- Email: [email protected]
Interface reactions between rutile coatings and molten aluminium or AlSi7Mg0.6 alloy
Accepted Manuscript Title: Interface reactions between rutile coatings and molten aluminium or AlSi7Mg0.6 alloy Authors: Anton Salomon, Lilit Amirkhanyan, Christiane Ullrich, Mykhaylo Motylenko, Olga Fabrichnaya, Jens Kortus, David Rafaja PII: DOI: Reference:
S0955-2219(18)30473-4 https://doi.org/10.1016/j.jeurceramsoc.2018.07.052 JECS 12020
To appear in:
Journal of the European Ceramic Society
Received date: Revised date: Accepted date:
20-6-2018 26-7-2018 27-7-2018
Please cite this article as: Salomon A, Amirkhanyan L, Ullrich C, Motylenko M, Fabrichnaya O, Kortus J, Rafaja D, Interface reactions between rutile coatings and molten aluminium or AlSi7Mg0.6 alloy, Journal of the European Ceramic Society (2018), https://doi.org/10.1016/j.jeurceramsoc.2018.07.052 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Interface reactions between rutile coatings and molten aluminium or AlSi7Mg0.6 alloy
Anton Salomona,*, Lilit Amirkhanyanb, Christiane Ullricha, Mykhaylo Motylenkoa, Olga
a
IP T
Fabrichnayaa, Jens Kortusb and David Rafajaa
Institute of Materials Science, TU Bergakademie Freiberg, Gustav-Zeuner-Straße 5, 09599 Freiberg, Germany
Institute of Theoretical Physics, TU Bergakademie Freiberg, Leipziger Str. 23, 09599
SC R
b
Freiberg, Germany
U
* Corresponding author:
N
Email: [email protected] Tel.: +493731/39-2671
M
ED
homepage: www.ww.tu-freiberg.de
A
Fax: +493731/39-2604
Abstract
PT
Rutile coatings deposited on corundum substrates are considered as promising functional ele-
CC E
ments improving the efficiency of the filtration of oxide inclusions out of aluminium melts. This contribution describes the reactions between rutile and two kinds of the aluminium melts and discusses the consequences of these reactions for the filtration process. It was found that
A
the contact of rutile coatings with molten aluminium leads to the formation of a corundum layer at the solid/liquid interface. The exposure of the rutile coatings to molten AlSi7Mg0.6 alloy produces an interface layer of MgTiO3. The interface layers possess defined orientation relationship to rutile which is characteristic for locally heteroepitaxial growth. The density functional theory calculations revealed that the TiO2/α-Al2O3 and TiO2/MgTiO3 interfaces with the orientation relationships observed experimentally have low interface energies. The
mechanisms of the interface layer formation and the impact of these layers on the degradation of the rutile coatings are discussed.
Keywords: Filter ceramics; Rutile coating; Reactive diffusion; Local heteroepitaxy; Interface
SC R
IP T
energy
1. Introduction
Ceramic filters containing corundum are traditionally used for filtration of metallic melts [1].
U
In order to improve their filtration efficiency, different approaches are pursued. One of them
N
is the development of hierarchical structures [2], another one is the functionalization of the
A
corundum surface by ceramics coatings that attract and anchor small inclusions and bond ox-
M
ygen solved in the melt [1]. Concurrently, the functional coatings should not have a negative
ED
impact on the melt flow rate and should not substantially corrode [3]. Furthermore, the functional coatings and the products of the reactions between the metallic melt and the functional
PT
coatings must not peel off and pollute the melt.
CC E
Some time ago, it was shown that rutile-coated ceramic foam filters provide high flow rates for AlSi7Mg melts and that they possess moderate filtration efficiency for different inclusion types [1]. Both properties were attributed to a specific wetting behaviour of the aluminium
A
alloy melt on the rutile surface during the filtration. The wetting and the filtration efficiency are typically affected by chemical reactions between the filter surface and the metallic melt [5, 6]. Recently, it was shown that rutile (tetragonal TiO2 crystallising in the space group P42/mnm) reacts with molten aluminium and aluminium alloys [4], and that it can be employed for the functionalisation of reactive filters [3].
2
Rutile coatings brought in contact with pure aluminium were found to form primarily a corundum layer (α-Al2O3, space group R3̅c) at the interface to the Al melt [4]. At long operation times, Ti2O3 (SG R3̅c) and Al3Ti (SG I4/mmm) formed additionally as minor phases. In the case of the aluminium alloy AlSi7Mg0.6, a layer of MgTiO3 (SG R3̅) grew predominantly on the surface of the TiO2 coatings. Additional phases occurring after long operation times were
IP T
α-Al2O3 (SG R3̅c) and (Al,Si)3Ti (SG I4/mmm).
SC R
The formation of α-Al2O3 or MgTiO3 as principal phases confirmed the aspired reactive be-
haviour of the rutile coatings, i.e., the capability of reducing the oxygen content in the aluminium melt or molten AlSi7Mg0.6 [4]. Additionally, the growth of the α-Al2O3 or MgTiO3 lay-
U
ers delayed the formation of Al3Ti or (Al,Si)3Ti, which could precipitate in the aluminium or
N
AlSi7Mg0.6 melt and act as unwanted inclusions. However, the α-Al2O3 and MgTiO3 layers
A
can peel off from the surface of the functionalised metal melt filters, which would contami-
M
nate the filtered melt as well. In order to understand the binding mechanisms and to improve the adhesion between the original rutile coating and the respective oxide layer, the reaction
ED
between rutile and molten pure Al or AlSi7Mg0.6 alloy and the growth of α-Al2O3 and
PT
MgTiO3 on TiO2 were investigated systematically. The central phenomena discussed in this study are the formation of reaction products at the
CC E
surface of functional rutile coatings brought in contact with Al or AlSi7Mg0.6 melt, their adhesion to the respective substrate and their ability to inhibit further reactions. In this context, a
A
heteroepitaxy between rutile and the corrosion products is believed to play a very important role, as the heteroepitaxial growth is considered both as a feature improving the adhesion of the corrosion products and as a factor influencing the reaction kinetics via lattice strain at the interface.
3
The reaction experiments were performed in a Spark Plasma Sintering (SPS) device to imitate the heat treatment conditions during standard aluminium casting but without the flow of the melt. Possible heteroepitaxy between TiO2 and the reaction products was concluded from specific orientation relationships between the adjacent phases. The orientation relationships were obtained from the electron backscatter diffraction (EBSD) patterns measured in a scanning
IP T
electron microscope and from the selected area electron diffraction (SAED) patterns measured
in a transmission electron microscope. For selected orientation relationships found experimen-
SC R
tally, specific interfacial energies and formation probabilities were determined using the Density Functional Theory (DFT) calculations. Calculated Gibbs energies of the chemical reac-
N
U
tions identified experimentally are negative and shown for comparison.
M
A
2. Materials and methods 2.1 Sample preparation
ED
The samples under study consisted of corundum substrates coated by rutile that were brought
PT
in contact with molten aluminium or molten aluminium alloy AlSi7Mg0.6 in an SPS device. Both, Al and AlSi7Mg0.6, were supplied in powder form. For metal melting, a special sample
CC E
setup was applied that consisted of a hollow corundum cylinder, whose interior was coated with rutile and filled with Al or AlSi7Mg0.6 powder. The corundum cylinders were prepared from Martoxid MR70 (high-purity ground alumina; Albemarle Corporation, USA) by slip
A
casting and subsequent conventional sintering at 1600 °C according to the procedure described in references [4] and [7], and had an outer diameter of 20 mm, an inner diameter of 15 mm and a length of 10 mm. The rutile coatings were applied via slip casting and conventionally sintered at 1300 °C [7]. This technique is commonly used for rutile deposition in industry. Its drawback is a thickness variation. In our case, the thickness of the rutile coating 4
varied between 10 to 25 µm. However, the thickness variation did not affect the results of this study. As starting material for the rutile deposition, powder R320 from Sachtleben, Germany (98.5 mass% TiO2, mean particle size 0.55 µm) was used. The powder particles of pure Al powder (99.9 %, Goodfellow, Germany) had a maximum size of 60 μm, the AlSi7Mg0.6 al-
IP T
loy powder particles (TLS Technik, Germany) had the size between 45 and 100 μm. The main impurities in the starting materials are summarised in Table 1. The chemical compositions of
SC R
the starting materials were provided by the respective manufacturer. The impurities in rutile
U
were identified and quantified by X-ray fluorescence analysis (XRF).
Source
Zn
Particle size [µm]
-
-
< 60
Goodfellow
0.02
0.28
45 – 100
TLS Technik
0.55 (d0.5)
Sachtleben
Al
Si
Mg
Fe
Cu
Pure Al
balance
0.04
0.0004
0.03
A
N
Elemental composition [mass%] Powder
AlSi7Mg0.6
balance
7.07
0.61
M
0.003
0.09
0.09
Mn
> 98.5 mass% of rutile * Rutile TiO2 R320
CC E
PT
ED
* XRF: 0.4 mass% Al2O3, 0.3 mass% of Fe2O3 and Cr2O3, 0.2 mass% of P2O5 and K2O, MgO < 0.02 mass% Table 1: Chemical compositions and particle sizes of the aluminium powder, aluminium alloy powder and rutile powder used for the sample production as specified by manufacturers. Secondary oxides in rutile were analysed additionally using XRF.
The functionalised cylinders (corundum with rutile coating) were filled with Al or AlSi7Mg0.6 powder and heated to 750 °C in a SPS unit HPD 25 (FCT Systeme GmbH, Ger-
A
many). The working atmosphere was residual air with a pressure between 0.5 and 5 Pa. The power-controlled heating resulted in heating rates of about 1500 K∙min-1. The dwell times at 750 °C were 1, 30, 60 and 300 minutes. After the heating was switched off, the samples were rapidly cooled by the water-cooled steel electrodes of the SPS apparatus. In this way, a cool-
5
ing rate of about 300 K∙min-1 was achieved. The sample set-up, the tooling geometries and further detailed information about the SPS heat treatment are given in reference [8].
2.2 Methods of sample characterisation
IP T
On the microscopic scale, the products of the reaction between the respective melt and the TiO2 coatings were analysed using scanning electron microscopy with energy dispersive X-
SC R
ray spectroscopy (SEM/EDX) and electron backscatter diffraction (SEM/EBSD), and using electron probe microanalysis with wavelength dispersive X-ray spectroscopy (EPMA/WDX).
U
For SEM/EDX/EBSD, a high-resolution SEM LEO-1530 (Carl Zeiss AG, Germany) with
N
field-emission cathode, an EDX detector (Bruker AXS) and a Nordlys II EBSD detector
A
(HKL Technology) was used. All experiments were carried out at an acceleration voltage of
M
20 kV. The working distance for EBSD was 15 mm, the tilting angle 70 ° and the step size 0.3 µm. The software package Channel 5 (HKL Technology) was employed for identification
ED
of the Kikuchi patterns and for evaluation of the measured data. An electron probe microanalyser JXA8900 RL (Jeol GmbH, Germany) with five crystal spectrometers was employed
PT
for EPMA/WDX.
CC E
On the nanoscale, the samples were characterised using transmission electron microscopy (TEM), selected area electron diffraction (SAED) and energy dispersive X-ray spectroscopy (EDX). The TEM analyses were done in a JEM 2200 FS transmission electron microscope
A
(JEOL, Japan) at an acceleration voltage of 200 kV. The TEM samples were prepared by the focussed ion beam method (FIB) with a Helios NanoLab 600i (FEI, USA) in form of thin slices.
2.3 Calculation of the Gibbs and interface energy 6
The Gibbs energies (ΔG°) of the chemical reactions found experimentally were calculated with the Thermo-calc software [9] using the SGTE unary database [10] combined with recently assessed data for the systems MgO-Al2O3 [11], Al-Ti-O [12] and Al2O3-MgO-TiO2 [13] were included. The activity of molten Mg in molten Al alloy was calculated using the SGTE
750 °C and at ambient pressure. No gaseous phase was involved.
IP T
solution database [14]. It was assumed that the reactions run at a constant temperature of
The interface energy (σ) has been calculated as a difference between the total energy of an
𝐸𝑠𝑙𝑎𝑏 − 𝐸𝑏𝑢𝑙𝑘1 −𝐸𝑏𝑢𝑙𝑘2 2𝐴
(1)
U
𝜎=
SC R
interface slab and the total energies of the corresponding bulk materials forming the interface:
N
In equation (1), A represents the interface area, and Eslab, Ebulk1 and Ebulk2 are the total energies
A
of a slab including the interface and the bulk structures, respectively. The total energies were
M
obtained from the DFT calculation as implemented in the Quantum ESPRESSO code [15]. The projector augmented wave (PAW) method was used together with the generalised gradi-
ED
ent approximation (GGA) for exchange correlation functional according to the Perdew, Burke
PT
& Ernzerhof (PBE) approach [16].
