A Model of Solidification Microstructures in Nickel-Based

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 17

Acta Materialia 51 (2003) 2971–2987

www.actamat-journals.com

A model of solidification microstructures in nickel-based


superalloys: predicting primary dendrite spacing selection
W. Wang, P.D. Lee ∗, M. McLean
Department of Materials, Imperial College, Prince Consort Road, London SW7 2BP, UK

Received 3 December 2002; accepted 11 February 2003

Abstract

A combined cellular automaton-finite difference (CA-FD) model has been developed to simulate solute diffusion
controlled solidification of binary alloys. Constitutional and curvature undercooling were both solved to determine the
growth velocity of the solid/liquid interface. A modified decentered square/octahedron (in two or three dimensions)
growth technique was implemented in the cellular automaton to account for the effect of crystallographic anisotropy.
The resulting model is capable of simulating the growth of equiaxed and columnar dendritic grains in 2D and 3D,
with the ⬍100⬎ directions either aligned or inclined with the grid. The algorithm used can also be used on coarser
grids, with a concomitant loss in resolution, allowing simulation of sufficiently large numbers of dendrites in 3D to
investigate the distribution of spacings, as well as average behavior.
Simulations were performed for directional solidification with a range of withdrawal velocities and nucleation con-
ditions, but a constant thermal gradient. The simulations capture the full microstructural development and primary
spacing selection by both branching and overgrowth mechanisms. The model illustrates that there is a range of possible
stable spacings, and that the final spacing is history dependent. It was also found that a minimum deviation from the
steady state dendrite spacing is required before the spacing adjustment mechanisms are activated. The influence of
perturbing the withdrawal velocity upon the stability of the spacing was also investigated. It was found that perturbations
significantly reduce the range of stable primary dendrite spacing.
 2003 Acta Materialia Inc. Published by Elsevier Science Ltd. All rights reserved.

Keywords: Cellular automaton; Solute diffusion; Directional solidification; Primary dendrite spacing

1. Introduction umnar dendrites where microsegregation of


alloying elements occurs. This segregation pattern
Many multi-component industrial alloys are sol- is characterized by primary dendrites parallel to,
idified in conditions (i.e. low velocity or high ther- and secondary dendrites perpendicular to, the
mal gradient) that produce a complex array of col- ⬍100⬎ directions closest to the macroscopic sol-
idification direction; these features can have sig-
nificant effects on the mechanical properties of sol-

Corresponding author. Tel.: +44-20-7594-6801; fax: +44- idified products [1,2]. In case of the nickel-based
20-7594-6758. superalloys used for gas turbine blades and discs,
E-mail address: p.d.lee@imperial.ac.uk (P.D. Lee). these features are of particular importance as their

1359-6454/03/$30.00  2003 Acta Materialia Inc. Published by Elsevier Science Ltd. All rights reserved.
doi:10.1016/S1359-6454(03)00110-1
2972 W. Wang et al. / Acta Materialia 51 (2003) 2971–2987

influence can persist through the subsequent growth mechanisms. Based on this concept,
thermo-mechanical and/or heat treatment processes numerical models have been proposed for the ste-
[3]. In order to improve the properties of the cast ady and transient growth of a cellular/dendritic
alloys, as well as controlling the formation of sol- array in binary alloys that predict dendrite spacing
idification defects, a fundamental knowledge of the and tip undercooling [11,12].
growth of the initial dendritic microstructure dur- Recently, a numerical technique, the phase field
ing solidification is required. Over the past thirty method, was proposed. Phase field simulations
years or so, considerable progress has been made accurately reproduce the morphology of
in understanding dendritic growth from both developing dendrites in simple thermal conditions,
experimental and theoretical studies. capturing most of the phenomena associated with
Several experimental studies have used trans- dendrite formation (e.g. dendrite tip kinetics, pre-
parent organic analogues to directly observe the ferred growth direction, coarsening, coalescence,
development of dendrites under well-defined ther- etc.) [13–15]. Initially developed for thermal dif-
mal conditions and have identified the mechanism fusion controlled growth of dendrites in two
for adjusting the dendrite spacing to a change in dimensions, it has been extended to solute-dif-
solidification condition [4–6]. These experimental fusion controlled dendrite growth [13] and to three
results show no unique primary cellular/dendritic dimensions [16].
spacing, but rather a range of spacing, for a given Another numerical technique used to simulate
processing condition. The development of the pri- solidification is the cellular automaton (CA) stoch-
mary spacing is apparently dependent not only on astic based method where rules are applied that in
the current processing parameters, but also on the the limit are solutions of the governing differential
sequence of previous solidification conditions to equations [17,18]. Although these techniques do
which the system has been subjected. Recently, the not capture the same level of detail, nor have they
effect of changing the solidification velocity during previously incorporated all of the growth phenom-
directional solidification has been investigated; the ena, they are able to simulate grain structures on
dendritic structure that develops depends on the much larger simulation domains than is possible
effectiveness of different selection mechanisms using the phase field method. CA models trace the
including dendrite branching and competitive position of a sharp solid/liquid interface, while the
growth [7,8]. phase field method expresses it as a transition layer
Many theoretical approaches have been used to with a finite thickness that separates the liquid and
calculate the morphology and size of dendrites. solid phases. However, molecular dynamic studies
Initially, analytical models of the primary dendrite show that this thickness should only extend over a
tip were proposed that assumed the dendrite to be few atomic dimensions [19]. To capture the inter-
ideally spaced, i.e. only an averaged interaction face, the phase field method requires a very fine
was assumed between the neighboring dendrites grid, leading to very high computational cost.
[9,10]. For the primary dendrite spacing, l1, most Compared to the phase field method, the CA mod-
of the theoretical treatments give the following els can run on a much coarser grid and, therefore,
functional form: are more computationally efficient when solved
over the same domain.
l1 ⫽ A·G⫺0.5V⫺0.25⌬T0.25
0 (1)
In CA models, kinetic undercooling is often neg-
where G is the temperature gradient, V the growth lected, and the velocity of the interface is determ-
velocity, ⌬T0 the freezing range of the alloy and ined by other methods. The most commonly
A is a function of the alloy. It should be noted that applied method of determining the interface velo-
during the growth of a dendritic array, solute is city is the application of the Kurz–Giovanola–Tre-
rejected by each dendrite so that the solute-fields vedi (KGT) model [20], which is based on the
of neighboring dendrites interact. This feature assumption that the dendrite tip has an ideal para-
determines the small range of dendrite spacings bolic shape and is advancing with a steady state
that are stable with respect to branching and over- velocity. The resulting relationship between the
W. Wang et al. / Acta Materialia 51 (2003) 2971–2987 2973

