Fundamental studies of ultrasonic melt processing

Ultrasonic (cavitation) melt processing attracts considerable interest from both academic and industrial com- munities as a promising route to provide clean, environment friendly and energy efficient solutions for some of the core issues of the metal casting industry, such as improving melt quality and providing structure refinement. In the last 5years, the authors undertook an extensive research programme into fundamental mechanisms of cavitation melt processing using state-of-the-art and unique facilities and methodologies. This overview sum- marises the recent results on the evaluation of acoustic pressure and melt flows in the treated melt, direct observations and quantitative analysis of cavitation in liquid aluminium alloys, in-situ and ex-situ studies of the nucleation, growth and fragmentation of intermetallics, and de-agglomeration of particles. These results provide valuable new insights and knowledge that are essential for upscaling ultrasonic melt processing to industrial level.


Introduction
Ultrasonic melt processing (USP) has been proven on the industrial scale to be an effective method for degassing, filtration, and grain refinement of aluminium and magnesium alloys [1][2][3][4]. In the past, the challenge of treating large melt volumes was met by using multiple ultrasonic sources [1,2]. This complex and energy-demanding approach works well in batch degassing operations in foundries or for grain refining or degassing a single cast billet/ingot in direct-chill casting. However, when it comes to casting multiple strands of extrusion billets, it becomes too technologically complicated and expensive. In addition, the introduction of other competitive techniques in industry in the 1980s (e.g. Ar-rotary degassing and Al5Ti1B grain refiner), has stalled the uptake of USP. Interest in USP was recently revived due to its attractive advantages such as environmental friendliness and energy efficiency. Current research concentrates on achieving high efficiency in processing large melt volumes by using a single (or limited number of) ultrasonic radiator with a well-designed flow arrangement. The success of such operation requires in-depth knowledge of the underlying physical mechanisms of USP, supported by adequate and validated numerical modelling.
In the past the USP was studied either by in-situ observations in transparent liquids or by correlation of ultrasonic parameters to post-mortem observations of as-cast alloy structure. The main phenomena and mechanisms established can be briefly described as follows.
High-frequency oscillations of an ultrasonic source (sonotrode) (typically in the range above 16-17 kHz) result in the formation of cavitation bubbles (typically on "cavitation nuclei" represented either by gas bubbles or inclusions) and in acoustic streaming in the processed liquids [3,5]. Reports on direct observations of cavitation and acoustic streaming are limited and most of those are done in transparent low temperature liquids. Cavitation bubbles nucleate in a region below the sonotrode that is called the "cavitation zone". To initiate cavitation, a certain pressure in the liquid phase, termed "cavitation threshold", needs to be overcome. The required pressure can be calculated theoretically or determined experimentally using hydrophones or cavitometers [5]. In real liquids, including metallic melts, the actual cavitation threshold is lower than the theoretical threshold due to the presence of the cavitation nuclei. With increasing ultrasonic frequency, liquid viscosity, and surface tension, the cavitation threshold increases [6]. The acoustic pressure decreases exponentially with distance from the source, an effect that is called "attenuation"; this effect has been predicted by modelling. As the cavitation develops, a significant portion of acoustic energy is consumed in the cavitation zone. Due to the very large number of cavitation bubbles, the cavitation zone becomes a lowdensity region with high volume occupied by gas/vapour. As a result, the cavitation zone "shields" the bulk of the volume from the ultrasonic waves and the attenuation becomes even more pronounced [5]. Acoustic pressure and flow measurements in liquid metals have been rather rudimentary and qualitative so far, with cavitation intensity (level of white noise from imploding cavitation bubbles) being the most widely used parameter for cavitation activity.
The theory of ultrasonic degassing has been developed in the 1970s for water with some measurements of oxygen concentration [7] and then applied to aluminium melts based on hydrogen measurements in as-cast alloys [1,2]. Analytical modelling established the main principles of rectified diffusion of dissolved gas into an oscillating bubble, its growth, and flotation.
One of the main applications of USP is related to microstructure refinement. There are a number of suggested mechanisms for this effect, most of which are based on theoretical estimates with indirect experimental evidence. A cavitation bubble can facilitate nucleation of a solid phase through changing the local equilibrium conditions upon bubble implosion. The extremely high pressure generated upon bubble collapse (in the range of GPa) increases the solidification temperature so that the liquid phase becomes greatly undercooled and may form a stable nucleus [8]. The survival of the nucleus is subject to the melt temperature and flow. This mechanism, though theoretically possible for homogeneous nucleation, may be more applicable to the conditions of heterogeneous nucleation when occurring close to the liquidus. In particular, it may facilitate heterogeneous nucleation during solidification [2,4]. In addition to the changed local conditions, heterogeneous nucleation during solidification is assumed to be related to the so-called "activation of inclusions" [2,4]. The activation can be a result of improved wetting of inclusions (e.g. oxides in liquid aluminium), which is also related to a sonocapillary effect [2,9,10]. Although the sonocapillary effect has been observed experimentally on the macroscopic scale, it has never been confirmed for the case of microscopic inclusions in liquid metals. There is also very limited and indirect evidence of the nucleation of intermetallics and matrix crystals on oxides, including the facilitation of this process by ultrasonic cavitation [2].
Another mechanism of structure refinement is fragmentation of growing crystals by ultrasonic cavitation. There are some direct observations of crystal fragmentation by cavitation bubbles in transparent analogues and with limited resolution [1,11]. No thorough analysis of the fragmentation mechanisms has been performed yet. The role of acoustic streaming in this process is reserved only for the distribution of fragments in the bulk liquid.
The modelling description of USP is very limited due to the numerical difficulties of addressing multi-scale and multi-physics phenomena, although the physics of single bubble oscillation and implosion are rather well known and numerically/analytically described. Realistic simulations of cavitation phenomena are also difficult due to the nonlinearity of the problem [12]. Recent attempts of cavitation modelling in water include models that couple the evolution of cavitation and acoustic fields by solving a nonlinear Helmholtz equation that accounts for bubble dynamics [13][14][15]. This nonlinear model can be used to predict acoustic streaming in the presence of cavitation [16].
In the last decade, a wealth of new knowledge on USP has been generated, to a great extent, due to the availability of new analytical and characterisation techniques (in situ synchrotron imaging and highspeed filming, acoustic measurements, particle-image velocimetry) and modelling approaches.
In this paper, we give an overview of our research in the last 5 years. The paper highlights the important phenomena in USP and illustrates those with our own results. For more detailed information on relevant experimental and modelling procedures and results, the reader is referred to the original papers.