For the calculation of the total energies of the bulk materials, the crystallographic models of
CC E
rutile (ICSD #66650), corundum (ICSD #73725) and MgTiO3 (ICSD #156217) were used. The interface slabs have been modelled using several lattice planes of the corresponding material, i.e., rutile and corundum or rutile and magnesium titanate, by considering the crystallo-
A
graphic orientation relationships obtained from our experiments and/or reported in literature [17] - [19]. In the starting slabs employed for the relaxation procedure, the lattice misfit between the counterparts was avoided by adjusting the in-plane lattice parameters of TiO2 (lattice parameters along the interface to α-Al2O3 or MgTiO3). The in-plane lattice parameters of α-Al2O3 or MgTiO3 were not changed, because both compounds have larger bulk modulus 7
than TiO2. However, the expansion of the in-plane TiO2 lattice parameter was smaller than 1 % in all cases. In this way, “supercells” of the space group P1 containing several formula units of TiO2 and either Al2O3 or MgTiO3 were created as the starting structures for the relaxation. In the next step, the individual supercells were fully optimised by relaxing the cell pa-
IP T
rameters and the atomic positions. The slab applied for description of the TiO2/MgTiO3 interface with the orientation relation-
ship (100)rutile || (110)titanate and [001]rutile || [11̅0]titanate [18] contained 159 atoms (18 formula
SC R
units of TiO2 and 21 formula units of MgTiO3). The interface between TiO2 and -Al2O3 with the orientation relationship (101)rutile || (012)corundum and [010]rutile || [100]corundum [19] was
U
modelled using 78 atoms (6 TiO2 and 12 Al2O3). The model of the TiO2/MgTiO3 interface
N
with the orientation relationship (001)rutile || (100)titanate and [100]rutile || [010]titanate interface
A
contained 45 atoms in the slab (5 TiO2 and 6 MgTiO3). Finally, the interface slab TiO2/-
M
Al2O3 with the orientation relationship (001)rutile || (100)corundum and [100]rutile || [010]corundum
Interfaces
k points
Ecut [Ry]
Ecut_rho [Ry]
(100)rutile || (110)titanate, [001]rutile || [11̅0]titanate [18]
3×3×3
80
320
(101)rutile || (012)corundum, [010]rutile || [100]corundum [19]
8×2×9
70
280
(001)rutile || (100)titanate, [010]rutile || [001]titanate *
12×3×14
95
380
(001)rutile || (100)corundum, [010]rutile || [001]corundum [17] *
8×2×9
60
240
CC E
PT
ED
[17] comprised 141 atoms (17 TiO2 and 18 Al2O3).
A
* experimentally observed in this study Table 2: Orientation relationships between the counterparts, number of the k points, the cut-off energy and the cut-off of kinetic energy for charge density and potential calculation used for the interface energy calculation.
The computational parameters used for individual interfaces are summarised in Table 2. The valence electron wave functions are expanded in a plane wave basis with a cut-off energy 8
(Ecut). The effective potential and the electron density are given in a numerical basis with Ecut_rho. The criterion to terminate the geometry relaxation of the interface slabs relaxing all the internal coordinates has been that all force components on the atoms were lower than
IP T
0.001 Ry Å-2.
3. Results
SC R
3.1 Rutile in contact with molten AlSi7Mg0.6
The reaction between the rutile coatings and the molten AlSi7Mg0.6 alloy at 750 °C leads to
U
the formation of MgTiO3 (geikielite, SG R3̅). The MgTiO3/TiO2 interface is strongly fringed
N
(Fig. 1), which results in a large variation of the MgTiO3 layer thickness. For the reaction time
A
of 60 min, the MgTiO3 layer thickness ranges between 2.0 µm and 11.9 µm (Fig. 1a), for the
M
reaction time of 300 min between 2.5 µm and 19.7 µm (Fig. 1b). These large variations of the MgTiO3 layer thickness are caused primarily by the needle-like or wedge-like shape of the
ED
MgTiO3 grains. Although MgTiO3 is not stable at the 750 °C (cf. Ref. [4]), its stoichiometry
PT
was confirmed by the EPMA/WDX analysis and its crystal structure by local diffraction methods (EBSD/SEM and SAED/TEM).
CC E
The EBSD measurements performed in the reaction zone containing rutile, the newly formed MgTiO3 layer and the solidified AlSi7Mg0.6 alloy (Fig. 2a) revealed that TiO2 and MgTiO3
A
possess a well-defined orientation relationship, which is very close to (001)rutile || (100)titanate and [010]rutile || [001]titanate. The average deviation of the measured orientation relationships from the ideal one was approximately 2.5 ° (Fig. 2b). The ideal orientation relationship between TiO2 and MgTiO3 was found in the small, stripe- or needle-like, dark grey MgTiO3 grains that penetrate into the TiO2 matrix. The MgTiO3 needles grew always anomalously in the [001]titanate direction, which was parallel to the [010] direction in rutile. 9
The orientation relationship close to (001)rutile || (100)titanate and [010]rutile || [001]titanate was also confirmed by the SAED patterns that were taken near the TiO2/MgTiO3 interface (Fig. 3a). The formation of MgTiO3 with this orientation relationship to TiO2 can be seen as reactive diffusion of MgO into TiO2. The anomalous growth of the MgTiO3 needles along the [001] direction (Fig. 2c) indicates that the diffusion takes place predominantly along the directions
IP T
[001]titanate and [010]rutile (Fig. 3b). From the point of view of the crystal symmetry, the pre-
ferred diffusion directions must lie in the (001)rutile plane. A faster diffusion within the
SC R
(001)rutile planes has already been described in references [20] and [21], where it was explained by the diffusion of cations over the interstitial positions (see also Section 3.3).
U
The chemical reaction behind the formation of MgTiO3 can be described as 1
TiO2 + [Mg] + 2 O2 → MgTiO3
(2)
A
N
(Δ°G = -512.5 kJ mol-1)
M
The source of magnesium and oxygen is the aluminium alloy. Magnesium is one of the alloying element in AlSi7Mg0.6. Even if the positive contribution (47.9 kJ mol-1) of the activity of
ED
Mg in the liquid Al alloy is taken into account, ΔG is still negative making it possible for
PT
above reaction to occur. Oxygen is originally adsorbed on the surface of the aluminium alloy powder particles and contained in the form of amorphous oxide skin [22] covering the powder
CC E
particles used for SPS melting. As the amorphous alumina films are less stable than the crystalline Al2O3 (corundum), they are decomposed during the SPS process, in particular if Mg is
A
present as alloying element in the aluminium melt (see Ref. [23] and [24]). In reaction (2), MgO is a possible intermediate phase. However, it was not found in our samples. Anyway, this reaction model implies that the amount of magnesium and oxygen in the AlSi7Mg0.6 melt is reduced by the contact of the melt with the functional TiO2 coating. This phenomenon was confirmed by the local chemical analysis of the solidified aluminium alloy.
10
In the proximity of the reaction zone, the amount of magnesium in the solidified melt was reduced below the detection limit of WDX. Possible habitus planes in Figure 3a are (311)rutile and (011)titanate. The lattice misfit that was calculated for the oxygen atoms located in these habitus planes using the lattice parameters arutile = 0.4653 nm, crutile = 0.2965 nm, atitanate = 0.5057 nm and ctitanate = 1.3903 nm (ICSD
IP T
#66650 and #156217) was smallest along the [1̅03]rutile || [100]titanate directions (0.24 %) but
SC R
significantly larger along the [1̅21]rutile || [11̅1]titanate directions (zone axes in Fig. 3a), where it
was about 10 %). The reference oxygen atoms used for the lattice misfit calculation had the Wyckoff positions 4f and 18f in TiO2 and MgTiO3, respectively. The lattice misfit between
U
TiO2 and MgTiO3 is compensated by geometrically necessary dislocations with almost equi-
N
distant spacing, which were found at the interface between TiO2 and MgTiO3 (Fig. 3a). The
A
dislocation ordering causes the slight deviation from the ideal orientation relationship
M
(001)rutile || (100)titanate and [010]rutile || [001]titanate discussed above. The inclination between the areas 1 and 2 in Figure 3a is about 1 – 2 ° and agrees well with the deviations found using
ED
EBSD (Fig. 2b).
PT
The mutual orientation (001)rutile || (100)titanate and [010]rutile || [001]titanate has not been explicitly reported in literature so far. However, the orientation relationships between rutile and phas-
CC E
es with ilmenite structure (SG R3̅) that can be found in literature, e.g., (010)rutile || (001)ilmenite and [101]rutile || [210]ilmenite [25] - [27], (101)rutile || (110)ilmenite and [100]rutile || [010]ilmenite [25],
A
[28] [29], or (301)rutile || (11̅0)ilmenite and [010]rutile || [001]ilmenite [25], [29], are strongly related to the observed one, as they result in similar atomic arrangements. These orientation relationships can hardly be distinguished from each other, because they are mutually inclined by less than 3 ° [26]. This phenomenon can also contribute to the mean deviation from the perfect orientation relationship (001)rutile || (100)titanate and [100]rutile || [010]titanate, which was observed by EBSD (cf. Fig. 2b). 11
As the growth of the MgTiO3 reaction layer changes the chemical composition of the AlSi7Mg0.6 melt, a deceleration of the reaction diffusion process is expected for longer diffusion times, in particular if the amount of Mg and O in the melt is limited and/or if the reaction diffusion process runs without melt convection. In the TiO2 coating that was in contact with the AlSi7Mg0.6 melt for 300 minutes at 750 °C, adjacent tiny stripes of MgTiO3 and Ti2O3
IP T
were found by transmission electron microscopy (Fig. 4). The chemical composition of Ti2O3 was proven by EDX in STEM, its crystal structure (SG R3̅c, corundum type) by SAED. Addi-
SC R
tionally, SAED revealed that MgTiO3 and Ti2O3 grow with the same crystal orientation (Fig. 4). In analogy to other materials combinations [30] - [34], the heteroepitaxial growth of
U
MgTiO3 on TiO2 and Ti2O3 is believed to improve the adhesion of the MgTiO3 layer to the
N
respective substrate and to protect the delamination of the reaction layer during the metal melt
A
filtration.
M
The presence of Ti2O3 instead of TiO2 implies that the oxygen needed for the formation of
ED
MgTiO3 is obtained from the reduction of TiO2 to Ti2O3: 3 TiO2 + [Mg] → Ti2 O3 + MgTiO3 ,
(ΔG° = -238.4 kJ mol-1)
(3)
PT
when oxygen contained in the aluminium alloy is spent or if the diffusion of Mg into TiO2 is
CC E
faster than the oxygen diffusion. The contribution of the Mg activity in the Al alloy is the same as mentioned above. Some deviations from the stoichiometric composition of the reaction products cannot be excluded, because MgTiO3 and Ti2O3 may form a solid solution [35].
A
Still, both (separated) phases were found in the TEM micrograph (Fig. 4) and confirmed by the TEM/EDX line scans. This change in the reaction mechanism, which is caused by the consumption of oxygen from the melt or by a slower oxygen diffusion in comparison with the magnesium diffusion, slows down the MgTiO3 formation significantly, as it can be seen on
12
the lower growth rate of the corrosion layer containing Ti2O3. For more details, see Section 3.3.
3.2 Rutile in contact with pure molten aluminium
IP T
In order to illustrate the effect of the magnesium concentration on the reaction between the
functional TiO2 coating and the aluminium alloy melt, analogous experiments as above were
SC R
performed in a system containing a very low amount of Mg impurities in the Al powder and
in the TiO2 powder used for production of the functionalised coatings. At negligible Mg con-
U
tents, the reaction between the rutile coating and the Al melt led to the formation of an -
N
Al2O3 layer at the surface of the TiO2 coating (Fig. 5). The stoichiometry of the layer was
A
confirmed by the EPMA/WDX analysis, its crystal structure by EBSD/SEM. After 60 min
M
and 300 min at 750 °C, the α-Al2O3 reaction layer had a thickness of about 0.7 ± 0.3 µm (see Fig. 5a) and 1.6 ± 0.4 µm (Fig. 5b), respectively.
ED
In all cases, the corundum layer adhered very well to the rutile coating, which indicates, in
PT
combination with the crack-free interface, a heteroepitaxial growth mechanism. The EBSD measurements performed in the vicinity of the original Al/TiO2 interface (Fig. 6) in the sam-
CC E
ple held for 60 min at 750°C revealed an orientation relationship between the newly formed α-Al2O3 grains and the TiO2 matrix, which can be expressed as (001)rutile || (100)corundum and [100]rutile || [010]corundum. This orientation relationship has already been reported in literature
A
(see Ref. [17]). It is worth noting that this orientation relationship is identical with the orientation relationship found for TiO2 and MgTiO3 (cf. Section 3.1 and Fig. 3b), because the phases having the corundum-type or the ilmenite-type structure exhibit generally similar crystallographic orientations towards rutile [25] - [29]. In our samples, this orientation relationship
13
appeared along the whole α-Al2O3/TiO2 interface (cf. Fig. 6a). The experimentally observed misorientations were below 3 ° (Figure 6b). In reference [17], the oriented growth of rutile on α-Al2O3 was explained by a similar arrangement of oxygen atoms in rutile (Wyckoff positions 4f) and in corundum (Wyckoff posi-
IP T
tions 18e) having the observed orientation relationship (001)rutile || (100)corundum and [100]rutile || [010]corundum. The lattice misfit calculated using the lattice parameters arutile = 0.4653 nm, crutile = 0.2965 nm (ICSD #66650), acorundum = 0.4759 nm and ccorundum = 1.2997 nm (ICSD #73725)
SC R
for the directions [100]rutile || [010]corundum and [010]rutile || [001]corundum within the above parallel lattice planes was -2.3 % and +7.1 %, respectively.
