1. Introduction
A great deal of work has been published concerning the interaction of insoluble particles with an advancing solid–liquid interface [
1,
2,
3,
4,
5,
6,
7] owing to its importance in various fields, including the processing of composites (metal matrix or polymer matrix) [
8,
9], preservation of biological materials [
10], and freezing of soils [
11,
12]. It has been generally recognized that certain types of particles are pushed by the solid–liquid interface advancing at small velocities. There is a critical growth velocity,
VC, or pushing-to-engulfment transition (PET) velocity, above which a particle ceases to be pushed but is engulfed in the growing crystal. Engulfment likely leads to a uniform distribution of particles within a solid crystal, while pushing results in particle segregation in the solidified material.
The PET velocity has been defined in various forms [
1,
2,
3,
4,
5,
6,
7], but generally, it is proportional to the difference in interfacial energies, Δ
σ, of the particle/material system, the minimum separation between the particle and the advancing solid–liquid interface,
h, and the reciprocal of particle radius,
R. For discussion purpose,
VC can be written as:
where
α is a constant with a value of 1, 3/2, or 4/3 from various models [
1,
2,
4], and Δ
σ is defined as:
Equation (2) is proposed based on interfacial energy interactions [
1,
2,
4], suggesting that when the solid–particle interfacial energy,
σSP, is greater than the sum of the solid–liquid interfacial energy,
σSL, and the particle–liquid interfacial energy,
σPL, a repulsive force arises that acts on the particle as the solid–liquid interface approaches the particle within
h. Analytical models in the form of Equation (1) predict that, for a 2-µm particle satisfying Equation (2) in a pure material,
VC is in the order of a few micrometers per second, which is in general agreement with experimental observation, but the
h value has to be altered in order to fit experimental data [
13].
In an alloy or in a pure material where impurity has to be considered, the boundary conditions at the advancing solid–liquid interface become too complicated to obtain simple analytical solutions on
VC. Pötschke and Rogge [
6] developed a numerical model to describe the influence of solute content on
VC. Kao et al. [
6] presented a more rigorous numerical model for the capturing of a foreign particle by an advancing solid–liquid interface in binary alloys. These models predict that
VC is in the form:
where
G is the temperature gradient,
C0 is the alloy composition, and
β and
ε are positive constants in the range of 0–1 [
4,
6,
7]. The values of
α,
β, and
ε are affected by Δ
σ [
7].
VC in a binary alloy can be orders of magnitude smaller than that in a pure material [
7]. One issue with the numerical models is the lack of accurate data for Δ
σ, especially when a solute is enriched in the thin gap between the particle and the advancing solid–liquid interface. Interface interactions related to Δ
σ define the morphology of the solid–liquid interface beneath the particle and the resultant
h, which, in turn, affects the force balance during particle pushing.
A major drawback in the numerical models is associated with the length scales of
h and that of interface morphology. The repulsive force is related to the van der Waals forces, which become substantial when
h is in the order of a few atomic diameters. In most binary alloys, the first solid crystal precipitating from a liquid usually consists of a non-faceted phase. The solid–liquid interface of such a non-faceted crystal is only a few atoms thick and is zigzag at the nanoscale. This means that the minimum separation,
h, is in the same order of interface roughness at the nanoscale level and should vary from atom to atom substantially. However, numerical models so far developed have been focusing on calculating a macroscopically smooth solid–liquid interface and a resultant constant
h using rigorous macroscopic boundary conditions. Consequently, these models may not be meaningful in describing the van der Walls forces since the
h value cannot be obtained simply by solving the macroscopic diffusion equations. It is noted that numerical models predict that particle pushing occurs when the depth of a depression on the solid–liquid interface is around 0.8–1.2
R [
7]. Such s deep depression on the advancing solid–liquid interface during steady-state particle pushing has not yet been observed experimentally. Much sophisticated theoretical work involving multi-length scale modeling is required to describe particle pushing.
