Art de Lucrat2

in

. Zshim

xbrid

Man

es, K

Oxfo

9 Oct26 N

ys. Hing oed thther et thee liqudissoency

melt after grain rener addition to achieve the columnar-to-equiaxed transition. 2014 Published by Elsevier Ltd. on behalf of Acta Materialia Inc.

exact mechanism for grain renement involving theaddition of AlTiB-based grain reners [5].

gested that Al3Ti maybe the nucleant. However, duringthe grain renement of Al alloys using AlTiB masteralloys, the amount of free Ti added to the melt via the mas-ter alloy is usually less than 0.01%, which is much lowerthan the Ti level required for peritectic reaction (0.15%).Corresponding author; e-mail: [email protected]

Available online at www.sciencedirect.com

ScienceDirectActa Materialia 84 (2015) 292Keywords: Grain renement; Nucleation; Aluminum; TiB2; Interfacial layer

1. Introduction

Grain renement of Al alloys is desirable since it facili-tates the casting process, reduces cast defects and thereforeoers improved casting quality [1]. To achieve a grain-rened cast microstructure, chemical inoculation by theaddition of grain reners has become a common industrialpractice [2]. The most widely used grain rener for Al alloysis the Al5Ti1B (all compositions are in wt.% unlessotherwise specied) master alloy, which contains both theAl3Ti intermetallic compound and TiB2 particles in an alu-minum matrix. The Al5Ti1B master alloy contains 2.8%excess Ti over the stoichiometric ratio of TiB2 (i.e., 2.2:1 inwt.%). The Al5Ti1B master alloy is very eective for thegrain renement of most of Al alloys with a nucleationundercooling of usually less than 1 K [2]. Since theintroduction of AlTiB-based grain reners over 60 yearsago [3], tremendous eorts have been made to understandthe mechanism of grain renement, which has been thesubject of a number of review papers (e.g. [2,4,5]).However, until now, there has been no consensus on the

In the 1950s, the carbideboride particle theory was rstproposed by Cibula [3]. This theory proposed that insolubleTiB2 particles in the melt acted as the heterogeneous nucle-ation sites. However, Mohanty and Gruzleski [6] observedthat the borides are pushed to grain boundaries, and nograin renement was observed in the absence of Ti solute.This contradicted the carbideboride particle theory.Similarly, AlB2 particles (less stable than TiB2 particles inthe Al melt) do not act as heterogeneous nucleation sites[7].

Parallel to the carbideboride particle theory is theperitectic theory proposed by Crossley and Mondolfo [8]in the early 1950s. This theory conjectured that Al3Ti, notTiB2, nucleated a-Al via the peritectic reaction. There wereexperimental observations that appeared to support theperitectic theory. For example, Marcantonio and Mon-dolfo [9], and Maxwell and Hellawell [10] showed that thenucleation of a-Al required a smaller undercooling in thepresence of Al3Ti compared with TiB2, indicating thatAl3Ti is a more potent nucleant than TiB2. Davies et al.[11] observed Al3Ti at the centres of the grains and sug-Grain rening mechanism

Z. Fan,a,Y. Wang,a Y. Zhang,a T. Qin,a X.R

T. HaaBCAST, Brunel University, U

bSchool of Materials, University ofcSuperSTEM, STFC Daresbury Laboratori

dDepartment of Materials, University of

Received 5 September 2014; revised 1Available online

AbstractAlTiB is the most widely used grain rener for many Al allo60 years of intensive research. This work aims to further our understandExtensive high-resolution electron microscopy investigation has conrmmost likely to be a (112) Al3Ti two-dimensional compound (2DC). FurTiB2 particles and the stability of the Al3Ti 2DC. Our results showed thamonolayer of Al3Ti 2DC on their surface. The Al3Ti 2DC forms at thdilute AlTi solutions, although the kinetics of both the formation andby the Al5Ti1B grain rener is directly attributed to the enhanced pothttp://dx.doi.org/10.1016/j.actamat.2014.10.0551359-6462/ 2014 Published by Elsevier Ltd. on behalf of Acta Materialia Ithe Al/AlTiB system

hou,b G.E. Thompson,b T. Pennycookc,d andotob

ge, Middlesex UB8 3PH, UK

chester, Manchester M13 9PL, UK

eckwick Lane, Warrington WA4 4AD, UK

rd, Parks Road, Oxford OX1 3PH, UK

ober 2014; accepted 21 October 2014ovember 2014

owever, the precise mechanism of grain renement is still not clear aftern the grain rening mechanism involving AlTiB-based grain reners.e existence of a Ti-rich monolayer on the (0001) TiB2 surface, which isxperimental investigation was carried out to understand the potency ofpotency of TiB2 particles is signicantly increased by the formation of aidAl/TiB2 interface in concentrated AlTi solutions, but dissolves inlution of Al3Ti 2DC are relatively sluggish. Eective grain renementof TiB2 particles with the Al3Ti 2DC and sucient free Ti solute in the

304

www.elsevier.com/locate/actamatnc.

terialTherefore, Al3Ti is not thermodynamically stable for thehypoperitectic case. For example, Johnsson [12] indicatedthat the aluminides probably take less than 1 min, but def-initely less than 5 min to dissolve at a holding temperatureof 775 C. Jones and Pearson [7] suggested that all alumi-nides dissolved in molten Al in less than 30 s. Althoughthe dissolution time depends on both the holding tempera-ture and the size of the Al3Ti particles, the resulting consen-sus is that Al3Ti dissolves rapidly above the Al liquidus andhence the peritectic reaction is not thermodynamicallyfeasible.

For Al3Ti to be responsible for grain renement duringsolidication of hypoperitectic alloys, other factors have tobe operational. Vader et al. [13] and Backerud et al. [14]proposed the peritectic hulk theory in the early 1990s. Thistheory recognized that Al3Ti is a more potent nucleant thanthe TiB2 and attempted to explain how the borides increasethe stability of aluminides. It was suggested that the boridesform a shell around the aluminides, and slow down disso-lution of the aluminides. The aluminides eventually dissolveand leave a cell of liquid with approximately the peritecticcomposition. The peritectic reaction can then take place toform the a-Al. In addition, Marcantonio and Mondolfo[15] suggested that boron additions could change the phasediagram considerably, so that Al3Ti could be stable even atlower concentration. However, a number of researchers(e.g., Refs. [7,16]) investigated the Al-rich corner of theAlTiB phase diagram and showed that boron had virtu-ally no eect on the AlTi phase diagram. Therefore, thenucleation process cannot be explained purely based ontheories which attempt to modify the conditions for theperitectic reaction to occur.

Both TiB2 and Al3Ti have been observed within individ-ual Al grains [6]. It has been generally recognized that bothcan act as nucleation sites for Al [17], although Al3Ti ismore potent than TiB2. It is also known that incorporatinga small amount of excess Ti (less than 0.15%) in the meltcan cause a drastic reduction of grain size compared withthe grain rening using solely TiB2 [3]. Subsequently, thenecessity for both TiB2 and the small amount of excess Tihas led to other theoretical developments.

