Recent innovations in laser additive manufacturing of titanium alloys

Up until now, the most common ways to optimize process parameters are based on trial-and-error methods, e.g. energy density criteria or processing window mapping. However, due to the complex interactions between the processing parameters and their effects on the material density, it is time-consuming and costly to optimize the processing parameters through trial-and-error methods. Machine learning has been shown as a suitable alternative technique for directly modelling the underlying relationship between various process parameters and part properties to address this challenge [32].

Machine learning models can model the relationship between process parameters and as-build part properties, such as relative density [33]. Maitra et al [34] adopted a Gaussian process regression (GPR) model for predicting the relative density of LPBF Ti-6Al-4V alloy. The inputs of the GPR model are five process parameters (i.e. P, v, h, t and E_V ), and the output is the relative density. Through training the GPR model with 2 900 data points collected from previous references, the relative density of LPBF Ti-6Al-4V alloy can be predicted accurately by inputing the process parameters (with a mean absolute error of 1.12%). The relationship between process parameters and relative density has been modelled as a second-order polynomial equation in their study, as shown in equation (5). Similarly, Nguyen et al [35] trained an artificial neural network (ANN) model to predict the relative density of LPBF Ti-6Al-4V alloy via using four independent process parameters (P, v, h and t). By jointly considering the deposition efficiency and the relative density, the optimal LPBF process parameters for Ti-6Al-4V alloy are determined as laser power of 180 W, scan speed of 900 mm·s⁻¹, layer thickness of 20 μm and hatch spacing of 80 μm. The experimental result indicated that a fully dense part with a relative density of 99.8% could be achieved under the optimal process parameters, demonstrating the effectiveness of the ANN prediction method. Cacace and Semeraro [36] proposed a fast optimization procedure for selecting L-PBF parameters based on a semi-analytical thermal model and a geometric-based defect model, aiming to achieve a balance between good solidification and productivity. The methodology is validated by producing AISI 316L specimens and demonstrates reliable performance in an industrial LPBF system. Gong et al [37] investigated the process-structure-property (PSP) linkages in the LAM of Ti-6Al-4V metal parts through a novel ML model, aiming to understand and predict machining behaviour. The model leverages reduced-dimensional representation of material characterization data, validated through various experimental techniques like scanning electron microscopy and x-ray diffraction. The study achieved extremely high prediction accuracy (>99%, statistically significant at a 95% confidence interval) and offers insights that can be integrated with existing models to optimize material behaviours such as machining, fatigue, and corrosion resistance. In terms of β-Ti alloy, Shin et al [38] adopted the ANN model for the processing parameters optimization of LPBF Ti-5Al-5V-5Mo-3Cr alloy. The inputs of the ANN model were P, v and h, and the output was the relative density. By training the ANN model with 60 experimental datasets, the ANN model showed a highly accurate performance on relative density prediction, as shown in figures 4(a) and (b). In addition, Lim et al [39] proposed a novel method for optimizing the processing parameters of LDED Ti-6Al-4V alloy via the single-track deposited surface colour and machine learning, as shown in figure 4(c). Through 135 single-track experiments with different combinations of laser power and scan speed, six surface colours were observed, including deep blue, blue-white, blue, brown, gold and silver, and the blue-white was determined as the best surface color. Validation experiment results indicated the random forest (RF) model exhibits excellent predictive performance on the single-track surface color, which could help workers to select appropriate manufacturing conditions.

$$\begin{align}\begin{aligned} Density & = 111.92993 + 0.027478P - 0.000986v\\ & \quad - 0.027404h - 0.42701t - 0.046009{E_V}\\ & \quad - 0.000173{P^2} - 0.00000160008{v^2}\\ & \quad + 0.000027{h^2} - 0.000884{t^2}\\ & \quad + 0.000047{E_V}^2 + 0.000022Pv\\ & \quad - 0.000112Ph + 0.001896Pt\\ & \quad - 0.000121P{E_V} - 0.000000529651vh\\ & \quad - 0.00011vt + 0.000015v{E_V} \hfill \\ & \quad + 0.000024ht + 0.000274h{E_V}\\ & \quad - 0.000201t{E_V} \hfill \\ \end{aligned} \end{align} \tag{ 1 }$$

