Strong and ductile titanium–oxygen–iron alloys by additive manufacturing

Titanium alloys are advanced lightweight materials, indispensable for many critical applications1,2. The mainstay of the titanium industry is the α–β titanium alloys, which are formulated through alloying additions that stabilize the α and β phases3–5. Our work focuses on harnessing two of the most powerful stabilizing elements and strengtheners for α–β titanium alloys, oxygen and iron1–5, which are readily abundant. However, the embrittling effect of oxygen6,7, described colloquially as ‘the kryptonite to titanium’8, and the microsegregation of iron9 have hindered their combination for the development of strong and ductile α–β titanium–oxygen–iron alloys. Here we integrate alloy design with additive manufacturing (AM) process design to demonstrate a series of titanium–oxygen–iron compositions that exhibit outstanding tensile properties. We explain the atomic-scale origins of these properties using various characterization techniques. The abundance of oxygen and iron and the process simplicity for net-shape or near-net-shape manufacturing by AM make these α–β titanium–oxygen–iron alloys attractive for a diverse range of applications. Furthermore, they offer promise for industrial-scale use of off-grade sponge titanium or sponge titanium–oxygen–iron10,11, an industrial waste product at present. The economic and environmental potential to reduce the carbon footprint of the energy-intensive sponge titanium production12 is substantial.

Bidirectional scan strategy for DED and the resulting thermal pulses (scan speed: 800 mm/min; layer interval: 15 s, see Extended Data Table 2  Schematic extraction of tensile specimens (12 mm × 3 mm × 2 mm as per Australian Standard AS 1391-2007) from the printed coupons. Each tensile specimen is extracted from the middle nine layers of the coupon. The temperature variation is within 30 C across these nine layers by simulation. As a result, the microstructure is expected to be consistent. At least five coupons are printed for each condition.    -flecks in titanium alloys and avoidance of Fe-stabilised -flecks by DED

-flecks in titanium alloys
Most information about the -flecks in near- Ti alloys is based on observations from billet or forged products. Assuming a uniform reduction during billet conversion, the width of -flecks in commercial Ti alloy ingots is estimated to be 1-4.4 mm 11 . The length can be up to 10 times the width, i.e. up to a few centimetres. -flecks form in the late-stage solidification when the solid fraction is greater than 0.8 or 0.9, calculated from the measured composition of the -flecks using the Scheil model 12 . The distribution of the -flecks is not uniform. Fig. 1 of Ref. [13] shows several cross-sectional views of the -flecks (white patches) in a rolled and forged tensile sample of the Ti-10V-2Fe-3V alloy. The number density of the -flecks (> 50 m in size) is about 200/cm 2 .
Systematic experimental data revealed that these -flecks reduce the tensile ductility of Ti-10V-2Fe-3V by more than 50% and the low cycle fatigue strength by more than 90% [13]. Large flecks are difficult to remove by annealing.
As already shown in Extended data Fig. 1 Fig. 1d). This provides a unique opportunity to study its ̇. Supplementary Fig. 6 shows the dendritic prior- grains observed in a pore cavity on the tensile fracture surface of this alloy. The secondary dendrite arm spacing (2) is measured to be 13.83  0.81 m (excluding tertiary dendrites).
Supplementary Fig. 6 Dendritic prior- grains in a pore cavity of the Cu-mould cast Ti-0.35O-3Fe alloy sample. The image is taken from a tensile fracture surface of the alloy. It is generally accepted that "For cooling rates up to 10 3 K/s, local equilibrium with compositional partitioning between the liquid and solid phases at the solidification interface is maintained. The interface undercooling is small. However, when the cooling rate increases above 10 3 K/s, nonequilibrium solidification occurs" 23 , namely solute trapping occurs above 10 3 C/s. Note that the Scheil equation or CALPHAD-numerical-Scheil is only applicable when the assumption of local equilibrium is valid. Therefore, at the cooling rate of ̇ = 89 C/s, no solute trapping is expected. Consequently, a significant accumulation of Fe is expected to occur in the remaining liquid. This is the key reason for the formation of the Fe-stabilised -flecks in our Cu-mould cast It should be emphasised that the solidification of an ingot always starts from the mould walls and proceeds towards the centre. This sequential solidification process leads to continuous accumulation of Fe in the remaining liquid, conducive to the formation of -flecks. This is another important contributing factor (melt volume is therefore important, which in turn affects the ̇) .
Conversely, in the DED process, this cumulative effect is much weaker due to the small melt pool (1.08 mm 3 ), which corresponds to a much faster ̇ (> ~210 3 C/s, see below).