growth velocity and the total undercooling of the an approach described as the decentered octa-
tip has been widely used by other researchers in hedron technique [23].
simulations of grain growth. Since kinetic These CA models are capable of simulating
undercooling is neglected, the only contributions many solidification processes on a macroscopic
to the total undercooling at the tip arise from the scale. However, if the solute diffusion equations
effect of solute pileup in front of the interface and are not solved, the dendritic structure remains
from surface tension. This model is a reasonable unknown. In order to simulate the dendrite mor-
approximation for the average behavior of a pri- phology and the extent of micro-segregation, the
mary tip, but cannot be used to simulate the inter- CA method can be coupled to a finite element or
action of dendrite tips that are not ideally spaced. the finite difference model to solve for solute dif-
Neither is it valid for simulating the development fusion, termed a modified cellular automaton
of the secondary and tertiary arms where the solute (MCA) model [24]. One of the greatest challenges
fields interact in a complex manner. of these MCA models is the method of determining
Crystallographic anisotropy is another important the growth velocity of solid/liquid interface at all
factor that affects the solidification process. For points on the dendrite, not just at the tip.
cubic metals, ⬍100⬎ are the preferential growth In this paper, a CA model is introduced in which
directions, and the growth velocity along these constitutional undercooling and curvature
directions is faster than other directions. Therefore, undercooling are related to the growth velocity of
in the directional solidification process, grains with the solid/liquid interface. The solute distribution in
one of the ⬍100⬎ directions closely aligned with the liquid causing constitutional undercooling is
the temperature gradient will overgrow those with calculated using a finite difference (FD) solution
larger deviations from ⬍100⬎. In order to account of solute diffusion. The crystallographic aspect of
for the effect of crystallographic anisotropy in CA, growth is considered by adapted implementation of
a neighborhood selection procedure within a the decentered square/octahedron growth algorithm
square grid was developed [21,22]. However, this developed by Gandin and Rappaz [23]. The
technique only allows limited deviations in the resulting model has several novel aspects: the treat-
growth direction that depend on the complexity of ment of the tip velocity using an iterative scheme;
the network; different types of neighborhood inter- grid independence; and scaling using a blurring of
actions will result in the prediction of different the solid–liquid interface. These novel aspects
occupancy of cells. allow the model to be applied to predict the den-
Another approach to accounting for crystallo- drite spacing selection mechanisms and the effect
graphic anisotropy is the decentered square growth of perturbation of solidification velocity on the
technique developed by Gandin and Rappaz [23]. dendritic microstructure including growth history
They assumed, in 2D simulations, that grains of dependence, using as an example a binary approxi-
cubic metals always grow in a square shape with mation of a nickel-based superalloy.
the diagonal aligned with the preferential ⬍10⬎
direction. The rate of growth of the tip (corner of
the square) was then calculated using the KGT 2. Model theory
approximation. These growing squares were super-
imposed upon a regular CA grid, but with the pre- The present model combines a continuous func-
ferred growth direction of the grain represented by tion cellular automaton description of grain growth
the misorientation of the square’s diagonal from with a finite difference computation of solute dif-
the grid direction. A growing square captures the fusion. Both the CA and FD components of the
neighboring cells when it touches their centers. At model run on the same regular spatial square grid
this stage, a new square, decentered from the old and with the same time step. Each spatial cell can
one, is then associated with the newly captured have three states: liquid, solid and growing (i.e. a
cell. The decentered square method was then mixture of solid and liquid). The model builds
extended for three-dimensional simulations using upon prior work by some of the authors [25–27],
2974 W. Wang et al. / Acta Materialia 51 (2003) 2971–2987

and novel extensions to track the tip location due 1 0


to solute and curvature terms in a grid-independent C∗L ⫽ C0 ⫹ [T ⫺⌫k⫺T], (8)
ml liq
manner within a CA framework are described
below. where C0 is the initial concentration, T0liq the liqui-
dus temperature at C0, ␬ the average curvature of
2.1. Solute diffusion analysis the solid/liquid interface, and T the local tempera-
ture. ␬ is calculated by an approximate method
Assuming the equilibrium partition coefficient, described by Nastac [28]:
k, is a constant, the partitioning of solute in the
growing cell is determined by: ␬⫽
1
⌬x 再
1⫺
2
f ⫹
N⫹1 S 冋 冘 册冎
N