Zoom In Zoom Out Reset image size

Machine learning can also optimize the processing parameters to achieve better mechanical performance. Figure 4(d) depicts the relative importance of parameters on the ductility of LPBF Ti-6Al-4V alloy calculated by the RF model [40]. Among all the parameters considered by Yap et al [40] to process Ti-6Al-4V by LPBF, they found that heat treatment temperature and hatch distance (h) are the most important parameters governing the ductility of the alloy, while post-heat treatment generally weakens its strength. Moreover, the partial dependence plot (figure 4(e)) predicted by the RF model indicates that the ductility initially increases and then decreases with increasing h. Considering this information provided by the RF model, the mechanical properties of LPBF Ti-6Al-4V are enhanced significantly by tailoring the h deliberately, as shown in figure 4(f). In addition, decision tree and ANN were used to classify various types of volumetric defects such as lack-of-fusion (LOF), gas-entrapped pores, and keyholes in LPBF Ti-6Al-4V alloy [41]. Morphological features such as dimensions, roundness and aspect ratio were extracted, and machine learning (ML) models can predict defect types with 99% accuracy. Advanced ML methods, including deep learning, have also shown promising results in the recent literature for processing parameter optimization. Kim et al [42] developed a conditional generative adversarial network (CGAN) to generate virtual surface morphology of Ti-6Ai-4V parts manufactured by LDED under given process parameters. The proposed AI guidance system can result in an optimized smooth surface design. The author illustrated that AI-recommended virtual surface morphology can aid in producing high-quality parts at a low cost.

Moreover, machine learning can create surrogate models for multiscale-multiphysics mechanistic simulations, constructing digital twins for the AM process. High-fidelity mechanistic models can reveal comprehensive information regarding the underlying complex physical mechanisms. However, the simulations are often computational-intensive and time-consuming, which creates significant barriers for the industry end-user. For multi-track and multi-layer samples [43], FEM-based numerical simulations for thermal history dataset construction can take up to three months. To resolve this challenge, Gunasegaram et al [44] discussed the exceptional value of digital twins in AM, highlighting the need for high-fidelity multiscale-multiphysics models and their faster-solving surrogates using ML approaches for real-time problem-solving.

Even when optimal processing parameters have been determined, the importance of in-situ monitoring and closed-loop control cannot be underestimated for ensuring the successful fabrication of high-quality, defect-free parts in LAM. Despite using optimal processing parameters, defects such as distortions, porosity, cracks, and microstructural heterogeneity may still occur as a result of factors such as inconsistent printing speeds [45] and localized heat accumulation [46, 47]. To guarantee a successful build and minimize defects, real-time monitoring of the LAM process is vital. This is achieved through the use of various sensing techniques (e.g. visual, thermal, acoustic, and x-ray imaging.) that gather data from the process [48]. By integrating machine learning models with sensor signals, early detection of potential defects becomes feasible, allowing for the immediate suspension of the process if defects are detected [49]. This proactive approach helps prevent further deterioration of quality and potential build failures. Numerous sensing techniques and machine learning methods have been implemented for the in-situ monitoring of LAM processes involving Ti alloys. Khanzadeh et al [50] proposed a novel in-situ porosity detection method based on the temperature distribution of the top surface of the melt pool. The method utilized Self-Organizing Maps (SOMs) to analyze two-dimensional melt pool image streams for identifying similar and dissimilar melt pools. X-ray tomography was employed to experimentally locate porosity within a Ti-6Al-4V thin-wall specimen, which was then compared with the predicted porosity locations based on the melt pool analysis. Results showed that incorporating thermal distribution significantly improved the accuracy of porosity prediction, with the proposed method predicting the location of porosity almost 96% of the time when the appropriate Self-Organizing Map model was selected. Wolff et al [51] introduced a piezo-driven powder deposition system for powder-blown LDED, allowing for imaging of individual powder particles flowing into a scanning melt pool. They performed the first-of-its-kind in-situ high-speed x-ray imaging of the powder-blown LDED process of Ti-6Al-4V powder particles, revealing how laser-matter interaction influences powder flow and porosity formation. On top of this system, Ren et al [52] developed a machine learning–aided approach for detecting stochastic keyhole porosity generation events in LPBF of Ti-6Al-4V. They used simultaneous high-speed synchrotron x-ray imaging and thermal imaging, coupled with multiphysics simulations, to discover two types of keyhole oscillation. Their approach achieved submillisecond temporal resolution and a near-perfect prediction rate, demonstrating a practical method for adopting the in-situ defect detection approach in commercial systems. Furthermore, multisensor monitoring offers enriched insights into the complex dynamics of LAM processes, thereby boosting defect detection accuracy [53]. Yet, the heterogeneity of data, sensor noise characteristics, and the varying relevance of different sensors to quality parameters complicate multi-sensor system deployment.