U
In the sample exposed to the Al melt for 300 min, areas with stripe-like contrasts were ob-
N
served at the interface between the newly-formed corundum and the rutile coating. The stripes
A
(marked by arrows in Fig. 5b) are almost perpendicular to the TiO2/Al2O3 interface. The na-
M
ture of the stripes was investigated using a combination of TEM imaging, SAED and EDX chemical analysis in the STEM mode (Fig. 7a). The stripes were identified as MgAl2O4 nee-
ED
dles (SG Fd3̅m) that penetrate into the former TiO2 coating, which contains additionally
PT
MgTiO3 regions (Fig. 7). It was discussed above that the formation of MgTiO3 from TiO2 requires magnesium and additional oxygen (equation (2)). Without additional oxygen, TiO2
CC E
would be partially reduced to Ti2O3 (equation (3)). Mg and O are fast diffusing and highly reactive species in TiO2. Their diffusivity is especially high along grain boundaries and along
A
specific crystallographic directions in rutile (see section 3.1). Because of the high reactivity of Mg, even extremely small amounts of this species as a contaminant of molten Al or in rutile seem to be sufficient to trigger the formation of the Mg-containing compounds (see also discussion in section 4). As no Ti2O3 was found in the vicinity of MgTiO3 in TiO2 being in contact with molten aluminium (Fig. 7b) in contrast to the AlSi7Mg0.6/TiO2 system (Fig. 4), it can be concluded that 14
a lower Mg content and consequently a higher [O]/[Mg] ratio in the system facilitate the chemical reaction according to equation (2). An analogous result can be obtained from the observed formation of MgAl2O4. This compound stems from the reaction of Al2O3 with magnesium and oxygen: 1
Al2 O3 + [Mg] + 2 𝑂2 → MgAl2 O4
(4)
IP T
(Δ°G = -520.3 kJ mol-1)
The alternative reaction without the additional oxygen supply would lead to the TiO2 reduc-
SC R
tion and to the production of Ti2O3: 2 TiO2 + Al2 O3 + [Mg] → MgAl2 O4 + Ti2 O3
(5)
U
(ΔG° = -246.2 kJ mol-1)
N
At the beginning of the reaction process, corundum growing between the Al melt and the
A
TiO2 coating (Fig. 5) might be formed from Al and O contained in the aluminium melt. Later
M
on, oxygen is provided mainly by the reduction of rutile. Prominent reduction mechanisms are:
PT
and
ED
6 TiO2 (𝑠) + 8 [Al] → 4 Al2 O3 (𝑠) + 6 [Ti]
(6)
(ΔG° = -531.6 kJ mol-1)
(7)
CC E
6 TiO2 (𝑠) + 2 [Al] → Al2 O3 (𝑠) + 3 Ti2 O3 (𝑠)
(ΔG° = -835.9 kJ mol-1)
The formation of Ti2O3 as an intermediate phase according to equation (7) was observed in
A
sintered powder mixtures of Al and TiO2 (mass ratio of 2:1) and confirmed by thermodynamic calculations [4]. However, it should be noted that the aluminium supply in the above powder mixture was limited because of the high amount of TiO2. In case of the aluminium surplus like in this study, the reduction of TiO2 to metallic titanium (equation (6)) is more probable. In the sintered powder mixtures of Al and TiO2 from reference [4], titanium stemming from the TiO2 reduction led to the formation of Al3Ti. In the present study, no Ti and Al3Ti were 15
detected, because TiO2 was not even reduced to Ti2O3. Thus, the reactions according to equations (6) or (7) did not contribute to the phase formation processes of this study. In the TiO2 coatings brought into contact with pure Al melt, rutile (TiO2), MgAl2O4 spinel and magnesium titanate (MgTiO3) also possess distinguished (and already reported) orientation relationships (010)rutile || (111)spinel, [100]rutile || [110]spinel [25], [28], and (001)titanate || (111)spinel,
IP T
[110]titanate || [110]spinel [26] [36], see Figure 8. A possible reason for this kind of heteroepitaxy is the parallel arrangement of close packed oxygen layers that was already discussed by Arm-
SC R
bruster [25]. In this regard, the appearance of the stripe-like contrast merely represents a lo-
calised change of the heteroepitaxially grown phases, but the presence of the stripes does not
A
N
U
deteriorate the adhesion of corundum to rutile.
M
3.3 Role of the diffusion processes in the growth of the interface layers The thickness of the corundum layer, which has formed at the interface between molten Al
ED
(750 °C) and TiO2, follows a parabolic time dependence. The average layer thickness was (0.7 ± 0.3) µm after 60 min and (1.6 ± 0.4) µm after 300 min, see Figure 5. The growth rate
PT
constant calculated from this layer thickness increase is (7.0 ± 1.3) × 10-17 m2s-1 (cf. Ref.
CC E
[37]). As no Ti2O3 or metallic titanium were observed at the Al2O3/TiO2 interface, it can be expected that the aluminium melt is the primary source of oxygen needed for the Al 2O3 formation and that the growth of the Al2O3 layer is controlled mainly by the diffusion of oxygen
A
in the melt. Furthermore, the Al2O3 layer can be regarded as a protective coating against the reduction of TiO2 by molten aluminium. The growth of the MgTiO3 layer at the interface between the AlSi7Mg0.6 melt and rutile is much faster than the growth of the Al2O3 layer from the Al melt, and depends strongly on the local orientation of the TiO2 grains. The minimum MgTiO3 layer thickness was 2.0 µm after 16
60 min and 2.5 µm after 300 min (Fig. 1). The maximum thickness of the MgTiO3 layer (11.9 µm and 19.7 µm after 60 min and 300 min, respectively) was found in the MgTiO3 needles that were oriented with their c axis along the diffusion direction. In contrast to the Al2O3 layer, the MgTiO3 layer grows at the expense of rutile and its growth cannot be described as parabolic. At the beginning of the chemical reaction between molten AlSi7Mg0.6 alloy and
IP T
the TiO2 coating, MgTiO3 grows much faster (approximately 10 times) than at longer reaction
times. This change of the reaction kinetics is caused by a change in the reaction mechanism,
SC R
which results from the different amounts of magnesium and oxygen in the melt and/or from the different diffusivities of both species in MgTiO3.
U
As long as enough oxygen is present in the melt (close to the AlSi7Mg0.6/TiO2 interface), the
N
MgTiO3 layer grows according to equation (2) from Mg and O stemming from the melt. After
A
oxygen in the melt is consumed and/or after a closed MgTiO3 layer possibly acting as a diffu-
M
sion barrier for oxygen is built on, mainly Mg diffuses from the AlSi7Mg0.6 melt through MgTiO3 into rutile, where it reduces TiO2 to Ti2O3, see equation (3). The rutile reduction and
ED
the diffusion of Mg to the reaction front at the MgTiO3/TiO2 interface are the reasons for the significantly reduced growth rate at longer dwell times. The change in the reaction kinetics
PT
might be seconded by the exhaustion of Mg in the aluminium alloy in the vicinity of the
CC E
AlSi7Mg0.6/MgTiO3 interface.
A
3.4 Interface energies of the materials combinations with experimentally identified orientation relationships
The energies of the TiO2/MgTiO3 and TiO2/Al2O3 interfaces with distinct orientation relationships between the counterparts (Figs. 9, 3b and 6c) were calculated using the interface models described in Section 2.3, and are summarised in Table 3. These interface energies take the 17
deformation energy, i.e., the energy needed for the lattice adaptation in each individual orientation, into account. Despite this positive energy contribution, the negative values of the most total interface energies indicate that the formation of such interfaces is much more favourable than the separate state of the two bulk structures. The calculated interface energies also show
IP T
a clear preference of the orientation relationships observed experimentally (Figs. 3b and 6c).
Interfaces and orientations
Contact surface A [Å2]
Fully relaxed interface energy [J m-2]
(100)rutile || (110)titanate, [001]rutile || [11̅0]titanate [18]
126.35
0.14
(101)rutile || (012)corundum, [010]rutile || [100]corundum [19]
41.58
(001)rutile || (100)titanate, [010]rutile || [001]titanate *
32.12
(001)rutile || (100)corundum, [010]rutile || [001]corundum [17] *
105.43
SC R
-4.15 -4.95 -6.78
A
N
U
* experimentally observed in this study Table 3: The interface energies of the investigated interfaces with the indicated crystallographic orientations as obtained using DFT.
M
4. Discussion
ED
In simulated metal melt filtration experiments performed without melt convection, silicon present in the AlSi7Mg0.6 melt did not enter any chemical reaction with the rutile coating. On
PT
the contrary, aluminium, magnesium and oxygen contained in molten aluminium alloys were found in nearly all reaction products. Though, the reaction mechanisms and the products of
CC E
the reaction between molten aluminium alloys and rutile are strongly affected by the magnesium and oxygen concentrations.
A
In magnesium-free aluminium melts, aluminium and oxygen from the melt react to Al2O3, which forms a continuous layer heteroepitaxially grown at the surface of the rutile coating (Fig. 5). Aluminium melts without oxygen would reduce TiO2, while the released oxygen would react again with Al to α-Al2O3 at the TiO2 surface. Anyway, the corundum layer acts as a barrier against a further corrosion of TiO2 by liquid Al. The heteroepitaxial growth of αAl2O3 layers on the TiO2 surfaces is facilitated by a relatively small lattice misfit between the 18
counterparts (Fig. 6) and by the reduction of the total energy of the α-Al2O3/TiO2 interface in comparison with total energy of the individual bulk compounds (Table 3). At magnesium concentrations in the melt, which are comparable with the Mg concentrations in the AlSi7Mg0.6 alloy (0.6 mass%), Mg reacts quickly with oxygen from the melt and with
IP T
the TiO2 coating to MgTiO3 (Fig. 1). Because of a strong dependence of the diffusivity of Mg and O in rutile on the crystallographic direction, the interface between MgTiO3 and TiO2 is
heavily fringed (Fig. 1). If oxygen from the melt is consumed or if it is transmitted too slowly
SC R
through the MgTiO3 layer with fully occupied oxygen sublattice, Mg still diffuses into TiO2 and produces MgTiO3. The oxygen needed for this reaction is obtained from TiO2, which is
U
reduced to Ti2O3 (Eq. (3) and Fig. 4). These findings illustrate an extreme reactivity of Mg
N
with O and TiO2 and a fast diffusivity of Mg in TiO2, which inhibit the formation of corun-
A
dum at the surface of the TiO2 coatings in contact with the aluminium melts containing mag-
M
nesium.
ED
Very low concentrations of Mg in the ‘contaminated’ aluminium melt (or in the rutile coating) lead to a competition between the formation of MgTiO3 and α-Al2O3 at the surface of the
PT
TiO2 coatings, where the formation of corundum is accompanied by the formation of MgAl2O4. As Mg (and O) diffuse very quickly into TiO2, the TiO2 coating is ‘filled’ with
CC E
magnesium (and oxygen) before a protective corundum layer can form at the coating surface. If the local Mg content (near the interface to the MgTiO3/TiO2 stack) is depleted, the corundum layer starts to grow. As α-Al2O3 acts as a barrier for the Mg (and O) diffusion into TiO2,
A
the presence of the α-Al2O3 layer inhibits further formation of MgTiO3, even if the local concentration of Mg at the interface between the melt and the corundum layer is increased again. In such a case, magnesium and oxygen from the melt react with the aluminium oxide to MgAl2O4. However, it is worth noting that the occurrence and sequence of these processes strongly depend on the [Mg]/[O] ratio that is altered by magnesium impurities present in the 19
aluminium alloy and/or by sporadic MgO particles or lumps present in the sintered rutile. Furthermore, as MgAl2O4 was found in the transition zone between the original TiO2 coating and the α-Al2O3 layer only (Fig. 5), where it was in contact with TiO2 and MgTiO3 (Fig. 7), it can be assumed that the growth of MgAl2O4 is facilitated by its structural similarity to MgTiO3
IP T
(Fig. 8). Observed reaction processes can be utilised for the filtration of oxygen from the aluminiumbased melts. Because of a possible heteroepitaxy between TiO2 and α-Al2O3 and because of a
SC R
negative energy of the TiO2/α-Al2O3 interfaces having favourable orientation relationships
between the counterparts, the functional TiO2 coating serves as a suitable docking layer for
U
the formation of corundum stemming from O (and Al) present in the melt. On the other hand,
N
rutile is open for magnesium and oxygen diffusion, and forms MgTiO3 with these elements.
A
MgTiO3 serves as diffusion barrier for oxygen but not for magnesium, thus the rapid reaction
M
diffusion of Mg in TiO2 leads to the reduction of TiO2 to Ti2O3, because the oxygen is needed for formation of MgTiO3. Also at very low Mg concentrations in molten Al alloys, magnesi-
ED
um is removed from the melt and forms two phases containing magnesium, i.e., MgTiO3 and MgAl2O4. Consequently, the rutile coatings extract magnesium from the melt if they are used
PT
as functional materials on the surface of metal melt filters intended for the filtration of alu-
CC E
minium alloys containing magnesium. The formation of intermediate phases discussed above is facilitated by the heteroepitaxy between TiO2 and α-Al2O3, between TiO2, MgTiO3 and MgAl2O4, and between MgTiO3 and
A
Ti2O3, as well as by the negative energy of the TiO2/ α-Al2O3 and TiO2/MgTiO3 interfaces (including a positive deformation energy stemming from the small lattice misfit). The heteroepitaxy and the negative interface energies also guarantee a good adhesion of the reaction layers to the rutile coating. Therefore, no peeling off of the reaction layers is expected.
20
Conclusions Rutile exposed to aluminium alloys at 750 °C was found to react differently for different magnesium concentrations in the melt. The contact of TiO2 with molten aluminium leads to the epitaxial growth of a thin layer of thermodynamically stable corundum. The orientation relationship between TiO2 and α-Al2O3 was described as (001)rutile || (100)corundum and [100]rutile
IP T
|| [010]corundum. The α-Al2O3 layer accommodates oxygen solved in the aluminium melt and
protects the TiO2 coating from corrosion by molten aluminium. In an oxygen-free melt, the α-
SC R
Al2O3 layer would be built up from oxygen achieved from the TiO2 reduction. The epitaxial growth of α-Al2O3 on TiO2 and a negative energy of the α-Al2O3/TiO2 interface calculated
U
using the density functional theory promise a good adhesion of the corundum layer to the ru-
N
tile coating. The contact of TiO2 with molten AlSi7Mg0.6 leads to the formation of magnesi-
A
um titanate (MgTiO3) with a distinct orientation relationship to rutile, which was described as
M
(001)rutile || (100)titanate and [010]rutile || [001]titanate. Because of the high diffusivity of magnesium in TiO2, the TiO2 coatings are heavily altered and react to MgTiO3 and Ti2O3. The latter
ED
compound is a product of the TiO2 reduction, which is needed to gain oxygen for the MgTiO3 production. Also in this case, the structural similarity of present phases and their possible epi-
CC E
PT
taxy favour the growth of interface layers and support the adhesion of the counterparts.