The hypernucleation theory was proposed by Jones [18],who suggested that Ti atoms in the melt segregated to themelt/TiB2 interface and subsequently promoted the forma-tion of a pseudo-crystal with a structure similar to that ofa-Al. This pseudo-crystal would be stable at temperaturesabove Al liquidus and could grow at temperatures belowthe liquidus, thus requiring little or no nucleation underco-oling. Since this theory was proposed, the stability of thepseudo-crystal above the Al liquidus has become the mainfocus of debate because it appears to violate the principlesof thermodynamics. However, the recent high-resolutiontransmission electron microscopy (HRTEM) study ofliquid Xe/a-Al [19] and liquid Al/a-Al2O3 [20] interfacesand molecular dynamics (MD) simulation [21] have shownthat there is considerable atomic ordering at the liquid/substrate interface, providing both experimental andtheoretical support for the existence of pseudo-crystals atthe liquid/substrate interface.

The duplex nucleation theory was proposed by Mohantyand Gruzleski [6], who suggested that Ti segregation to themelt/TiB2 interface can reach such a level that the forma-tion of Al Ti would be favored, implying that the Ti con-

Z. Fan et al. / Acta Ma3

tent in the melt close to the interface would exceed 0.15%to allow the peritectic reaction to take place. However, thisconcept was dismissed by Sigworth [22] based on thermo-dynamic argument. In order to verify the duplex nucleationtheory, Schumacher et al. [23,24] used a melt spinning tech-nique to produce amorphous Al85Y8Ni5Co2 alloy ribbonscontaining TiB2 particles. They observed that a thin crystal-line phase of 3 nm thickness with a lattice spacing close tothat of Al3Ti phase existed between TiB2 and the amor-phous Al. However, it was dicult to precisely identifythe Al3Ti phase. It is possible that the observed 3 nm crystallayer on the TiB2 surface is a-Al formed in the solid statesince the amorphous Al85Y8Ni5Co2 alloy is highly metasta-ble and crystallization may take place in the solid state.

The solute theory [10,25] suggested that both the nucle-ating particles and the solute elements in the alloy meltwere important factors aecting grain renement. The eectof solute elements on heterogeneous nucleation was takeninto account through their eect on growth restriction toallow more nucleating substrates to be active beforerecalescence [26]. Easton and StJohn [26,27] proposed asemi-empirical model based on the constitutional underco-oling theory. Their analysis showed that grain size can berelated closely to the growth restriction factor. In such anapproach, the grain renement due to the increased solutecontents was explained by the increased nucleation eventsthrough delayed recalescence [27,28]. Further developmentalong this direction has seen a more analytical approach tothe solute eect [29,30] and the postulation of the interde-pendence theory by StJohn et al. [31]. They proposed thatgrain formation was the result of repeated cycles of growthand nucleation events moving towards the thermal centreof the casting. However, it is likely that the solute theoryand its extension may have over-estimated the eect of sol-ute on grain renement. Recently, Du and Li [32] extendedthe KampmannWagner numerical model [33] for solidstate precipitation to predict grain size in solidied multi-component alloys. Their simulation results showed thatthe solute suppressed nucleation (SSN) eect has a negligi-ble inuence on the nucleation behavior and the nal grainsize during isothermal melt solidication.

One of the major challenges in nucleation research is theexperimental diculties in examining the nucleation processin situ since nucleation occurs in a non-opaque medium athigh temperature and in an extremely short time scale.Consequently, nucleation research is usually conducted bypostmortem examination of the solidied materials com-bined with some speculations. Alternatively, modeling tech-niques have been employed for investigating heterogeneousnucleation at the atomic level, such as the density functionaltheory (DFT) and the MD simulations. DFT calculationssuggested that an Al3Ti-like thin layer could be thermody-namically stable on the TiB2 surface prior to Al solidica-tion [34], but the ab initio MD simulations failed to showthe existence of such a layer of Al3Ti on the top of TiB2 sur-face [35], which may have been due to the time limitation ofab initio MD. However, the more recent MD simulation byQin and Fan [36] has conrmed that an atomic monolayerof (112) Al3Ti could be stable at the liquid Al/TiB2 interfaceat temperatures above the melting point of Al.

A particularly important aspect of grain renement isthe crystallographic matching across the solid/substrateinterface at the moment of heterogeneous nucleation.Relevant to this is epitaxial growth of a thin layer of onematerial on the surface of another material (the substrate)

ia 84 (2015) 292304 293[37]. The scientic basis for epitaxial growth of strainedlayers on a substrate originates from the theory of Frank

and Van der Merwe [38]. Interestingly, Frank and Van derMerwes theory was applied to study heterogeneous nucle-ation of a solid on a crystalline substrate during solidica-tion of a metallic melt by Turnbull and Vonnegut [39] asearly as 1952. They based their analysis on classical hetero-geneous nucleation theory [40], which we now know breaks

particles (>99% purity) used in this work had a particlesize range of 110 lm and an average particle size of3.1 lm. The detailed chemical compositions of thematerials used are listed in Table 1.

2.2. Intensive melt shearing

294 Z. Fan et al. / Acta Materialia 84 (2015) 2923042.1. Materials

Commercial-purity aluminum (hereafter referred to asCPAl) is commonly used as a standard material forassessing the eectiveness of grain reners [2,45]. In thiswork CPAl (>99.86% Al) was used as the base materialfor all experiments. Commercial Al5Ti1B (equivalentto Al2.8Ti3.2TiB2) grain rener, supplied by LSM(Rotherham, UK), was used as the reference grain renerfor assessing the eectiveness of other grain reners.According to Quested and Greer [46], commercialAl5Ti1B grain rener contains TiB2 particles with a sizerange of 0.56 lm, an average size of 0.72 lm and a par-ticle number density of 7.3 1012 m3 with 0.1% addi-tion. A stoichiometric grain rener, Al2.2Ti1B(equivalent to Al3.2 TiB2), supplied by LSM, was alsoused in this work, which provided the same TiB2 particlenumber density as the commercial Al5Ti1B grain renerbut with no excess Ti. In addition, an Al10Ti masteralloy, supplied by LSM, was used to vary the Ti concen-tration and to prepare new grain reners based on syn-thetic TiB2 particles (unlike TiB2 particles formed in situduring grain rener production). The synthetic TiB2

Table 1. Compositions (wt.%) of the materials used in this work.

Alloy Ti B Fe Si V

CPAl 0.006 0.08 0.03 Al-10Ti 9.90 0.10 0.02 down for systems with a small contact angle [41], such assystems where grain renement is relevant. Nevertheless,their analysis suggested that nucleation undercoolingincreased sharply with increasing lattice mist betweenthe solid and the substrate. Further realization of theimportance of lattice matching at the solid/substrate inter-face in more recent years has led to the development of theedge-to-edge matching approach for the identication ofnew substrates for grain renement [42,43] and the develop-ment of a more general epitaxial nucleation model [44].

In this paper we present our experimental evidence forthe existence of an atomic monolayer of (112) Al3Ti two-dimensional compound (2DC) on the (0001) TiB2 surfaceof commercial AlTiB-based grain reners. We will dem-onstrate experimentally that the Al3Ti 2DC forms at theliquid Al/TiB2 interface in concentrated AlTi solutions,but dissolves in dilute AlTi solutions, although the kineticsof both formation and dissolution of Al3Ti 2DC are rela-tively sluggish. The eective grain renement of Al5Ti1B grain rener can be attributed to the enhanced potencyof TiB2 particles with the Al3Ti 2DC monolayer in placeand adequate free Ti solute in the melt after grain reneraddition to achieve the columnar-to-equiaxed transition.

