The Role of Oxygen in α 2 Formation in the Titanium Model Alloy Ti-7Al

The mechanical and microstructural response of a model Ti-7wt.%Al alloy doped with 500 wppm oxygen has been examined in detail as a funcon of ageing. Three ageing condions, ice-water quenched, and aged at 550 o C for 10 or 49 days were examined in detail as a funcon of ageing. Nanoindentaon was used to measure hardness as a funcon of ageing, while Atom Probe Tomography (APT) also revealed α 2 precipitaon in the aged samples. The paroning preference of oxygen to the α matrix instead of the α 2 precipitates has been directly measured.


Introduction
Titanium is widely used in the aerospace industry owing to its high specific fa gue strength and good corrosion resistance. Approximately 35% of a jet engine is made from tanium [1], with other components being made from nickel, steel, aluminium and composites. Titanium components include fan blades/ discs and compressor blades/ discs, all of which are located towards the front end of a jet engine, ahead of the combus on chamber [2]. Fan blades operate at lower temperatures, up to about 315 o C, and are therefore o�en made from the common workhorse alloy Ti-6Al-4V whereas compressor discs are made from the alloy IMI 834, designed for improved creep resistance over Ti-6Al-4V, which can operate at temperatures approaching 600 o C [2].
Even trace amounts of oxygen are known to strengthen, but also embri�le tanium alloys [3]- [5]. Specifically, oxygen has been shown to have a detrimental effect on mechanical proper es by encouraging forma on of the embri�ling α 2 phase, Ti 3 Al. The α 2 phase typically forms via chemical ordering of alpha tanium during thermal ageing at temperatures between 500-700 o C [6], although the precise mechanism is not well understood. It has been suggested that this chemical ordering causes disloca ons to travel in pairs through the ordered regions, as evidenced by a transi on from homogeneous to planar slip [7]. This causes strain localisa on. Thus it is thought that forma on of the ordered Ti 3 Al phase as a result of ageing at intermediate temperatures, such as those experienced during the processing of fan blades, could make the alloy more suscep ble to cold dwell fa gue [8].
Previous papers [6], [9] have qualita vely described the effect of oxygen on α 2 forma on but not quan fied the amount of oxygen needed to form α 2 , nor defined the mechanism by which oxygen actually encourages α 2 precipita on. In this work we directly quan fy the oxygen concentra on in both the α and α 2 phase to determine the loca on and quan ty of oxygen within the alloy microstructure and relate this to changes in the hardness measured through nanoindenta on.

Materials
Ti-7wt.%Al, a model alloy with composi on similar to that of the alpha phase of Ti-6Al-4V, was supplied by TIMET Wi�on, Birmingham, UK. The material was melted, using an Arc furnace, at Imperial College London and 500 wppm oxygen was introduced into the melt via addi on of TiO 2 powder. The material was rolled at 900 o C and then recrystallised at 980 o C for 1 hour before being quenched in ice water.
Pieces of the ice water quenched material were treated for 10 days at 550 o C and 49 days at 550 o C. Samples were encapsulated under argon before that treatment.

Nanoindentation
Nanoindenta on was performed on samples that had been ground using 600-4000 grit SiC paper, polished with 3 µm to 1 µm diamond suspension and then polished with colloidal silica solu on neutralised with hydrogen peroxide.
A NanoIndenterXP (MTS) was used to create arrays of 2 µm deep indents using a diamond Berkovich indenter. The contact s ffness was con nually measured using the con nuous s ffness measurement (CSM) technique (2 nm amplitude, 50 Hz oscilla on) and was used to assess the quality of the indenta on data. All indents used here have a near linear depth vs. load/s ffness, which indicates suitable sample moun ng and good contact between p and surface. Hardness values reported are derived from the CSM data and averaged between depths of 1250-1750 nm.