Acoustic measurements
USP is inherently related to the propagation of acoustic waves in the liquid, succession of negative and positive acoustic pressure cycles, and acoustic cavitation. The measurement of acoustic spectra and pressure in a liquid is often made by using hydrophones in low-temperature liquids. Direct measurements in liquid metals (including aluminium) were only recently made possible due to the development and calibration of a high-temperature cavitometer [17]. Originally designed for qualitative measurements of acoustic noise and cavitation intensity, this cavitometer was calibrated in the National Physical Laboratory (NPL, UK) and applied to acquire accurate acoustic spectra and measure acoustic pressure in a wide range of frequencies (from 15 kHz up to 10 MHz). It became, therefore, possible to quantify for the first time the distribution of acoustic pressure in the treated melt volume, and better understand the attenuation of acoustic waves in the melt. In the series of experiments presented here, we used a 1-kW piezoelectric transducer coupled with a ceramic sonotrode 48 mm in diameter at a driving frequency of 20 kHz. A charge of 2 L (5.2 kg of commercially pure liquid Al or 2 kg of water) was held inside a clay-graphite glazed crucible 150 mm in diameter and 180 mm deep. The liquid height was 120 mm. The sonotrode was submerged 10 mm below the liquid surface in the middle of the crucible. The measurements were made with a cavitometer placed 70 mm below the liquid surface in three positions: under the sonotrode, at half-radius of the crucible, and close to the crucible wall. The temperature changed from 26°C to 49°C in water (due to the introduction of acoustic energy; however, repeated checks showed no effect of this change on the measured results in these series of experiments) and maintained at 703 ± 3°C in aluminium. The power of the transducer was varied; and the transmitted acoustic power and the corresponding maximum acoustic pressures under the sonotrode are listed in Table 1. Examples of measured acoustic emissions and pressure in water and liquid aluminium are shown in Fig. 1. These studies led to the following new findings: (a) the highest ultrasonic power is not necessarily the best one for treating larger volumes (Fig. 1c); (b) sound waves attenuate much stronger with distance and more uniformly with regard to the acoustic power in water (80% attenuation, Fig. 1b) than in liquid aluminium (30-50%, Fig. 1c), which may reflect more developed cavitation in water due to abundance of vapour bubbles and, hence, more substantial shielding of acoustic waves by the cavitation zone; and (c) acoustic pressures are significantly higher in liquid Al than in water [18,19].
The acoustic pressure measured in most liquids at the driving frequency under cavitation conditions is several times lower than theoretical values in the absence cavitation. This is a direct result of the shielding effect. In water, root mean square (RMS) acoustic pressures at 1 MHz in the acquired acoustic spectrum (associated with acoustic emissions from the collapse of cavitating bubbles) were found to be 2-3 times lower than in liquid aluminium [18], indicating that the potential energy is stored in relatively long living Al cavitation bubbles as shown in [20] and then released back into the bulk in a form of collapsing events. The results shown in Fig. 1c confirm that cavitation treatment in liquid aluminium can be performed at some distance/volume around the cavitation zone. A lower acoustic power (at least in liquid aluminium) maybe beneficial due to a lesser shielding effect and a wider propagation of powerful acoustic waves outside the cavitation zone. Acoustic pressures in liquid aluminium are certainly high enough to break clusters of particles and facilitate the de-agglomeration process and wetting through the sono-capillary effect [4,21].
Systematic measurements of cavitation spectra and pressure in bulk volume of liquid aluminium [22] demonstrated that, the most important USP parameter that determines the intensity of acoustic cavitation is the distance from the sonotrode. Melt temperature also affects cavitation albeit to a lower extent: the lower the temperature, the stronger the cavitation. The source acoustic power has an ambivalent influence: more cavitation is expected with higher powers, but the shielding effect of the developed cavitation zone may increase attenuation, thereby limiting the affected liquid volume. It was also experimentally shown that physical properties of the treated liquid may significantly change the appearance, dynamics and extent of cavitation [19]. For example, the cavitation region in high-viscosity glycerine is very well defined and confined in volume with very strong (though slow) acoustic streaming and recirculation. Cavitation bubbles are formed only close to the sonotrode surface. In a low-density and lowviscosity ethanol, the cavitation development occurs in a larger volume without any stable or well defined cavitation zone and acoustic streaming pattern. The cavitation bubbles are, however, rather stable over the considerable liquid volume, slowly expanding before floating to the liquid surface. The experiments also confirmed that water qualitatively resembled liquid aluminium in cavitation development, exhibiting very similar cavitation behaviour as compared to other tested liquids: hence, water can be used as a transparent analogue for aluminium (except for the attenuation phenomenon).
Acoustic measurements were also applied in testing a new scheme of ultrasonic degassing with a plate sonotrode. It was confirmed that this novel design of a sonotrode indeed generated acoustic cavitation with the formation of large stable bubbles suitable for degassing [23]. The new sonotrode design allowed for 25-30% increase in degassing efficiency as compared with a traditional cylindrical sonotrode due to the multiple locations of cavitation excitation along the length of a plate vibrating in a flexural mode [23].
In an attempt to understand the fundamentals of USP, an acoustic cavitation model was developed to predict the evolution of the bubble cloud structure and acoustic pressure inside a sonicated liquid. Based on the nonlinear bubbly equations proposed by van Wijngaarden [24] Table 1 for the acoustic power in W corresponding to the transducer power in %.