Approximate minimum ̇ to help avoid -flecks in the DED process used in this study
Before studying our Ti-O-Fe alloys, we first investigated the -fleck issue in three binary Ti-Fe alloys: Ti-3Fe, Ti-5Fe and Ti-7Fe (all containing 0.13-0.14 O after DED). Supplementary Table   5 lists the predicted cooling rates in Ti-3Fe rectangular coupons built on a 10 mm-thick Ti-6Al-4V plate under different DED conditions. Each sample has 25 layers. Due to the lack of similar data for Ti-Fe alloys, we used the temperature-dependent thermophysical data of Ti-6Al-4V.
Room temperature thermophysical property data cannot be used because these properties vary considerably between Troom and Tliquidus (e.g. 4-6 times) 24 .
As expected, ̇ decreases with increasing laser energy density but is influenced by the layer interval time. In each 25-layer build, the centres of the layers 18, 20 and 22 undergo slower cooling rates, with the slowest ̇ being predicted to be 1994 C/s.
To evaluate the influence of the cooling rate on the formation of Fe-stabilised -flecks, we printed rectangular coupons (40  10  5 mm 3 ; thickness: 5 mm) of Ti-3Fe, Ti-5Fe and Ti-7Fe alloys.
Each composition was printed with three sets of DED conditions, listed in Supplementary Table   6. Each printed sample was examined layer by layer. No -flecks were observed in any of these alloy samples. We first discuss the Ti-7Fe alloy ( Supplementary Fig. 7) and then briefly discuss the Ti-3Fe and

Supplementary
Ti-5Fe alloys ( Supplementary Fig. 8). All micrographs were taken from around the 18 th layer of each sample. The highest energy density (50 J/mm 2 , v = 400 mm/min) produced the coarsest -  Supplementary Fig. 7e-f). Supplementary Fig. 8 shows a brief view of the microstructures of the Ti-3Fe and Ti-5Fe alloys printed under three DED conditions. At higher magnifications, they all consist of ultrafine (100-250 nm thick) - lamellae.
In summary, the DED conditions investigated in Supplementary Table 5 were all successful in avoiding the -flecks, including in the Ti-7Fe alloy. All these DED conditions fall in the green zone of Fig. 1c. The approximate slowest ̇ identified by simulations for these alloy samples is Note that this approximate slowest solidification cooling rate (~2000 C/s) and those predicted for the layers of 18, 20 and 22 in each sample (2000 -3500 C/s, Supplementary Table 5) are clearly slower than the cooling rates (10 4 -10 5 C/s) indicated earlier for complete solute trapping.
Therefore, these cooling rates allow only partial solute (Fe) trapping to occur, reducing Fe accumulation in the remaining liquid, which helps avoid the formation of Fe-stabilised -flecks.
As emphasized earlier, the small melt pool (1.08 mm 3 ) in the DED process limits Fe accumulation in the remaining liquid (fs > 0.8) compared to ingot solidification. The actual accumulation of Fe is much weaker. This is another important factor to mitigate the formation of Fe-stabilised -flecks.
Finally, the multiple thermal pulses and significant cyclic heating effects of the DED process, unlike conventional annealing, may decompose some of the Fe-rich -phases. According to our literature review, this effect has not been investigated and in our opinion should not be neglected.
Therefore, we propose that the complete avoidance of Fe-stabilised -flecks in these Ti-  Supplementary Fig. 14).

Chemical homogeneity and microstructure uniformity
We used a laser spot size of 1.5 mm, which facilitates chemical homogeneity. Supplementary Fig.   9 shows half of a semi-ellipsoidal melt pool geometry resulting from the default DED conditions used in this work (laser power: 500 W; laser spot size: substances or elements, chemical inhomogeneities are expected if the sample size is limited to a few hundred mixed powder particles. However, when the "sample" size exceeds 6000 mixed powder particles, it can be assumed that after a long mixing time, the average composition of each such "sample" (> 6000 powder particles) is consistent or varies within an acceptable narrow range.
In addition, the melt pool depth is 0.6 mm (three layers). This means that each portion of the build will be remelted twice after the initial melting. This further helps to mitigate chemical inhomogeneity.
The above analyses were confirmed by our experimental measurements of the Fe content from different build heights shown in Supplementary Fig. 10. The measured compositions are consistent.
Among all the samples we have fabricated, only a small batch-to-batch variation was observed in the Fe content from 3.12% to 3.36 wt.% (Extended Data Table 1). Commercial - Ti alloys normally allow for a much wider range of variations for each  or  stabilizer, e.g. Ti-6Al-4V