2. ExperimentalAl-2.2Ti-1B 2.26 0.96 0.12 0.11 0.09Al-5Ti-1B 4.80 0.85 0.09 0.08 0.04Zn Ni Cu Cr Al Supplier

0.003 0.005 0.001 0.001 Bal. Norton Bal. LSMIntensive melt shearing was achieved by a rotorstatorhigh-shear device [47], which provides both dispersivemixing at the gap between the rotor and the stator anddistributive mixing of the dispersed particles in the melt.The rotation speed between 5000 and 15,000 rpm providesa local shear rate as high as 105 s1. The key functions ofthe high-shear device include eective forced wetting ofsolid particles by the melt, enhanced kinetic conditionsfor any chemical reaction involving the liquid phase, eec-tive dispersion of particle agglomerates and the uniformdistribution of the dispersed solid particles in the melt. Inthis work, intensive melt shearing was used to wet, disperseand distribute synthetic TiB2 particles in CPAl, and toenhance the kinetic condition for the interaction between soluteTi and TiB2 particles. For the intensive melt shearing in thiswork, the high-shear device was immersed in the melt andoperated at 8000 rpm for a pre-determined period of time.

2.3. Preparation of AlTiB grain reners from syntheticTiB2particles

To investigate the eect of processing conditions on theeectiveness of AlTiB grain reners, two non-standardAlTiB grain reners were prepared under laboratoryconditions using synthetic TiB2 particles and Al10Timaster alloy. These were Al5.5Ti2.5B (equivalent tostoichiometric Al8TiB2) and Al1Ti0.2B (equivalent toAl0.56Ti0.64TiB2).

To prepare the stoichiometric Al5.5Ti2.5B grain re-ner, the high-shear device was used to wet, disperse and dis-tribute the synthetic TiB2 particles in the CPAl melt.Synthetic TiB2 particles were preheated to 300 C andsieved to eliminate large particle agglomerates, and thenwrapped in pure Al foil before being fed into the meltunderneath the high-shear device. Intensive shearing wasconducted at 750 C and 8000 rpm for 15 min after theaddition of synthetic TiB2 particles. The preparedAl8TiB2 melt was then cast into 10 mm diameter grainrener rods.

The non-standard Al1Ti0.2B grain rener was pre-pared in two sequential steps. In the rst step, theAl0.636TiB2 melt was prepared following the procedurefor the stoichiometric Al5.5Ti2.5B (Al8TiB2) grainrener. The prepared Al0.636TiB2 melt was then heatedto 850 C, and the Al10Ti master alloy was added into themelt to provide the required level of free Ti. The Al1Ti0.2B melt was then isothermally held at 850 C for 1 hfollowed by intensive melt shearing for a pre-determined Bal. LSM Bal. LSM

terialperiod of time before being cast into 10 mm diameter grainrener rods.

2.4. TP-1 test and grain size assessment

In order to assess the eectiveness of dierent grainreners, a standard TP-1 test [48] was used to provide con-sistent solidication conditions. All the TP-1 tests wereoperated in conditions to provide a consistent cooling rateof 3.5 Ks1 at the central region of a cross-section 38 mmfrom the bottom of the TP-1 sample. For all TP-1 tests,the melt with and without grain rener addition was alwayspoured at 720 5 C into the TP-1 mould pre-heated to350 C, which was then placed onto the water spray witha controlled water ow rate of 3.8 lmin1. The additionof commercial Al5Ti1B grain rener was made at 0.2%,and followed by isothermal holding for 20 min at 720 Cwith occasional stirring with a ceramic rod to preventparticle settling. For other grain reners, the addition levelwas chosen so that the number density of TiB2 particles waskept consistent at a level equivalent to that of 0.2% com-mercial Al5Ti1B addition (1013m3).

The specimens for grain structure examination weresectioned from a cross-section 38 mm from the base ofthe TP-1 sample, providing a cross-section of 43.3 mmdiameter. The specimens for grain size assessment weretaken from the centre region of the cross-section and pre-pared using a standard metallographic technique. Electro-polishing was performed prior to anodizing with Barkersreagent (4% HBF4 in distilled water) to show the grainboundaries. All the samples were examined under polarizedlight, using a Zeiss optical microscope tted with the AxioVision 4.3 image analysis system. The mean linear intercepttechnique was used to quantify grain size with at least 500grains. In addition, both the cross-section (38 mm from thebase) and the vertical section of the lower part of the TP-1sample (38 mm in height) were polished, etched and photo-graphed to review the grain structure of the TP-1 sample.

2.5. HRTEM and high-resolution scanning transmissionelectron microscopy (HRSTEM)

A pressurized melt ltration technique [49] was used tocollect the TiB2 particles in aluminum melt for examinationby high-resolution TEM and STEM. The solidiedmaterial immediately above the lter, which contained thelocally concentrated TiB2 particles, was sectioned andprepared for metallographic examinations.

To prepare thin foils for conventional HRTEM, andHRSTEM examinations, slices from the ltered residuematerial above the lter were mechanically ground andcut into 3 mm diameter discs. The discs were then handground to a thickness of less than 70 lm and dimpled,and then ion-beam-thinned using a Gatan precision ionpolishing system under conditions of 1.55.0 kV and anincident angle of 36. Conventional TEM and HRTEManalyses were conducted using a Tecnai FEG F30 micro-scope operated at an accelerating voltage of 300 kV.HRSTEM with high-angle annular dark eld (HAADF)imaging (Z-contrast) and atomic resolution electron energyloss spectroscopy (EELS) mapping across the TiB2/Alinterface were carried out using an aberration corrected

Z. Fan et al. / Acta MaNion UltraSTEM 100 instrument (second generation)operated at an accelerating voltage of 100 kV.3. Results

3.1. Reference microstructures for grain renement

For assessing the eectiveness of dierent grain reners,CPAl has been frequently used as a reference material[2,45]. However, it is also well appreciated that variationin impurity level in CPAl and in solidication conditionsduring the TP-1 test will lead to changes in grain structurein the TP-1 sample. Therefore, it is necessary to establishreference microstructures using the same base materialsand the same solidication conditions for all the TP-1 testsin this work. Fig. 1a and b presents the grain structure ofCPAl solidied in the TP-1 mould. It shows a completelycoarse columnar grain structure. The solidication of CPAlbegan by heterogeneous nucleation in the melt near the wallof the TP-1 mould followed by columnar growth withoutany signicant further heterogeneous nucleation. Thiscoarse columnar grain structure will be taken as the refer-ence for a non-grain-rened structure in this work.

Another reference in this work is the fully grain-renedmicrostructure. Fig. 1c and 1d shows the grain structure ofCPAl with the addition of 0.2% commercial Al5Ti1Bgrain rener. Both the cross-section and the vertical sectionshow a ne and equiaxed grain structure. The detailedmicrostructural analysis conrmed that the average grainsize is 175 lm, suggesting that TiB2 particles in theAl5Ti1B grain rener are very eective for grainrenement. This ne and equiaxed microstructure (Fig. 1cand d) was taken as the reference for fully grain-renedmicrostructures.