Atom Probe Tomography (APT)
Samples for atom probe tomography (APT) were prepared on a Zeiss NVision 40 FIB-SEM (focussed ion beam scanning electron microscope) using the standard li�out method [10]. APT samples were then analysed with a CAMECA LEAP 5000 HR system, using a laser energy of 40 pJ at a stage temperature of 50 K with a pulse frequency of 200 kHz. Addi onal APT samples (not shown) were also analysed in voltage mode (20% pulse frac on) to confirm any laser artefacts were not affec ng composi on measurements. APT data analysis was performed using the commercially available Integrated Visualisa on and Analysis So�ware (IVAS) so�ware package (Cameca) and the open access 3Depict so�ware (threedepict.sourceforge.net, version 0.0.21).
Atom probe data is collected by using a voltage or laser pulse to field ion evaporate single atoms from a sharp (<100 nm diameter) needle of the material of interest. Evaporated ions are accelerated by the applied electric field towards a detector. Since the me of flight, posi on of impact on the detector and order in which each ion is detected are measured, a 3D reconstruc on of the evaporated needle can be created. The reconstruc on consists of a 3D point cloud in which each point represents an ion for which the chemical iden ty and spa al loca on is known.

Precipitate identification
The Pearson coefficient, μ, was chosen as a metric to describe devia on from a random solid solu on within this point cloud data, since devia on from a random solu on may be an indicator of precipita on [11]. The Pearson coefficient derives from chi-squared sta s cs, and is used to describe the departure from randomness of an experimental distribu on from the theore cal, binomial distribu on. It should be noted that μ can be used to compare changes in degrees of clustering/precipita on across atom probe reconstruc ons of different sizes. However, quan ta ve comparisons of the degree of clustering are o�en not possible [12].
Once devia on from random had been established, precipitates were iden fied and extracted using the isoconcentra on surface method. Proximity histograms (proxigrams [12]) were used to select an appropriate isoconcentra on se ng by adjus ng the value un l the posi on of the interface coincided with the point of inflec on of the aluminium concentra on profile on the proxigram. Ions contained within the isoconcentra on surface were then extracted as precipitates. The concentra on of the extracted precipitates was calculated using a MATLAB mass spectrum peak deconvolu on tool (AtomProbeLab v0.1.4, (2019), h�p://atomprobelab.sourceforge.net/).

Results
Hardness tests were performed across the Ti-7Al + 500 wppm O samples in the ice water quenched, 10 day aged and 49 day aged states. Figure 1 shows that there is reasonable agreement between the predicted hardness for unaged Ti-64 + 500 wppm O calculated using the model from [13], and experimental measurements. There is also no statistically significant hardness increase with ageing time.    To investigate if any microstructural changes are taking place during ageing, APT samples of each of the three Ti-7Al+500 wppm O samples were analysed, the atom maps for which are shown in figure 2. From these, it can be seen that there is clear clustering of aluminium atoms in figure 2 (c). However, it is hard to determine by eye if clustering is occurring in the other samples. Calculation of the Pearson coefficient for the three datasets shows an increase in non-randomness with ageing time; the Pearson value increases from 0.034 to 0.47 to 0.59 from the ice water quenched to 10 day aged to 49 day aged samples, as is shown in figure 3. For each sample, the Pearson coefficient was also calculated for the reshuffled dataset. The reshuffled dataset contains the same data points in the same spatial locations as the original dataset, except that the mass-to-charge labels of the points have been randomly reshuffled. This means that when the calculation is repeated on this reshuffled data, any non-randomness detected is due to density artefacts in the atom probe data, and not true chemical clustering. The reshuffled dataset can thus be viewed as a control dataset. Ideally, the μ value of the reshuffled dataset should be close to zero, indicative of a random solid solution, which it is in all 3 cases.
Isoconcentration surfaces were used to extract precipitates, using 1 nm voxels with a delocalisation value of 3 nm. The isoconcentration analysis was performed on both the original ('experimental') dataset and the reshuffled dataset. In both of the aged samples, the number of precipitates found in the reshuffled dataset was negligible. The ice water quenched sample was not found to contain precipitates. The precipitate ions extracted from the 10 day aged and 49 day aged samples can be seen in figure 4. α 2 volume fractions were calculated using the Lever rule (eq. 1).
C n = Cα 2 Vα 2 +Cα (1 -Vα 2 ) Equation 1 where C n = overall composition of dataset; Vα 2 = α 2 volume fraction; Cα 2 = α 2 composition; Cα = matrix composition. This assumes a two phase system, such that Vα = 1-Vα 2. Rearranging eq. 1 and plotting C n -Cα against Cα 2 -Cα gives a straight line with gradient equal to α 2 volume fraction. It can be seen from figure 5 that the gradient of the Lever rule plot increases from 0.087 to 0.13 from the 10 day aged to the 49 day aged sample, showing that the volume fraction of α 2 increases with ageing time.  Figure 6 shows the APT-measured oxygen concentration in the bulk (whole APT needle), precipitates and matrix for each of the IWQ, 10 day aged and 49 days samples. The bulk oxygen concentration is consistent between the three samples and for both of the aged samples, the oxygen concentration is higher in the matrix than the precipitates. The precipitate compositions for each of the aged samples are displayed in Table 1, and oxygen partition coefficients are displayed in Table 2.

