h
2
R   RT 
(19)
where L is the average grain size at time t, L 0 is the initial average grain size, g is the
grain boundary energy, A is the accommodation probability, Z is the average number of
atoms per unit area at grain boundary, V m is the atomic molar volume, Na is Avagadro’s
number, h is Planck’s constant, DSa is the activation entropy, Q is the activation energy
for grain growth, T is the absolute temperature, R is the gas constant, and t is time.
On the other hand, through the MC simulation, an empirical relation between the
simulated grain size and the MC simulation time can be obtained as:28
 L  K1     t MCS 
n1

(18)
where L is the dimensionless simulated grain size measured by mean grain intercepts, l is
the discrete grid point spacing in MC technique, t MCS is MC simulation time or MC
simulation iteration steps, K1 and n1 are the model constants, which are obtained by
regression analysis of the data generated from MC simulation. Equating the mean grain
intercept of equation *.* to that of equation *.* and integrating t mcs over the entire
thermal cycle, the tmcs for simulation of grain growth in the HAZ can be expressed as:
2
 L  1  4AZVm

2  m
 S a   Q 
( t MCS ) 2 n2
 0     exp       t i  exp(  )
 K 1   K 1  2  2
 Na h  R   i1 RTi 

2.3.3.2.C Experimental Data Base (EDB) Model

Based on the data from isothermal grain growth experiments, a relation can be
established between the grain size (L), temperature (T) and holding time (t), which can
be expressed as:
Q
Ln  Ln0  K  t  exp(  )
RT
As shown in equation *.*, the relation between the grain size and the t MCS can also be
established by Monte Carlo simulation at constant temperature. Equating equations *.*
and *.*, the relation between the t MCS for the isothermal grain growth and the real time
temperature can be established as:
n
n  n1  L  K Q
( t MCS )  0    t  exp(  )
 K 1   K 1  n RT

Similarly, the temperature-time curve in the HAZ needs to be divided into many intervals
and each step is assumed to be isothermal. Equation *.* can then be transformed into:
n
 L  1 m Q 
( t MCS ) n  n1   0      t i  exp(  )
 K 1   K1  i1
n RTi 

Because of the small size of the atoms and the limitation of computer resource,
the atomic model can only be used to simulate grain growth in the structure with small
assemblies of atoms such as nanocrystals. In the GBM and EDM models, the grid points
are treated as a block of physical area (in 2D) or volume (in 3D) in the material.
Generally, these two models can be used to simulate grain growth regardless of grain
size. When the isothermal grain growth kinetics of a material are available, the EDB
model can be applied to simulate the HAZ grain growth in welds. When the isothermal
grain growth data for the given material are insufficient, the GBM model is a good
approach for grain growth simulation. However, the physical properties of the material
system must be available.

2.3.3.3 Previous Investigations

In recent years, some efforts have been made to apply MC technique in

simulation of grain growth in the HAZ. A major advantage of using the MC technique
over other analytical techniques for simulation of grain growth is that the MC simulation
presents both microstructural and kinetic information. MC simulation is particularly
attractive for studying grain growth in the HAZ with steep temperature gradient because
the spatial variation of grain boundary mobility can be easily incorporated into the
algorithm.

Gao et al. simulated the grain structure in the HAZs of different welds by using
MC technique in two dimensions.66 The investigated materials were nanocrystalline
nickel thin film, aluminum plate, and 0.5Cr-Mo-V steel. Depending on the material
systems investigated, different kinetic models as described in section 2.3.3.2 (e.g.
Atomistic Model, GBM model, and EDB model) were applied in establishing the
relations between the MC step, tmcs, and the real time-temperature. It was found that the
Monte Carlo based simulation technique can be successfully used in simulating normal
grain growth in the presence of thermal gradients as in the HAZ of welds. When the
isothermal grain growth data were available such as 0.5Cr-Mo-V steel, good results were
obtained from the experimental data based model. For the case where experimental data
were insufficient but data on physical properties were available such as pure aluminum
plate, the grain boundary migration model was used as an alternative to simulate grain
growth.