Machine learning-assisted in-situ monitoring is critical in both LPBF and LDED for real-time defect detection and process optimization [54]. However, the two methods differ regarding material build manner and rate, thermal history, and environment, which may necessitate different sensor types and ML modelling techniques. Several factors need to be considered when determining the applicability of in-situ monitoring for LPBF and LDED. For example, acoustic noise, a key impediment in acoustic-based monitoring, manifests differently in the two processes. LPBF predominantly encounters noise from the protective gas flow, which is mostly low-frequency and readily filterable [55]. In contrast, LDED grapples with a more complex noise content due to the impact of protective gas flow, powder stream and platform movement via different axes, thereby complicating acoustic signal analysis [56]. In addition, the two processes also differ in their need for temporal resolution, particularly for vision-based monitoring. In LPBF, where laser scan speed is much higher, sensors must have superior temporal resolution to capture subtle changes in the melt pool. Standard coaxial CCD cameras with sampling rates between 30 Hz and 100 Hz, often adequate for LDED, may fall short in LPBF. High-speed cameras in LPBF can even reach acquisition rates up to 10 kHz to capture subtleties when speed inconsistencies arise [45]. However, higher sampling frequencies present challenges in efficient data management. Another essential distinction lies in the post-process validation methods employed for in-situ monitoring applications. For training data and model development, LDED studies often utilize single tracks or single bead walls. In contrast, LPBF typically fabricates small cubes for this purpose. The subsequent defect analysis techniques also diverge: LPBF commonly uses x-ray imaging to identify porosity defects, whereas LDED commonly employs cross-sectional optical microscopy. An alternative approach for defect annotation in both processes involves employing a separate visual camera to monitor the plume and spatters. The severity of these spatters and the area covered by the plume can indicate the build quality [57].

Despite these challenges and differences, LPBF and LDED share core physical phenomena like melting and solidification. This similarity offers the potential transferability for the ML model between LPBF and LDED [58]. Transfer learning techniques can leverage LPBF's advanced AI applications to inform less mature processes like LDED. For instance, a machine learning model designed for porosity detection in LPBF could be adapted for LDED with some recalibration and retraining. Conversely, techniques like high-speed x-ray imaging presented by Wolff et al [51], more specific to LDED's powder-blown deposition, may not translate directly to LPBF.

Closed-loop control systems can also be established based on in-situ captured sensor data. For example, Gibson et al [59] demonstrated multiple modes of closed-loop melt pool size control in laser-wire-based DED of Ti-6Al-4V. They reported a closed-loop melt pool size control method through laser power modulation for intralayer control of bead geometry. They also developed a controller that modulates the print speed and deposition rate on a per-layer basis, allowing control of average melt pool size or average laser power in coordination with real-time melt pool size control. This research demonstrate that accumulated heat in components under deposition can be exploited to maintain process stability as print speed and deposition rate are automatically increased under closed-loop control, with significant implications for overall production efficiency.

The advancements in in-situ monitoring and closed-loop control have significantly enhanced the reliability and quality consistency of LAM-produced Ti alloys. Table 1 summarises the state-of-the-art ML-assisted process optimization method adopted for LAM of Ti alloy. By leveraging state-of-the-art sensing techniques and machine learning methods, researchers have demonstrated the ability to detect and mitigate defects in real time, preventing potential build failures and ensuring optimal part quality. As the field continues to evolve, further development and refinement of these monitoring and control systems will play a critical role in the widespread adoption of LAM for the fabrication of high-performance components across various industries. Ultimately, the integration of advanced machine learning-assisted in-situ monitoring and closed-loop control systems will pave the way for more efficient, reliable, and cost-effective LAM solutions.