It should be pointed out that it is not adequate to assessthe grain structure only using the cross-section of the TP-1sample as described in the standard of the TP-1 test [48].For example, Fig. 1a shows columnar grains at the edgeand equiaxed grains at the central region of the sample,suggesting that a columnar-to-equiaxed transition occurredat some stage of the solidication process. This is mislead-ing, since the vertical section of the same sample in Fig. 1bshows a fully columnar grain structure. The equiaxed grainmorphology in Fig. 1a is a result of sectioning of thecolumnar grains at the centre of the TP-1 sample. For thisreason, vertical sections were used for assessing grain struc-tures and cross-sections were only used for assessing thegrain size of fully equiaxed microstructures, i.e., fullygrain-rened structures.

3.2. Grain rening with TiB2 particles and free Ti

A number of investigations in the literature (e.g. [6]) sug-gested that TiB2 particles are not eective substrates forheterogeneous nucleation of the a-Al. However, a fullycolumnar grain structure is not conclusive evidence of theineectiveness of TiB2 particles since, for eective grainrenement, the solid particles not only need to have a goodlattice matching [44], but also need to be completely wetted,well dispersed and uniformly distributed in the liquidphase, and need to have sucient particle number density[44,50]. In this work, intensive melt shearing was used toensure complete wetting, eective dispersion and uniformdistribution of the synthetic TiB2 particles during the prep-aration of Al8TiB2 grain rener without any free Ti.

ia 84 (2015) 292304 295Fig. 2a shows the grain structure of the CPAl TP-1 samplewith the addition of Al8TiB2 grain rener to provide TiB2

terial296 Z. Fan et al. / Acta Maparticles with a number density equivalent to that of 0.2%addition of commercial Al5Ti1B grain rener. This TP-1 sample exhibits a fully columnar grain structure, althoughthe width of the columnar grains is ner (Fig. 2a) than thatof the TP-1 sample without addition of any grain rener(Fig. 1b). A similar result was obtained by the addition of0.2% stoichiometric Al3.2TiB2 grain rener, which con-tains TiB2 particles formed by in situ chemical reactionbut without any excess Ti, (Fig. 2b) and by simultaneousaddition of Al8TiB2 grain rener and Al10Ti masteralloy to provide the equivalent TiB2 particle number den-sity (1013m3) and free Ti (56 ppm) to that of the additionof 0.2% commercial Al5Ti1B grain rener (see Fig. 2c).These results conrmed that TiB2 particles are not eectivefor nucleating a-Al, even with the simultaneous addition offree Ti.

3.3. Identication of Ti atoms adsorbed at the (0001) TiB2surface

To understand the grain rening mechanism of Al5Ti1B grain rener, we focused our eort on the investigationof the Al/TiB2 interfaces in the commercial Al5Ti1Bgrain rener. Fig. 3a is a SEM secondary electron imageshowing the morphology of the TiB2 particles in a deepetched sample of commercial Al5Ti1B grain rener.The boride particles present a typical hexagonal plate-likeshape with the average size being 0.21.5 lm. The TEMbright eld image in Fig. 3b, with the incident electronbeam being perpendicular to the [0001] direction of TiB2,

Fig. 1. Macrographs showing the grain structures of commercial purity alumgrain rener addition, and (c, d) are for CPAl with addition of 0.2% commvertical sections. Without grain rener addition, CPAl has a coarse and fully cnon-grain-rened structure; with 0.2% addition of commercial Al-5Ti-1B grwhich will be taken as a reference for the fully grain-rened structure.ia 84 (2015) 292304shows two such TiB2 particles. The TiB2 particles are facet-ted with their {0001} basal and {1100} prism planes dis-played. Heterogeneous nucleation of Al is believed to occuron the {0001} surfaces of the boride particles [23].

The TEM observation of the interface between the TiB2particle and Al was focused on the surface of the faceted{0001} planes of multiple boride particles. An exampleof such observation using high-resolution lattice imagingis given in Fig. 4a, where the incident electron beam is par-allel to the [1120] zone direction of TiB2 (bottom part)and the [011] zone direction of the adjacent Al grain(upper part). This gives the following well-dened orienta-tion relationship (OR) between the two crystals:

000111-20TiB2; == 1110-11Al:This observation of the much expected OR in a solidied

AlTiB2 sample provides for the rst time evidence to ver-ify that the TiB2 particle does nucleate a-Al. The above ORwas also conrmed by the selected area electron diraction(SAED) pattern, as shown in Fig. 4b that is indexed inFig. 4c.

TEM observation showed that the majority of the TiB2particles were not in the above OR, indicating that themajority of the boride particles from the commercialAlTiB grain rener did not contribute to the nucleationprocess during the solidication. This is consistent withthe conclusion derived from the free growth model thatonly 0.11% of the added TiB2 particles are active for het-erogeneous nucleation [50]. However, careful investigationsby HRTEM at high magnications on multiple TiB2

inum (CPAl) solidied in the TP-1 mould. (a, b) are for CPAl withoutercial Al-5Ti-1B grain rener. (a, c) are cross-sections, and (b, d) areolumnar grain structure (a, b), which will be taken as a reference for theain rener, CPAl has a ne and fully equiaxed grain structure (c, d),

terialZ. Fan et al. / Acta Maparticles showed that there was apparently a monolayer onthe {0001} surface of all of the observed boride particles.Fig. 5a gives an example of such a phenomenon, wherethe incident electron beam is parallel to the [1120] TiB2zone direction, but is o from any low index zone axis ofthe adjacent Al grain. It is seen that an atomic monolayeris uniformly located on the TiB2 surface, with theposition of every atom column in the monolayer shiftedin the [1100] TiB2 direction for a distance of a certainfraction of the (1100) d-spacing (0.26 nm).

To understand the chemical nature of this atomicmonolayer at the Al/TiB2 interface, atomic resolutionsuperSTEM was carried out. Fig. 5b is a STEM HAADFZ-contrast image showing the detail of the Al/TiB2 inter-face, with the incident electron beam being parallel to

Fig. 2. Macrographs showing the fully columnar grain structure ofCPAl inoculated with the TiB2 particles and solidied in the TP-1mould. For all the cases, the number density of TiB2 particles isequivalent to that of 0.2% addition of commercial Al-5Ti-1B grainrener. (a) with synthetic TiB2 particles (via Al-8TiB2 grain rener); (b)with addition of 0.2% stoichiometric Al-2.2Ti-1B master alloy (withoutfree Ti); and (c) with simultaneous addition of synthetic TiB2 particles(via Al-8TiB2 grain rener) and free Ti (56 ppm). These results suggestthat synthetic TiB2 particles have no grain rening power even with thesimultaneous addition of free Ti.500nm

(a)

(b)

ia 84 (2015) 292304 297[1120] TiB2 but o from any low-index zone axis of Al.It shows clearly the atomic monolayer in which the atomcolumns have a brightness comparable to that of the Tiatom columns in TiB2, suggesting that the atomic columnsin the monolayer are likely to be Ti-rich. It is also notedthat the spacing between the atomic monolayer and theTi (0001) plane of TiB2 is 0.25 nm, being shorter thanthe (0001) TiB2 spacing (0.32 nm), as indicated inFig. 5b. This suggests that the monolayer is dierent fromthe Ti-terminated TiB2 surface.

Atomic resolution EELS mapping in superSTEM wasperformed in order to characterize the chemical nature ofthis atomic monolayer. EELS mapping was carried outon a small area (4 6 d-spacing2 as marked by the greenbox in Fig. 6a) across the Al/TiB2 interface. The localHAADF Z-contrast image, Ti K-edge map (green) andsuperimposition of the local HAADF image and Ti K-edgemap are shown in Fig. 6bd, respectively. The atom col-umns with blue circles were identied as Ti-rich columns,whilst those with red circles were Ti columns of TiB2.The EELS mapping conrms that the atomic monolayercontains Ti atoms.

