Professional Documents
Culture Documents
art ic l e i nf o a b s t r a c t
Available online 22 September 2014 The effects of a transverse flow on the dynamics of a directional solidification interface are studied
Keywords: experimentally in a thin sample. The set-up enables a non-intrusive visualization of the interface and an
A1. Directional solidification independent control of both the flow and the solidification conditions. The flow is forced in the sample
A1. Fluid flows from an external thermosiphon which provides an accurate steady velocity up to 1.2 mm/s. A
A1. Convection transparent melt of succinonitrile is used with a sample depth allowing the solidification of a single
A1. Wave layer of microstructures. Downstream inclinations of microstructures and downstream promotion of
A1. Thin sample dendritic sidebranching are observed. Surprisingly, large scale traveling waves involving a wavelength of
A1. Striation several cells or dendrites are evidenced on the interface in a large range of conditions. Two kinds of
waves are evidenced, one involving a slow velocity, a weak amplitude and a sinusoidal profile, the other
a large velocity, a large amplitude and a non-linear profile. Both result from the coupling between
solidification and flow and induce striations in the solid phase.
& 2014 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.jcrysgro.2014.09.026
0022-0248/& 2014 Elsevier B.V. All rights reserved.
38 T. Jiang et al. / Journal of Crystal Growth 417 (2015) 37–43
study, the evidence of flow-induced interfacial waves, and we To improve the microstructure homogeneity, care has been
document their main features. We finally discuss the relevance of taken to prepare the sample in a single crystal state. This has been
these results and conclude about this study. achieved by making a selected grain invade the whole sample by a
The experimental set-up is designed to achieve directional spatial control of fusion/solidification prior to the sample insertion
solidification in homogeneous and controlled conditions [8–10] in the experimental set-up. The chosen orientation was such that
together with a permanent visualization of the interface dynamics. the principal axes of the cubic crystal of succinonitrile were
To this aim, solidification proceeds in thin samples that are filled aligned on the thermal gradient direction, the mean interface
with a transparent material and whose depth corresponds to the direction and the cell depth. Control of the orientation was
typical thickness of a layer of microstructures. Samples are then achieved from the symmetry displayed by the sidebranches of
pushed at a controlled speed in between heaters and coolers that freely growing germs and of dendrites rapidly growing in a
set a uniform thermal gradient (Fig. 1a). Heaters and coolers are thermal gradient.
made of top and bottom metallic blocks that are electronically An interesting feature of thin samples is to inhibit natural
regulated at 100 1C and 10 1C respectively to an accuracy better convection owing to the large viscous dissipation imposed by their
than 0.1 1C . They yield a thermal gradient G of either 70 K cm 1 or small depth. This situation, which forbids the occurrence of
140 K cm 1 on the solidification interface, depending on the gap spurious uncontrolled flows, is thus especially suitable for achiev-
imposed between the blocks by suitable spacers. The pushing ing a control of flows. It however requires forcing flows either
stage is provided by a micro-stepper motor which drives a linear mechanically by moving solid parts or hydrodynamically by
ball-screw attached to a translating stage. Pushing velocities up to injecting flows. Here, we preferred using the latter mean owing
50 μm s 1 may be achieved with a relative accuracy better than to the difficulty in controlling mechanical parts in a thin sample. In
73% on long duration. addition, we looked for a permanent continuous injection in
Samples consist of two glass plates separated by spacers. They contrast with the discrete injections used in [6]. This was provided
are filled with the mixture to solidify, here a dilute alloy of by connecting the sample to an external thermosiphon that is
succinonitrile with acrylonitrile as solute. As the melting tempera- known to induce low amplitude controlled flows in similar
ture of succinonitrile is 58 1C when pure, the solidification inter- configurations [16]. It consists in two vertical branches that are
face sets around the middle of the gap between heaters and differentially heated so that the fluid is lighter in one vertical
coolers. This ensures that the sample translation negligibly modi- branch than in the other. These branches are then connected to
fies the thermal gradient on the interface [11]. Samples are wide each side of the sample by flexible tubes to yield a closed circuit.