D.G. Eskin et al.
Ultrasonics -Sonochemistry xxx (xxxx) xxx-xxx and mathematically derived by Caflisch et al. [25] using Foldy's method [26], the model couples the wave equation (governing sound propagation in a moving fluid) and the Rayleigh-Plesset equation (for bubble dynamics). To minimise the effects of numerical dispersion, the wave equation is discretized in space using a 6-point stencil and in time using a 4-point stencil. For stability, the bubble dynamics equation is solved using a multi-stage explicit solver. Fig. 2 demonstrates the modelling results for USP of aluminium at 17.7 kHz in a crucible, assuming a bubble density of 1 × 10 11 m −3 and initial radii of 1 µm. The pressuredistance relationship obeys a power law, and the attenuation of pressure with distance agrees well with experimental results [22]. Such a macroscopic model is essential for the optimization of USP in scaled-up application, e.g. ultrasonic degassing in a ladle/launder or grain refinement upon direct-chill casting.
In parallel, a theoretical study of the interaction of hydrogen bubbles in liquid aluminium was undertaken to understand the interplay between bubble resonance and degassing [27]. By adapting the equations derived by Doinikov [28] to aluminium, it was found that large forcing signals in the order of MPa were required for fast degassing, especially when the frequency of the forcing signal is less than 22 kHz (favourable frequencies for degassing aluminium melt).