Acknowledgement
A
This work was financially supported by the German Research Foundation (DFG) in the frame of the Collaborative Research Centre SFB 920. The authors would like to thank Dr. D. Heger (Institute of Materials Science, TU Bergakademie Freiberg) for the EPMA analyses and Dr. A. Plessow (Institute of Mineralogy, TU Bergakademie Freiberg) for the XRF analysis.
21
References
A
CC E
PT
ED
M
A
N
U
SC R
IP T
[1] C. Voigt, E. Jäckel, F. Taina, T. Zienert, A. Salomon, G. Wolf, C. Aneziris, P. LeBrun, Filtration Efficiency of Functionalized Ceramic Foam Filters for Aluminum Melt Filtration, Metall. Mater. Trans. B 48 (2016) 497-505. [2] H. Hashimoto, S. Kojima, T. Sasaki, H. Asoh, α-Alumina membrane having a hierarchical structure of straight macropores and mesopores inside the pore wall, J. Eur. Ceram. Soc. 38 (2018) 1836-1840. [3] C. Aneziris, S. Dudczig, M. Emmel, H. Berek, G. Schmidt, J. Hubalkova, Reactive filters for steel melt filtration, Adv. Eng. Mat. 15 (2013) 46-59. [4] A. Salomon, C. Voigt, O. Fabrichnaya, C. G. Aneziris, D. Rafaja, Formation of Corundum, Magnesium Titanate, and Titanium(III) Oxide at the Interface between Rutile and Molten Al or AlSi7Mg0.6 Alloy, Adv. Eng. Mat. 19 (2017) 1700106. [5] M. Ibarra Castro, J. Almanza Robles, D. Cortés Hernandez, J. Escobedo Bocardo, R. MartinezLopez, Chemical interaction between SrAl2Si2O8 and molten aluminum, J. Eur. Ceram. Soc. 35 (2015) 4287-4292. [6] W. Yan, A. Schmidt, S. Dudczig, T. Wetzig, Y. Wei, Y. Li, S. Schafföner, C. Aneziris, Wettability phenomena of molten steel in contact with alumina substrates with alumina and alumina-carbon coatings, J. Eur. Ceram. Soc. 38 (2018) 2164-2178. [7] C. Voigt, T. Zienert, P. Schubert, C. Aneziris, J. Hubalkova, Reticulated porous foam ceramics with different surface chemistries, J. Am. Ceram. Soc. 97 (2014) 2046-2053. [8] A. Salomon, T. Zienert, C. Voigt, E. Jäckel, O. Fabrichnaya, D. Rafaja, C. Aneziris, Comparison of interfacial reactions between AlSi7Mg and alumina filter after casting and spark plasma sintering, Adv. Eng. Mat. 15 (2013) 1206-1215. [9] J.-O. Andersson, T. Helander, L. Höglund, P. Shi, B. Sundmann, Thermo-Calc & DICTRA, computational tools for materials science, Calphad 26 (2002) 273-312. [10] A.T. Dinsdale, SGTE data for pure elements, Calphad 15 (1991) 317-425. [11] T. Zienert, O. Fabrichnaya, Thermodynamic Assessment and Experiments in the system MgOAl2O3, Calphad 40 (2013) 1-9. [12] M. Ilatovskaia, G. Savinykh, O. Fabrichnaya, Thermodynamic Description of the Ti-Al-O System Based on Experimental Data, J. Phase Equilib. Diff. 38 (2017) 175-184. [13] M. Ilatovskaia, I. Saenko, G. Savinykh, O. Fabrichnaya, Experimental study of phase equilibria in the Al2O3-MgO-TiO2 system and thermodynamic assessment of the MgO-TiO2 system, J. Am. Ceram. Soc. (2018), https://doi.org/10.1111/jace.15748. [14] SGTE solution database, 2015SGSOL. http://www.sgte.net, 2018 (accessed 20 June 2018). [15] P. Giannozzi, S. Baroni, N. Bonini, M. Calandra, R. Car, C. Cavazzoni, D. Ceresoli, G.L. Chiarotti, M. Cococcioni, I. Dabo, A. Dal Corso, S. De Gironcoli, S. Fabris, G. Fratesi, R. Gebauer, U. Gerstmann, C. Gougoussis, A. Kokalj, M. Lazzeri, L. Martin-Samos, N. Marzari, F. Mauri, R. Mazzarello, S. Paolini, A. Pasquarello, L. Paulatto, C. Sbraccia, S. Scandolo, G. Sclauzero, A.P. Seitsonen, A. Smogunov, P. Umari, R.M. Wentzcovitch, QUANTUM ESPRESSO: A modular and open-source software project for quantum simulations of materials, J. Phys. Condens. Matter 21 (2009) 395502-119. [16] J. Perdew, K. Burke, M. Ernzerhof, Generalized gradient approximation made simple, Phys. Rev. Lett. 77 (1996) 3865-3868. [17] C. Chen, A. Korotcov, Y.-S. Huang, W. Chung, D. Tsai, K. Tiong, Growth and characterization of well-aligned rutile TiO2 nanocrystals on sapphire substrates via metal organic vapour deposition, J. Mat. Sci.: Mat. Elect. 20 (2009) S332-S335. [18] A. Lotnyk, S. H. D. Senz, Orientation relationships of SrTiO3 and MgTiO3 thin films grown by vapor-solid reactions on (100) and (110) TiO2 (Rutile) single crystals, J. Phys. Chem. C 111 (2007) 6372-6379. [19] R. Engel-Herbert, B. Jalan, J. Cagnon, S. Stemmer, Microstructure of epitaxial rutile TiO2 films grown by molecular beam epitaxy on r-plane Al2O3, J. Cryst. Growth 312 (2009) 149-153. [20] K. Hoshino, N. Peterson, C. Wiley, Diffusion and point defects in TiO2-x, J. Phys. Chem. Solids 46 (1985) 1397-1411. [21] J. Sasaki, N. Peterson, K. Hoshino, Tracer impurity diffusion in single-crystal rutile (TiO2-x), J. Phys. Chem. Solids 46 (1985) 1267-1283. 22
A
CC E
PT
ED
M
A
N
U
SC R
IP T
[22] L. Jeurgens, W. Sloof, F. Tichelaar, E. Mittemeijer, Growth kinetics and mechanisms of aluminum-oxide films formed by thermal oxidation of aluminium, J. Appl. Phys. 92 (2002) 1649-1656. [23] G. Xie, O. Ohashi, T. Sato, N. Yamaguchi, M. Song, K. Mitsuishi, K. Furuya, Effect of Mg on the sintering of Al-Mg alloy powders by pulse electric-current sintering process, Mater. Trans. 45 (2002) 904-909. [24] G. Xie, O. Ohashi, N. Yamaguchi, M. Song, K. Mitsuishi, K. Furuya, T. Noda, Reduction of surface oxide films in Al-Mg alloy powders by pulse electric current sintering, J. Mater. Res. 19 (2004) 815-819. [25] T. Armbruster, On the origin of sagenites: structural coherency of rutile with hematite and spinel structures, N. Jb. Miner. Mh. 7 (1981) 328-334. [26] E. Force, R. Richards, K. Scott, P. Valentine, N. Fishman, Mineral intergrowths replaced by "elbow-twinned" rutile in altered rocks, Can. Mineral. 34 (1996) 605-614. [27] A. Recnik, N. Stankovic, N. Daneu, Topotaxial reactions during the genesis of oriented rutile/hematite intergrowths from Mwinilunga (Zambia), Contrib. Mineral. Petr. 169 (2015) 19. [28] F. Wong, S. Ramanathan, Heteroepitaxy of distorted rutile-structure WO2 and NbO2 thin films, J. Mater. Res. 28 (2013) 2555-2563. [29] N. Daneu, A. Recnik, W. Mader, Atomic structure and formation mechanism of (101) rutile twins from Diamantina (Brazil), Am. Mineral. 99 (2014) 612-624. [30] D. Rafaja, T. Schucknecht, V. Klemm, A. Paul, H. Berek, Microstructural characterisation of titanium coatings deposited using cold gas spraying on Al2O3 substrates, Surf. Coat. Tech. 203 (2009) 3206-3213. [31] J. Kutzner, D. Rafaja, J. Kortus, On the possibility of a hetero-epitaxial Fe/ZrO2 interface: A density-functional-theory point of view, Model. Simul. Mater. Sci. Eng. 20 (2012) 045011. [32] R. Drehmann, T. Grund, T. Lampke, B. Wielage, K. Manygoats, T. Schucknecht, D. Rafaja, Splat formation and adhesion mechanisms of cold gas-sprayed al coatings on Al2O3 substrates, J. Therm. Spray Techn. 23 (2014) 68-75. [33] B. Matthey, S. Höhn, A.-K. Wolfrum, U. Mühle, M. Motylenko, D. Rafaja, A. Michaelis, M. Herrmann, Microstructural investigation of diamond-SiC composites produced by pressureless silicon infiltration, J. Eur. Ceram. Soc. 37 (2017) 1917-1928. [34] C. Wüstefeld, D. Rafaja, M. Motylenko, C. Ullrich, R. Drehmann, T. Grund, T. Lampke, B. Wielage, Local heteroepitaxy as an adhesion mechanism in aluminium coatings cold gas sprayed on AlN substrates, Acta Mater. 128 (2017) 418-427. [35] A. Sheikh, J. Irvine, Phase Formation and Electronic Transport Properties in the Corundum (Ti2O3)-Ilmenite (MgTiO3) System, J. Solid State Chem. 103 (1993) 30-37. [36] P. Robinson, F. Langenhorst, S. McEnroe, K. Fabian, T. Ballaran, Ferroan geikielite and coupled spinel-rutile exsolution from titanohematite: Interface characterization and magnetic properties, Am. Mineral. 99 (2014) 1694-1712. [37] A. Heuer, Oxygen and aluminum diffusion in alpha-Al2O3: How much do we really understand?, J. Eur. Ceram. Soc. 28 (2008) 1495-1507. [38] K. Momma, F. Izumi, VESTA 3 for three-dimensional visualization of crystal, volumetric and morphology data, J. Appl. Crystallogr. 44 (2011) 1272-1276.
23
Figure captions
Figure 1:
A
CC E
PT
ED
M
A
N
U
SC R
IP T
SEM micrographs (BSE contrast) of rutile coatings, which were brought in contact with molten AlSi7Mg0.6 alloy for 60 min (a) and 300 min (b) at 750 °C. At the interface between the coating and the liquid alloy, a MgTiO3 layer formed.
24
Figure 2:
A
CC E
PT
ED
M
A
N
U
SC R
IP T
Results of the EBSD analysis performed on the rutile coating from Fig. 1a. (a) Phase map showing the distribution of rutile (yellow) and MgTiO3 (blue). Pores/voids are reproduced in black, the grey region at the bottom is the solidified alloy AlSi7Mg0.6. The green lines mark the interfaces between the TiO2 and MgTiO3 grains having the orientation relationship (001)rutile || (100)titanate and [010]rutile || [001]titanate. A histogram of the local deviations from this orientation relationship is shown in (b). The above orientation relationship is illustrated in figure (c).
25
Figure 3:
A
CC E
PT
ED
M
A
N
U
SC R
IP T
(a) A TEM micrograph of the MgTiO3/TiO2 interface. A phase boundary with geometrically necessary misfit dislocations is marked by white arrows. The orientation relationship (001) rutile || (100)titanate and [010]rutile || [001]titanate shown in panel (b) was verified by SAED. Crystal structures were created with the VESTA 3 software [38].
26
Figure 4:
A
CC E
PT
ED
M
A
N
U
SC R
IP T
TEM and SAED of the MgTiO3/Ti2O3 interfaces. The growth direction is perpendicular to the plane of the image.
27
Figure 5:
A
CC E
PT
ED
M
A
N
U
SC R
IP T
SEM micrographs (BSE contrast) of rutile coatings that were in contact with molten Al (at 750 °C) for 60 min (a) and 300 min (b). The main reaction layer contains corundum (α-Al2O3). The dotted box in (b) shows the position of a FIB lamella, which was investigated by TEM and SAED in order to explain the nature of the stripes marked by black arrows.
28
Figure 6:
A
CC E
PT
ED
M
A
N
U
SC R
IP T
(a) EBSD phase map of the TiO2/Al interface from Fig. 5a. Rutile is plotted in yellow, corundum in red. Black areas within the colourised region are pores, non-indexed bottom region is solidified aluminium. The green lines mark the interfaces between rutile and corundum crystallites having the orientation relationship (001)rutile || (100)corundum and [010]rutile || [001]corundum. A histogram of the local deviations from this orientation relationship is shown in (b). (c) Model of rutile and corundum in the above orientation relationship plotted using VESTA 3 [38].
29
Figure 7:
A
CC E
PT
ED
M
A
N
U
SC R
IP T
Element maps (a) and TEM micrograph (b) of the stripes from Fig. 5b. Individual phases were assigned using a combination of chemical analysis (EDX) and SAED. The orientation relationships are summarised in the text. The viewing direction is perpendicular to the sample surface and corresponds to the growth direction of corundum into rutile.
30
Figure 8:
A
CC E
PT
ED
M
A
N
U
SC R
IP T
Crystal structure models of TiO2, MgAl2O4 and MgTiO3 mutually oriented according to the orientation relationships that were identified using SAED (cf. Fig. 7). The parallel planes highlighted in green mark close-packed oxygen sublattice planes in each structure (highly distorted in rutile), arrows mark the corresponding parallel directions. The crystal structures were created with VESTA 3 [38].
31
Figure 9:
A
CC E
PT
ED
M
A
N
U
SC R
IP T
Orientation relationships found in literature (see Refs. [18] and [19] for (a) and (b), respectively) and used for comparative DFT calculations of interface energies. Parallel lattice planes are highlighted in green and arrows mark the corresponding parallel directions. Only atoms lying in or behind the parallel planes are depicted, cf. figures 3b, 6c and 8. The crystal structures were created with VESTA 3 [38].