As suggested by our previous MD simulation [36], amonolayer of Al3Ti 2DC can be stable at the liquidAl/TiB2 interface. Therefore, it is likely that Al3Ti 2DC

Fig. 3. (a) Scanning electron micrograph showing the hexagonalmorphology of the TiB2 particles in a deep etched sample ofcommercial Al-5Ti-1B grain rener; and (b) transmission electronmicroscopy image showing the morphology of the TiB2 particles withtheir typically faceted (0001) crystal planes.

Al

0.32nm

(b)

0.25nm

terialAl

0001TiB2-1100TiB2

200Al111Al

TiB2

2 nm

(111)

(0001)

TiB2[11-20])

Al [0-11])

(a)

(b) (c)

298 Z. Fan et al. / Acta Maexists at the solid Al/TiB2 interface. Fig. 7 shows sche-matically the atomic arrangement in the (112) plane ofAl3Ti with the [201] and [241] directions beingmarked. It is shown that the [201] columns in the(112) Al3Ti contain 25 at.% Ti, which is in qualitativeagreement with the Ti-rich conclusion from Fig. 6, wherethe brightness of the [201] columns is uniform and lessbright than that of the Ti columns on the TiB2 surface.In addition, the spacing of the [201] columns in the(112) Al3Ti (0.25 nm measured from Figs. 5b and 6a)is also in good agreement with the atomic spacing alongthe [241] direction in the bulk Al3Ti. Based on theabove analysis, it can be concluded that the observedatomic monolayer at the Al/TiB2 interface is most likelyto be the (112) Al3Ti 2DC, with the following ORsbetween TiB2, Al3Ti 2DC and a-Al:

000111-20TiB2 == 112-201Al3Ti 2DC == 1110-11Al:In the following section, we will assume the observed Ti-

rich monolayer at the Al/TiB2 interface is (112) Al3Ti 2DCand examine experimentally its stability under dierentconditions.

3.4. Formation of Al3Ti 2DC on TiB2 particles

The bulk Al3Ti phase (three-dimensional) is only stablein liquid Al at Ti concentrations greater than 0.15% [3].However, at lower Ti concentrations (

To check the formation of Al Ti 2DC on the TiB par-

terialia 84 (2015) 292304 299Z. Fan et al. / Acta Mapossibly due to the low concentration of Ti in the meltand/or the short duration for the interaction betweenTiB2 particles and the solute Ti atoms.

3 2

ticles at a higher concentration of Ti, a special grain rener,Al1Ti0.2B, was prepared at 850 C using synthetic TiB2particles and Al10Ti master alloy under intensive meltshearing but with dierent shearing times. Intensive meltshearing was used to enhance mass transport in the melt,and therefore to accelerate the adsorption of solute Ti atthe liquid Al/TiB2 interface. The obtained Al1Ti0.2Bgrain rener was used to grain-rene CPAl at an additionlevel equivalent to 0.2% commercialAl5Ti1B grainrener. The resultant grain structures of the TP-1 samplesare compared in Fig. 8 for addition of grain reners pre-pared by intensive melt shearing for 2 and 10 min. TheTP-1 sample grain-rened with the grain rener with2 min shearing exhibited a fully columnar grain structure,while that with 10 min shearing showed a ne and nearlyequiaxed grain structure. This demonstrates that intensivemelt shearing accelerates the formation of Al3Ti 2DC athigher Ti concentration in the AlTi liquid. These resultsconrm that Al3Ti 2DC is thermodynamically stable at850 C in concentrated AlTi solution (Al0.56Ti in thiscase).

3.5. Dissolution of the Al3Ti 2DC adsorbed on the TiB2particles

It is important to point out that an enhanced kineticcondition can only shorten the time for the formation ofan equilibrium phase; it should not aect the stability ofa phase. It is expected that the Al3Ti 2DC is only stable

Ti containing column Ti

(b) (c) (d)

(a)

TiB2[11-20]

Al

1 nm

Fig. 6. (a) SuperSTEM Z-contrast HAADF image; (b) local Z-contrast HAADF across the Al/TiB2 interface; (c) atomic resolutionEELS map of Ti K-edge (green); and (d) superimposition of the localZ-contrast image and Ti K-edge map. The atom columns with bluecircles are Ti-rich columns, and those with red circles are Ti columns.The EELS mapping conrms that the atomic monolayer contains Tiatoms. (For interpretation of the references to colour in this gurelegend, the reader is referred to the web version of this article.)

Fig. 7. Schematic diagram showing the atomic arrangement of Al(lime) and Ti (blue) atoms in the (112) plane of the bulk Al3Tiintermetallic compound. Also marked here are the [201] and[241] directions of Al3Ti.when the Ti concentration is beyond a critical level. Thismeans that Al3Ti 2DC formed on TiB2 particles at high

Fig. 8. Macrographs showing the grain structure of CPAl withaddition of the lab-prepared Al-1Ti-0.2B grain rener from syntheticTiB2 particles (via the Al-8TiB2 grain rener) and free Ti (via the Al-10Ti master alloy) to provide the TiB2 particle number density(1013m3) and free Ti (56 ppm) equivalent to that of addition of 0.2%

commercial Al-5Ti-1B grain rener. The Al-1Ti-0.2B grain rener wasprepared by intensive melt shearing at 850 C during the preparationfor (a) 2 min and (b) 10 min respectively.

Ti concentration should dissolve in dilute AlTi solutions.To conrm this, isothermal holding experiments were car-ried out. 0.2% addition of commercial Al5Ti1B grainrener was made to CPAl at 800 C for a predeterminedperiod of time, and stirred using a ceramic rod before cast-ing into a TP-1 mould at 720 C for grain size assessment.Fig. 9 compares the grain structures of 2 TP-1 samples; onewas cast after isothermal holding for 1 h (Fig. 9a) and theother one was cast after isothermal holding for 78 h(Fig. 9b). Isothermal holding for 1 h had little eect onthe Al3Ti 2DC formed on TiB2 particles, and the TP-1 sam-ple showed a ne and fully columnar grain structure, indi-cating that some solute Ti may have been lost duringisothermal holding. However, isothermal holding for 78 hremoved some of the Al3Ti 2DC formed on TiB2 particles,resulting in the formation of a much coarser columnargrain structure. The isothermal holding experiments(Fig. 10) show that the Al3Ti 2DC formed on TiB2 particlessurvives isothermal holding for 48 h at 800 C, but was notstable in dilute AlTi solution and will dissolve aftersuciently long isothermal holding at high temperature.However, it should be noted that the dissolution of Al3Ti2DC is a very slow process, and that a few hours isother-mal holding in Al with a ppm level of Ti has no eect on thepotency of TiB2 particles covered with Al3Ti 2DC.