Synchrotron observations of cavitation
The development of state-of-the-art synchrotron X-ray facilities, as well as image acquisition capability and processing software, made it possible to study the dynamics of ultrasonic processing of liquid and solidifying metals [30,31], e.g. observing in-situ the implosion and oscillation of ultrasonic bubbles [32], as well as size distribution of cavitation bubbles in relation to the acoustic power and frequency; their interaction with intermetallics, solid particles and solidification front; acoustic flows; and sono-capillary effect. Fig. 3a shows bubble oscillation and implosion in a liquid Bi-8% Zn alloy (427°C, an ultrasound power of 20 W) taken using the ultrafast (up to 271,554 fps) synchrotron X-ray imaging at 32-ID-B of Advanced Photon Source. Bubble implosion occurred immediately below the sonotrode under an ultrasonic intensity of 276 W/mm 2 . Fig. 3b shows the bubble radius measured from the images, and their correlation to the acoustic pressure applied into the liquid metal. The dashed lines indicate the sudden drop of bubble radius at the instance of bubble implosion [32]. Although in this experiment due to the limited resolution of 1-2 µm it was not possible to observe the immediate aftermath of bubble implosion, earlier experiments on water and silicon oil demonstrated that the initial bubble collapsed into many tiny bubbles, and then those tiny bubbles acted as the bubble nuclei to grow again in the next cycles of acoustic wave [33]. Fig. 4 shows selected images of a transient bubble (framed) in a liquid Al-10% Cu alloy taken in by synchrotron X-rays. The size of this bubble is in good agreement with the Minnaert equation [34] that gives 70 µm for the bubble resonance radius at the 30 kHz driving frequency used.
A statistical size distribution of stable cavitation bubbles observed in similar experiments and their growth kinetics are illustrated in Fig. 5.
Macroscopic acoustic streaming originated from the cavitation zone can be also captured in liquid metals using synchrotron X-ray imaging. An example shown in Fig. 6 gives a typical flow pattern with a downstream flow and two recirculation vortices on either side. There is a clear similarity with the acoustic streaming observed in water and glycerine (see Fig. 8 below). As this image was taken during solidification of a Sn-Cu alloy, the fragmented dendrites drawn into the melt bulk by the acoustic flow are also visible as (shown by arrows). This observation clearly confirms one of the grain refinement mechanisms of USP, i.e. fragmentation and dispersion of solid phase by cavitation and acoustic streaming.
One of the current applications of USP is the introduction of nanoparticles in liquid aluminium or magnesium, with the aim of manufacturing nanocomposite materials [35,36]. USP is used to improve wetting of nanoparticles, de-agglomerate their conglomerates, and improve the dispersion of nanoparticles. Synchrotron X-ray studies allowed, for the first time, the observation and quantification of the sonocapillary effect that is responsible for de-agglomeration, improved filtration and wetting (and possibly for the activation of inclusions) [21,37]. These studies also demonstrated that the presence of nanoparticles in the melt significantly intensified acoustic flows and cavitation as demonstrated in Fig. 7 [38].

