Phase separation and up-hill diffusion in the ordered α2 compound of a γ- Ti-Al-Nb alloy

γ-TiAl alloys are intermetallic materials that are designed for applications in air craft engines. Oftentimes alloying elements are added to improve the corrosion resistance and mechanical properties. Here we investigate microstructural changes that take place when Nb is added to a binary γ-TiAl alloy. Previous studies of the alloy Ti-42Al-8.5Nb have shown that upon Nb addition the orthorhombic O-phase forms out of the parent α2 phase within the alloy’s microstructure. During annealing Nb segregates to the newly formed O-phase while Al and Ti partition to the remaining α2-like matrix phase. The changes these newly formed O-phase domains undergo with time at elevated temperature are not known. Here we use Transmission Electron Microscopy techniques to monitor the distribution of Nb and the size of O-phase domains during long-term annealing between 1 week and 4 weeks at 550°C. The results show that the size of O-phase domains shrinks with annealing time. At the same time the Nb concentration remains largely constant or possibly shows a small decrease.


Introduction
γ-Titanium aluminides have a high specific strength at elevated temperatures, which makes them an attractive structural light-weight material for applications in aircraft engines. At an Al-content of 40-45 at.-% the main constituents in the microstructure of these alloys are the hexagonal α 2 phase (Ti 3 Al, D019) that forms out of the high temperature hcp α phase and the tetragonal y phase (TiAl, L12) that precipitates within α 2 grains during further cooling. A negligible percentage of β 0 phase (or ω 0 phase) may form at triple junctions if the content of β-stabilizing elements (e.g. Nb, Mo, Ta) is sufficiently high. During cooling a near lamellar microstructure forms, where the γ and α 2 phases are aligned parallel to their close packed lattice planes in lamellar (α 2 + γ) colonies. The crystallographic orientation relationship has been described by Blackburn [1]. Third generation γ-TiAl-alloys are often alloyed with niobium for improved creep strength and oxidation resistance [2]. The addition of Nb can introduce an additional, orthorhombic phase to the alloy's microstructure that forms inside α 2 lamellae in the temperature range between 450 and 670°C. The thermal stability range of this O-phase is limited: above 700°C it re-transforms to hexagonal α 2 phase. Furthermore, its formation can be largely suppressed by air-cooling from temperatures above 700°C. We investigated the formation of O-phase in the alloy Ti-42Al-8.5Nb (at.-%) by in-situ heating HEXRD and studied its location within lamellar (α2+γ) colonies by TEM [3,4]. Initial ambiguity on the nature of the orthorhombic phase was resolved and it was identified as sub-stoichiometric O-phase (Ti2AlNb, Cmcm) [5,6]. Similar findings were reported by [7,8]. The orthorhombic O-phase has been described for the first time much earlier by Banerjee in Nb-rich, Al-lean Ti-Al compositions and it has been studied intensely [9][10][11]. However until the first reports by Appel et al. it was unknown in γbased TiAl-alloys and it was a novelty to identify O-phase at this small, nm size scale within α2 lamellae. By now the existence of O-phase is accepted and more details of its appearance -the size and the shape of O-phase domains including the distribution of Nb-are investigated [12,13]. In the present paper the microstructure of two annealed states is compared (T= 550°C, duration: 1 week and 4 weeks). We show that Nb agglomerates in small domains separated by Ti and Al rich channels (1 week). The lattice within these Nbrich domains is distorted with respect to the hexagonal lattice of α 2 and is identified as the orthorhombic O-phase. In contrast, the Nb-lean channels inbetween domains show only little distortion. After 4 weeks of annealing the O-phase domains split into smaller islands separated by wider channels. At the same time, there is only little change in the Nb concentrations observed in the domains.