In order to further conrm the above nding, extensivemelt shearing was employed to accelerate the dissolution ofthe Al3Ti 2DC formed on TiB2 particles in a dilute AlTisolution. CPAl melt with 0.2% addition of commercialAl5Ti1B grain rener was intensively sheared at

The grain structure of the resultant TP-1 sample isshown in Fig. 11. A simple comparison between Figs. 1band 11 reveals that intensive melt shearing for 45 min hasmade the commercial Al5Ti1B grain rener ineectivefor grain renement. This suggests that the Al3Ti 2DCformed on TiB2 particles is completely stripped by 45 minof intensive melt shearing. It is interesting to note that onthe one hand intensive melt shearing accelerates the forma-tion of Al3Ti 2DC on the TiB2 particles in the concentratedAlTi solution while, on the other hand, it accelerates thede-adsorption of Al3Ti 2DC on the TiB2 particles in diluteAlTi solutions.

3.6. Eect of free Ti in relation to Al3Ti 2DC adsorbed onTiB2 particles

It is generally believed that eective grain renementrequires both potent particles as substrates for heteroge-neous nucleation and solute elements for eective growthrestriction, which allows for larger undercooling to beachieved so that more particles have a chance to participate

Fig. 11. Grain structure of CPAl with addition of 0.2% commercialAl-5Ti-1B grain rener, showing the eect of intensive melt shearingon the eectiveness of commercial Al-5Ti-1B grain rener. The meltwith the grain rener addition was intensively sheared at 720 C for45 min before pouring into the TP-1 mould for solidication. It is seenthat the commercial grain rener was no longer eective for grainrenement.

300 Z. Fan et al. / Acta Material720 C for 45 min before pouring into the TP-1 mould forsolidication.

Fig. 9. Grain structures of CPAl with addition of 0.2% commercial Al-5Ti-1B grain rener solidied in the TP-1 mould, showing the eect ofisothermal holding time at 800 C on the eectiveness of the commer-

cial Al-5Ti-1B grain rener. The melt with the grain rener additionwas held at 800 C for (a) 1 h and (b) 78 h respectively, before beingstirred and poured into the TP-1 mould for solidication.120

150

180

210

240

1.0 1.5 2.0 2.5 3.0 3.5 4.0

Gra

in s

ize,

m

log(t), min

Fig. 10. Average grain size of CPAl with addition of 0.2% commercialAl-5Ti-1B grain rener as a function of holding time at 800 C beforebeing stirred and poured into the TP-1 mould for solidication.

ia 84 (2015) 292304in nucleation at the solidication front [27,28,31] causing acolumnar-to-equiaxed transition [52]. However, in the case

grain renement.

terialof addition of Al5Ti1B grain rener, the contributionfrom the solute Ti to eective grain renement may alsocome from the adsorbed Al3Ti 2DC on TiB2 particles, forwhich only a small amount of excess Ti is required. Thisprompted us to carry out the following experiments toassess the grain rening eciency of TiB2 particles withadsorbed Al3Ti 2DC without further free Ti.

5% commercial Al5Ti1B grain rener was added toCPAl at 720 C to form a 5 kg melt. The melt containinggrain rener was isothermally held for 2 h to allow the set-tlement of TiB2 particles before furnace cooling in the samecrucible without stirring. The bottom part of the solidiedpiece (1 kg) containing the settled TiB2 particles was addedto the fresh CPAl melt to form another 5 kg melt, and thesettling and solidifying experiments was repeated severaltimes to achieve a grain rener containing only TiB2 parti-cles covered with Al3Ti 2DC but with nominally zero freeTi. Such a grain rener was then used to grain-rene CPAlwith appropriate addition level to provide a TiB2 particlenumber density equivalent to 0.2% addition of commercialAl5Ti1B grain rener. The resultant TP-1 sample showeda very ne but columnar grain structure with an averagecolumnar width of 196 lm (Fig. 12). This result demon-strated that the extracted TiB2 particles with Al3Ti 2DCcan be very potent nucleating substrates, resulting in thene columnar grain structure. However, without adequatesolute in the melt a columnar-to-equiaxed transition cannotFig. 12. Grain structure of CPAl with addition of TiB2 particlescollected from the commercial Al-5Ti-1B grain rener with very littlefree Ti, showing that TiB2 particles with Al3Ti 2DC are eective for

Z. Fan et al. / Acta Maoccur, leading to a non-grain-rened microstructure.

4. Discussion

4.1. The existence of Al3Ti 2DC adsorbed on the TiB2particles

In our previous work [36], MD simulation has con-rmed that Al3Ti 2DC can exist at the liquid Al/TiB2 inter-face at temperatures above the Al liquidus. The MDsimulation also revealed that Al3Ti 2DC contains a triangu-lar network of dislocations to accommodate the relativelylarge lattice mist between Al3Ti and TiB2, and that forma-tion of Al3Ti 2DC lowers the liquid Al/TiB2 interfacialenergy compared to that without Al3Ti 2DC [36]. In thispaper, we used an experimental approach to conrm theexistence of Al3Ti 2DC.

Our experimental conrmation of the existence of theAl3Ti 2DC comes from both direct TEM observationsand indirect conrmations from the switch-on andswitch-o experiments with the assistance of intensivemelt shearing [47]. HRTEM and HRSTEM observationshave allowed us for the rst time to identify a Ti-richatomic monolayer at the Al/TiB2 interface in thecommercial Al5Ti1B grain rener (Figs. 5 and 6). Withthe structural analysis (Fig. 7) and the previous MD simu-lations [36], this atomic monolayer is consistent with (112)Al3Ti 2DC, which is most likely formed during the grainrener production process (characterized by a high concen-tration of Ti, high temperature and long reaction time).Our experimental results have also conrmed that TiB2particles without Al3Ti 2DC are not potent substrates forheterogeneous nucleation (Fig. 2), and that the nucleationpotency of TiB2 particles can be signicantly improved bythe formation of the Al3Ti 2DC on the TiB2 surface(Fig. 8). In addition, we have demonstrated experimentallythat Al3Ti 2DC can be made to form on synthetic TiB2particles in a concentrated AlTi melt by intensive meltshearing (switch-on, Fig. 8b) and can be either dissolvedin dilute AlTi solutions by prolonged isothermal holding(switch-o, Figs. 9b and 10) or stripped from the TiB2particles in dilute AlTi solutions by intensive melt shearing(switch-o, Fig. 11).These results have conrmed that theformation of Al3Ti 2DC on the TiB2 surface is crucial forthe eectiveness of AlTiB-based grain reners.

4.2. Formation of Al3Ti 2DC adsorbed on TiB2 surface

Thermodynamically, bulk Al3Ti (three-dimensional) canonly be stable in AlTi alloys with Ti contents greater than0.15% [3]. However, Al3Ti 2DC is a monolayer formed onthe (0001) TiB2 surface through adsorption of Ti atomsfrom the AlTi solution. The thermodynamic stability ofAl3Ti 2DC may not follow the AlTi phase diagram; com-pared with bulk Al3Ti, Al3Ti 2DC may be stable at a lowerconcentration of Ti or even at temperatures above the alloyliquidus. The driving force for the formation of Al3Ti 2DCis the reduction of interfacial energy at the liquid Al/TiB2interface. This means that the formation of Al3Ti 2DC doesnot violate the second law of thermodynamics if the inter-facial energy is taken into consideration for thermodynamicequilibrium.

The Al3Ti 2DC adsorbed on the (0001) TiB2 surfacecan only be stable in AlTi solutions with a Ti concentra-tion greater than a critical concentration. This is supportedby the following experimental observations: (i) Al3Ti 2DCdissolves in dilute AlTi solution (56 ppm Ti) duringisothermal holding (Figs. 9 and 10); (ii) Al3Ti 2DC canbe stripped in dilute AlTi solution by intensive meltshearing (Fig. 11); (iii) Al3Ti 2DC can form in concentratedAlTi alloys (e.g., Al0.56Ti in Fig. 8).