Influence of porosity on tensile ductility
As previously indicated, our DED process (laser spot size: 1. This observation is similar for other samples reported in our Extended Data Fig. 2(a, b).
Metallographically, pores were rarely observed in polished cross-sections, as exemplified in Supplementary Fig. 14 for the Ti-0.35O-3Fe alloy that exhibited the lowest ductility.
As Supplementary Fig. 13a confirms, five pores in the size range of 16-46 m were randomly distributed on the entire tensile fracture surface of this lowest tensile ductility sample (f = 2.2 ± 0.6 %). Compared to the same alloy samples in Supplementary Fig. 13 b-c, this sample has the least and smallest pores, but the lowest tensile ductility. Clearly, porosity is not the major factor controlling ductility here.
Compared with Supplementary Fig. 13a, seven larger pores in the size range of 45-72 m were observed on the entire tensile fracture surface of the sample with f = 14.0 ± 0.7 % ( Supplementary   Fig. 13b). However, the tensile ductility was six times higher (more pores with larger sizes). This reaffirms that porosity is not related to the substantially low ductility of the sample shown in Supplementary Fig. 13a.
The last sample (the same alloy) shown in Supplementary Fig. 13c exhibited the highest tensile ductility (f = 21.9 ± 2.2 %). Six pores (red circles) in the size range of 20-65 m were observed on the fracture surface. Similarly, there were more pores with larger sizes than those shown in Supplementary Fig. 13a. However, the tensile ductility of this sample was almost 10 times higher.
Again, this indicates that porosity is not responsible for the substantially low ductility of the sample shown in Supplementary Fig. 13a. In fact, this high tensile ductility (21.9 ± 2.2 %) may suggest that spherical pores in the size range of 20-65 m (small quantity) are not significantly detrimental to the tensile ductility of these titanium alloys at the strain rates tested here.  Although this alloy exhibited the lowest tensile ductility (f = 2.2 ± 0.6 %), it is not due to porosity.
There are some irregular black dots in (a), which are not pores, but coarser -phase particles, as shown in (b) (right edge).  Table 2) for our simulations.

Melt pool shape and size
In other words, the basic input conditions for simulations are based on the single-track DED experiments obtained under various DED conditions.
As emphasized earlier, due to the lack of similar data for Ti-O-Fe alloys, we used the temperaturedependent thermophysical data of Ti-6Al-4V (the room-temperature data cannot be used because the thermophysical properties vary considerably between Troom and Tliquidus (4-6 times) 22 .
Therefore, to best assess the predictability of the Simufact Welding (DED) used for this study, we chose to compare the simulation results with the single-track DED experiments for Ti-6Al-4V.
As emphasized earlier, due to the lack of similar data for Ti-O-Fe alloys, the simulations used the temperature-dependent thermophysical data of Ti-6Al-4V (the room temperature data cannot be used because the thermophysical properties vary considerably between Troom and Tliquidus (4-6 times) 22 . Therefore, to best assess the predictability of the Simufact Welding (DED) used for this study, we chose to compare the simulation results with the single-track DED experiments for Ti-6Al-4V.
It is noteworthy that the density of Ti (Ti) increases profoundly when cooled from the melt pool temperature (up to 3100 °C by simulation for DED) to room temperature (RT, 4.51 g/cm 3 ). The authors of Ref. [28] reviewed the density measurements of molten Ti and experimentally determined the following relationship for the density of molten Ti up to 2127 C (2400 K) Ti = 4.14 -2.1510 -4 (T -Tm) -3.7110 -8 (T -Tm) 2 (2) The latest study of the density of molten Ti (up to 1817 C) 29 is consistent with Eq. (2).
Extrapolating Eq. (2) yields Ti = 3.756 g/cm 3 at 3100 C (Tm = 1668 C). This means a 20% increase in density when cooled to RT. Therefore, the simulated melt pool volume is expected to be at least 20% larger than that observed at RT. In other words, if the difference is within 20-30%, it should be considered highly consistent. Our Supplementary Fig. 15 and Table 7 confirm the high predictability (on the micron length scale) of the DED simulation module used in this work.