Discussion
The hardness measurements reported in figure 1 show that as ageing me is increased, there is no sta s cally significant change in hardness. However, the increase in Pearson coefficient with increasing ageing me seen in figure 3, shows that the alloy is becoming increasingly non-random as ageing progresses. The high spa al resolu on of APT reveals that this increase in non-randomness is clearly due to precipita on, which can be seen in the atom maps in figure 4. The precipitates were iden fied as α 2 based on their composi on (Table 1) and based on the fact that the me and temperature of the ageing treatments lies within the α 2 phase field of the TTT curve in figure 7. The lack of sta s cally relevant hardness increase upon ageing suggests that α 2 precipita on is not making a significant hardness contribu on, in agreement with work by Liu et al. [5]  In both the 10 day aged and 49 day aged sample, the precipitates are substoichiometric. This could be because the ordered phase is defec ve, with tanium atoms on some aluminium sites, as described by Liew et al. [14] in their proposed condi onal spinoidal mechanism for α 2 forma on.
It has previously been suggested in the literature that oxygen par ons to the α 2 phase. Gehlen et al. [15] report that the octahedral inters al site where oxygen resides is larger in the Ti 3 Al la ce than the regular HCP la ce. Ardakani et al. [6] propose that this causes oxygen to par on to, and stabilise, the α 2 phase. In their work on aged Ti-8Al, Lim et al [9] also assume that oxygen must par on to and stabilise the α 2 phase based on hardness measurements. However, the current work has directly measured the oxygen concentra on in both the α 2 precipitates and the matrix, and found the oxygen preferen ally par ons to the matrix, as can be seen in figure 6. This agrees with the higher maximum solubility of oxygen in α [16] than in α 2 [17], [18], and is in line with α 2 precipitate composi ons independently measured using APT by Bagot et al. [19] in Ti-6Al-4V.
The oxygen par on coefficient calculated from the APT-measured α and α 2 composi ons reported by Bagot et al [19] is 1.41. The study by Bagot et al [19] was performed on Ti-64 aged for 24 hours at 800 o C , such that the α and α 2 phases were approaching their corresponding theore cal oxygen solubility limits (34 at. % [16] in α and 10-14 at. % in α 2 [17], [18]). Consequently, this oxygen par on coefficient value can be considered as an upper limit. Thus, the lower oxygen par on coefficients measured in this current study (table 2) suggest that the precipitates have not yet reached equilibrium. The quan fied increase in α 2 volume frac on from 0.087 to 0.13 with ageing me supports this, and suggests that the composi ons and volume frac on of both phases, sensi vely depend on not only exposure condi ons but also long-term ageing treatments. This has clear implica ons for understanding vital changes in mechanical proper es in any in-service components.

Conclusions
Precipitates formed during ageing of Ti-7Al containing 500 wppm O have been analysed using atom probe tomography (APT) and iden fied as the ordered α 2 phase, Ti 3 Al. The following conclusions can be drawn: Oxygen par ons preferen ally to the matrix rather than the α 2 precipitates. This is in agreement with previous APT studies [19] and with the theore cal solubility limits of oxygen in α and α 2 [16]- [18].
There is no sta s cally significant increase in hardness between the quenched and aged Ti-7Al + 500 wppm O sample, sugges ng α 2 precipita on is not providing a signinficant contribu on to hardening. α 2 forma on is dependent on both exposure condi ons and long term ageing treatments, as demonstrated by the increase in α 2 volume frac on from 0.087 to 0.13 from the 10 day aged to 49 day aged sample.
This study lays the founda on for future work currently being undertaken to be�er understand the role of oxygen in α 2 forma on, and to enable be�er predic on of embri�lement in these vitally important alloys for the aerospace industry.