j⫽1
fSj , (9)
CS ⫽ kCL, (2)
where N is the number of neighboring cells.
where CS and CL are the average solute concen- The solute concentration gradient, GC, which is
trations of the solid and liquid, respectively, in a calculated using values from a previous time step,
specific cell. An equivalent concentration, CE, is is then substituted into Eq. (7) to calculate the con-
defined as: centration in the growing cells at the current time
CE ⫽ CL(1⫺fS) ⫹ CSfS ⫽ CL[1⫺(1⫺k)fS], (3) step. Eq. (6) is solved using an explicit FD scheme,
allowing the change in the solid fraction in the
where fS is the solid fraction. The diffusion equa- growing cells and the concentration in other cells
tions in the liquid and solid can now be expressed to be calculated. Although nucleation models are
in a general form: included in the code, this paper only presents
∂CE results for growth of seeded grains. Each seed has
⫽ ⵜ·(DEⵜCL), (4) a crystallographic orientation, q, associated with it.
∂t
where DE is an equivalent diffusion coefficient, 2.2. Grain growth in 2D
expressed as:
DE ⫽ (1⫺fS)DL ⫹ fSkDS. (5) The 2D decentered square algorithm developed
by Gandin and Rappaz [23] is not directly appli-
DL and DS are diffusion coefficients in liquid and cable for simulating grain growth from the liquid
solid, respectively. The equivalent diffusion Eq. (4) when solute diffusion is considered. In order to
can be rewritten as: solve the diffusion equation, the solid fraction
∂fs within cells must be allowed to change continu-
CL(1⫺k) ⫽ ⫺ⵜ·(DEⵜCL) ⫹ [1⫺(1 (6) ously and smoothly from 0 to 1 if numerical insta-
∂t
bilities are to be avoided. The KGT model is no
∂CL longer appropriate since the model has already
⫺k)fS] .
∂t taken into account the effect of solute diffusion. A
novel modified 2D decentered square growth
The solute concentration, CL, in the liquid within algorithm, that has been developed to overcome
a growing cell, is given by: these problems, is presented below. A schematic
1⫺fS illustration of the modified 2D decentered square
CL ⫽ C∗L ⫺ ⌬xGC, (7) growth algorithm is shown in Fig. 1. Only nine
2
cells of a whole CA network are displayed for clar-
where C∗L is the concentration in the liquid at the ity. A grain with misorientation angle, q, has
solid/liquid interface, ⌬x the cell size, and GC the nucleated at the central cell and is growing into a
concentration gradient in front of the solid/liquid square shape. The increment of the half width of
interface. C∗L can be determined from a linearised the square is related to the increment of fraction
equilibrium phase relation: solid by:
W. Wang et al. / Acta Materialia 51 (2003) 2971–2987 2975

Fig. 1. Schematic diagram of modified 2D decentered square CA growth algorithm: (a) a grain nucleated in the central cell is
growing within a square envelope; (b) the grain has captured the four neighboring cells and four new squares are associated with
them (only parts of these squares are shown for clarity); (c) the grain captures more cells.

⌬L ⫽ ⌬x·⌬fS. (10) hedron along the diagonal towards the corner point
that touches the neighboring cell generates the new
When the misorientated square envelope asso-
octahedron that can keep growing along the same
ciated with the solid in the central cell touches its
⬍100⬎ directions as the nuclei.
four neighboring cells, their state indices change
Thus, constitutional and curvature undercooling,
to that of the original. In each of these ‘touched’
together with crystallographic orientation, are all
neighboring cells, a new square is created with the
incorporated in the model through the application
same orientation as the solid cell that grew into it.
of this modified decentered square/octahedron
One corner of each of the new squares overlaps
technique within a CA-FD framework.
the original square, hence the term “decentered
square” method. The growth velocities of these
new squares can be different from the original 2.4. Solution techniques
square depending on the solute diffusion or local
temperature. Parts of the growing new squares are Simulations were performed on a regular square
shown in Fig. 1(b). The misoriented solid squares grid in 2D and a regular cubic grid in 3D with a
progressively encroach on more reference cells, as cell edge length of 5 µm. A range of solidification
illustrated in Fig. 1(c). rates with a constant thermal gradient (varying in
the vertical direction) were used in the simulations.
2.3. Grain growth in 3D In the 2D simulations, periodic boundary con-
ditions were applied to the left and right sides of
The 2D modified decentered square growth tech- the domain, while a zero flux boundary condition
nique was also extended into three dimensions, was applied to the top and bottom of the domain.
termed the 3D modified decentered octahedron In the 3D simulations, a periodic boundary con-
growth technique. The underlying approach is the dition was applied to the left and right, front and
same. A misoriented grain is considered to back surfaces of the domain, respectively, while a
nucleate and grow within a regular array of cubic zero flux boundary condition was applied at the top
cells having an octahedral shape. When the solid and bottom.
grows along one of the crystallographic cube direc- A moving frame of reference technique was
tions to touch any of the six neighboring cubic applied in the simulations of directional solidifi-
cells, growth continues from that point along cation. When the temperature at the bottom
⬍100⬎ directions into the captured cells by gener- decreased to a predetermined value, the bottom
ating a new octahedron. Shifting the original octa- layer of cells was removed, all the other layers are
2976 W. Wang et al. / Acta Materialia 51 (2003) 2971–2987