32
S0955-2219(18)30473-4 https://doi.org/10.1016/j.jeurceramsoc.2018.07.052 JECS 12020
To appear in:
Journal of the European Ceramic Society
Received date: Revised date: Accepted date:
20-6-2018 26-7-2018 27-7-2018
Please cite this article as: Salomon A, Amirkhanyan L, Ullrich C, Motylenko M, Fabrichnaya O, Kortus J, Rafaja D, Interface reactions between rutile coatings and molten aluminium or AlSi7Mg0.6 alloy, Journal of the European Ceramic Society (2018), https://doi.org/10.1016/j.jeurceramsoc.2018.07.052 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Interface reactions between rutile coatings and molten aluminium or AlSi7Mg0.6 alloy
Anton Salomona,*, Lilit Amirkhanyanb, Christiane Ullricha, Mykhaylo Motylenkoa, Olga
a
IP T
Fabrichnayaa, Jens Kortusb and David Rafajaa
Institute of Materials Science, TU Bergakademie Freiberg, Gustav-Zeuner-Straße 5, 09599 Freiberg, Germany
Institute of Theoretical Physics, TU Bergakademie Freiberg, Leipziger Str. 23, 09599
SC R
b
Freiberg, Germany
U
* Corresponding author:
N
Email: [email protected] Tel.: +493731/39-2671
M
ED
homepage: www.ww.tu-freiberg.de
A
Fax: +493731/39-2604
Abstract
PT
Rutile coatings deposited on corundum substrates are considered as promising functional ele-
CC E
ments improving the efficiency of the filtration of oxide inclusions out of aluminium melts. This contribution describes the reactions between rutile and two kinds of the aluminium melts and discusses the consequences of these reactions for the filtration process. It was found that
A
the contact of rutile coatings with molten aluminium leads to the formation of a corundum layer at the solid/liquid interface. The exposure of the rutile coatings to molten AlSi7Mg0.6 alloy produces an interface layer of MgTiO3. The interface layers possess defined orientation relationship to rutile which is characteristic for locally heteroepitaxial growth. The density functional theory calculations revealed that the TiO2/α-Al2O3 and TiO2/MgTiO3 interfaces with the orientation relationships observed experimentally have low interface energies. The
mechanisms of the interface layer formation and the impact of these layers on the degradation of the rutile coatings are discussed.
Keywords: Filter ceramics; Rutile coating; Reactive diffusion; Local heteroepitaxy; Interface
SC R
IP T
energy
1. Introduction
Ceramic filters containing corundum are traditionally used for filtration of metallic melts [1].
U
In order to improve their filtration efficiency, different approaches are pursued. One of them
N
is the development of hierarchical structures [2], another one is the functionalization of the
A
corundum surface by ceramics coatings that attract and anchor small inclusions and bond ox-
M
ygen solved in the melt [1]. Concurrently, the functional coatings should not have a negative
ED
impact on the melt flow rate and should not substantially corrode [3]. Furthermore, the functional coatings and the products of the reactions between the metallic melt and the functional
PT
coatings must not peel off and pollute the melt.
CC E
Some time ago, it was shown that rutile-coated ceramic foam filters provide high flow rates for AlSi7Mg melts and that they possess moderate filtration efficiency for different inclusion types [1]. Both properties were attributed to a specific wetting behaviour of the aluminium
A
alloy melt on the rutile surface during the filtration. The wetting and the filtration efficiency are typically affected by chemical reactions between the filter surface and the metallic melt [5, 6]. Recently, it was shown that rutile (tetragonal TiO2 crystallising in the space group P42/mnm) reacts with molten aluminium and aluminium alloys [4], and that it can be employed for the functionalisation of reactive filters [3].
2
Rutile coatings brought in contact with pure aluminium were found to form primarily a corundum layer (α-Al2O3, space group R3̅c) at the interface to the Al melt [4]. At long operation times, Ti2O3 (SG R3̅c) and Al3Ti (SG I4/mmm) formed additionally as minor phases. In the case of the aluminium alloy AlSi7Mg0.6, a layer of MgTiO3 (SG R3̅) grew predominantly on the surface of the TiO2 coatings. Additional phases occurring after long operation times were
IP T
α-Al2O3 (SG R3̅c) and (Al,Si)3Ti (SG I4/mmm).
SC R
The formation of α-Al2O3 or MgTiO3 as principal phases confirmed the aspired reactive be-
haviour of the rutile coatings, i.e., the capability of reducing the oxygen content in the aluminium melt or molten AlSi7Mg0.6 [4]. Additionally, the growth of the α-Al2O3 or MgTiO3 lay-
U
ers delayed the formation of Al3Ti or (Al,Si)3Ti, which could precipitate in the aluminium or
N
AlSi7Mg0.6 melt and act as unwanted inclusions. However, the α-Al2O3 and MgTiO3 layers
A
can peel off from the surface of the functionalised metal melt filters, which would contami-
M
nate the filtered melt as well. In order to understand the binding mechanisms and to improve the adhesion between the original rutile coating and the respective oxide layer, the reaction
ED
between rutile and molten pure Al or AlSi7Mg0.6 alloy and the growth of α-Al2O3 and
PT
MgTiO3 on TiO2 were investigated systematically. The central phenomena discussed in this study are the formation of reaction products at the
CC E
surface of functional rutile coatings brought in contact with Al or AlSi7Mg0.6 melt, their adhesion to the respective substrate and their ability to inhibit further reactions. In this context, a
A
heteroepitaxy between rutile and the corrosion products is believed to play a very important role, as the heteroepitaxial growth is considered both as a feature improving the adhesion of the corrosion products and as a factor influencing the reaction kinetics via lattice strain at the interface.
3
The reaction experiments were performed in a Spark Plasma Sintering (SPS) device to imitate the heat treatment conditions during standard aluminium casting but without the flow of the melt. Possible heteroepitaxy between TiO2 and the reaction products was concluded from specific orientation relationships between the adjacent phases. The orientation relationships were obtained from the electron backscatter diffraction (EBSD) patterns measured in a scanning
IP T
electron microscope and from the selected area electron diffraction (SAED) patterns measured
in a transmission electron microscope. For selected orientation relationships found experimen-
SC R
tally, specific interfacial energies and formation probabilities were determined using the Density Functional Theory (DFT) calculations. Calculated Gibbs energies of the chemical reac-
N
U
tions identified experimentally are negative and shown for comparison.
M
A
2. Materials and methods 2.1 Sample preparation
ED
The samples under study consisted of corundum substrates coated by rutile that were brought
PT
in contact with molten aluminium or molten aluminium alloy AlSi7Mg0.6 in an SPS device. Both, Al and AlSi7Mg0.6, were supplied in powder form. For metal melting, a special sample
CC E
setup was applied that consisted of a hollow corundum cylinder, whose interior was coated with rutile and filled with Al or AlSi7Mg0.6 powder. The corundum cylinders were prepared from Martoxid MR70 (high-purity ground alumina; Albemarle Corporation, USA) by slip
A
casting and subsequent conventional sintering at 1600 °C according to the procedure described in references [4] and [7], and had an outer diameter of 20 mm, an inner diameter of 15 mm and a length of 10 mm. The rutile coatings were applied via slip casting and conventionally sintered at 1300 °C [7]. This technique is commonly used for rutile deposition in industry. Its drawback is a thickness variation. In our case, the thickness of the rutile coating 4
varied between 10 to 25 µm. However, the thickness variation did not affect the results of this study. As starting material for the rutile deposition, powder R320 from Sachtleben, Germany (98.5 mass% TiO2, mean particle size 0.55 µm) was used. The powder particles of pure Al powder (99.9 %, Goodfellow, Germany) had a maximum size of 60 μm, the AlSi7Mg0.6 al-
IP T
loy powder particles (TLS Technik, Germany) had the size between 45 and 100 μm. The main impurities in the starting materials are summarised in Table 1. The chemical compositions of
SC R
the starting materials were provided by the respective manufacturer. The impurities in rutile
U
were identified and quantified by X-ray fluorescence analysis (XRF).
Source
Zn
Particle size [µm]
-
-
< 60
Goodfellow
0.02
0.28
45 – 100
TLS Technik
0.55 (d0.5)
Sachtleben
Al
Si
Mg
Fe
Cu
Pure Al
balance
0.04
0.0004
0.03
A
N
Elemental composition [mass%] Powder
AlSi7Mg0.6
balance
7.07
0.61
M
0.003
0.09
0.09
Mn
> 98.5 mass% of rutile * Rutile TiO2 R320
CC E
PT
ED
* XRF: 0.4 mass% Al2O3, 0.3 mass% of Fe2O3 and Cr2O3, 0.2 mass% of P2O5 and K2O, MgO < 0.02 mass% Table 1: Chemical compositions and particle sizes of the aluminium powder, aluminium alloy powder and rutile powder used for the sample production as specified by manufacturers. Secondary oxides in rutile were analysed additionally using XRF.
The functionalised cylinders (corundum with rutile coating) were filled with Al or AlSi7Mg0.6 powder and heated to 750 °C in a SPS unit HPD 25 (FCT Systeme GmbH, Ger-
A
many). The working atmosphere was residual air with a pressure between 0.5 and 5 Pa. The power-controlled heating resulted in heating rates of about 1500 K∙min-1. The dwell times at 750 °C were 1, 30, 60 and 300 minutes. After the heating was switched off, the samples were rapidly cooled by the water-cooled steel electrodes of the SPS apparatus. In this way, a cool-
5
ing rate of about 300 K∙min-1 was achieved. The sample set-up, the tooling geometries and further detailed information about the SPS heat treatment are given in reference [8].
2.2 Methods of sample characterisation
IP T
On the microscopic scale, the products of the reaction between the respective melt and the TiO2 coatings were analysed using scanning electron microscopy with energy dispersive X-
SC R
ray spectroscopy (SEM/EDX) and electron backscatter diffraction (SEM/EBSD), and using electron probe microanalysis with wavelength dispersive X-ray spectroscopy (EPMA/WDX).
U
For SEM/EDX/EBSD, a high-resolution SEM LEO-1530 (Carl Zeiss AG, Germany) with
N
field-emission cathode, an EDX detector (Bruker AXS) and a Nordlys II EBSD detector
A
(HKL Technology) was used. All experiments were carried out at an acceleration voltage of
M
20 kV. The working distance for EBSD was 15 mm, the tilting angle 70 ° and the step size 0.3 µm. The software package Channel 5 (HKL Technology) was employed for identification
ED
of the Kikuchi patterns and for evaluation of the measured data. An electron probe microanalyser JXA8900 RL (Jeol GmbH, Germany) with five crystal spectrometers was employed
PT
for EPMA/WDX.
CC E
On the nanoscale, the samples were characterised using transmission electron microscopy (TEM), selected area electron diffraction (SAED) and energy dispersive X-ray spectroscopy (EDX). The TEM analyses were done in a JEM 2200 FS transmission electron microscope
A
(JEOL, Japan) at an acceleration voltage of 200 kV. The TEM samples were prepared by the focussed ion beam method (FIB) with a Helios NanoLab 600i (FEI, USA) in form of thin slices.
2.3 Calculation of the Gibbs and interface energy 6
The Gibbs energies (ΔG°) of the chemical reactions found experimentally were calculated with the Thermo-calc software [9] using the SGTE unary database [10] combined with recently assessed data for the systems MgO-Al2O3 [11], Al-Ti-O [12] and Al2O3-MgO-TiO2 [13] were included. The activity of molten Mg in molten Al alloy was calculated using the SGTE
750 °C and at ambient pressure. No gaseous phase was involved.
IP T
solution database [14]. It was assumed that the reactions run at a constant temperature of
The interface energy (σ) has been calculated as a difference between the total energy of an
𝐸𝑠𝑙𝑎𝑏 − 𝐸𝑏𝑢𝑙𝑘1 −𝐸𝑏𝑢𝑙𝑘2 2𝐴
(1)
U
𝜎=
SC R
interface slab and the total energies of the corresponding bulk materials forming the interface:
N
In equation (1), A represents the interface area, and Eslab, Ebulk1 and Ebulk2 are the total energies
A
of a slab including the interface and the bulk structures, respectively. The total energies were
M
obtained from the DFT calculation as implemented in the Quantum ESPRESSO code [15]. The projector augmented wave (PAW) method was used together with the generalised gradi-
ED
ent approximation (GGA) for exchange correlation functional according to the Perdew, Burke
PT
& Ernzerhof (PBE) approach [16].