Experimental validation of model prediction yielded conflicting results. The predicted
VC is in the order of a few micrometers per second in a pure material [
1,
2,
3,
4,
5,
6,
7] and orders of magnitude smaller than that in alloys [
6,
7]. Such a
VC is much smaller than the average growth rates of solid solidified under normal casting conditions. As a result, particles should have been engulfed by the growing solid during the normal solidification of composite materials. Experimental data on particle pushing during dendritic solidification in metal matrix composites do not support theoretical predictions [
14,
15,
16,
17,
18]. Instead of being engulfed, particles are actually pushed by dendrites growing at rates a few orders of magnitude greater than predicted [
14,
15,
16,
17,
18,
19]. In fact, except for particles that nucleate the growing solid phase, i.e., a particle that exists in the center of a growing grain, there is a lack of convincing experimental observation conforming that particles being actually engulfed within a dendrite arm under cooling conditions, so far tested regardless of Δ
σ being positive [
14,
15] or negative [
17,
18,
19]. Obviously, convection in liquid plays a significant role in particle pushing [
20,
21,
22]. However, convection or mechanical disturbance in the melt has been difficult to avoid in most of the experiments performed so far.
Engulfment of insoluble particles by advancing solid–liquid interface is still not fully understood owing to the deficiencies both in modeling and in experiments described above. The purpose of this work was to examine the effect of solute on particle pushing/engulfment in an alloy under convection-less conditions. Simple analytical equations were derived considering only the first-order phenomena governing particle pushing so that the boundary conditions do not need to be as rigorous as that used in numerical modeling [
6,
7,
23,
24]. A unique experimental method, which minimized convection and mechanical disturbance, was designed to capture particles within dendrite arms. These simple equations can be used for estimating
VC in binary alloys without using interfacial energies. The method can be used for observing the capturing of particles by an advancing solid–liquid interface under convection-less conditions.
2. Analysis on the Pet Velocity
As shown in
Figure 1, a depression should be gradually formed on a growing dendrite arm as the arm approaches an insoluble foreign particle. The existence of the particle in the vicinity of the dendrite arm will certainly affect solute diffusion ahead of the advancing solid–liquid interface of the arm. In the gap between the particle and the solid–liquid interface, solute rejected by the growing dendrite arm is difficult to be transferred out of the gap. The enrichment of solute in the gap decreases the local melting temperature if the partition coefficient of the solute element,
k, is smaller than one. As a result, the solid–liquid interface under the gap will be concave, forming a depression on the dendrite arm to allow the particle to dwell. Such type of depression formation has been confirmed by Sen et al. [
25] in molten aluminum by means of X-ray imaging.
Consider the condition of quasi-steady particle pushing, i.e., a particle that dwells on an advancing solid–liquid interface advancing at a constant rate is pushed steadily by the solid–liquid interface over a certain distance [
1]. In order to push the particle under such conditions, fresh liquid has to constantly flow into the gap and freeze on the growing solid. At the same time, in an alloy, the solute must diffuse against this physical flow at a rate sufficient to prevent being built up in the gap.
Figure 2 depicts the temperature and solute fields in the vicinity of the solid–liquid interface. Following Pötschke and Rogge [
6], the balance of mass and solute under steady-state conditions in the gap region can be described. At any point in the gap shown in
Figure 2, the mass of liquid,
ML, flowing into the gap per unit of time is:
where
VL is the flow rate,
ρL is the density of the liquid, and
θ is the angle shown in
Figure 2.
The mass of liquid,
ML, which is transformed into solid growing at a velocity,
VS, per unit of time in the region defined by any
θ is:
where
ρS is the density of the solid. Mass balance defined by equating Equations (4) and (5) gives:
Ignoring the solute entering the growing solid, one can write the solute balance equation at any location in the gap as [
5]:
Under directional solidification conditions, the temperature field is fixed so the temperature distribution can be estimated using:
where
G is the temperature gradient in the solid,
TI is the temperature at the planar solid–liquid interface far away from the particle, and
Z is the vertical distance from any point in the solid to the planar solid–liquid interface.