shifted down by one cell and a new layer with the was assumed that no nucleation occurs in the bulk
initial concentration added to the top of the simul- liquid; only the growth of the seeds placed at the
ation domain. This technique allows the dendrites base of the directionally solidified ingot is con-
to grow to a sufficient length to ensure a stable sidered. Therefore, the final microstructure
state is reached, without increasing the domain size develops only by branching and overgrowth mech-
and hence reducing the computational resources anisms from these initial seeds. The undercooling
required. for the nucleation of the seeds was taken to be 4
K throughout this work. Several simulations were
run for directional solidification from different
3. Two dimensional simulations numbers of seeds (from 2–60) and the preferred
growth direction of all the seeds, ⬍10⬎, were per-
Numerical simulations of directional solidifi- fectly aligned with the grid. Different growth
cation of a binary approximation of a nickel-based phenomena are observed in the simulations with
alloy were performed to investigate the influence different numbers of seeds, which can be roughly
of a number of material parameters and process divided into three categories: i) steady growth; ii)
variables on the development of the dendritic struc- dendrite branching; and iii) competitive growth.
ture. The material properties and model parameters Fig. 2 shows the dendritic structure after 40 s of
used in the simulations are given in Table 1. The growth in these simulations with: (a) 8 seeds (375
simulations were carried out in both two and three µm initial spacing); (b) 12 seeds (250 µm initial
dimensions, with the results presented and dis- spacing); and (c) 20 seeds (150 µm initial spacing).
cussed in this and the next section, respectively. In these simulations, the secondary dendrites are
In this section, 2D simulations, the effects on the shown to develop perpendicular to the macroscopic
dendritic structure of the initial nucleation density, solidification direction and to coarsen with distance
rate of solidification and of perturbation of the sol- from the primary dendrite tips, as shown in Fig.
idification rate are considered in turn. 2(b1) and (b2). Unlike the primary dendrites from
which they emanate, the secondary dendrites are
3.1. The effect of nucleation density not identically arranged nor are they of uniform
size, indicating a degree of randomness in their
In the following simulations the same imposed nucleation and competition in their growth. In case
solidification conditions were used: i) a thermal (a), where the primary dendrite spacing is 375 µm,
gradient of 12 K/mm, and ii) a growth velocity of some tertiary dendrites are observed emanating
150 µm/s (giving a cooling rate of ⫺1.8 K/s). It from secondary dendrites and growing in the same

Table 1
Material properties and model parameters used in the simulations

Property Variable Value

Liquidus temperature TLIQ 1609 K


Liquidus slope m ⫺10.9 K/wt%
Partition coefficient k 0.48
Diffusion coefficient in liquid DL 3.0×10⫺9 m2/s
Diffusion coefficient in solid DS 3.0×10⫺12 m2/s
Gibbs Thomson Coefficient ⌫ 1.0×10⫺7 K·m
Initial concentration C0 4.85 wt%
Cell size x 5 µm

Time step t
1
ⱕ min
5 冉
⌬x ⌬x2
,
Vmax DL冊S
W. Wang et al. / Acta Materialia 51 (2003) 2971–2987 2977

Fig. 2. Simulation results of columnar dendrites developing form (a) 6, (b) 12 and (c) 20 seeds after 40 s of growth, under conditions
of a temperature gradient of 12 K/mm and a solidification rate of 150 µm/s. (a1) is an amplified (×2) section from (a) showing a
tertiary dendrite emanating from a secondary dendrite and blocked by another secondary dendrite. (b1) and (b2) show amplified (×2)
dendritic structure at the tip and about 2 mm below the tip position from (b). (c1) shows amplified (×2) dendritic structure about
2mm below the tip position from (c).

direction of the primary dendrites, but they are all where the initial primary spacing is even smaller
blocked by other secondary dendrites, with one (150 µm), the secondary dendrites are less
example shown in Fig. 2(a1). In case (b), no appar- developed and have a smaller spacing than in case
ent tertiary dendrites are observed due to smaller (b). In these simulations, there is no indication of
primary dendrite spacing (250 µm). In case (c), dendrite branching or overgrowth and, conse-
2978 W. Wang et al. / Acta Materialia 51 (2003) 2971–2987