Experimental
The alloy Ti-42Al-8.5Nb was produced by powder metallurgy starting from pure elements following the method described in [14]. After compacting by hot-isostatic pressing (200 MPa, 1250 °C, 2 h, Ar atmosphere) the alloy was annealed at 550°C for 1 week (168 h) and for 4 weeks (672 h) to form the orthorhombic phase. Specimens for TEM investigations were prepared by standard methods, see [3]. TEM and HAADF STEM investigations were performed at the Helmholtz-Zentrum in Geesthacht with the Titan 80-300 kV and the CM 200 operated at 300 kV and at 200 kV, respectively. Atomic resolution HAADF STEM and EDX mapping were employed to characterize the size/shape and elemental distribution within α 2 /O-phase lamellae. The atomic resolution HAADF STEM images were obtained at the CNMS, at Oakridge National Laboratory using the Nion UltraSTEM 100 dedicated aberration-corrected STEM at 100 kV. Elemental EDX mapping with the signals Al-K, Nb-L, Ti-K was performed at the BEEM facility of the Technical University Hamburg using a SuperX detector system attached to an FEI-Talos F200XTEM operated at 200 kV with a probe current of 1 nA. The average size of O-phase domains was measured in Adobe® photoshop on a grid superimposed onto STEM HAADF images. The image magnification was 160,000x (0.85 nm pixels size) and 320,000x (0.42 nm pixels size) and 5 images of different lamellar colonies were measured in each annealing state. The size of the grid was chosen so that every domain was measured only once in vertical and horizontal image direction. The total number of measurements was 351 and 343 for the 1 week and the 4 week annealing state. For the measurement of matrix channel width intensity-scaled Nb maps were used (image intensity proportional to Nb wt.-%, 0.4 nm pixel size). The measurements were performed in Digital Micrograph™ on intensity profiles oriented perpendicular to channels between Nb rich regions. Per annealing state 13 measurements were performed. For the 1 week specimen Nb maps recorded from 2 lamellar colonies were used; for the 4 week specimen measurements were performed in one grain. At the same time, the Nb concentrations were measured in a 5 nm wide region above of the line profile as illustrated in Fig. 5.

Results and Discussion
An overview of the alloy's microstructure consisting of lamellar (α 2 + γ) colonies is given in the SEM image in Fig. 1a. The schematic sketch in Fig. 1b illustrates the orientation relationship between the phases inside the lamellar colonies where the α 2 -(0001) plane is parallel to the α 2 /γ interface. The O-phase that is investigated here forms within the α 2 phase lamellae by a displacement of atom positions in α 2 -(0001) planes. As a result of the displacement, the lattice of the O-phase is distorted with respect to the lattice of the parent α 2 phase. All following micrographs are taken in α 2 -[0001] viewing direction. The α 2 lamellae are called α 2 /O-phase lamellae hereafter because after the heat treatment at 550°C both phases co-exist in them.  Fig. 1a, are no longer visible. Instead a pattern of dark lines on a brighter background is discernable that suggests that heavy and light elements concentrate in different regions (in z-contrast elements with a higher atomic number z appear brighter). In Fig. 2c, taken in atomic resolution HAADF STEM mode, the darker and brighter regions of the α 2 /O-phase lamellae can be seen more clearly. In the bright regions the lattice is elongated along <1120> directions of the α 2 lattice reducing the six-fold hexagonal α 2 symmetry to the two-fold symmetry of an orthorhombic lattice. These bright regions are O-phase domains. The darker regions between the O-phase domains are formed by lighter elements. Close inspection also shows that these regions are less distorted than the O-phase domains. Based on the almost hexagonal appearance and the predominance of Al and Ti (see below) we call these regions α 2 -like matrix even though the symmetry is not perfectly hexagonal. There is no sharp border between O-phase domains and α 2 -matrix but rather a smooth transition. The nature of the z-contrast observed in HAADF STEM images can be identified in the EDX maps in Fig. 3. Elemental maps taken with the signals of Al-K, Ti-K and Nb-L are shown in Figs. 3a-c. A comparison to Fig. 2b. and 2c. confirms agglomerations of Al and Ti in α 2 -like matrix channels whereas Nb concentrates within O-phase domains.