Study of cavitation in transparent liquids
Despite recent advances in synchrotron studies of liquid metals, transparent media are still widely used for in situ observations, allowing for more detail to be captured. Water is proven to be similar to liquid aluminium in ultrasonic cavitation development and is frequently used as a transparent alternative in USP studies. The study of cavitation processing also goes beyond liquid metal applications and is relevant to other USP areas such as food, chemical, drug, and fuel manufacturing where transparent liquids of different properties can be used.
We have studied the cavitation development in three liquids with significant differences in viscosity, vapour pressure, surface tension, and density. These parameters are key factors that determine cavitation behaviour [4]. This research demonstrated vast differences in cavitation kinetics, dynamics, macroscopic patterns, as well as in associated acoustic spectra. Fig. 8 illustrates the geometry and kinetics of development for cavitation zones and acoustic streaming patterns in de-ionised water, ethanol and glycerine.
The active cavitation zone outline is clearly observed in water (conical bubble structure) and in glycerine (thick round layer), while in ethanol the cavitation bubbles are dispersed and vigorously oscillate towards the free surface (similar to liquid degassing) rather than creating complicated bubbly structures and multiple collapsing events. The observations demonstrated that the bubble oscillation and implosion may happen outside the cavitation zone (where the bubbles are generated) being powered by the acoustic pressure wave.
Acoustic measurements [19] reveal that the cavitation intensity (measured in dBu) are similar in water and glycerine but considerably lower in ethanol. On the other hand, bubbles in ethanol survive longer, thus giving rise to distinctive pressure peaks at MHz frequencies. It is Fig. 2. Variation of maximum pressure with distance from sonotrode [29] and comparison with experimental cavitation intensity [22].

D.G. Eskin et al.
Ultrasonics -Sonochemistry xxx (xxxx) xxx-xxx also demonstrated that water shows the closest cavitation behaviour to liquid aluminium, as indicated by the spectra analysis and selected dimensionless quantities; and can therefore be used as a transparent analogue of liquid Al. There is, however, a potential to design transparent liquids closer matching properties to liquid metals using cavitometer measurements. Acoustic streaming in transparent liquids can be successfully studied using particle-image velocimetry (PIV) [39]. Recently, we have established an interesting acoustic flow pattern in water at lower acoustic powers using a 20 kHz transducer coupled with a titanium sonotrode of 20 mm in diameter submerged 2 cm below liquid surface. The dominant axial flow direction is upwards towards the centre of the sonotrode tip (24 μm p-p amplitude; 50% power; 20 kHz) with the average flow velocity 0.03-0.1 m/s (Fig. 9a). At a higher power (51 μm p-p amplitude; 100% power, 20 kHz), a downward jet with an average velocity of 0.1-0.5 m/s (Fig. 9b) is observed. Interestingly, in the case of the lower power, dominant vortices with velocities up to 0.09 m/s (comparable on the same scale to the centreline stream) are formed near the cavitation zone maximising the effect of recirculating and processing of floating phases (Fig. 9c); in contrast to the higher power, where vortices significantly weaker than the downstream centreline jet are formed further away from the tip of the sonotrode (Fig. 9d). Numerical analysis may shed light to this interesting mechanism and possibly reveal the nature of this inverse flow. Some preliminary computer simulations indicate that this phenomenon is related to the intensity of cavitation, i.e. the number density and size of cavitation bubbles in the cavitation zone.