The relationship between solute and temperature on a phase diagram is described by the phase diagram of the alloy system. For an alloy with a diluted solute content, the liquidus slope,
m, on the phase diagram is usually a constant. The change in temperature is related to the change in composition by:
Under steady-state growth conditions and ignoring the kinetic growth undercooling, the temperature,
TI, at the planar solid–liquid interface far away from the particle is:
where
C0 is the bulk solute content,
T0 is the liquidus temperature of the alloy, and
k is the partition coefficient. The composition at the liquid side of the planar solid–liquid interface is:
Assuming that the depression shown in
Figure 2 has a constant radius,
a, the temperature at any given
θ is:
where ∆
Ta is curvature undercooling and ∆
TK is kinetic undercooling. Under the assumption of a constant gap radius, ∆
Ta and ∆
TK are constants in the gap region. The temperature difference between
θ0 and
θ in the gap region can be calculated in the following equation by either using Equation (9) or Equation (12).
Combining Equations (8) and (12), one can write the vertical distance, ∆
Z, between
θ0 and
θ in the gap as:
i.e.,
Let the depth of the depression be
h0, the vertical distance from the planar solid–liquid interface to the center of the particle is
R-
h0, as shown in
Figure 2, and the edge of the planar interface neighboring the gap is the location corresponding to
θ0. At any given
θ, the vertical distance from the point corresponding to
θ in the gap to the planar interface is given by:
Combining Equations (15) and (16) gives:
where
Differentiating Equation (17) gives:
Substituting Equations (17) and (19) into Equation (7) yields:
Combining Equations (6), (11) and (20) gives:
Considering mass balance at the entrance of the gap where:
and ignoring
h in the
R +
h and
R +
h/2 terms since
h is only a few atomic layers thick [
1,
2,
3,
4,
5,
6] and is much smaller than
R, one can rewrite Equation (21) as:
Rearranging Equation (23) gives:
In deriving Equation (6), solute entering the growing solid is ignored. To account for this amount of solute, the PET velocity,
VC, of the solid has to be smaller than
VS described in Equation (24) in order to maintain the particle being pushed under steady-state, i.e.,
Similar to Equation (2), the PET velocity in Equation (25) is proportional to h and the reciprocal of R. The uniqueness of Equation (25) is that the PET velocity is a linear function of composition, diffusion coefficient, and temperature gradient as well. Compared to Equation (3), the values of α, β, and ε in Equation (25) are different, and the term of interfacial energies is not included because of the assumption that the radius of the gap is a constant. Still, the minimum separation between the particle and solid–liquid interface has to be determined by the forces acting on the particle. Furthermore, m and k in Equation (25) should not be zero because of the division operations performed in obtaining these equations. When m = 0, no depression will form on the growing solid behind the particle because TI = T0 and CL = C0. Thus, the solute should have no effect on particle pushing. This is also held true for C0 = 0 and k = 1. Under such conditions, solute diffusion is not a dominant factor during the engulfment of a particle, so particle pushing is governed by surface energy interactions when this is no convection in the liquid.
4. Discussion
The analytical models of this work are derived under simplified boundary conditions in order to obtain analytical solutions. The macroscopic morphology of the solid–liquid interface is not solved because it cannot represent the morphologies of the solid–liquid interface and the resultant separation,
h, at the atomic scale. As a result, the curvature undercooling of the interface has to be estimated, assuming a constant radius for the solid–liquid interface in the gap region. Another simplification in obtaining the analytical models is that the solute composition at the edges of the gap is assumed to be that at the planar solid–liquid interface, i.e., Equation (18). In spite of these oversimplified assumptions and less rigorous boundary conditions, results shown in
Figure 3,
Figure 4,
Figure 5 and
Figure 6 indicate that the predictions made by Equation (25) are generally in agreement with experimental observations in distilled water during directional solidification and in an aluminum alloy during isothermal coarsening.