quently, no noticeable deviation from the initial strong that competitive growth occurs, with the
spacings during 40 s of solidification. The primary spacing increasing by an overgrowth mechanism.
spacings remain unchanged when the three simula- lS1 = l01: when lS falls between 130–430 µm, the
tions were continued to 200 s. simulation results show no change in primary spac-
From Fig. 2, a conclusion can be drawn that the ing, with a final l1 equal to l01; therefore, the initial
model predicts that columnar dendrites can pro- spacing is stable for the given thermal conditions.
ceed in a stable state 40 s after nucleation with a lS1 ⬍ l01: when lS is larger than 430 µm, the gaps
quite wide range of primary spacings, from 150– between the original dendrites are so large that ter-
375 µm for the same thermal conditions. This tiary dendrites can form reducing the spacing by a
raises the question of how wide a range is possible, branching mechanism.
and what will happen if the initial seed spacing is
outside this range. Fig. 3(a) and (b) show the den- 3.2. Effect of solidification rate
drite development during the initial growth stage
in two simulations from 2 and 60 seeds, respect- The stable primary dendrite spacing reached in
ively. In Fig. 3(a) the dendrites develop from a the simulations starting from a wide range of initial
sparse distribution of seeds with an initial spacing seed spacings with a constant solidification rate of
of 1500 µm. As solidification proceeds, many 150 µm/s is shown in Fig. 4. A series of simula-
small tertiary dendrites emanate from the second- tions with the same initial nucleation conditions
ary dendrites where the undercooling is progress- were then carried out with a range of solidification
ively increased. Some of the tertiary dendrites rates from 10–300 µm/s, all at the same constant
grow very quickly and catch up with the two orig- temperature gradient of 12 K/mm. The distribution
inal primaries. As a result, l1 is significantly of calculated values of the average dimensionless
reduced, reaching 214 µm after 40 s. Fig. 3(b) primary dendrite spacing, defined as l⬘ = lmC0 /
shows the dendrite development from a much more ⌫, against dimensionless solidification rate, defined
closely spaced set of seeds that have an initial as V⬘ = V⌫ / (DLmC0), for various initial nucleation
spacing of 50 µm. Within 12 s, a few dendrites conditions at different solidification rates is shown
have grown ahead, the others blocking their growth in Fig. 5. The graph is plotted on a logarithmic
by the formation of secondary dendrites. After 40 scale, with two sets of axis, one for dimensionless
s of growth the dendrite spacing l1 increases also values and the other for dimensional values. The
to 214 µm. When the two simulations continue to points at each solidification rate represent the final
200 s, l1 in case (a) increases slightly to 231 µm stable primary dendrite spacing in the simulation
while in case (b) it remains unchanged. Although with different numbers of seeds. It is noticed in
the initial spacings in the two simulations are very the figure that fewer points are obtained at some
different, one is 1500 µm and the other only 50 solidification rates that others, because simulations
µm, the final dendritic structures show little differ- starting with different numbers of seeds could
ence, and with a similar primary dendrite spacing. result in the same final primary dendrite spacing.
More simulations with different initial seed spa- A noticeable characteristic of the results is that all
cings (l01) were run to explore the relationship these points are all located within a narrow band
between the final primary dendrite spacing and demarcated by two parallel lines:
initial seed spacing, with the results given in Fig.
4. The initial seed spacing ranged from 50–1500 l⬘1max ⫽ 1.49 ⫻ 104(V⬘)⫺0.288
µm, resulting in predicted final stable primary den- . (11)
l⬘1min ⫽ 4.51 ⫻ 103(V⬘)⫺0.288
drite spacing (lS1) of 130–430 µm, or an upper limit
of approximately 3 times the lower limit. The These two lines give the upper and lower limits
stable spacing can be divided into three categories: of the allowable primary dendrite spacing, both of
which are power functions of solidification rate.
lS1 ⬎ l01: when l01 is less than 130 µm, the diffusion The value of the exponent is marginally higher
field interaction of neighboring dendrites is so than the value of 0.25 predicted by early analytical
W. Wang et al. / Acta Materialia 51 (2003) 2971–2987 2979

Fig. 3. Predicted evolution of dendritic structure in directional solidification processes with (a) 2 and (b) 60 seeds nucleated at the
bottom under conditions of a temperature gradient of 12 K/mm and a solidification rate of 150 µm/s.

solutions of dendrite growth (Eq. (1)), but well whereas Hunt and Lu [12] suggested a factor of
within the range of values predicted by Hunt and approximately two. In practical solidification there
Lu [12] using a quasi-3D simulation (i.e. 2D are inevitable fluctuations in the processing con-
axisymmetric). The upper limit of the l⬘1 distri- ditions, which may affect the dendrite growth and
bution is about three times that of the lower limit, thereby change the final l⬘1.
2980 W. Wang et al. / Acta Materialia 51 (2003) 2971–2987

V = 150 µm / s; (ii) 40–50 s, V = 300 µm / s; (iii)


50–60 s, V = 75 µm / s; and (iv) 60–75 s, V =
150 µm / s. The primary dendrite spacing remains
unchanged in the first 40 s of growth, maintaining
the initial seed spacing of 500 µm (see Fig. 6(a)).
Doubling the solidification rate causes almost
immediate dendrite branching and a rapid increase
in the density of primary dendrites reducing l1
from 500 to 188 µm (see Fig. 6(b)). Subsequent
reduction in the growth rate to half of the original
value, leads to a very slow increase in dendrite
spacing, which continues on returning to original
growth velocity to approach a stable state spacing
of 231 µm (see Fig. 6(c) and (d)). Fig. 6(e) shows
the global structure, from which the generation and
annihilation of dendrites can be clearly observed.
Similar perturbations of the solidification rate
were then applied to all the simulations with differ-
Fig. 4. Final stable primary dendrite spacing (lS1) versus the ent numbers of seeds at each solidification rate.
initial seed spacing (l01). The predicted dimensionless average primary den-
drite spacing l⬘1 against dimensionless solidifi-
cation rate V⬘ from different nucleation conditions
after perturbations of the withdrawal velocity are
plotted in Fig. 7. The upper and lower limits of
the l⬘1 distribution after perturbation of the solidi-
fication rate are expressed as:
l⬘1max ⫽ 1.22 ⫻ 104(V⬘)⫺0.291
. (12)
l⬘1min ⫽ 5.24 ⫻ 103(V⬘)⫺0.291
Comparing the results to those without pertur-
bation (Eq. (11)), the values of the exponents are
similar, but the upper limit is now about twice that
of the lower limit, which is in better agreement
with the suggested values of Hunt and Lu [12]. The
change in the ratio between the upper limit and the
Fig. 5. Distribution of primary dendrite spacing against solidi-
fication rate. lower limit shows that perturbation of the growth
rate reduces the dendrite spacing to a narrower
range than the steady state growth condition.
3.3. The effect of perturbations of solidification The Kurz–Fisher prediction [10] and previous
rate experimental results [7,29,30] are also shown for
comparison. However, the simulated primary den-
Since perturbations of the control parameters in drite spacings are higher than experimental results.
practical solidification are usually unavoidable, This discrepancy may be due to the experiments
artificial perturbation of the solidification rate has being carried out in 3D, and the simulations in 2D.
been incorporated in the simulations to investigate In order to carry out direct comparison between
its effect. Fig. 6 shows the change of the dendritic simulations and experiments, some 3D simulations
structure in the simulation with 6 seeds, undergo- were carried out, with results displayed and dis-
ing the following cycle of perturbation: (i) 0–40 s, cussed in the next section.
W. Wang et al. / Acta Materialia 51 (2003) 2971–2987 2981