Enhanced heterogeneous nucleation
Refinement of primary grains and intermetallics by USP is one of the major practical applications. The two main mechanisms are usually considered: cavitation-enhanced heterogeneous nucleation and fragmentation [4]. The role of insoluble inclusions in heterogeneous nucleation assisted by ultrasonic cavitation was considered already in the 1970s [40] and the mechanisms of inclusion activation were developed for aluminium alloys in the 1980s [2]. At this time, however, the hypotheses were based on indirect evidence and some theoretical considerations. Only recently the direct proof of heterogeneous nucleation of primary phases on oxide particles has been obtained [41].
We studied the formation of intermetallic particles Al 3 Zr and Al 3 Ti during USP above their liquidus and found out that the primary particles were refined [42,43], Fig. 10a and b. In almost all cases, a particle was found in the centre of a primary crystal as shown in Fig. 10c. Thorough studies using dissolution of the aluminium matrix, focused ion beam cutting of the crystals, electron diffraction and energy-dispersive X-ray spectroscopy with transmission electron microscopy unambiguously demonstrated that the nucleation of the intermetallics occurred on alumina inclusions with well-defined orientation relationships. The significantly increased amount of primary particles after USP testified for the improved nucleating ability of the indigenous alumina inclusions caused by ultrasonic cavitation.
These results are obtained by post-mortem analysis of the structure. It would be, however, very useful to observe the phenomenon of nucleation on oxide particles as it happens, i.e. in situ. This was achieved by observation of solidification of an Al-35% Cu alloy with the Al 2 Cu phase as a primary intermetallic [41]. In this experiment, the molten alloy has been treated with ultrasonic cavitation and broken pieces of a surface oxide film were entrained into the melt. After the end of USP, the melt was allowed to cool down and the primary Al 2 Cu particles started to nucleate and form on the fragments of alumina film (multiple particles as the film was polycrystalline) as shown in Fig. 11. This, to the best of our knowledge, is the first direct evidence of the long-debated issue of nucleation on oxides facilitated by ultrasonic processing. Practical implementation of this mechanism can be found in recent works on oxide-containing master alloys [44,45]. It was demonstrated that in-situ formed alumina or spinel particles have good potency for nucleating primary aluminium with a synergetic effect of USP that enhances the grain refining effect. Fig. 12 illustrates this for aluminaand spinel-containing master alloys added to aluminium alloys. The application of USP boosts the nucleating ability of oxides and results in additional grain refining. This is due to two effects: (i) improved wettability of oxide particles and (ii) dispersion of their agglomerates.

Fragmentation of primary crystals
As has been already mentioned, another powerful mechanism of structure refinement upon USP is fragmentation of the solid phase crystals. This mechanism works while the ultrasonic processing is applied in the solidification range of the phase of interest, i.e. below the liquidus. On many occasions, it was shown that fragmentation of primary dendrites of the matrix phase (e.g. Al or Mg), although quite impressive, is confined to a relatively small volume of the treated solidifying alloy, limited by the rapid attenuation of acoustic power and very restricted acoustic streaming in the presence of the solid phase. There are two practically important cases when this mechanism can work on a larger scale, i.e. direct-chill casting of light alloys [2] and arcremelting of a vibrating rod [3]. In these cases, the spatial position of the liquidus and solidus are well defined and can be kept constant with regard to the ultrasonic source. Even in these technologies, however, the scaling up of the process required the use of several or more powerful ultrasonic transducers. In other casting processes where the solidification front is moving, this mechanism is not applicable at all.
One of the ways out of this impasse would be treating the melt in a fluid state, i.e. in a melt flow, targeting the fragmentation of primary intermetallics that may act as nucleation substrates for the matrix phase. Ultrasonic processing of aluminium alloys containing Zr and Ti was shown to produce a remarkable grain refining effect when applied   6. Acoustic streaming and dendrite fragments (arrows) observed in a synchrotron using a Sn-30 wt% Cu alloy and 100 W acoustic power.