THERMAL COOLING
The above assessment can be used as a valid evaluation of the predicted thermal cooling predictions. Direct and accurate measurements of the cooling rate (solidification or solid state) remain a challenge for a typical DED process due to the small melt pool and the layer additive manufacturing nature (the location of interest does not pre-exist).
An indirect assessment of the cooling rate for DED of Ti alloys is to use the relationship between the secondary dendrite arm spacing (2) or the prior- grain size (1) and the solidification cooling rate ̇. Unlike the Cu-mould cast ingots ( Supplementary Fig. 6), no prior- dendrites were observed in any DED-fabricated Ti-O-Fe alloy samples of this work (all being elongated or equiaxed prior- grains). Therefore, we focus on the prior- grain size 1.
Broderick et al. 19 have established that the prior- grain size 1 (m) of Ti-6Al-4V under rapid solidification can be described as a function of the solidification cooling rate ̇ (K/s), i.e.  Table 5, the exponent of ̇ was found to be around 1.15. Therefore, we used the following Eq. (4) to further evaluate the solidification rate ̇ of our Ti-O-Fe alloys studied in this work and then compare the results with Simufact Welding (DED) predictions: We measured the prior- grain size in the surface layer of the Ti-0.35O-3Fe alloy samples printed under four sets of DED conditions. Representative prior- grain structures observed in the surface layer of these samples are shown in Supplementary Fig. 16 for each selected DED condition, while Supplementary Table 8 summarises the predictions and measurements.
The equiaxed prior- grains are not uniform in the surface layer of each sample, Supplementary   Fig. 16, featured by a large standard deviation. We therefore focused on the mean prior- grain size. As shown in Supplementary Table 8, in each case, the prior- grain size obtained from Eq.  Table 2.

Advantages of AM in producing the designed Ti-O-Fe alloys
Compared with ingot metallurgy-based manufacturing techniques and shape casting, the DED (laser powder) process investigated offers the following advantages: • • Capability and flexibility of tuning mechanical properties within a broad processing window or along the build height or across the wall thickness through adjusting the scan speed and/or layer-to-layer interval time.
Sintering-based net-shape or near net-shape powder metallurgy (PM) processes, including conventional powder metallurgy (C-PM), metal injection moulding (MIM) and hot isostatic pressing (HIP), can all avoid -fleck formation by using either elemental powder blends (for C-PM) or pre-alloyed spherical powder (for MIM and HIP). However, the high temperature pressureless sintering process (1200-1300C for 1-4 hours) for C-PM and MIM usually leads to high residual porosity (~2 vol.%), coarse  grains (due to long isothermal holding), and coarse − lamellae (due to the subsequent slow furnace cooling). Consequently, the resulting tensile mechanical properties are usually only comparable to their as-cast counterparts.
HIP is a possible option but not comparable to AM in terms of net shape formation. In addition, our HIP experience with Ti-6Al-4V indicates that even with a low HIP temperature (820 C) and a high HIP pressure (200 MPa), HIP still yields very thick -laths (~2 m thick), resulting in low tensile strengths. Therefore, AM is the best net-shape fabrication method for these alloys which can also ensure excellent or outstanding tensile properties.

The -phase fraction in Ti-O-Fe alloys
Increasing the O content resulted in an increase in the -phase fraction in our Ti-O-Fe alloys ( Supplementary Fig. 2). This is due to the partitioning effect of O and Fe between the  and  2) Aerospace, marine, defence, chemical processing, pulp and paper production (where austenitic stainless steels only last for 2~3 months) sectors We limit our predictions to non-fatigue critical applications at this point of time.
Due to their excellent tensile properties in the as-fabricated state, we envisage that these simple Ti-O-Fe alloys, when manufactured in net or near-net shapes by DED, will be attractive and competitive for a broad range of room temperature structural applications that currently use Ti-6Al-4V or ATI 425® or Ti-3Al-2.5V across various sectors. In addition, since our as-fabricated Ti-0.35O-3Fe alloy has already reached the tensile properties of cold-rolled and hot-rolled ATI 425® for ballistic applications, it may be evaluated for ballistic applications as well. Also, they do offer some cost advantages (see below). The volume of the off-grade sponge Ti accounts for 10-20% of the total sponge production [32][33][34] (an average of 10% of the sponge production is reasonable). Due to their high O and Fe content, this substantial amount of off-grade sponge Ti is currently used as the raw material of ferrotitanium for the steel industry [32][33][34] or even to make fireworks.
Sponge titanium production is extremely energy-intensive (423 GJ/ton) 35 , which is about 5-10 times the energy consumption of primary aluminium production (which is already considered energy-intensive). It is therefore important to revitalize these off-grade sponge Ti materials. Additionally, high-oxygen scrap CP-Ti Grade 3 and Grade 4 can be used in the same way for much higher value creation.