Comparison between microstructures after 1 and 4 weeks of annealing at 550°C
With prolonged annealing time the general features of the microstructure do not change. In both annealing states Nb-rich O-phase domains are observed that are separated by α 2 -like matrix channels enriched in Al and Ti. At close inspection, it appears that the size of O-phase domains and the widths of matrix channels have changed after 4 weeks of annealing. This is illustrated in Fig. 4, where Nb maps recorded after 1 week and 4 weeks of annealing are shown side by side. The inset Fig. 4a represents the annealing state after 1 week whereas the larger part of the image (Fig. 4b) corresponds to 4 weeks of annealing. Comparison shows that the larger regions in Fig. 4a are divided into smaller islands in Fig. 4b. At the same time, the channels between the bright regions appear wider.
To confirm these impressions, the average domain sizes were measured on HAADF STEM images of both annealing states. The matrix channels widths and the Nb-concentrations were measured using concentration profiles on Nb maps as illustrated in Fig. 5.
The measurements of domain sizes on STEM-HAADF images give average values of 10 nm for the 1 week annealing state and 7.0 nm for the 4 weeks annealing state. The margins of error (95% confidence interval for population mean) are 1 nm and 0.5 nm respectively (rounded). The large variation in domain sizes is in agreement with the observations in Fig. 4. There, after 1 week of annealing, elongated, rhomboidal or rectangular Nb rich regions are observed that have large aspect ratios. For comparison after 4 weeks of annealing the brighter Nb islands are smaller in size with an aspect ratio closer to one. Statistical analysis (t-test) indicates that the difference in means between the two populations is significant (p-value < 10-3). This supports the conclusion that the domain size shrank during the extra time the 4 week specimen was exposed to the annealing temperature. This result is rather surprising as one would expect continued growth with longer annealing time. The mechanism of shrinkage is splitting of domains as illustrated in Fig. 5. There, the splitting of a domain is heralded by a darker image contrast in the center of the domain and a drop in Nb concentration at the corresponding position in the profile. The two corresponding positions are marked by arrows.   The results of the measurements of Nb concentrations and of matrix channel widths are summarized in table 1. The mean channel width grows from 2.1 ± 0.6 nm (standard deviation) after 1 week of annealing to 4.6 ± 0.6 nm (standard deviation) after 4 weeks annealing time. Even though the standard deviation is rather large (about 30 %) there is no overlap between the measured values. The difference is also apparent in Fig. 4. A comparison of the Nb concentrations between the 2 annealing states shows that higher values were measured in bothmatrix channels and domains-after the shorter annealing time. This is unexpected. 20 ± 1 wt.-% are measured in domains after 1 week of annealing compared to 17 ± 1 wt.-% after 4 week of annealing. Similarly, 16 ± 1 wt.-% are measured in matrix channels after 1 week of annealing compared to 11 ± 1 wt.-% after 4 weeks of annealing. Likely explanations are that the Nb concentrations differ between separate grains or that Nb segregates to the interface with γ or moves into the γphase. These points will have to be checked. In previous measurements the difference in Nb concentration between lamellar colonies was 0.8 -3.2 wt.-% which is close to the observed difference in the present experiments [3]. A factor that certainly influences the present results is the irregular shape of the Nb enrichment within the O-phase domains. In the 1 week specimen Nb rich ridges appear near the domain edge in Fig. 4a but not after 4 weeks in Fig. 4b. This may explain higher Nb values in the 1 week specimen. Clearly, more measurements are needed and the measured Nb concentrations should be regarded with caution. Nevertheless, the finding that there is no distinct increase in Nb concentration in O-phase domains after 4 weeks of annealing is remarkable. Especially considering the history of the microstructural evolution: The Nb concentration was homogeneous in the starting material and after 8 hours of annealing at 550°C. It should be mentioned that after 8 hours of annealing O-phase had already formed [12]. Only after annealing for 1 week at 550°C Nb up-hill diffusion to O-phase domains was observed (Figs. 2, 3c). In the continuation of this trend one might expect even higher values after 4 weeks of annealing. But this is not found here. For comparison to the present measurements on elemental maps the Nb concentration measured in larger regions within α 2 /O-phase lamellae in this alloy was 9 at.-% or 19 wt.-% in previous experiments [3]. The stoichiometric O-phase Ti2AlNb has a Nb concentration of 43 wt.-%.

Conclusions
We monitored size and composition of O-phase domains in α 2 /O-phase-lamellae in the alloy Ti-42Al-8.5Nb after annealing at 550°C for a duration of 1 and 4 weeks. Atomic resolution HAADF-STEM images and elemental maps show that Nb rich O-phase domains are embedded in an α 2 -like matrix in both annealing states. The orthorhombic, Nb-rich lattice of the Ophase shows a distinct distortion while the lattice in the α 2 -like matrix channels is the less distorted-almost hexagonal. The size of O-phase domains shrinks with prolonged annealing time by a splitting mechanism. Surprisingly, the Nb concentrations within O-phase domains in both annealing states remain roughly constant or even show a small decrease. Taken together, the microstructural evolution between 1 and 4 weeks of annealing at 550°C indicates that in absence of a concentration gradient, Nb diffusion in this system might be influenced by internal stresses that act between O-phase domains.

Acknowledgements
Atomic resolution HAADF STEM imaging was conducted using the Nion UltraSTEM 100 (U100) dedicated aberrationcorrected STEM at the Center for Nanophase Materials Sciences, which is a DOE Office of Science User Facility.