Herein, eutectoid (α + β)-Ti alloys refer to Ti alloys with β-eutectoid elements addition. The β-eutectoid elements (e.g. Ni, Co, Cu, Mn, Fe, Cr) are both effective grain refiners for Ti alloy due to their high grain restriction factor (Q) (table 2). According to the Hunt criterion, the β-eutectoid element addition is beneficial for achieving columnar to equiaxed grain transition (CET) in Ti alloys [64]. Alloying with solutes of large Q values also results in the rapid development of the amount of constitutional supercooling in front of the growing solid that provides the nucleation undercooling (ΔT_cs) [65], restriction of columnar grain growth and promotion of heterogeneous nucleation at the solidification front. As a result, considerable work has been done in recent years to investigate the effects of eutectoid elements addition on LAM Ti alloys. Note that the studies on the LAM of eutectoid Ti alloys mainly focused on hypoeutectoid Ti alloys. The effects of eutectoid elements addition on microstructure and properties of LAM Ti alloys are overviewed systematically in this section. The mechanical properties of LAM-processed eutectoid (α + β)-Ti alloys are summarized in table 3.

Table 3. Tensile properties of customized (α + β) Ti alloys for LAM. The increments are compared with the base alloy (without eutectoid elements added) prepared with the same processing parameters.

Alloying element	LAM	Composition	Microstructure	YS/MPa	YS increment	UTS/MPa	UTS increment	/%	increment	References
Cu	LDED	Ti-3.5Cu	α + Ti2Cu	747	—	867	—	14.9	—	[65]
		Ti-6.5Cu	α + Ti2Cu	964	—	1 073	—	5.5	—
		Ti-8.5Cu	α + Ti2Cu	1 023	—	1 180	—	2.1	—
	LPBF	Ti-1.25Cu	α'+ Ti2Cu	724	—	842	—	12.0	—	[72]
		Ti-2.5Cu	α + Ti2Cu	826	—	1 002	—	5.3	—
		Ti-5Cu	α + Ti2Cu	1 036	—	1 242	—	2.5	—
	LPBF	Ti-6Al-4V	—	—	—	1 243	—	6.6	—	[73]
		Ti-6Al-4V-1.38Cu	—	—	—	1 550	+24.7%	4.5	−31.4%
	LDED	Ti-6Al-4V	α + β	—	—	1 030	—	14.5	—	[74]
		Ti-6Al-4V-5.4Cu	α + β + Ti2Cu	—	—	1 614	+56.7%	13.7	−5.5%
		Ti-6Al-4V-6.8Cu	α + β + Ti2Cu	—	—	967	−6.1%	4.9	−66.2%
Ni	LPBF	Ti	α'	670	—	745	—	18.5	—	[75]
		Ti-0.4Ni	α'	1 020	+52.2%	1 120	+50.3%	8.8	−52.4%
	LDED	Ti-5Ni	α + Ti2Ni + discontinuous αGB	750	—	937	—	5.4	—	[76]
		Ti-5Ni (800 °C/24 h)	α + α'+ Ti2Ni	1 040	—	1 225	—	5.4	—
	LDED	Ti-6Al-4V	α + β	1 034	—	1 144	—	8.1	—	[77]
		Ti-6Al-4V-0.9Ni	α + β	1 310	+26.6%	1 356	+18.5%	5.2	−35.8%
	LDED	Ti-6Al-4V	α + β	939	—	1 054	—	17.6	—	[78]
		Ti-6Al-4V-1.1Ni	α + β + Ti2Ni	998	+6.2%	1 079	+2.4%	9.4	−46.6%
		Ti-6Al-4V-1.7Ni	α + β + Ti2Ni	1 080	+15.0%	1 172	+11.2%	8.3	−52.8%
		Ti-6Al-4V-2.5Ni	α + β + Ti2Ni	1 155	+23.0%	1 228	+16.5%	4.4	−75.0%
	LPBF	Ti-1.2Ni-0.06B	Equiaxed α	—	—	970	—	10.7	—	[79]
Co	LDED	Ti-6Al-4V	α + β	1 131	—	1 184	—	2.7	—	[80]
		Ti-6Al-4V-5CoCrMo	α + β + continuous αGB	1 246	+10.2%	—	—	—	—
		Ti-6Al-4V-10CoCrMo	β	957	−15.4%	1 029	−13.1%	13.7	+407%
Fe	LPBF	Ti-5Fe-0.2Y	α'+ α + β	—	—	700	—	0.7	—	[81]
		Ti-5Fe-0.2Y (500 °C/1 h)	α'+ α + β	870	—	945	—	4.5	—
		Ti-5Fe-0.2Y (760 °C/1 h)	α'+ α + β	780	—	865	—	12.2	—
	LDED	Ti-2Fe-0.1B	α + β + TiB	756	—	779	—	18.0	—	[82]
	LDED	Ti-6Al-4V	α + β	1 053	—	1 132	—	7.0	—	[83]
		Ti-6Al-4V-1Fe	α + β	1 070	+1.7%	1 152	+1.9%	10.5	+50%
		Ti-6Al-4V-2Fe	α + β	1 226	+16.5%	1 294	+14.4%	8.0	+14.3%
		Ti-6Al-4V-3Fe	α + β	1 299	+23.5%	1 353	+19.6%	5.3	−24.3%
		Ti-6Al-4V-4Fe	α + β	1 413	+34.3%	1 457	+28.8%	3.0	−57.1%
Cr	LDED	Ti-6Al-4V	α + β	1 053	—	1 132	—	7.0	—	[83]
		Ti-6Al-4V-1Cr	α + β	1 045	−0.6%	1 121	−0.9%	8.0	+14.3%
		Ti-6Al-4V-2Cr	α + β	1 113	+5.9%	1 190	+5.2%	7.2	+2.9%
		Ti-6Al-4V-3Cr	α + β	1 246	+18.6%	1 323	+17.0%	5.4	−22.9%
		Ti-6Al-4V-4Cr	α + β	1 340	+27.5%	1 404	+24.1%	4.1	−41.4%
CoCrNi	LDED	Ti-6Al-4V	α + β	1 059.1	—	1 166	—	6.0	—	[84]
		Ti-6Al-4V-1CoCrNi	α + β	1 260	+19.0%	1 336	+14.6%	5.1	−14.9%
316 l	LPBF	Ti-6Al-4V-4.5(%)316l	α' +β	984	—	1 297	—	8.8	—	[85]
Re	LPBF	Ti	α'	461	—	592	—	21	—	[86]
		Ti-0.5Re	α'	653	+41.6%	781	+31.9%	14.6	−30.5%
		Ti-1.0Re	α'	770	+67.0%	906	+53.0%	10.2	−51.4%
		Ti-1.5Re	α'	905	+96.3%	1 095	+85.0%	4.7	−77.6%