For the calculation of the total energies of the bulk materials, the crystallographic models of
CC E
rutile (ICSD #66650), corundum (ICSD #73725) and MgTiO3 (ICSD #156217) were used. The interface slabs have been modelled using several lattice planes of the corresponding material, i.e., rutile and corundum or rutile and magnesium titanate, by considering the crystallo-
A
graphic orientation relationships obtained from our experiments and/or reported in literature [17] - [19]. In the starting slabs employed for the relaxation procedure, the lattice misfit between the counterparts was avoided by adjusting the in-plane lattice parameters of TiO2 (lattice parameters along the interface to α-Al2O3 or MgTiO3). The in-plane lattice parameters of α-Al2O3 or MgTiO3 were not changed, because both compounds have larger bulk modulus 7
than TiO2. However, the expansion of the in-plane TiO2 lattice parameter was smaller than 1 % in all cases. In this way, “supercells” of the space group P1 containing several formula units of TiO2 and either Al2O3 or MgTiO3 were created as the starting structures for the relaxation. In the next step, the individual supercells were fully optimised by relaxing the cell pa-
IP T
rameters and the atomic positions. The slab applied for description of the TiO2/MgTiO3 interface with the orientation relation-
ship (100)rutile || (110)titanate and [001]rutile || [11̅0]titanate [18] contained 159 atoms (18 formula
SC R
units of TiO2 and 21 formula units of MgTiO3). The interface between TiO2 and -Al2O3 with the orientation relationship (101)rutile || (012)corundum and [010]rutile || [100]corundum [19] was
U
modelled using 78 atoms (6 TiO2 and 12 Al2O3). The model of the TiO2/MgTiO3 interface
N
with the orientation relationship (001)rutile || (100)titanate and [100]rutile || [010]titanate interface
A
contained 45 atoms in the slab (5 TiO2 and 6 MgTiO3). Finally, the interface slab TiO2/-
M
Al2O3 with the orientation relationship (001)rutile || (100)corundum and [100]rutile || [010]corundum
Interfaces
k points
Ecut [Ry]
Ecut_rho [Ry]
(100)rutile || (110)titanate, [001]rutile || [11̅0]titanate [18]
3×3×3
80
320
(101)rutile || (012)corundum, [010]rutile || [100]corundum [19]
8×2×9
70
280
(001)rutile || (100)titanate, [010]rutile || [001]titanate *
12×3×14
95
380
(001)rutile || (100)corundum, [010]rutile || [001]corundum [17] *
8×2×9
60
240
CC E
PT
ED
[17] comprised 141 atoms (17 TiO2 and 18 Al2O3).
A
* experimentally observed in this study Table 2: Orientation relationships between the counterparts, number of the k points, the cut-off energy and the cut-off of kinetic energy for charge density and potential calculation used for the interface energy calculation.
The computational parameters used for individual interfaces are summarised in Table 2. The valence electron wave functions are expanded in a plane wave basis with a cut-off energy 8
(Ecut). The effective potential and the electron density are given in a numerical basis with Ecut_rho. The criterion to terminate the geometry relaxation of the interface slabs relaxing all the internal coordinates has been that all force components on the atoms were lower than
IP T
0.001 Ry Å-2.
3. Results
SC R
3.1 Rutile in contact with molten AlSi7Mg0.6
The reaction between the rutile coatings and the molten AlSi7Mg0.6 alloy at 750 °C leads to
U
the formation of MgTiO3 (geikielite, SG R3̅). The MgTiO3/TiO2 interface is strongly fringed
N
(Fig. 1), which results in a large variation of the MgTiO3 layer thickness. For the reaction time
A
of 60 min, the MgTiO3 layer thickness ranges between 2.0 µm and 11.9 µm (Fig. 1a), for the
M
reaction time of 300 min between 2.5 µm and 19.7 µm (Fig. 1b). These large variations of the MgTiO3 layer thickness are caused primarily by the needle-like or wedge-like shape of the
ED
MgTiO3 grains. Although MgTiO3 is not stable at the 750 °C (cf. Ref. [4]), its stoichiometry
PT
was confirmed by the EPMA/WDX analysis and its crystal structure by local diffraction methods (EBSD/SEM and SAED/TEM).
CC E
The EBSD measurements performed in the reaction zone containing rutile, the newly formed MgTiO3 layer and the solidified AlSi7Mg0.6 alloy (Fig. 2a) revealed that TiO2 and MgTiO3
A
possess a well-defined orientation relationship, which is very close to (001)rutile || (100)titanate and [010]rutile || [001]titanate. The average deviation of the measured orientation relationships from the ideal one was approximately 2.5 ° (Fig. 2b). The ideal orientation relationship between TiO2 and MgTiO3 was found in the small, stripe- or needle-like, dark grey MgTiO3 grains that penetrate into the TiO2 matrix. The MgTiO3 needles grew always anomalously in the [001]titanate direction, which was parallel to the [010] direction in rutile. 9
The orientation relationship close to (001)rutile || (100)titanate and [010]rutile || [001]titanate was also confirmed by the SAED patterns that were taken near the TiO2/MgTiO3 interface (Fig. 3a). The formation of MgTiO3 with this orientation relationship to TiO2 can be seen as reactive diffusion of MgO into TiO2. The anomalous growth of the MgTiO3 needles along the [001] direction (Fig. 2c) indicates that the diffusion takes place predominantly along the directions
IP T
[001]titanate and [010]rutile (Fig. 3b). From the point of view of the crystal symmetry, the pre-
ferred diffusion directions must lie in the (001)rutile plane. A faster diffusion within the
SC R
(001)rutile planes has already been described in references [20] and [21], where it was explained by the diffusion of cations over the interstitial positions (see also Section 3.3).
U
The chemical reaction behind the formation of MgTiO3 can be described as 1
TiO2 + [Mg] + 2 O2 → MgTiO3
(2)
A
N
(Δ°G = -512.5 kJ mol-1)
M
The source of magnesium and oxygen is the aluminium alloy. Magnesium is one of the alloying element in AlSi7Mg0.6. Even if the positive contribution (47.9 kJ mol-1) of the activity of
ED
Mg in the liquid Al alloy is taken into account, ΔG is still negative making it possible for
PT
above reaction to occur. Oxygen is originally adsorbed on the surface of the aluminium alloy powder particles and contained in the form of amorphous oxide skin [22] covering the powder
CC E
particles used for SPS melting. As the amorphous alumina films are less stable than the crystalline Al2O3 (corundum), they are decomposed during the SPS process, in particular if Mg is
A
present as alloying element in the aluminium melt (see Ref. [23] and [24]). In reaction (2), MgO is a possible intermediate phase. However, it was not found in our samples. Anyway, this reaction model implies that the amount of magnesium and oxygen in the AlSi7Mg0.6 melt is reduced by the contact of the melt with the functional TiO2 coating. This phenomenon was confirmed by the local chemical analysis of the solidified aluminium alloy.
10
In the proximity of the reaction zone, the amount of magnesium in the solidified melt was reduced below the detection limit of WDX. Possible habitus planes in Figure 3a are (311)rutile and (011)titanate. The lattice misfit that was calculated for the oxygen atoms located in these habitus planes using the lattice parameters arutile = 0.4653 nm, crutile = 0.2965 nm, atitanate = 0.5057 nm and ctitanate = 1.3903 nm (ICSD
IP T
#66650 and #156217) was smallest along the [1̅03]rutile || [100]titanate directions (0.24 %) but
SC R
significantly larger along the [1̅21]rutile || [11̅1]titanate directions (zone axes in Fig. 3a), where it
was about 10 %). The reference oxygen atoms used for the lattice misfit calculation had the Wyckoff positions 4f and 18f in TiO2 and MgTiO3, respectively. The lattice misfit between
U
TiO2 and MgTiO3 is compensated by geometrically necessary dislocations with almost equi-
N
distant spacing, which were found at the interface between TiO2 and MgTiO3 (Fig. 3a). The
A
dislocation ordering causes the slight deviation from the ideal orientation relationship
M
(001)rutile || (100)titanate and [010]rutile || [001]titanate discussed above. The inclination between the areas 1 and 2 in Figure 3a is about 1 – 2 ° and agrees well with the deviations found using
ED
EBSD (Fig. 2b).
PT
The mutual orientation (001)rutile || (100)titanate and [010]rutile || [001]titanate has not been explicitly reported in literature so far. However, the orientation relationships between rutile and phas-
CC E
es with ilmenite structure (SG R3̅) that can be found in literature, e.g., (010)rutile || (001)ilmenite and [101]rutile || [210]ilmenite [25] - [27], (101)rutile || (110)ilmenite and [100]rutile || [010]ilmenite [25],
A
[28] [29], or (301)rutile || (11̅0)ilmenite and [010]rutile || [001]ilmenite [25], [29], are strongly related to the observed one, as they result in similar atomic arrangements. These orientation relationships can hardly be distinguished from each other, because they are mutually inclined by less than 3 ° [26]. This phenomenon can also contribute to the mean deviation from the perfect orientation relationship (001)rutile || (100)titanate and [100]rutile || [010]titanate, which was observed by EBSD (cf. Fig. 2b). 11
As the growth of the MgTiO3 reaction layer changes the chemical composition of the AlSi7Mg0.6 melt, a deceleration of the reaction diffusion process is expected for longer diffusion times, in particular if the amount of Mg and O in the melt is limited and/or if the reaction diffusion process runs without melt convection. In the TiO2 coating that was in contact with the AlSi7Mg0.6 melt for 300 minutes at 750 °C, adjacent tiny stripes of MgTiO3 and Ti2O3
IP T
were found by transmission electron microscopy (Fig. 4). The chemical composition of Ti2O3 was proven by EDX in STEM, its crystal structure (SG R3̅c, corundum type) by SAED. Addi-
SC R
tionally, SAED revealed that MgTiO3 and Ti2O3 grow with the same crystal orientation (Fig. 4). In analogy to other materials combinations [30] - [34], the heteroepitaxial growth of
U
MgTiO3 on TiO2 and Ti2O3 is believed to improve the adhesion of the MgTiO3 layer to the
N
respective substrate and to protect the delamination of the reaction layer during the metal melt
A
filtration.
M
The presence of Ti2O3 instead of TiO2 implies that the oxygen needed for the formation of
ED
MgTiO3 is obtained from the reduction of TiO2 to Ti2O3: 3 TiO2 + [Mg] → Ti2 O3 + MgTiO3 ,
(ΔG° = -238.4 kJ mol-1)
(3)
PT
when oxygen contained in the aluminium alloy is spent or if the diffusion of Mg into TiO2 is
CC E
faster than the oxygen diffusion. The contribution of the Mg activity in the Al alloy is the same as mentioned above. Some deviations from the stoichiometric composition of the reaction products cannot be excluded, because MgTiO3 and Ti2O3 may form a solid solution [35].
A
Still, both (separated) phases were found in the TEM micrograph (Fig. 4) and confirmed by the TEM/EDX line scans. This change in the reaction mechanism, which is caused by the consumption of oxygen from the melt or by a slower oxygen diffusion in comparison with the magnesium diffusion, slows down the MgTiO3 formation significantly, as it can be seen on
12
the lower growth rate of the corrosion layer containing Ti2O3. For more details, see Section 3.3.
3.2 Rutile in contact with pure molten aluminium
IP T
In order to illustrate the effect of the magnesium concentration on the reaction between the
functional TiO2 coating and the aluminium alloy melt, analogous experiments as above were
SC R
performed in a system containing a very low amount of Mg impurities in the Al powder and
in the TiO2 powder used for production of the functionalised coatings. At negligible Mg con-
U
tents, the reaction between the rutile coating and the Al melt led to the formation of an -
N
Al2O3 layer at the surface of the TiO2 coating (Fig. 5). The stoichiometry of the layer was
A
confirmed by the EPMA/WDX analysis, its crystal structure by EBSD/SEM. After 60 min
M
and 300 min at 750 °C, the α-Al2O3 reaction layer had a thickness of about 0.7 ± 0.3 µm (see Fig. 5a) and 1.6 ± 0.4 µm (Fig. 5b), respectively.
ED
In all cases, the corundum layer adhered very well to the rutile coating, which indicates, in
PT
combination with the crack-free interface, a heteroepitaxial growth mechanism. The EBSD measurements performed in the vicinity of the original Al/TiO2 interface (Fig. 6) in the sam-
CC E
ple held for 60 min at 750°C revealed an orientation relationship between the newly formed α-Al2O3 grains and the TiO2 matrix, which can be expressed as (001)rutile || (100)corundum and [100]rutile || [010]corundum. This orientation relationship has already been reported in literature
A
(see Ref. [17]). It is worth noting that this orientation relationship is identical with the orientation relationship found for TiO2 and MgTiO3 (cf. Section 3.1 and Fig. 3b), because the phases having the corundum-type or the ilmenite-type structure exhibit generally similar crystallographic orientations towards rutile [25] - [29]. In our samples, this orientation relationship
13
appeared along the whole α-Al2O3/TiO2 interface (cf. Fig. 6a). The experimentally observed misorientations were below 3 ° (Figure 6b). In reference [17], the oriented growth of rutile on α-Al2O3 was explained by a similar arrangement of oxygen atoms in rutile (Wyckoff positions 4f) and in corundum (Wyckoff posi-
IP T
tions 18e) having the observed orientation relationship (001)rutile || (100)corundum and [100]rutile || [010]corundum. The lattice misfit calculated using the lattice parameters arutile = 0.4653 nm, crutile = 0.2965 nm (ICSD #66650), acorundum = 0.4759 nm and ccorundum = 1.2997 nm (ICSD #73725)
SC R
for the directions [100]rutile || [010]corundum and [010]rutile || [001]corundum within the above parallel lattice planes was -2.3 % and +7.1 %, respectively.