(4.5 cm) and long (15 cm) enough to provide solidification They then provide a hydrostatic pressure difference between the
domains far from boundary disturbances. Their thickness was two sides of the sample that sets the fluid into motion in between.
chosen small enough to avoid the emergence of a second layer Within the sample, the flow is then parallel to the solidification
of microstructures and large enough to provide a 3D behavior of interface (Fig. 1b) and, following the small sample depth d, it
microstructures. This ensured that microstructures behave similar stands in a Stokes regime with a Reynolds number Re ¼ dU=ν of
to those of a multi-layer solidification, in particular regarding order 10 3 in our range, ν denoting the fluid kinematic viscosity
forms [12,13], undercooling [12,14], sidebranching [8] or instabil- and U the flow amplitude. It then displays a Poiseuille profile.
ities [15]. Thanks to the transparency of the filled sample, A simple modeling of the thermosiphon can be obtained from
visualization of the solidification interface is achieved from the the Navier–Stokes equation in the Stokes regime by considering its
slight optical aberrations undergone by a parallel light beam circulation along a streamline.
crossing the sample, which reveal the interface as a sharp contrast I I I
line. The corresponding images are recorded by a camera involving ∇p dl þ ρg dlþ μΔv dl ¼ 0 ð1Þ
1024 768 pixels. In the grooves, images sometimes show dark
where p denotes the fluid pressure, ρ the fluid volumic mass, μ the
channels which correspond to a gaseous phase that does not
fluid dynamic viscosity, g the acceleration of gravity and v the fluid
couple with the interface dynamics.
velocity. The circuit being horizontal, except in the thermosiphon,
the contribution of gravity reduces to the differential effects of
thermal dilation in the thermosiphon branches. Following the
Poiseuille flow profile, the contribution of dissipation is propor-
tional to the constant net mass flow D. The proportionality factor,
called the hydrodynamic resistance Rh, depends on the geometry
of the cross section. As the circulation of the pressure gradient
vanishes, one finally obtains :
ρ0 αΔTgH ¼ D∑Rh;i ð2Þ
i
where ρ0, α, H and ΔT are the fluid volumic mass at the freezing
temperature, its thermal dilation factor, the length of the thermo-
siphon branches and their temperature difference. The index “i”
labels the different parts of the closed loop and Rh;i stands for their
hydrodynamical resistance. In particular, for a cylindrical section of
length lc and diameter Dc, one gets Rh ¼ 128=πμlc Dc 4 and for the
rectangular liquid domain ahead of the interface of length L, width
3
l and depth d, Rh ¼ 12μd L 1 l. The net mass flow D and thus the
Fig. 1. (a) Sketch of the usual solidification set-up. A motor pushes a sample within flow amplitude U are then directly proportional to the tempera-
a thermal gradient induced by heaters and coolers. Observation is achieved by ture difference between the thermosiphon branches. This ensures
ombroscopy on a parallel light beam crossing the sample at the interface location. their accurate tune, independent of the growth conditions. With a
(b) Thermosiphon added to force a controlled flow U in the sample. This flow
sweeps the solidification interface and induces, above a growth velocity V, a
maximum temperature difference of ΔT ¼ 32 1C, one obtains this
coupled solidification-hydrodynamic instability that yields traveling waves and way a maximal flow amplitude ranging experimentally from
striations (Fig. 3). 300 μm s 1 at d ¼ 150 μm to 1200 μm s 1 at d ¼ 400 μm. We
T. Jiang et al. / Journal of Crystal Growth 417 (2015) 37–43 39
note that it keeps low enough to enable the fluid coming from the prevents to study the sole effect of flow advection on the planar
thermosiphon to equilibrate its temperature with the surrounding front instability [3]. We note, however, that this limitation is not
medium before entering the solidification domain. specific to our experimental configuration but is common to all
Direct measurements of the mixture properties provided a sol- those that involve a localized flow entrance at a definite solute
utal diffusivity D ¼ 13507 50 μm2 s 1 and a partition coefficient concentration.
K¼ 0.2970.05 [8]. The flow amplitude was determined from the Nevertheless, farther from onset, when microstructures have
advection of tiny particles observed in the liquid phase and was developed, their mean interface stands at higher temperature,
found to agree with the value estimated from the thermosiphon closer to the liquidus line. Following the heuristic view of Brody,
model. The typical ranges of the experiment were the following: Bower and Flemmings (BBF) [17], the mean solidification front
fixed thermal gradient G ¼ 70 K cm 1 or G ¼ 140 K cm 1 , micro- then rejects a diffusive flux similar to that of an effective planar
structure spacings Λ about 50–200 μm, pushing velocities V up to front located at the same place. It may thus be affected with an
50 μm s 1 and flow velocities U up to 1200 μm s 1 . effective partition coefficient close to unity. In this case, the
entrance length le ¼ K 1 DU=V 2 decreases first because of the raise
of K and also because of the raise of the growth velocity V. At
2. Onset of planar instability and homogeneity V ¼ 10 μm s 1 , it then falls to less than 3 mm and may thus be
neglected compared to the sample width, 15 times larger. Accord-
When increasing the growth velocity V at a fixed flow ampli- ingly, the system accommodates quickly the entrance condition in
tude U, we observed front destabilization into cells first near the the cellular and dendritic regimes so that their dynamics will
flow exit and then closer to the entrance. On a range of growth actually be relevant to those of an infinitely large solidification
velocities, the front was thus splitted into a destabilized part on interface, swept by a uniform flow.