Ni is a typical active eutectoid-forming element with a large Q (Q_Ni = 14.3c₀ ) and is therefore an effective grain refiner for Ti alloys. As a result, Ni alloying is of great interest for LAM of Ti alloys [75–78]. According to the phase diagrams, the maximum solubility of Ni in α-Ti is only 0.3 wt% [87]. Excessive addition of Ni could lead to the formation of the intermetallic phase Ti₂Ni via the following eutectoid reaction (occurring at ∼760 °C), which thereby deteriorates the ductility:

$$\begin{align}\beta \to \alpha + T{i_2}Ni.\end{align} \tag{ 2 }$$

Xiong et al [75] reported that the addition of 0.4 wt% Ni into LPBF Ti enhanced the strength significantly without the formation of the undesirable Ti₂Ni phase. When the Ni content reaches 1.6 wt% and 3.0 wt%, however, the brittle eutectoid Ti₂Ni phase is formed, reducing the ductility significantly. For LDED, on the other hand, Narayana et al [76] reported that the LDED Ti-5Ni alloy still shows a good ductility of 5.4%. The reason for the difference between these two studies could be that the lower cooling rates during LDED retards the martensite transformation and leads to more β phase formation than that in LPBF. Due to the much higher solubility of Ni in β-Ti than in α-Ti, the formation of brittle Ti₂Ni can be inhibited efficiently in the LDED-processed Ti-Ni alloy. After a sub-critical annealing treatment (800 °C/24 h + WQ), the LDED Ti-5Ni alloy transformed into a pre-eutectoid α + α' + Ti₂Ni microstructure. Under this circumstance, a superior tensile strength (1 225 MPa) and proper elongation (5.4%) can be achieved [76]. Furthermore, adding B into LAM Ti-Ni alloy can further provide extra-constitutional supercooling, thereby further refining the prior-β grains and enhancing the mechanical performance [79, 88]. In addition, Ni addition has also been used to modify the LDED Ti-6Al-4V alloy [77, 78]. The LDED Ti-6Al-4V alloy has long been known to suffer from the formation of coarse and columnar prior-β grains because Al and V almost do not contribute to grain refinement [89]. Sui et al [77] used the calculation of phase diagrams (CALPHAD) calculations combined with the Scheil solidification model to predict the solidification behaviour of LDED Ti-6Al-4V-xNi alloys, and they indicated that the Ti₂Ni phase form when the Ni content reaches approximately 1.0 wt%. The experimental results demonstrated that the microstructure of the LAM Ti-6Al-4V could be highly refined with 0.9 wt% Ni addition. The LDED Ti-6Al-4V-0.9Ni alloy possesses an ultrahigh yield strength (1 309 MPa) and proper ductility (5.2%) without forming undesirable intermetallic phases. In another work, Sui et al [78] investigated the influence of Ni content on the microstructure and mechanical properties of LDED Ti-6Al-4V-xNi alloys, as illustrated in figure 6. The introduction of Ni led to a substantial refinement of both the prior-β grains and α lath. Nevertheless, owing to the formation of the brittle Ti₂Ni Laves phase within the Ti-6Al-4V-xNi alloys, a noteworthy reduction in alloy ductility was observed.