Fig. 6. Simulated dendrite structures undergoing a cycle of perturbation of the withdrawal velocity around a mean value. (a) t =
40 s, V = 150 µm / s, (b) t = 50 s, V = 300 µm / s, (c) t = 60 s, V = 75 µm/ s, (d) t = 75 s, V = 150 µm/ s, (e) the global structure.

4. Three dimensional simulations composed of 100 × 100 × 100 cells with a cell size
of 5 µm. The grain is nucleated at the center of
A 3D simulation was first performed for a single the box, with an initial misorientation characterized
grain nucleated at a undercooling of 5 K and grow- by three Euler angles (0°, 20°, 0°). The dendritic
ing under conditions of a temperature gradient of morphology of the grain after 1 s of growth is
12 K/mm and a constant cooling rate of ⫺1.8 K/s. shown in Fig. 8(c). The envelope of the dendritic
The simulation was run on a regular cubic grid grain shows excellent accordance with both the
2982 W. Wang et al. / Acta Materialia 51 (2003) 2971–2987

analytical and numerical predictions of a dendritic


grain envelope by Gandin and Rappaz [23], as
shown in Fig. 8(a) and (b). A major improvement
of the current model is its capability of reproducing
the micro-scale dendritic structure, such as primary
and secondary dendrites, and hence allows the
interaction of the solute fields of both dendrite tips
and between grains to be studied.
Using this model, simulations of directional sol-
idification were then performed on a domain of
150 × 50 × 250 cells with the same cell size, under
conditions of a temperature gradient of 12 K/mm
and a constant solidification rate of 150 µm/s. The
Fig. 7. Distribution of primary dendrite spacing against solidi-
fication rate after a cycle of perturbation of the withdrawal velo-
effects of the initial nucleation density at the base
city. and perturbations of the solidification rate on the
dendritic structure are considered in turn.

Fig. 8. 3D views of (a) analytical and (b) numerical predictions of a dendritic grain envelope without considering solute diffusion
(after [23]), and (c) simulated dendritic structure coupled with solute diffusion. Simulation (c) was run on a regular cubic grid with
a cell size of 5 µm under conditions of a temperature gradient of 12 K/mm and a cooling rate of ⫺1.8 K/s. The original misorientation
of the grain is characterized by three Euler angles (0°, 20°, 0°).
W. Wang et al. / Acta Materialia 51 (2003) 2971–2987 2983

4.1. Effect of initial nucleation density emanating from the secondary arms at the initial
growth stage, they are blocked by the secondary
Similar to the preliminary 2D simulations, the dendrites growing from the primary dendrites, as
initial seed spacing has an important effect on the shown in Fig. 9(a1). The three primary dendrites
development of columnar dendrites. Fig. 9 shows proceed at the same velocity along the z direction,
the development of three dendrites at different time and the primary dendrite spacing remains
steps: (a) 2 s, (b) 4 s, (c) 6 s and (d) 40 s. Three unchanged (see Fig. 9(b), and (c)). At 6 s the mov-
seeds are nucleated at the bottom, equally spaced ing frame of reference technique is activated and
along the x direction, with the ⬍100⬎ directions the whole domain is cooled down at a constant rate
perfectly aligned with the coordinate axis. of ⫺1.8 K/s. The dendrites keep growing, but their
Although some tertiary dendrites are observed tip position remains roughly unchanged relatively
to the moving reference frame. This simulation was
continued until 40 s from the start of growth, and
no apparent change was observed in the primary
dendrite spacing. Because of the periodic boundary
conditions along x and y directions, these primary
dendrites are of quadratic arrangement in the trans-
verse section perpendicular to the macroscopic sol-
idification direction. This can be viewed by rep-
licating the transverse slide shown in Fig. 9(d2)
along both the x and y directions infinitely. The
primary dendrite spacing along the y direction is
equal to the box size along y direction, 250 µm,
and the primary dendrite spacing along x direction
is the box size along x direction divided by 3,
which is also 250 µm.
A 3D example of steady growth for a simulation
with 3 seeds has been shown in Fig. 9, and two
examples of transient growth are shown in Fig. 10
(1 seed) and Fig. 11 (15 seeds). When only one
seed is placed at the bottom (see Fig. 10(a)), it
grows freely along all the directions at the initial
solidification stage. The secondary dendrites along
the x direction grow faster than in the z direction,
because of greater tip undercooling. Many small
tertiary dendrites were observed emanating from
the fast growing secondary dendrites, growing in
the same direction as the original primary dendrite.
The newly formed tertiary dendrites have a random
initial spacing, and some of them grow very
quickly at the expense of others (see Fig. 10(b)).
After 6 s of solidification, only 6 primary dendrites
survive the competitive growth; among which 5
Fig. 9. 3D simulation for 3 seeds at a solidification time of have developed from the tertiary dendrites and 1
(a1) 2 s, (b1) 4 s, (c1) 6 s, and (d1) 40 s. (a2) transverse slice
at z = 150µm, t = 2 s, (b2) transverse slice at z = 450 µm, t
directly from the seed (see Fig. 10(b)). The 6 den-
= 4 s, (c2) transverse slice at z = 750 µm, t = 6 s, and (d2) drites all continue to grow for 40 s, without any
transverse slice at z = 750 µm, t = 40 s. obvious change in the structure (see Fig. 10(b)),
2984 W. Wang et al. / Acta Materialia 51 (2003) 2971–2987