U
In the sample exposed to the Al melt for 300 min, areas with stripe-like contrasts were ob-
N
served at the interface between the newly-formed corundum and the rutile coating. The stripes
A
(marked by arrows in Fig. 5b) are almost perpendicular to the TiO2/Al2O3 interface. The na-
M
ture of the stripes was investigated using a combination of TEM imaging, SAED and EDX chemical analysis in the STEM mode (Fig. 7a). The stripes were identified as MgAl2O4 nee-
ED
dles (SG Fd3̅m) that penetrate into the former TiO2 coating, which contains additionally
PT
MgTiO3 regions (Fig. 7). It was discussed above that the formation of MgTiO3 from TiO2 requires magnesium and additional oxygen (equation (2)). Without additional oxygen, TiO2
CC E
would be partially reduced to Ti2O3 (equation (3)). Mg and O are fast diffusing and highly reactive species in TiO2. Their diffusivity is especially high along grain boundaries and along
A
specific crystallographic directions in rutile (see section 3.1). Because of the high reactivity of Mg, even extremely small amounts of this species as a contaminant of molten Al or in rutile seem to be sufficient to trigger the formation of the Mg-containing compounds (see also discussion in section 4). As no Ti2O3 was found in the vicinity of MgTiO3 in TiO2 being in contact with molten aluminium (Fig. 7b) in contrast to the AlSi7Mg0.6/TiO2 system (Fig. 4), it can be concluded that 14
a lower Mg content and consequently a higher [O]/[Mg] ratio in the system facilitate the chemical reaction according to equation (2). An analogous result can be obtained from the observed formation of MgAl2O4. This compound stems from the reaction of Al2O3 with magnesium and oxygen: 1
Al2 O3 + [Mg] + 2 𝑂2 → MgAl2 O4
(4)
IP T
(Δ°G = -520.3 kJ mol-1)
The alternative reaction without the additional oxygen supply would lead to the TiO2 reduc-
SC R
tion and to the production of Ti2O3: 2 TiO2 + Al2 O3 + [Mg] → MgAl2 O4 + Ti2 O3
(5)
U
(ΔG° = -246.2 kJ mol-1)
N
At the beginning of the reaction process, corundum growing between the Al melt and the
A
TiO2 coating (Fig. 5) might be formed from Al and O contained in the aluminium melt. Later
M
on, oxygen is provided mainly by the reduction of rutile. Prominent reduction mechanisms are:
PT
and
ED
6 TiO2 (𝑠) + 8 [Al] → 4 Al2 O3 (𝑠) + 6 [Ti]
(6)
(ΔG° = -531.6 kJ mol-1)
(7)
CC E
6 TiO2 (𝑠) + 2 [Al] → Al2 O3 (𝑠) + 3 Ti2 O3 (𝑠)
(ΔG° = -835.9 kJ mol-1)
The formation of Ti2O3 as an intermediate phase according to equation (7) was observed in
A
sintered powder mixtures of Al and TiO2 (mass ratio of 2:1) and confirmed by thermodynamic calculations [4]. However, it should be noted that the aluminium supply in the above powder mixture was limited because of the high amount of TiO2. In case of the aluminium surplus like in this study, the reduction of TiO2 to metallic titanium (equation (6)) is more probable. In the sintered powder mixtures of Al and TiO2 from reference [4], titanium stemming from the TiO2 reduction led to the formation of Al3Ti. In the present study, no Ti and Al3Ti were 15
detected, because TiO2 was not even reduced to Ti2O3. Thus, the reactions according to equations (6) or (7) did not contribute to the phase formation processes of this study. In the TiO2 coatings brought into contact with pure Al melt, rutile (TiO2), MgAl2O4 spinel and magnesium titanate (MgTiO3) also possess distinguished (and already reported) orientation relationships (010)rutile || (111)spinel, [100]rutile || [110]spinel [25], [28], and (001)titanate || (111)spinel,
IP T
[110]titanate || [110]spinel [26] [36], see Figure 8. A possible reason for this kind of heteroepitaxy is the parallel arrangement of close packed oxygen layers that was already discussed by Arm-
SC R
bruster [25]. In this regard, the appearance of the stripe-like contrast merely represents a lo-
calised change of the heteroepitaxially grown phases, but the presence of the stripes does not
A
N
U
deteriorate the adhesion of corundum to rutile.
M
3.3 Role of the diffusion processes in the growth of the interface layers The thickness of the corundum layer, which has formed at the interface between molten Al
ED
(750 °C) and TiO2, follows a parabolic time dependence. The average layer thickness was (0.7 ± 0.3) µm after 60 min and (1.6 ± 0.4) µm after 300 min, see Figure 5. The growth rate
PT
constant calculated from this layer thickness increase is (7.0 ± 1.3) × 10-17 m2s-1 (cf. Ref.
CC E
[37]). As no Ti2O3 or metallic titanium were observed at the Al2O3/TiO2 interface, it can be expected that the aluminium melt is the primary source of oxygen needed for the Al 2O3 formation and that the growth of the Al2O3 layer is controlled mainly by the diffusion of oxygen
A
in the melt. Furthermore, the Al2O3 layer can be regarded as a protective coating against the reduction of TiO2 by molten aluminium. The growth of the MgTiO3 layer at the interface between the AlSi7Mg0.6 melt and rutile is much faster than the growth of the Al2O3 layer from the Al melt, and depends strongly on the local orientation of the TiO2 grains. The minimum MgTiO3 layer thickness was 2.0 µm after 16
60 min and 2.5 µm after 300 min (Fig. 1). The maximum thickness of the MgTiO3 layer (11.9 µm and 19.7 µm after 60 min and 300 min, respectively) was found in the MgTiO3 needles that were oriented with their c axis along the diffusion direction. In contrast to the Al2O3 layer, the MgTiO3 layer grows at the expense of rutile and its growth cannot be described as parabolic. At the beginning of the chemical reaction between molten AlSi7Mg0.6 alloy and
IP T
the TiO2 coating, MgTiO3 grows much faster (approximately 10 times) than at longer reaction
times. This change of the reaction kinetics is caused by a change in the reaction mechanism,
SC R
which results from the different amounts of magnesium and oxygen in the melt and/or from the different diffusivities of both species in MgTiO3.
U
As long as enough oxygen is present in the melt (close to the AlSi7Mg0.6/TiO2 interface), the
N
MgTiO3 layer grows according to equation (2) from Mg and O stemming from the melt. After
A
oxygen in the melt is consumed and/or after a closed MgTiO3 layer possibly acting as a diffu-
M
sion barrier for oxygen is built on, mainly Mg diffuses from the AlSi7Mg0.6 melt through MgTiO3 into rutile, where it reduces TiO2 to Ti2O3, see equation (3). The rutile reduction and
ED
the diffusion of Mg to the reaction front at the MgTiO3/TiO2 interface are the reasons for the significantly reduced growth rate at longer dwell times. The change in the reaction kinetics
PT
might be seconded by the exhaustion of Mg in the aluminium alloy in the vicinity of the
CC E
AlSi7Mg0.6/MgTiO3 interface.
A
3.4 Interface energies of the materials combinations with experimentally identified orientation relationships
The energies of the TiO2/MgTiO3 and TiO2/Al2O3 interfaces with distinct orientation relationships between the counterparts (Figs. 9, 3b and 6c) were calculated using the interface models described in Section 2.3, and are summarised in Table 3. These interface energies take the 17
deformation energy, i.e., the energy needed for the lattice adaptation in each individual orientation, into account. Despite this positive energy contribution, the negative values of the most total interface energies indicate that the formation of such interfaces is much more favourable than the separate state of the two bulk structures. The calculated interface energies also show
IP T
a clear preference of the orientation relationships observed experimentally (Figs. 3b and 6c).
Interfaces and orientations
Contact surface A [Å2]
Fully relaxed interface energy [J m-2]
(100)rutile || (110)titanate, [001]rutile || [11̅0]titanate [18]
126.35
0.14
(101)rutile || (012)corundum, [010]rutile || [100]corundum [19]
41.58
(001)rutile || (100)titanate, [010]rutile || [001]titanate *
32.12
(001)rutile || (100)corundum, [010]rutile || [001]corundum [17] *
105.43
SC R
-4.15 -4.95 -6.78
A
N
U
* experimentally observed in this study Table 3: The interface energies of the investigated interfaces with the indicated crystallographic orientations as obtained using DFT.
M
4. Discussion
ED
In simulated metal melt filtration experiments performed without melt convection, silicon present in the AlSi7Mg0.6 melt did not enter any chemical reaction with the rutile coating. On
PT
the contrary, aluminium, magnesium and oxygen contained in molten aluminium alloys were found in nearly all reaction products. Though, the reaction mechanisms and the products of
CC E
the reaction between molten aluminium alloys and rutile are strongly affected by the magnesium and oxygen concentrations.
A
In magnesium-free aluminium melts, aluminium and oxygen from the melt react to Al2O3, which forms a continuous layer heteroepitaxially grown at the surface of the rutile coating (Fig. 5). Aluminium melts without oxygen would reduce TiO2, while the released oxygen would react again with Al to α-Al2O3 at the TiO2 surface. Anyway, the corundum layer acts as a barrier against a further corrosion of TiO2 by liquid Al. The heteroepitaxial growth of αAl2O3 layers on the TiO2 surfaces is facilitated by a relatively small lattice misfit between the 18
counterparts (Fig. 6) and by the reduction of the total energy of the α-Al2O3/TiO2 interface in comparison with total energy of the individual bulk compounds (Table 3). At magnesium concentrations in the melt, which are comparable with the Mg concentrations in the AlSi7Mg0.6 alloy (0.6 mass%), Mg reacts quickly with oxygen from the melt and with
IP T
the TiO2 coating to MgTiO3 (Fig. 1). Because of a strong dependence of the diffusivity of Mg and O in rutile on the crystallographic direction, the interface between MgTiO3 and TiO2 is
heavily fringed (Fig. 1). If oxygen from the melt is consumed or if it is transmitted too slowly
SC R
through the MgTiO3 layer with fully occupied oxygen sublattice, Mg still diffuses into TiO2 and produces MgTiO3. The oxygen needed for this reaction is obtained from TiO2, which is
U
reduced to Ti2O3 (Eq. (3) and Fig. 4). These findings illustrate an extreme reactivity of Mg
N
with O and TiO2 and a fast diffusivity of Mg in TiO2, which inhibit the formation of corun-
A
dum at the surface of the TiO2 coatings in contact with the aluminium melts containing mag-
M
nesium.
ED
Very low concentrations of Mg in the ‘contaminated’ aluminium melt (or in the rutile coating) lead to a competition between the formation of MgTiO3 and α-Al2O3 at the surface of the
PT
TiO2 coatings, where the formation of corundum is accompanied by the formation of MgAl2O4. As Mg (and O) diffuse very quickly into TiO2, the TiO2 coating is ‘filled’ with
CC E
magnesium (and oxygen) before a protective corundum layer can form at the coating surface. If the local Mg content (near the interface to the MgTiO3/TiO2 stack) is depleted, the corundum layer starts to grow. As α-Al2O3 acts as a barrier for the Mg (and O) diffusion into TiO2,
A
the presence of the α-Al2O3 layer inhibits further formation of MgTiO3, even if the local concentration of Mg at the interface between the melt and the corundum layer is increased again. In such a case, magnesium and oxygen from the melt react with the aluminium oxide to MgAl2O4. However, it is worth noting that the occurrence and sequence of these processes strongly depend on the [Mg]/[O] ratio that is altered by magnesium impurities present in the 19
aluminium alloy and/or by sporadic MgO particles or lumps present in the sintered rutile. Furthermore, as MgAl2O4 was found in the transition zone between the original TiO2 coating and the α-Al2O3 layer only (Fig. 5), where it was in contact with TiO2 and MgTiO3 (Fig. 7), it can be assumed that the growth of MgAl2O4 is facilitated by its structural similarity to MgTiO3
IP T
(Fig. 8). Observed reaction processes can be utilised for the filtration of oxygen from the aluminiumbased melts. Because of a possible heteroepitaxy between TiO2 and α-Al2O3 and because of a
SC R
negative energy of the TiO2/α-Al2O3 interfaces having favourable orientation relationships
between the counterparts, the functional TiO2 coating serves as a suitable docking layer for
U
the formation of corundum stemming from O (and Al) present in the melt. On the other hand,
N
rutile is open for magnesium and oxygen diffusion, and forms MgTiO3 with these elements.
A
MgTiO3 serves as diffusion barrier for oxygen but not for magnesium, thus the rapid reaction
M
diffusion of Mg in TiO2 leads to the reduction of TiO2 to Ti2O3, because the oxygen is needed for formation of MgTiO3. Also at very low Mg concentrations in molten Al alloys, magnesi-
ED
um is removed from the melt and forms two phases containing magnesium, i.e., MgTiO3 and MgAl2O4. Consequently, the rutile coatings extract magnesium from the melt if they are used
PT
as functional materials on the surface of metal melt filters intended for the filtration of alu-
CC E
minium alloys containing magnesium. The formation of intermediate phases discussed above is facilitated by the heteroepitaxy between TiO2 and α-Al2O3, between TiO2, MgTiO3 and MgAl2O4, and between MgTiO3 and
A
Ti2O3, as well as by the negative energy of the TiO2/ α-Al2O3 and TiO2/MgTiO3 interfaces (including a positive deformation energy stemming from the small lattice misfit). The heteroepitaxy and the negative interface energies also guarantee a good adhesion of the reaction layers to the rutile coating. Therefore, no peeling off of the reaction layers is expected.
20
Conclusions Rutile exposed to aluminium alloys at 750 °C was found to react differently for different magnesium concentrations in the melt. The contact of TiO2 with molten aluminium leads to the epitaxial growth of a thin layer of thermodynamically stable corundum. The orientation relationship between TiO2 and α-Al2O3 was described as (001)rutile || (100)corundum and [100]rutile
IP T
|| [010]corundum. The α-Al2O3 layer accommodates oxygen solved in the aluminium melt and
protects the TiO2 coating from corrosion by molten aluminium. In an oxygen-free melt, the α-
SC R
Al2O3 layer would be built up from oxygen achieved from the TiO2 reduction. The epitaxial growth of α-Al2O3 on TiO2 and a negative energy of the α-Al2O3/TiO2 interface calculated
U
using the density functional theory promise a good adhesion of the corundum layer to the ru-
N
tile coating. The contact of TiO2 with molten AlSi7Mg0.6 leads to the formation of magnesi-
A
um titanate (MgTiO3) with a distinct orientation relationship to rutile, which was described as
M
(001)rutile || (100)titanate and [010]rutile || [001]titanate. Because of the high diffusivity of magnesium in TiO2, the TiO2 coatings are heavily altered and react to MgTiO3 and Ti2O3. The latter
ED
compound is a product of the TiO2 reduction, which is needed to gain oxygen for the MgTiO3 production. Also in this case, the structural similarity of present phases and their possible epi-
CC E
PT
taxy favour the growth of interface layers and support the adhesion of the counterparts.