However, the kinetics for both formation and dissolu-tion of Al3Ti 2DC are extremely slow. It took 78 h toachieve some noticeable dissolution of Al3Ti 2DC in anAl0.01Ti dilute solution during isothermal holding at800 C (Figs. 9 and 10), and 45 min to strip Al3Ti 2DCfrom TiB2 particles in an Al0.01Ti dilute solution at720 C by intensive melt shearing (Fig. 11). It was alsofound that the formation of Al3Ti 2DC took 10 min, evenunder intensive melt shearing for a signicantly enhancedkinetic condition for mass transport (Fig. 8b). These resultssuggest that Al Ti 2DC is not easy to form on the TiB sur-

ia 84 (2015) 292304 3013 2

face by adsorption of Ti atoms from the concentrated AlTisolutions although Al3Ti 2DC is thermodynamically

favorable. Additionally, Al3Ti 2DC is not easy to dissolvein dilute AlTi solution where Al3Ti 2DC is not thermody-namically stable.

From both the thermodynamic and kinetic points ofview, it can be concluded that Al3Ti 2DC cannot be formedduring the grain rening process. This is because a 0.2%addition of Al5Ti1B grain rener only provides 56 ppmextra free Ti in the alloy melt, which is normally isother-mally held for a short time (20 min in our case). In addi-tion, the fact that Al3Ti 2DC can be stripped from theTiB2 surface in dilute AlTi solution by intensive meltshearing suggests that Al3Ti 2DC is not thermodynamicallystable. Therefore, Al3Ti 2DC on TiB2 particles can onlyform during the grain rener production process. AlTiB-based grain reners are usually produced by in situ chem-ical reaction in a liquid Al bath containing K2TiF6 andKBF4 salts at temperatures between 700 and 900 C underelectromagnetic stirring [53]. At the end of the process, theAl melt contains TiB2 particles, Al3Ti solid phase and somefree Ti. In this process the formation Al3Ti 2DC is favoredby a high concentration of free Ti and a good kinetic con-dition for mass transport provided by electromagneticstirring.

4.3. Grain rening mechanism in the Al/AlTiB system

Eective grain renement requires not only potentnucleating particles, but also adequate solute elements.Potent nucleating particles ensure copious heterogeneousnucleation events while the solute elements provide growth

to nucleate and to cause eventually columnar-to-equiaxedtransition. Solidication with only potent particles andwithout adequate solute elements produces a fully colum-nar grain structure with a ne columnar spacing, such asthe one shown in Fig. 12. Conversely, solidication witha high concentration of solute, but without adequate potentnucleating particles, will result in a coarse equiaxed grainstructure.

Potency of a nucleating system with given solid particlesand alloy composition can be measured by the lattice mistbetween the nucleating substrate and the nucleated solid atthe moment of nucleation, and is therefore a physical prop-erty of the nucleating system and is independent of thesolidication conditions [44]. The calculated lattice mistbetween TiB2 and the a-Al is 4.22% at 660 C. With theformation of Al3Ti 2DC at the liquid Al/TiB2 interface,the lattice mist is reduced to 0.09% by assuming that theAl3Ti 2DC is a perfect (112) plane of the bulk Al3Ti. Thismeans that the adsorbed Al3Ti 2DC on the TiB2 surfacecan signicantly alter the potency of TiB2 particles. Itchanges the moderately potent TiB2 particles to extremelypotent TiB2 particles covered with Al3Ti 2DC. Our experi-mental results suggest that Al3Ti 2DC is so rmly attachedto the TiB2 surface that it can be regarded as an integralpart of TiB2.

Ti has the largest growth restriction factor among allcommonly used alloying elements for Al alloys [29,46]. Tiis therefore very eective for both generating constitutionalundercooling to allow more TiB2 particles to participate innucleation and eventually causing the columnar-to-equiaxed

Al3T

(b

302 Z. Fan et al. / Acta Materialia 84 (2015) 292304restriction to the nucleated crystals and constitutional und-ercooling at the solidication front to allow more particles

(0001) TiB2 surface (112)

(a)

TiAlBFig. 13. Schematic illustration of the mechanism for heterogeneous nucleatioAl3Ti 2DC monolayer. Arrangements of atoms in (a) Ti-terminated (0001) plof Al. (d) Atomic matching at the TiB2/Al3Ti 2DC/Al interfaces.transition. In addition to the above eects, in this work wehave identied another important eect of Ti on grain

i 2DC (111) Al

) (c)

(d) n of aluminum on (0001) surface of a TiB2 particle which has a (112)ane of TiB2 surface; (b) (112) plane of Al3Ti 2DC; and (c) (111) plane

In summary, the eective grain renement of Al alloysby the Al5Ti1B grain rener can be attributed to the fol-

Al3Ti 2DC in liquid Al is extremely sluggish, andintensive melt shearing can accelerate both the for-

terialmation and dissolution kinetics.(5) Excess Ti aects the eectiveness of AlTiB grain

rener by the formation of Al3Ti 2DC to increasethe potency of TiB2 particles and the provision offree Ti in the inoculated melt to cause columnar-to-equiaxed transition.

(6) Eective grain renement of Al alloys by Al5Ti1B grain rener can be attributed to the increasedpotency of TiB2 particles by the formation of Al3Ti2DC during grain rener production and adequateTi solute in the inoculated melt to cause columnar-to-equiaxed transition.

Acknowledgements

EPSRC-UK is gratefully acknowledged for supporting theEPSRC Centre LiME under grant EP/H026177/1 and the LAT-EST2 Programme Grant.

References

[1] G.K. Sigworth, in: ASM Handbook, Casting. Metals Park,vol. 15, ASM, OH, 2008, p. 255.lowing factors: formation of Al3Ti 2DC on the surface of TiB2 particles

during the grain rener production process, which sig-nicantly increased the potency of TiB2 for nucleationof the a-Al, as schematically illustrated in Fig. 13;

excess Ti in the alloy melt after grain rener addition gen-erates eective growth restriction, which increases theeectiveness of the grain rener by allowing more TiB2particles to be active for heterogeneous nucleation, whichin turn promotes the columnar-to-equiaxed transition;

suitable size, size distribution and adequate number den-sity of TiB2 particles in the Al5Ti1B grain rener canalso be contributing factors to eective grain renement.Although the AlTiB-based grain reners were devel-oped over 60 years, mainly by trial and error, theirdegree of optimization is surprisingly high. It is antici-pated that further improvement of grain rening e-ciency will be very dicult, if at all possible.

5. Conclusions

(1) TiB2 particles can nucleate the a-Al, but they haveonly a moderate potency.

(2) The nucleation potency of TiB2 particles can be sig-nicantly improved by the formation of (112)Al3Ti 2DC on the (0001) TiB2 surface.

(3) Al3Ti 2DC is stable in concentrated AlTi solution,but is unstable and dissolves in dilute AlTisolution.

(4) The kinetics of both formation and dissolution ofrenement, i.e., increasing the potency of TiB2 particles bythe formation of Al3Ti 2DC through adsorption of Ti fromthe melt.