D.G. Eskin et al.
Ultrasonics -Sonochemistry xxx (xxxx) xxx-xxx above the liquidus of aluminium [2,4], and the explanation lies in the refinement of primary Al3Zr particles down to a few microns so that they become efficient nuclei for aluminium, Ti acting as a growth restriction element [46,47]. What remained unclear is how the fragmentation of a primary phase occurs under cavitation and streaming conditions. Only in-situ studies using synchrotron and high-speed imaging could help clarifying this. The first works in this direction showed that fracture of primary phases in metal solidification was mainly due to the shock wave (force) emitted at bubble implosion and the fatigue effect due to vigorously  oscillating ultrasonic bubbles [48,49]. We studied the fragmentation of intermetallics by cavitation bubbles in water using a high-speed camera [50]. The analysis of images revealed that the mechanism of cavitation-induced fragmentation is fatigue-like (and not instantaneous as it has been believed), with many cycles of oscillations required for an intermetallic particle to develop a crack and then break in a brittle fashion as shown in Fig. 13. The in-situ crack growth was analysed using the literature data on the fracture toughness of intermetallics and the pressure created by either an imploding or pulsating bubble [50]. It was estimated that for the observed  Ultrasonics -Sonochemistry xxx (xxxx) xxx-xxx critical crack size of 22 μm the stress required to fracture a primary Al 3 Ti crystal is between 215 and 322 MPa (at room temperature). Such a pressure can be produced upon the collapse and rebound of a cavitation bubble close to the solid interface in water [51]. A steadily pulsating bubble, however, will produce only about 10 MPa of pressure, which is not sufficient for the fracture of the intermetallic crystal (it would require a crack size in excess of a critical value of centimetres). For the matrix dendrites, however, in addition to the fragmentation of imploding bubbles, quasi-steady bubble oscillations may also result in the cyclic fatigue facilitated by acoustic flow (in the range of 0.5-0.6 m/s). The fatigue fracture of the Zn phase and the erosion of liquid-solid interface due to the oscillating bubbles observed in in-situ synchrotron experiments agreed very well with the theoretical calculations [32].
Acoustic flow plays an important role in fragmentation of elongated crystals and disruption of the solidification front (solid-liquid interface) as has been discovered by in-situ synchrotron observations of primary Al 2 Cu in an Al-35% Cu alloy. This primary phase forms thin and elongated dendrites that look like a sea weed on a cross section, Fig. 14a. Under ultrasonic irradiation the long crystals vibrate in tune with acoustic wave and sway under recirculating acoustic streaming, with eventually separating dendrite branches that are flown away with the flow, Fig. 14b [52]. Acoustic streaming also brings a hotter melt into the semi-solid region, facilitating remelting of dendrite branches and their thinning. In this case, the stress induced by the acoustic streaming (0.5 m/s) on a thin (22 μm diameter) crystal reaches 22-24 MPa, which exceeds the yield strength of the crystal (20 MPa) [52]. The detached branches are transported by the recirculating acoustic flow upward towards the cavitation zone where they experience further fragmentation by imploding bubbles, producing small dendrite fragments that act as nuclei for the intermetallic phase in the subsequent solidification, with eventual structure refinement, Fig. 14c.   Fig. 11. In-situ synchrotron observation of the nucleation of primary Al 2 Cu crystals (dark) on the fragments of alumina film (bright) in an Al-35 wt% Cu alloy. Temperature decreases from left to right as 550.7°C; 549.9°C and 549.1°C, respectively.

Fig. 12.
Effect of oxides and USP on the grain refining in an A357 (Al-Si-Mg) aluminium alloy: a, A357 cast with USP, no grain refiner; b, A357 cast with an addition of 1% alumina-containing master alloy; c, same as (b) but with USP; d, effects of spinel-containing master alloy addition and USP on the grain size in an A357 alloy. Fig. 13. Images of a primary Al 3 Ti crystal fragmented by a pulsating bubble (seen on the first frame). Arrows show the fatigues crack initiation and propagation. The field of view is 337.28 × 739.65 µm 2 and the images were recorded at 10 5 fps [50].

Ultrasonics -Sonochemistry xxx (xxxx) xxx-xxx
This mechanism is further illustrated in Fig. 15.