Fig. 10. 3D simulation for 1 seed at a solidification time of Fig. 11. 3D simulation for 15 seeds at a solidification time of
(a1) 2 s, (b1) 4 s, (c1) 6 s, and (d1) 40 s. (a2) transverse slice (a1) 2 s, (b1) 4 s, (c1) 6 s, and (d1) 40 s. (a2) transverse slice
at z = 150 µm, t = 2 s, (b2) transverse slice at z = 450 µm, t at z = 150 µm, t = 2 s, (b2) transverse slice at z = 450 µm, t
= 4 s, (c2) transverse slice at z = 750 µm, t = 6 s, and (d2) = 4 s, (c2) transverse slice at z = 750 µm, t = 6 s, and (d2)
transverse slice at z = 750 µm, t = 40 s. transverse slice at z = 750 µm, t = 40 s.

resulting in an average primary dendrite spacing


along x direction of about 250 µm.
Fig. 10 gives an example of solidification with metrical development along both the x and y direc-
sparse nucleation, during which both branching tions (see Fig. 11(b2)). As solidification continues,
and competitive growth phenomena are observed. the growth velocities of the 15 dendrites begin to
A contrary example with dense nucleation is vary, leading to some of the dendrites to slow
shown in Fig. 11, where 15 seeds were set at the down (see Fig. 11(c1)). This competitive growth
bottom (50 µm initial spacing). Because of the results in only 10 dendrites surviving after 40 s
dense packing along x direction, the growth of sec- of solidification (see Fig. 11(c1)), which gives an
ondary dendrites along this direction is suppressed, average primary dendrite spacing along x direction
and the secondary dendrites show severe asym- of about 75 µm.
W. Wang et al. / Acta Materialia 51 (2003) 2971–2987 2985

4.2. Effect of perturbation on pulling velocity to the reference frame, as shown in Fig. 12(b1).
Since the primary dendrite spacing along the x and
Perturbations of the withdrawal velocity were y directions are already very small, no branching
also added to the three 3D simulations. All the 3D was observed. On the contrary, two dendrites slow
simulations were subjected to the same cycle of down and die out, as shown in Fig. 12(b1). It is
perturbations as described for the 2D simulations. because of the decrease in the pulling velocity that
Fig. 12 shows the change of the dendritic structure the relatively stable state the dendrites have achi-
in the simulation starting with 15 seeds. When the eved at 40 s has broken down and introduced devi-
velocity is increased, the tip undercooling also ations in spacings that lead the dendrites to select
increases, thus the tip position falls back relative a more stable spacing. When the pulling velocity
is decreased 10 s later, one more dendrite slows
down, as shown in Fig. 12(c1). Finally, 7 of the
10 dendrites survive the perturbation of the pulling
velocity, and the average primary dendrite spacing
along x direction increases to 107 µm
In 3D simulations, the partitioned solute at the
dendrite tip can diffuse away more efficiently than
in 2D simulations; therefore the primary dendrite
spacing in 3D is also smaller than in 2D. The upper
and lower limits of dimensionless primary spacing
l⬘1 against dimensionless solidification rate V⬘ in
3D may be obtained simply through modification
of the predicted limit lines in 2D (Eq. (12)) by mul-
tiplying by a factor, which is chosen as a factor of
√2 / 2 in this work:
l⬘1max ⫽ 8.63 ⫻ 103(V⬘)⫺0.291
. (13)
l⬘1min ⫽ 3.70 ⫻ 103(V⬘)⫺0.291
These two lines are plotted in Fig. 13, together
with the predicted average primary dendrite spac-

Fig. 12. 3D simulation results for 15 seeds under perturbation


of withdrawal velocity at a solidification time of (a1) 40 s, (b1)
43 s, (c1) 60 s and (d1) 75 s. (a2) transverse slice at z = 750 Fig. 13. Predicted 3D upper and lower limits of the distri-
µm, t = 40 s, (b2) transverse slice at z = 625 µm, t = 50 s, (c2) bution of primary dendrite spacing, with the 3D simulation
transverse slice at z = 875 µm, t = 60 s, and (d2) transverse results. They are compared with Hunt–Lu model [12] and pre-
slice at z = 750 µm, t = 750 s. vious experimental results [7,29,30].
2986 W. Wang et al. / Acta Materialia 51 (2003) 2971–2987