Acknowledgement
A
This work was financially supported by the German Research Foundation (DFG) in the frame of the Collaborative Research Centre SFB 920. The authors would like to thank Dr. D. Heger (Institute of Materials Science, TU Bergakademie Freiberg) for the EPMA analyses and Dr. A. Plessow (Institute of Mineralogy, TU Bergakademie Freiberg) for the XRF analysis.
21
References
A
CC E
PT
ED
M
A
N
U
SC R
IP T
[1] C. Voigt, E. Jäckel, F. Taina, T. Zienert, A. Salomon, G. Wolf, C. Aneziris, P. LeBrun, Filtration Efficiency of Functionalized Ceramic Foam Filters for Aluminum Melt Filtration, Metall. Mater. Trans. B 48 (2016) 497-505. [2] H. Hashimoto, S. Kojima, T. Sasaki, H. Asoh, α-Alumina membrane having a hierarchical structure of straight macropores and mesopores inside the pore wall, J. Eur. Ceram. Soc. 38 (2018) 1836-1840. [3] C. Aneziris, S. Dudczig, M. Emmel, H. Berek, G. Schmidt, J. Hubalkova, Reactive filters for steel melt filtration, Adv. Eng. Mat. 15 (2013) 46-59. [4] A. Salomon, C. Voigt, O. Fabrichnaya, C. G. Aneziris, D. Rafaja, Formation of Corundum, Magnesium Titanate, and Titanium(III) Oxide at the Interface between Rutile and Molten Al or AlSi7Mg0.6 Alloy, Adv. Eng. Mat. 19 (2017) 1700106. [5] M. Ibarra Castro, J. Almanza Robles, D. Cortés Hernandez, J. Escobedo Bocardo, R. MartinezLopez, Chemical interaction between SrAl2Si2O8 and molten aluminum, J. Eur. Ceram. Soc. 35 (2015) 4287-4292. [6] W. Yan, A. Schmidt, S. Dudczig, T. Wetzig, Y. Wei, Y. Li, S. Schafföner, C. Aneziris, Wettability phenomena of molten steel in contact with alumina substrates with alumina and alumina-carbon coatings, J. Eur. Ceram. Soc. 38 (2018) 2164-2178. [7] C. Voigt, T. Zienert, P. Schubert, C. Aneziris, J. Hubalkova, Reticulated porous foam ceramics with different surface chemistries, J. Am. Ceram. Soc. 97 (2014) 2046-2053. [8] A. Salomon, T. Zienert, C. Voigt, E. Jäckel, O. Fabrichnaya, D. Rafaja, C. Aneziris, Comparison of interfacial reactions between AlSi7Mg and alumina filter after casting and spark plasma sintering, Adv. Eng. Mat. 15 (2013) 1206-1215. [9] J.-O. Andersson, T. Helander, L. Höglund, P. Shi, B. Sundmann, Thermo-Calc & DICTRA, computational tools for materials science, Calphad 26 (2002) 273-312. [10] A.T. Dinsdale, SGTE data for pure elements, Calphad 15 (1991) 317-425. [11] T. Zienert, O. Fabrichnaya, Thermodynamic Assessment and Experiments in the system MgOAl2O3, Calphad 40 (2013) 1-9. [12] M. Ilatovskaia, G. Savinykh, O. Fabrichnaya, Thermodynamic Description of the Ti-Al-O System Based on Experimental Data, J. Phase Equilib. Diff. 38 (2017) 175-184. [13] M. Ilatovskaia, I. Saenko, G. Savinykh, O. Fabrichnaya, Experimental study of phase equilibria in the Al2O3-MgO-TiO2 system and thermodynamic assessment of the MgO-TiO2 system, J. Am. Ceram. Soc. (2018), https://doi.org/10.1111/jace.15748. [14] SGTE solution database, 2015SGSOL. http://www.sgte.net, 2018 (accessed 20 June 2018). [15] P. Giannozzi, S. Baroni, N. Bonini, M. Calandra, R. Car, C. Cavazzoni, D. Ceresoli, G.L. Chiarotti, M. Cococcioni, I. Dabo, A. Dal Corso, S. De Gironcoli, S. Fabris, G. Fratesi, R. Gebauer, U. Gerstmann, C. Gougoussis, A. Kokalj, M. Lazzeri, L. Martin-Samos, N. Marzari, F. Mauri, R. Mazzarello, S. Paolini, A. Pasquarello, L. Paulatto, C. Sbraccia, S. Scandolo, G. Sclauzero, A.P. Seitsonen, A. Smogunov, P. Umari, R.M. Wentzcovitch, QUANTUM ESPRESSO: A modular and open-source software project for quantum simulations of materials, J. Phys. Condens. Matter 21 (2009) 395502-119. [16] J. Perdew, K. Burke, M. Ernzerhof, Generalized gradient approximation made simple, Phys. Rev. Lett. 77 (1996) 3865-3868. [17] C. Chen, A. Korotcov, Y.-S. Huang, W. Chung, D. Tsai, K. Tiong, Growth and characterization of well-aligned rutile TiO2 nanocrystals on sapphire substrates via metal organic vapour deposition, J. Mat. Sci.: Mat. Elect. 20 (2009) S332-S335. [18] A. Lotnyk, S. H. D. Senz, Orientation relationships of SrTiO3 and MgTiO3 thin films grown by vapor-solid reactions on (100) and (110) TiO2 (Rutile) single crystals, J. Phys. Chem. C 111 (2007) 6372-6379. [19] R. Engel-Herbert, B. Jalan, J. Cagnon, S. Stemmer, Microstructure of epitaxial rutile TiO2 films grown by molecular beam epitaxy on r-plane Al2O3, J. Cryst. Growth 312 (2009) 149-153. [20] K. Hoshino, N. Peterson, C. Wiley, Diffusion and point defects in TiO2-x, J. Phys. Chem. Solids 46 (1985) 1397-1411. [21] J. Sasaki, N. Peterson, K. Hoshino, Tracer impurity diffusion in single-crystal rutile (TiO2-x), J. Phys. Chem. Solids 46 (1985) 1267-1283. 22
A
CC E
PT
ED
M
A
N
U
SC R
IP T
[22] L. Jeurgens, W. Sloof, F. Tichelaar, E. Mittemeijer, Growth kinetics and mechanisms of aluminum-oxide films formed by thermal oxidation of aluminium, J. Appl. Phys. 92 (2002) 1649-1656. [23] G. Xie, O. Ohashi, T. Sato, N. Yamaguchi, M. Song, K. Mitsuishi, K. Furuya, Effect of Mg on the sintering of Al-Mg alloy powders by pulse electric-current sintering process, Mater. Trans. 45 (2002) 904-909. [24] G. Xie, O. Ohashi, N. Yamaguchi, M. Song, K. Mitsuishi, K. Furuya, T. Noda, Reduction of surface oxide films in Al-Mg alloy powders by pulse electric current sintering, J. Mater. Res. 19 (2004) 815-819. [25] T. Armbruster, On the origin of sagenites: structural coherency of rutile with hematite and spinel structures, N. Jb. Miner. Mh. 7 (1981) 328-334. [26] E. Force, R. Richards, K. Scott, P. Valentine, N. Fishman, Mineral intergrowths replaced by "elbow-twinned" rutile in altered rocks, Can. Mineral. 34 (1996) 605-614. [27] A. Recnik, N. Stankovic, N. Daneu, Topotaxial reactions during the genesis of oriented rutile/hematite intergrowths from Mwinilunga (Zambia), Contrib. Mineral. Petr. 169 (2015) 19. [28] F. Wong, S. Ramanathan, Heteroepitaxy of distorted rutile-structure WO2 and NbO2 thin films, J. Mater. Res. 28 (2013) 2555-2563. [29] N. Daneu, A. Recnik, W. Mader, Atomic structure and formation mechanism of (101) rutile twins from Diamantina (Brazil), Am. Mineral. 99 (2014) 612-624. [30] D. Rafaja, T. Schucknecht, V. Klemm, A. Paul, H. Berek, Microstructural characterisation of titanium coatings deposited using cold gas spraying on Al2O3 substrates, Surf. Coat. Tech. 203 (2009) 3206-3213. [31] J. Kutzner, D. Rafaja, J. Kortus, On the possibility of a hetero-epitaxial Fe/ZrO2 interface: A density-functional-theory point of view, Model. Simul. Mater. Sci. Eng. 20 (2012) 045011. [32] R. Drehmann, T. Grund, T. Lampke, B. Wielage, K. Manygoats, T. Schucknecht, D. Rafaja, Splat formation and adhesion mechanisms of cold gas-sprayed al coatings on Al2O3 substrates, J. Therm. Spray Techn. 23 (2014) 68-75. [33] B. Matthey, S. Höhn, A.-K. Wolfrum, U. Mühle, M. Motylenko, D. Rafaja, A. Michaelis, M. Herrmann, Microstructural investigation of diamond-SiC composites produced by pressureless silicon infiltration, J. Eur. Ceram. Soc. 37 (2017) 1917-1928. [34] C. Wüstefeld, D. Rafaja, M. Motylenko, C. Ullrich, R. Drehmann, T. Grund, T. Lampke, B. Wielage, Local heteroepitaxy as an adhesion mechanism in aluminium coatings cold gas sprayed on AlN substrates, Acta Mater. 128 (2017) 418-427. [35] A. Sheikh, J. Irvine, Phase Formation and Electronic Transport Properties in the Corundum (Ti2O3)-Ilmenite (MgTiO3) System, J. Solid State Chem. 103 (1993) 30-37. [36] P. Robinson, F. Langenhorst, S. McEnroe, K. Fabian, T. Ballaran, Ferroan geikielite and coupled spinel-rutile exsolution from titanohematite: Interface characterization and magnetic properties, Am. Mineral. 99 (2014) 1694-1712. [37] A. Heuer, Oxygen and aluminum diffusion in alpha-Al2O3: How much do we really understand?, J. Eur. Ceram. Soc. 28 (2008) 1495-1507. [38] K. Momma, F. Izumi, VESTA 3 for three-dimensional visualization of crystal, volumetric and morphology data, J. Appl. Crystallogr. 44 (2011) 1272-1276.
23
Figure captions
Figure 1:
A
CC E
PT
ED
M
A
N
U
SC R
IP T
SEM micrographs (BSE contrast) of rutile coatings, which were brought in contact with molten AlSi7Mg0.6 alloy for 60 min (a) and 300 min (b) at 750 °C. At the interface between the coating and the liquid alloy, a MgTiO3 layer formed.
24
Figure 2:
A
CC E
PT
ED
M
A
N
U
SC R
IP T
Results of the EBSD analysis performed on the rutile coating from Fig. 1a. (a) Phase map showing the distribution of rutile (yellow) and MgTiO3 (blue). Pores/voids are reproduced in black, the grey region at the bottom is the solidified alloy AlSi7Mg0.6. The green lines mark the interfaces between the TiO2 and MgTiO3 grains having the orientation relationship (001)rutile || (100)titanate and [010]rutile || [001]titanate. A histogram of the local deviations from this orientation relationship is shown in (b). The above orientation relationship is illustrated in figure (c).
25
Figure 3:
A
CC E
PT
ED
M
A
N
U
SC R
IP T
(a) A TEM micrograph of the MgTiO3/TiO2 interface. A phase boundary with geometrically necessary misfit dislocations is marked by white arrows. The orientation relationship (001) rutile || (100)titanate and [010]rutile || [001]titanate shown in panel (b) was verified by SAED. Crystal structures were created with the VESTA 3 software [38].
26
Figure 4:
A
CC E
PT
ED
M
A
N
U
SC R
IP T
TEM and SAED of the MgTiO3/Ti2O3 interfaces. The growth direction is perpendicular to the plane of the image.
27
Figure 5:
A
CC E
PT
ED
M
A
N
U
SC R
IP T
SEM micrographs (BSE contrast) of rutile coatings that were in contact with molten Al (at 750 °C) for 60 min (a) and 300 min (b). The main reaction layer contains corundum (α-Al2O3). The dotted box in (b) shows the position of a FIB lamella, which was investigated by TEM and SAED in order to explain the nature of the stripes marked by black arrows.
28
Figure 6:
A
CC E
PT
ED
M
A
N
U
SC R
IP T
(a) EBSD phase map of the TiO2/Al interface from Fig. 5a. Rutile is plotted in yellow, corundum in red. Black areas within the colourised region are pores, non-indexed bottom region is solidified aluminium. The green lines mark the interfaces between rutile and corundum crystallites having the orientation relationship (001)rutile || (100)corundum and [010]rutile || [001]corundum. A histogram of the local deviations from this orientation relationship is shown in (b). (c) Model of rutile and corundum in the above orientation relationship plotted using VESTA 3 [38].
29
Figure 7:
A
CC E
PT
ED
M
A
N
U
SC R
IP T
Element maps (a) and TEM micrograph (b) of the stripes from Fig. 5b. Individual phases were assigned using a combination of chemical analysis (EDX) and SAED. The orientation relationships are summarised in the text. The viewing direction is perpendicular to the sample surface and corresponds to the growth direction of corundum into rutile.
30
Figure 8:
A
CC E
PT
ED
M
A
N
U
SC R
IP T
Crystal structure models of TiO2, MgAl2O4 and MgTiO3 mutually oriented according to the orientation relationships that were identified using SAED (cf. Fig. 7). The parallel planes highlighted in green mark close-packed oxygen sublattice planes in each structure (highly distorted in rutile), arrows mark the corresponding parallel directions. The crystal structures were created with VESTA 3 [38].
31
Figure 9:
A
CC E
PT
ED
M
A
N
U
SC R
IP T
Orientation relationships found in literature (see Refs. [18] and [19] for (a) and (b), respectively) and used for comparative DFT calculations of interface energies. Parallel lattice planes are highlighted in green and arrows mark the corresponding parallel directions. Only atoms lying in or behind the parallel planes are depicted, cf. figures 3b, 6c and 8. The crystal structures were created with VESTA 3 [38].
32