De-agglomeration
Ultrasonic cavitation and acoustic streaming play a vital role in deagglomerating and dispersing particles in liquids. With regard to liquid metal processing, these processes are important for improving the efficiency of grain refinement (de-agglomeration of TiB 2 agglomerates from commercial master alloys) [53] and manufacturing of metal-matrix composites (de-agglomeration and dispersion of reinforcing particles including nanoparticles) [36,54]. The mechanisms of de-agglomeration were thought to include the penetration of the liquid metal into the agglomerates (overcoming capillary pressure), followed by implosion of cavitation bubbles from inside the agglomerate, affectively dispersing the particles in the bulk. It is known, however, that dispersion of clusters, especially consisting of nanoparticles, takes a lot of time, e.g. 25-30 min for a half-kg of a Mg-based composite [55]. This is inconsistent with an explosive destruction of agglomerates from within (at least for agglomerates of a considerable size).
Recently a numerical model has been developed that describes the de-agglomeration in an ultrasonic field and demonstrates the role of particle size, acoustic pressure and liquid flow on the de-agglomeration kinetics [56]. This model does not, however, predict the processing time that is required for complete de-agglomeration.
Recent in-situ high-speed observations of cavitation-induced de-agglomeration showed that the process actually happens from the surface of agglomerates rather than from within. Fig. 16a and b demonstrate this phenomenon for MgO clusters in water. The process more resembles chipping off individual particles from the surface rather than rupture of the agglomerate from within its bulk. This observation clearly explains why the de-agglomeration takes long time, which should be the function of the size of the agglomerates, the size of the particles, and the pressure conditions.
It can be demonstrated [37,57] that the pressure difference Δp between the acoustic pressure from an imploding bubble close to the agglomerate and the capillary pressure inside the channel between particles is the controlling parameter for the de-agglomeration time. As these pressures are of the same order of magnitude (at least for micronsize channels), the excess pressure p ex is rather small (tens of Pa), and when p ex > Δp, the capillary will be filled with liquid metal after some time t wt . This time can be obtained by integrating Eq.  . 14b); (c) further fragmentation as the acoustic flow brings the fragments back into the cavitation zone; (d) after USP is stopped, the surviving small fragments grow as equiaxed dendrites (see Fig. 14c); (e) the equiaxed dendrites grow further leading to structure refinement.
where R cr is the capillary channel radius, η l is the viscosity of the melt, l cr is the capillary length filled with the melt, p is the acoustic pressure, D is the average particle diameter, S m is particles' specific surface, and ρ p is the density of a particle. Typical values for nanoparticles from tens of nanometers to several microns in size are listed in Table 2 [58]. It is interesting that the experimental data are counterintuitive and show that particles several microns in size may have larger specific area than nanoparticles. This is because the surface of the former is very uneven and contains a lot of folds, pores and crevices, while the surface of the latter is rather smooth. Fig. 16c shows the dependence of the time required for de-agglomeration on the l cr /R cr ratio for particles with the sizes from several microns to hundreds of microns. In this case the larger the particles, the lower their specific surface area and the shorter the time required for de-agglomeration. The agglomerates of small particles would require several minutes to de-agglomerate matching the experimental results (for alumina particles in aluminium: 3-5 min for 10-20 µm in diameter and 12-15 min for 0.1-10 μm in diameter [4]). In the case of nanoparticles, however, the suggested approach may not work as the capillary pressure (between the particles in a cluster) becomes so large that it outweighs the decrease in the specific surface area and may require either significantly higher pressures from cavitating bubbles (e.g. implosion pressure in liquid aluminium is larger than in water) or much longer times for de-agglomeration (which is experienced in practice).

Conclusion
Application of modern characterisation techniques allows researchers to get an insight into the mechanisms of ultrasonic melt processing, both from physical and mechanical points of view. These insights are invaluable for establishing reliable models of USP and upscaling USP to various industrial applications such as solidification and casting of metallic alloys, manufacturing composite materials, emulsification in a wide range of organic and inorganic systems, manufacturing of nanomaterials.
In the last 5 years we carried out extensive experimental and numerical studies, and this work established that: -Acoustic shielding plays a very important role in the propagation of sound waves and active cavitation throughout the volume of a treated liquid. -The acoustic pressure in the treated volume, outside the cavitation zone, can be higher at a lower driving power of the transducer, which is very important for practical applications. -Although water can be used as a suitable transparent analogue of liquid aluminium for qualitative studies of ultrasonic processing, the acoustic pressure, both in a driving-frequency and in the MHz range of acoustic spectrum, is much higher in liquid aluminium. -Acoustic streaming induced by a sonotrode can have different direction (towards, or away from its surface) depending on the cavitation intensity and the amount of bubbles in the cavitation zone.   : (a, b) in situ observation of de-agglomeration of MgO agglomerates by cavitation bubbles (shown by arrow) (a, stable bubbles before the onset of cavitation and b, imploding and rebounding bubbles producing de-agglomeration); (c) dependence of the penetration time of melt into the channels between the particles on the ratio of length to radius of the channel for various values of the differential pressure Δp (after [37]).

Table 2
Size, specific surface of powder particles (experimental, [58]) and the ratio of the length of the channels to the particle radius. phenomena supplemented by acoustic measurements give invaluable input to numerical modelling as well as provide validation of the models.