the exit side and a still stable part on the entrance side. The
destabilized part invaded the stable part as V increases until the
whole front became unstable. At a given location, the growth 3. Forms close to onset
velocity at which the front destabilizes grew with the flow
amplitude U almost linearly. For any sweeping flow U, we shall see that a wavy dynamics of the
According to these observations, front destabilization was an solidification interface occurs above some growth velocity V(U).
inhomogeneous process, presumably because of the specific con- Beyond it, the forms and the sidebranching dynamics of microstruc-
ditions imposed at the flow entrance. To model this effect, we tures are then dominated by the wave modulations. On the opposite,
consider a fluid particle at the flow entrance and follow it during below it, the cells and dendrites that appear closer to onset involve a
its advection by the flow U towards the exit. Meanwhile, it is filled steady tip temperature, as without flow. The implication of a sweeping
with solute rejection by the interface and with solute diffusion flow on their forms may then be considered in comparison to those
from the surrounding medium. Assuming that the evolution of displayed without flow.
solute concentration on the x-direction parallel to the interface Two major changes in form are noticeable. First an inclination
takes place on a larger scale le than on the growth direction z, i.e. of microstructures towards the downstream direction with an
le b lD ¼ D=V, we may neglect diffusion on the x-direction com- angle α that increases with U and decreases with V. This is
pared to that on the z-direction. The advected fluid particle then displayed in Fig. 2 on both cells (a) and dendrites (b) for the same
experiences the same evolution of concentration than a steady flow amplitude but different growth velocities. Cells then display a
particle on a starting solidification front, the correspondence larger inclination (4.3 1) than dendrites (1.2 1). This angle of
between space and time being x¼ Ut. As the built-up of a diffusive inclination increases almost linearly with the ratio U/V as shown
layer takes place on a characteristic time-scale τ ¼ K 1 D=V 2 , the in Fig. 2c. The second change implied by the flow is an asymmetry
length scale le over which the fluid particle reaches an asymptotic of sidebranching, sidebranches on the downstream side of the
value of solute concentration reads le ¼ K 1 DU=V 2 . dendrite being promoted (Fig. 2c).
On the other hand, as the cell depth is, in our experimental
range, smaller or at least of the same order than the diffusion
length lD ¼ D=V, the concentration field is homogeneous in the
depth direction. This legitimates a depth-averaged analysis of the
solutal field dynamics according to which the depth-averaged
value U ¼ 2U=3 of the flow has to be used instead of its maximal
value U. In agreement with a slow spatial variation of concentra-
tion on the x-direction, we may now invoke the supercooling
criterion locally at each front location. As the concentration on the
planar front raises from c1 at the flow entrance to c1 =K far from it
when the front has relaxed to equilibrium, one then concludes
that the critical velocity then decreases from V c =K near the flow
entrance to Vc far from it, Vc denoting the usual critical velocity at
nominal concentration c1 . Although this could be easily antici-
pated, the important thing is that this relaxation of local critical
velocity occurs on the length scale le ¼ K 1 DU=V 2 which may thus
be taken as the entrance length of the system. For a typical mean
flow amplitude of 100 μm s 1 and a velocity of about 3:5 μm s 1
at the onset of planar velocity, le is about 7 cm, i.e. larger than the
sample width. Then, the whole front involves a convective–
diffusive layer which departs from the usual diffusive layer that
Fig. 2. Downstream inclination and sidebranch asymmetry. G ¼ 70 K cm 1 , U ¼ 140
would only be recovered at larger distance from the flow entrance. μm s 1 . (a) V ¼ 6 μm s 1 . Cells with an inclination α ¼ 4:3 1. (b) V ¼ 18 μm s 1 .
This spatial transient effect means that the system remains Dendrites with an inclination α¼ 1.2 1 and downstream sidebranching. (c) Increase of
inhomogeneous at all time in this velocity regime. This therefore the inclination angle α with U/V.