Z. Fan et al. / Acta Ma[2] B.S. Murty, S.A. Kori, M. Chakraborty, Int. Mater. Rev. 47(2002) 3.[3] A. Cibula, J. Inst. Met. 80 (1951) 1.[4] D.G. McCartney, Int. Mater. Rev. 34 (1989) 247.[5] T.E. Quested, Mater. Sci. Tech. 20 (2004) 1357.[6] P.S. Mohanty, J.E. Gruzleski, Acta Mater. 43 (1995) 2001.[7] G.P. Jones, J. Pearson, Metall. Mater. Trans. B 7 (1976) 223.[8] F.A. Crossley, L.F. Mondolfo, AIME Trans. 191 (1951) 1143.[9] J.A. Marcantonio, L.F. Mondolfo, J. Inst. Met. 98 (1970) 23.[10] I. Maxwell, A. Hellawell, Acta Metall. 23 (1975) 229.[11] I.G. Davies, J.M. Dennis, A. Hellawell, Metall. Mater. Trans.

B 1 (1970) 275.[12] M. Johnsson. A critical survey of the grain rening mecha-

nisms of Al (PhD thesis), Stockholm University,1993.[13] M. Vader, J. Noordegraaf, P.C. Van Wiggen, in: E.L. Rooy

(Ed.), Light Metals, TMS, Warrendale, PA, 1991, pp. 11231130.

[14] L. Backerud, P. Gustafson, M. Johnsson, Aluminum 67(1991) 910.

[15] J.A. Marcantonio, L.F. Mondolfo, Metall. Mater. Trans. B 2(1971) 465.

[16] G.K. Sigworth, Metall. Mater. Trans. A 15 (1984) 277.[17] G.S. Vinod Kumar, B.S. Murty, M. Chakraborty, J. Mater.

Sci. 45 (2010) 2921.[18] G.P. Jones, in: J. Beech, H. Jones (Eds.), Solidication

Processing, Sheeld: University of Sheeld, 1987, p. 496.[19] S.E. Donnelly, R.C. Birtcher, C.W. Allen, I. Morrison, K.

Furuya, M. Song, K. Mitsuishi, U. Dahmen, Science 296(2002) 507.

[20] S.H. Oh, Y. Kaumann, C. Scheu, W.D. Kaplan, M. Ruhle,Science 310 (2005) 661.

[21] H. Men, Z. Fan, Comput. Mater. Sci. 85 (2014) 1.[22] G.K. Sigworth, Scr. Mater. 34 (1996) 919.[23] P. Schumacher, A.L. Greer, J. Worth, P.V. Evans, M.A.

Kearns, P. Fisher, A.H. Green, Mater. Sci. Tech. 14 (1998)394.

[24] P. Schumacher, A.L. Greer, Mater. Sci. Eng. A 181 (1994)1335.

[25] M. Johnsson, L. Backrud, G.K. Sigworth, Metall. Trans. A24A (1993) 481.

[26] M.A. Easton, D.H. StJohn, Metall. Mater. Trans. A 30 (1999)1613.

[27] M.A. Easton, D.H. StJohn, Acta Mater. 49 (2001) 1867.[28] H. Men, Z. Fan, Acta Mater. 59 (2011) 2704.[29] M.A. Easton, D.H. StJohn, Metall. Mater. Trans. A 36 (2005)

1911.[30] M. Qian, P. Cao, M.A. Easton, S.D. McDonald, D.H.

StJohn, Acta Mater. 58 (2010) 3262.[31] D.H. StJohn, M. Qian, M.A. Easton, P. Cao, Acta Mater. 59

(2011) 4907.[32] Q. Du, Y. Li, Acta Meter. 71 (2014) 380.[33] R. Wagner, R. Kampmann, Homogeneous second phase

precipitation, in: R.W. Cahn (Ed.), Materials Science andTechnology: A Comprehensive Treatment, Wiley, Weinheim,1991.

[34] Y. Han, Y. Dai, D. Shu, J. Wang, B. Sun, Appl. Phys. Lett. 89(2006) 144107.

[35] J. Wang, A. Horseld, U. Schwingenschlogl, P.D. Lee, Phys.Rev. B 82 (2010) 184203.

[36] T. Qin, Z. Fan, IOP Conf. Sere Mater. Sci. Eng. 27 (2011)012007.

[37] M. Ohring, Materials Science of Thin Films, Second ed.,Academic Press, San Diego, CA, 2002.

[38] F.C. Frank, J.H. Van der Merwe, Proc. R. Soc. London A198 (1949) 216.

[39] D. Turnbull, B. Vonnegut, Indus. Eng. Chem. 44 (1952) 1292.[40] M. Volmer, A. Webber, Z. Phys. Chem. 119 (1926) 277.[41] B. Cantor, Phil. Trans. R. Soc. London A 361 (2003) 409.[42] M.X. Zhang, P.M. Kelly, M.A. Easton, J.A. Taylor, Acta

Mater. 53 (2005) 1427.[43] M.X. Zhang, P.M. Kelly, M. Qian, J.A. Taylor, Acta Mater.

ia 84 (2015) 292304 30353 (2005) 3261.[44] Z. Fan, Metall. Mater. Trans. A 44 (2013) 1409.[45] T.E. Quested, A.L. Greer, Acta Mater. 53 (2005) 4643.

[46] T.E. Quested, A.L. Greer, Acta Mater. 52 (2004) 3859.[47] Z. Fan, Y.B. Zuo, B. Jiang, Mater. Sci. Forum 690 (2011)

141.[48] Standard test procedure for aluminium alloy grain reners

(TP-1). Washington, DC: The Aluminium Association; 1990.[49] Z. Fan, Y. Wang, M. Xia, S. Arumuganathar, Acta Mater. 57

(2009) 4891.

[50] A.L. Greer, A.M. Bunn, A. Tronche, P.V. Evans, D.J.Bristow, Acta Mater. 48 (2000) 2823.

[51] J. Gibbs, The Collected Works of J. Willard Gibbs, vol. 1,Longman, Green and Co., New York, 1928.

[52] J.D. Hunt, Mater. Sci. Eng. A 65 (1984) 75.[53] M.A. Kearns, S.R. Thistlewaite, P.S. Cooper, in: W. Hale

(Ed.), LightMetals, TMS,Warrendale, PA, 1996, pp. 713720.

304 Z. Fan et al. / Acta Materialia 84 (2015) 292304

Grain refining mechanism in the Al/AlTiB system1 Introduction2 Experimental2.1 Materials2.2 Intensive melt shearing2.3 Preparation of AlTiB grain refiners from synthetic TiB2particles2.4 TP-1 test and grain size assessment2.5 HRTEM and high-resolution scanning transmission electron microscopy (HRSTEM)

3 Results3.1 Reference microstructures for grain refinement3.2 Grain refining with TiB2 particles and free Ti3.3 Identification of Ti atoms adsorbed at the (0001) TiB2 surface3.4 Formation of Al3Ti 2DC on TiB2 particles3.5 Dissolution of the Al3Ti 2DC adsorbed on the TiB2 particles3.6 Effect of free Ti in relation to Al3Ti 2DC adsorbed on TiB2 particles

4 Discussion4.1 The existence of Al3Ti 2DC adsorbed on the TiB2 particles4.2 Formation of Al3Ti 2DC adsorbed on TiB2 surface4.3 Grain refining mechanism in the Al/AlTiB system

5 ConclusionsAcknowledgementsReferences