Monte Carlo based model have also been used to predict the “thermal pinning”
phenomena in the HAZ.66 This phenomena was first observed by Alberry in the weld
HAZ of a 0.5 Mo-Cr-V steel. In comparing the grain growth in the HAZ of real welds
with that in thermally simulated specimens, it was found that the maximum grain size is
larger in simulated specimens than that in the HAZ of a weld when comparison is made
on the basis of a similar thermal cycle. A possible explanation of this observation is the
phenomenon of “thermal pinning” which can be attributed to the presence of steep
temperature gradients in a weld HAZ. This effect has been disregarded in the previous
analytical analysis,66-88 but can be readily incorporated in Monte Carlo technique.
Radhakrishnan et al. have demonstrated that the retarded grain growth kinetics due to the
presence to steep temperature gradient can be simulated by using MC technique. Fig. *.*
shows the kinetics of grain growth at the 120 micron HAZ location and the kinetics for
bulk heating using an identical thermal cycle. It can be observed that the grain growth
kinetics in the HAZ location is lower than that for bulking heating using an identical
thermal cycle.

Although the thermal pinning effect can be simulated by the MC technique, the
magnitude of the pinning effect predicted by the simulations was significantly lower than
in the actual weld HAZ, i.e. the predicted grain size is larger than that in the actual weld
HAZ.66 This indicates that there is an additional inhibition effect except for the “thermal
pinning”. The formation of the grain boundary liquid is believed to be responsible for
this additional pinning.66 The grain boundary liquid in the HAZ may be formed by
different mechanisms. The grain boundary can be formed by the sub-solidus liquidation
caused by the presence of soluble second particles.66 The grain boundary liquidation may
also be formed by intergranular penetration of liquid and/or solute from the fusion zone
along grain boundaries in the HAZ. 66 The dominance of one mechanism over another
depends on the investigated system.
 The grain boundary liquidation can occur in the HAZ above the solidus
temperature and the grain growth in the HAZ can be significantly retarded due to the
presence of liquid on grain boundaries.66 Abnormal grain size distribution can be resulted
due to the grain boundary liquidation. In analyzing the grain size distribution in the HAZ
of Maraging steels, Pepe et al. found that the average grain diameter was 54 mm at the
location of 100 mm from the fusion line, while that at the location of the fusion line was
only 48 mm. Similar trend of grain size distribution was observed in the experiments by
Wilson et al.,66 in which the grain size distribution in the HAZs of the HSLA-80 and
HSLA-100 welds were analyzed systematically. Significant reductions in grain size in
the vicinity of the fusion line were observed in most cases. The pinning effect due to the
liquidation of grain boundary has been simulated using MC technique by Radhakrishnan
et al. and Wilson et al.. Since the grain growth was assumed completely pinned by the
grain boundary liquid, the corresponding portion of the thermal cycle was not considered
in the calculation of the MCS in their simulations. Thus, the MCS values closer to the
fusion line and the corresponding simulated grain size decreases. Fig. X.x shows the
comparison of the simulated and the experimental results.

2.3.5 Summary

The multivariable characteristics of grain growth in the HAZ during welding can
not be sufficiently described by using analytical modeling techniques. The MC based
grain growth simulation technique is able to effectively simulate the unique features of
the grain growth in the HAZ, in which the grains are subdivided into many discrete small
units and each unit can have its own atomic mobility depending on the local temperature.
In addition to the grain growth kinetics, the MC based technique can also model exact
shapes and sizes in topological connected microstructures.

Previous simulations of the grain growth in the HAZ were limited to two
dimensional calculations.66 Considering the significant local thermal and topological
environment change in the HAZ, simulation of grain growth in 2D is not sufficient for
grain growth occurring in real welds which are three dimensional. Furthermore, the
width of the HAZ usually varies with the location in the actual weldment. This
temperature dependence of spatial dimension variation was ignored in the previous 2D
simulations. For more realistic simulation, a 3D MC model needs to be established, in
which the whole region of the HAZ around the weld pool is considered. Only under this
condition, can the one-to-one correlation between the sites in the simulated domain and
those in the actual welds be established. By incorporating the calculated thermal history
data at each site into the 3D MC model, the simulated grain growth kinetics at each site
will thus be strictly based on the corresponding thermal history.