40 T. Jiang et al. / Journal of Crystal Growth 417 (2015) 37–43
4. Interfacial waves features distinguish several domains referring to small wave alone, rapid wave
alone, a coexistence of these waves or an absence of any wave. It then
Beyond some growth velocity V that depends on U, a spectacular appears that, in the scanned domain, rapid waves are the dominant
phenomenon with major implications on the interface form and on waves. They exist on a bounded range of growth velocity V that
segregation sets in. It consists in the occurrence of interfacial waves increases with the flow amplitude U.
propagating downstream and which make the microstructures under- Close to onset, there also exists slow waves on a much shorter
cooling periodically vary as waves pass through (Fig. 3). Microstruc- range of velocity V which shrinks when reducing the flow amplitude
tures tips thus move back and forth in the thermal gradient, together U. These waves, possibly important for the modeling following their
with noticeable changes of their form that imprint the solid phase. sinusoidal nature, therefore appear in practice somewhat as marginal
These waves involve a large wavelength compared to the in comparison to the rapid waves. Interestingly, they however
microstructure width. They thus correspond to a collective oscilla- involve a coexistence domain with the rapid wave as evidenced in
tion of many cells or dendrites. Surprisingly, two distinct types of Fig. 5. This double wavy solution provides a definite show-up of their
waves are in order, which may sometimes coexist despite their differences for the same growth parameters. In particular, Fig. 5
differences. One involves almost sinusoidal oscillations with small reveals two kinds of striations that differ by their inclinations. This
amplitudes (Figs. 3a and 4a) ; the other involves mostly large will be linked in Section 4.2.4 to the differences in wave velocities.
amplitudes and non-linear oscillations (Figs. 3b and 4b). However, The phase diagram of Fig. 6 reveals no upper bound in the
their most different feature is their velocities C which is small for scanned domain. This suggests that waves may exist for any flow
the former wave and large for the latter wave. Accordingly, we amplitudes U and an ever increasing growth velocity so that they
shall denote them slow wave and rapid wave respectively. might occur in a number of practical situations. On the other hand,
Figs. 3 and 4 show snapshots of these traveling waves on about wave occurrence involves a slight transient of a few diffusion
several wavelengths and a wavelength respectively. They highlight the times when putting on the pushing velocity V or the flow
difference of forms outlined above, the slow wave being nearly magnitude U. They thus correspond to a robust phenomenon that
sinusoidal (a) and the rapid wave non-linear with a marked asymmetry quickly sets in on a large range of growth conditions, as soon as a
(b). Interestingly, the periodic modulations brought about on micro- sweeping flow is in order.
structures leave permanent structures in the solid phase. They appear as
striations that may yield important implications on microsegregation.
We report below on the domain of existence of both kinds of 4.2. Wave features
waves and on their main features. We shall index slow wave by “s”
and rapid wave by “r”. 4.2.1. Period
Despite their differences, both waves surprisingly show the
same period TðV ; UÞ. In addition, this period evolves with the
4.1. Phase diagram growth velocity V so that the length δ ¼ VT is a constant,
independent of U, as displayed in Fig. 7. The common length δ
Fig. 6 shows the experimental scan of the parametric space (U,V) amounts to about 350 μm here. It stands geometrically as the
performed at a thermal gradient G ¼ 70 K cm 1 . It enables us to distance of successive iso-phase events along the heat flow
Fig. 3. Traveling waves on a solidification interface at flow velocity U ¼ 306 μm s 1 and sample depth e ¼ 150 μm. Flow goes from right to left. Iso-phase lines of waves
provide repetitive modulations in the solid phase. (a) Slow wave, G ¼ 70 K cm 1 , V ¼ 6 μm s 1 , (b) rapid wave, G ¼ 140 K cm 1 , V ¼ 16 μm s 1 .
Fig. 4. Wave profile and implications on the solid phase. Flow goes from right to left. G ¼ 70 K cm 1 , e ¼ 150 μm : (a) slow wave, U ¼ 306 μm s 1 , V ¼ 6 μm s 1 , (b) rapid
wave, U ¼ 229 μm s 1 , V ¼ 16 μm s 1 . Both waves yield striation in the solid phase due to the periodic modification of microstructures.
T. Jiang et al. / Journal of Crystal Growth 417 (2015) 37–43 41
Fig. 5. Coexistence of both kinds of waves, G ¼ 70 K cm 1 , V ¼ 10 μm:s 1 , U ¼ 306 μm s 1 . Flow goes from right to left. Striations with different inclinations are noticeable in
the solid phase. The largest inclinations are provided by the slow wave and the smallest inclinations by the rapid wave.