ing direction in the 3D simulations both without rower distribution after perturbation. The upper
and with perturbation. The Hunt–Lu prediction limit of the distribution of primary dendrite arm
[12] and previous experimental results [7,29,30] spacing is about two times of the lower limit. The
are also shown for comparison. It can be seen that average primary spacing was found to be depen-
the Hunt–Lu prediction is nearly a midline between dent not only on the current growth conditions, but
the prediction upper and lower limit lines, and the also on the way in which those conditions were
predicted 3D results and the experimental values reached.
fit well.
This result suggests that for applications such as
single crystal turbine blades, where as consistent a Acknowledgements
spacing as possible is desirable, then perturbation
in the process parameters early on in the formation The authors would like to thank Special Metals
of the dendritic microstructure might result in an Wiggin Ltd, Rolls-Royce plc, Wyman-Gordon Ltd,
improved product. Defects may also be reduced if QinetiQ and the EPSRC (GR/N14132) for financial
the spacing is in the center of the range. For support together with the provision of both
example, if the spacing is near the upper limit of materials and information. The authors gratefully
the stable range, the solute fields interact less, and acknowledge the assistance of their colleagues in
hence the region of undercooling between the den- the Materials Processing Group, ICSTM,
drites is larger, which may lead to defects such as especially Drs A. Kermanpur and R. Atwood.
stray grains, since any heterogeneous nuclei are
more likely to be activated.
References
5. Conclusions
[1] Flemings MC. Solidification processing. New York:
McGraw-Hill, Inc, 1974.
A combined CA-FD model was developed to [2] Kurz W, Fisher DJ. Fundamentals of solidification., 3rd
simulate the solidification process of binary alloys. ed Aedermannsdorf: Trans Tech Pub, 1992.
The velocity of the solid/liquid interface, determ- [3] McLean M. Directionally solidified materials for high
ined by the equilibrium conditions at the interface, temperature service. London: The Metals Society, 1983.
was calculated by considering both the solute and [4] Somboonsuk K, Trivedi R. Acta Mater. 1985;33(6):1051.
[5] Huang W, Geng X, Zhou Y. J. Cryst. Growth
curvature undercooling terms. A modified decent- 1993;134:105.
ered square/octahedron growth technique was [6] Wan X, Han Q, Hunt JD. Acta Mater. 1997;45(10):3975.
implemented to account for the effect of crystallo- [7] Wang W, Kermanpur A, Lee PD, McLean M, Wang X,
graphic anisotropy. Ward RM, Jacobs MH. An Investigation of the Effect of
Two and three dimensional simulations were Perturbations on Dendritic Growth in Nickel Base Super-
alloys. in Proceeding of the 2001 International Symposium
carried out for the directional solidification of a on Liquid Metal Processing and Casting. 2001. Santa Fe,
binary alloy with properties approximating a New Mexico AVS.
nickel-based superalloy. The initial seed density [8] Ma D. Metall. Mater. Trans. B 2002;33:223.
had a direct effect on the initial growth stage of [9] Hunt JD. Keynote Address: Cellular and primary dendrite
columnar dendrites, but had little effect on the pri- spacings. Solidification Cast. Met., Proc. Int. Conf. Solidi-
fication, Met. Society, 1979:3.
mary dendrite arm spacing and tip undercooling [10] Kurz W, Fisher DJ. Acta Mater. 1981;29:11.
after the dendrite growth reached a stable state. It [11] McCartney DG, Hunt JD. Metall. Trans. A
was found that the dendrite spacing selection had 1984;15(6):983.
a hysteresis whose value in dendrite branching is [12] Hunt JD, Lu S-Z. Metall. Mater. Trans. A 1996;27(3):611.
larger than that of dendrite competition. Pertur- [13] Warren JA, Boettinger WJ. Acta Mater. 1995;43(2):689.
[14] Boettinger WJ, Warren JA. Metall. Mater. Trans. A
bation on the growth velocity affects both the pri- 1996;27:657.
mary dendrite arm spacing and tip undercooling. [15] Loginova I, Amberg G, Agren J. Acta Mater. 2001;49:573.
The primary dendrite arm spacing has a much nar- [16] Karma A, Rappel W-J. Phys. Rev. E 1998;57(4):4323.
W. Wang et al. / Acta Materialia 51 (2003) 2971–2987 2987

[17] Gandin C-A, Rappaz M, Tintillier R. Metall. Trans. A Processes. Numerical simulation of dendrite mophology
1993;24(2):467. and grain growth with modified cellular automata. San
[18] Spittle JA, Brown SGR. J. Mater. Sci. 1995;30:3989. Diego, California: The Minerals, Metals & Materials
[19] Broughton JQ, Bonissent A, Abraham FF. J. Chem. Society; 1998.
Phys. 1981;74:4029. [25] Lee PD. In: Cantor B, O’Reilly K, editors. Solidification
[20] Kurz W, Giovanola B, Trivedi R. Acta Metall. and Casting. Bristol: Inst. of Physics, 2002. p. 121.
1986;34(5):823. [26] Lee PD, Atwood RC, Dashwood RJ, Nagaumi H. Mater.
[21] Xu X, Zhang W, Atwood RC, Sridhar S, Lee PD, McLean Sci. Eng. A 2002;328:213.
M, Drummings B, Ward RM, Jacobs MH. In: Proc. Int. [27] Xu X, Zhang W, Lee PD. Metallurgical and Materials
Sym. on Liq. Metal Proc. & Casting. Grain size predic- Transactions a—Physical Metallurgy and Materials
tions in VAR: A critical comparison of micromodelling Science 2002;33(6):1805.
approaches. Santa Fe, New Mexico: AVS; 1999. [28] Nastac L. Acta Mater. 1999;47(17):4253.
[22] Nastac L, Stefanescu DM. Modelling Simul. Mater. Sci. [29] Davies HA, Shohojin N, Warrington DH. Rapid solidifi-
Eng. 1997;5:391. cation processing principles and technologies II. Claitor’s
[23] Gandin C-A, Rappaz M. Acta mater. 1997;45(5):2187. Publishing Div, 1980.
[24] Dilthey U, Pavlik V. In: Proceedings of the Eighth Inter- [30] Kermanpur A, Varahraam N, Engilehei E, Mohammadza-
national Conference on Modeling of Casting and Welding deh M, Davami P. Mater. Sci. Tech. 2000;16:579.

You might also like