Data shown in
Figure 6 include only those obtained in distilled water. Experimental data in other particle/matrix systems [
1,
3,
28] are not plotted because each system would require its unique
h and
mC0/
k data in order to use Equation (25). Still, for a particle of 1 µm in radius, the PET velocities are in the range of between a few micrometers per second to a few tens of micrometers per second. Such growth rates are comparable to our model predictions since the viscosity and the interfacial energies are quite different from that of the particle/distilled water system. Experimental data in systems containing sub-millimeter-sized particles [
29,
30] are not considered in this study because the thermal conductivities of the materials in the systems are not included in our study. These physical properties have less effect on particle pushing for small particles [
1,
2,
5] but could have a significant effect on large particles [
6,
25,
29,
30,
31].
In comparison with experimental data in particle/distilled water systems, the
h value used in Equation (21) is in the range of 0.005 to 0.02 µm, which is one to two orders of magnitude greater than an interatomic spacing [
32]. Such a minimum separation between the particle and the solid–liquid interface is in agreement with what is reported in the literature [
1,
2,
3]. PET velocity increases with increasing
h because it would be less difficult for solute elements or impurities to diffuse out of a thicker gap than a thinner one. Solute or impurity elements have to diffuse out of the gap, against the feeding current, to maintain steady-state particle pushing. Otherwise, the particle should be engulfed by the growing solid.
The dependence of the PET velocity on particle size is also related to the diffusion of solutes/impurities out of the gap between the particle and the growing solid crystal. The length of the gap and the resultant diffusion length are proportional to the particle size. Thus, it would be more difficult for the solutes/impurities to diffuse out of a longer gap than a shorter one, given a minimum separation,
h. Our diffusion model, i.e., Equation (25), indicates that
VC is proportional to 1/
R. This is quite different from the models based on interface interactions where
VC is proportional to 1/
Rα and
α > 1 [
6,
7]. The dependence of
VC to 1/
R seems to fit with experimental data better than
VC to 1/
Rα where
α > 1.
The solute content has a major effect on
VC. Unfortunately, experiments carried out in alloys on particle pushing under diffusion-only conditions, i.e., no convection, are scarce. The purity of matrix material used is usually of high purity (about 99.999 wt. %). One experiment by Körber et al. [
28] indicates that the addition of 0.56% NaMnO
4 in water does not alter the PET velocity of water containing latex particles. However, no phase diagram of the water/NaMnO
4 system is available, so it would be impossible to determine the values of
m and
k using Equation (25). A solute element should have no effect on
VC if
m = 0 or
k = 1.
The PET velocity for a 2-µm particle is around a few micron meters per second for a particle in the pure matrix, usually of 99.999 wt.% purity. When the impurity or solute content is increased by a few orders of magnitude, the PET velocity should decrease by a few orders of magnitude according to Equation (25). For an alloy containing 1 wt. % of solute, the PET velocity for pushing a 2-µm particle could be in the nanometer per second range, which is a few orders of magnitude smaller than the average growth velocity of solids during normal cooling conditions for normal gravity casting. This would mean that particles should be engulfed during dendritic solidification in a casting. Experimental results in alloys suggest otherwise. Particles are pushed by the growing dendrites under casting conditions so far tested [
14,
15,
18]. Clearly, the fluid flow has a major effect on particle pushing during solidification. A particle traveling in the melt can bounce off the solid–liquid interface and thus is prevented from being engulfed by the growing solid [
18,
22]. A particle resting on a depression can be dislodged from the depression on the growing solid crystal by a rolling/sliding mechanism [
18,
20,
21]. Vibration or other mechanical disturbance is also capable of dislodging a particle from a depression [
18,
20]. As long as the particle is dislodged from a depression and starts motion, it should not be engulfed until it settles on a depression again.
Without convection and other mechanical disturbance in the liquid, a small particle (large enough than what is subjected to Brownian motion) should be engulfed by the growing solid in an alloy if it rests on a depression on a growing crystal under normal casting conditions. Convection can be suppressed in an alloy during isothermal coarsening or during the late stage of solidification when the growth rates of the solid are slow, and the fraction solid is relatively large. Indeed, we have found that particles are engulfed during isothermal coarsening of Al-4.5 wt. %Cu alloy in this study.