4.2.2. Velocity
As claimed above, the wave velocities largely differ either at the
same parameters or by their mean order of magnitude.
Those of short waves, Cs, keep below the low value 40 μm s 1 ,
whatever the pushing velocity and the flow magnitude. They seem
rather independent of the growth velocity V (Fig. 8a), but the
Fig. 6. Domain of existence of waves as a function of growth velocity V and flow relevant velocity range is nevertheless small to definitely conclude.
intensity U at G ¼ 70 K cm 1 . The planar instability, which is observed in the Although the range of variation of Cs is short, it seems to rise with
middle of the sample, shows a slight raise with U. U towards an asymptote in Fig. 8b.
In comparison, the velocity Cr of rapid waves is usually
significantly larger, ranging from 50 to 400 μm s 1 in our experi-
mental domain. Whereas Cr shows no noticeable variation with V
(Fig. 8a), it displays a linear variation with U as C r 0:4U (Fig. 8b).
Taking either the maximal flow velocity U or its depth averaged
U ¼ 2U=3, it thus remains that Cr is smaller than any of these flow
intensities. Accordingly, although rapid waves propagate down-
stream in the flow direction, they always travel at a significantly
smaller speed than either the flow or the mean flow.
4.2.3. Wavelength
Following the existence of the invariant δ, the wavelength λ of
waves writes λ ¼ CT ¼ ðC=VÞδ. As shown in Fig. 9, this simple
relationship well agrees with data for both waves on the whole
observation range and yields a refined value of δ : δ ¼ 366 μm. As
may be seen in this figure, the appropriate unit for λ, the
millimeter, stresses the large wavelength of waves compared to
microstructure width (about 50–200 μm). In addition, the wave-
lengths of rapid waves appear far larger than those of slow waves,
in agreement with Figs. 3, 4, and 5.
sidebranching, our results also differ from the literature since in thin
samples a promotion of sidebranching on the upstream side was
reported in either a SCN-acetone alloy [4], a cyclohexanol alloy [18]
or a Ga-In alloy [19]. However, we also stress that a sidebranch
promotion on the downstream side of dendrites was noticed in a
thin sample of a SCN-acetone alloy [6].
Accordingly, although upstream inclination and upstream pro-
motion of sidebranching is usually reported, the reverse tendency is
also found, as in the present experiment. Actually, both cases may
be achieved depending on whether the flow penetrates largely or
weakly in the microstructures grooves. In the former case, solute
iso-concentration lines are compressed on the upstream side of
microstructures yielding a promotion of growth and thus both an
inclination and an enhanced sidebranching. In the latter case, flow
only advects the poor-solute liquid surrounding microstructure tips
on the downstream direction which, after diffusion in the grooves,
induces inclination and sidebranch promotion on this downstream
side. More detailed analysis of each configuration is thus required to
converge towards a synthetic understanding of the effects of flows
on microstructure form and direction.
The novel phenomena revealed by our experiment are the
generation of interfacial traveling waves when a flow sweeps the
solidification interface in the cellular or dendritic regime. These
waves make the microstructure undercooling oscillate with notice-
able implications on their forms and on the microsegregation in the
solid phase. In particular, waves induce a striation of the solid phase
in the form of equally distant lines which correspond to iso-phase
lines of the waves. Their inclination depends on the ratio of the
growth velocity to the wave velocity. As this phenomenon is robust,
spectacular and dominant in our experimental range, one may
Fig. 8. Wave velocity C for both kinds of waves as a function of V at fixed U ¼
306 mm s 1 (a) and as a function of U at various V (b). Whereas C appears
wonder whether it could be at the origin of the striations some-
independent of V for both waves, it rises as 0:4 U on rapid waves. times observed in post-mortem analysis of cast materials.
On the fundamental ground, these waves, which travel down-
stream are analogous to the waves predicted on a planar front by
linear analysis [3]. However, they occur here on cellular or
dendritic interfaces and involve two kinds of waves with distinct
features, especially their velocity. They thus refer to a non-linear
phenomenon and display non-linear features, in particular an
asymmetric triangular profile for the rapid waves. Some progress
in the non-linear modeling of the solidification-flow interaction
will thus be required to understand their mechanism. Interest-
ingly, we note that waves involve a universal, constant length, the
distance between their striation lines in the thermal gradient
direction, which will be a landmark for the understanding of their
underlying mechanism.
6. Conclusion