Implications for sectors other than AM
With a focus on net-shape manufacturing, the findings of this study are best applied to the metal AM sector. Without considering net-shape manufacturing (high buy-to-fly ratio), we envisage that our − Ti-O-Fe alloys can be manufactured with significant tensile properties through the combination of PM, elaborate thermo-mechanical processing and machining.
First, ingots of Ti-O-Fe alloys can be made from hydride-dihydride (HDH) high-oxygen Ti powder with Fe powder by cold isostatic pressing (CIP) and sintering. Second, through β-field forging + multi-axial α−β field forging + extrusion/rolling, the pre-sintered Ti-O-Fe alloy ingots can be manufactured into billets or plates with fine equiaxed  grains (20-40 m, much finer than the as-deposited, article file Fig. 1d-1g). Finally, heat treatments could be explored to produce fine and short α−β lamellae in the fine equiaxed  grains. These Ti-O-Fe microstructures are expected to possess significant tensile properties. Machining will be the ultimate net-shape formation process. The allowed use of high-oxygen HDH Ti powders (low value powder) based on this work is a significant advantage.
Integrating alloy design with AM process design via high-fidelity simulations has allowed us to This work offers a promising approach to revitalise large amounts of off-grade sponge Ti (which is sponge Ti-O-Fe) and scrap high-oxygen Ti. These materials could be utilized as feedstock for powder production for AM. Sponge Ti production is highly energy-intensive (5-10 times that of primary aluminium production). Utilising off-grade sponge Ti represents a revitalisation with significant economic and environmental benefits. The same is expected for similarly produced off-grade sponge zirconium (Zr).
Oxygen embrittlement occurs not only in HCP and BCC Ti, but also in other BCC metals (e.g. Nb 36 and Mo 37 ), presenting a significant metallurgical challenge. However, here we reveal a unique distribution of oxygen in these AM-fabricated − Ti-O-Fe alloys. The high tensile ductility of these high-oxygen Ti-O-Fe alloys (f = 21.9 ± 2.2%) provides a template for future interstitial engineering in AM-fabricated alloys.
We also discuss the pathfinding potential of this study. For example, zirconium (Zr), exhibits similar physical metallurgy to Ti, and similar metallurgical design approaches could be explored for the development of a new class of strong and ductile − Zr-O-Fe alloys.
Moreover, a new interstitial engineering opportunity could arise from the results demonstrated here through AM. For instance, similar to oxygen, nitrogen embrittlement also occurs in Ti. As a result, N is tightly controlled to  0.05%N by conventional manufacturing ( 0.05%N), even though N is more potent than O as both -phase stabiliser and -phase strengthener.
We have shown that the unique partitioning of oxygen and iron in the two phases ( and ) is fundamental to the success of these Ti-O-Fe alloys (a suitable -phase volume fraction and the suppression of -flecks are also important). Following this pathfinding strategy, systematic DFT calculations and predictions could be performed to map out the combination of N with each stabiliser (Mo, V, Cr, Fe, Mn, Ni, Co, Nb, Ta, and W) for similar partitioning in the  and  phases.
These combinations could lead to the design of potentially attractive new − titanium alloys.
The same approach could be applied to nitrogen in zirconium.
Therefore, this work is not only highly significant in terms of our report of a new class of strong and ductile − Ti-O-Fe alloys with unique partitioning of O and Fe, but also it offers a potential pathfinding approach to various other new alloy systems.

Repeatability of tensile stress-strain curves
In this section, we briefly analyse the repeatability issue of the tensile stress-strain curves produced in this work (Extended Data Fig. 2) by focusing on the Ti-0.14O-3.23Fe alloy, whose tensile stress-strain curves displayed the widest gap. Let us refer to Nonetheless, the Sr value should be as small as possible.
As shown in Supplementary Note 2, the influence of significant chemical inhomogeneity can be excluded. To further substantiate this point, we have deposited the Ti-185 (Ti-1Al-8V-5Fe) alloy using the same DED system with top-quality pre-alloyed Ti-185 powder ordered from a commercial supplier, produced using a high-speed (17,000 rpm) plasma rotating electrode process