Zoom In Zoom Out Reset image size

Co is also an effective eutectoid-forming element for Ti alloy with a large Q value (Q_Co = 8.8c₀ ). Choi et al [80] incorporated 5 wt% and 10 wt% of Co-28Cr-6Mo (CCM) into Ti-6Al-4V via in-situ alloying during LDED. No intermetallic phases were found in the Ti-6Al-4V-5CCM and Ti-6Al-4V-10CCM alloys, which is ascribed to the fact that the β phase is stabilized by the synergetic addition of Co, Cr and Mo, and Co has a much higher solubility in β (17.3 wt%) than in α (1 wt%). Therefore, the undesirable intermetallic phase formation is suppressed successfully. With the addition of 10 wt% CCM into Ti-6Al-4V alloy, a full β microstructure with equiaxed morphology was achieved (figures 7(a) and (b)). Under this circumstance, favourable isotropic tensile properties and superior ductility (13 ∼ 15%) were achieved, as shown in figure 7(c).

Zoom In Zoom Out Reset image size

Cu is also a typical eutectoid-forming element in Ti alloys (Q_Cu = 6.5c₀ ), and the eutectoid reaction occurs at ∼792 °C [90].

$$\begin{align}\beta \to \alpha + T{i_2}Cu.\end{align} \tag{ 3 }$$

Due to the rapid diffusion of Cu in Ti, the eutectoid reaction is hard to inhibit during LAM. Zhang et al [65] investigated the microstructure and mechanical properties of LDED Ti-xCu (x = 3.5, 6.5 and 8.5 wt%) alloys. An ultrafine eutectoid α + Ti₂Cu lamella microstructure with equiaxed prior β grains was obtained. The significant grain refinement mainly originates from the considerably high Q of Ti-Cu alloys, providing sufficient constitutional supercooling during solidification. Moreover, the intrinsic cycling heat input during LDED also promoted the formation of eutectoid lamella via reverse martensitic transformation. In addition, the 9R phase along with nano-twins has been observed in an LPBF Ti-5Cu alloy, and the reason was attributed to the Cu segregation at the lath boundaries, which reduced the local stacking fault energy [91]. Both the YS and UTS of the LPBF and LDED Ti-Cu alloys can be enhanced with Cu addition; however, the elongation is decreased simultaneously, especially for the hyper-eutectoid Ti-xCu alloys (e.g. Ti-8.5Cu) since the coarse hyper-eutectoid Ti₂Cu particles reduce the ductility significantly due to the stress concentration in these particles. Cu has also been incorporated into Ti-6Al-4V alloy via LPBF and LDED [73, 74, 92]. Li et al [92] observed a notable refinement of the α' lath structure in LPBF Ti-6Al-4V upon the introduction of Cu, as shown in figures 8(a)–(f). Concurrently, the LPBF Ti-6Al-4V-xCu alloys exhibited the formation of high-density nano-twins, which can be attributed to the elevated thermal stresses encountered during the LPBF process. Due to the refined α' lath and high-density nano-twins, the tensile strength of LPBF-built Ti-6Al-4V-xCu alloys increases significantly with increasing Cu content, as shown in figure 8(g). Wang et al [74] reported that α' of LDED Ti-6Al-4V-5.4Cu decomposes into α and nodular Ti₂Cu phases due to the intrinsic heat treatment, thereby achieving a superior combination of strength and ductility. When the Cu content reaches 6.8 wt%, however, the β phase directly transforms into eutectoid α + Ti₂Cu lamellar during solidification. During the successive thermal cyclic of LDED, the eutectoid Ti₂Cu coarsens significantly, resulting in weakened ductility.

Zoom In Zoom Out Reset image size

Different from the above active eutectoid elements, the Fe, Cr and Mn elements are sluggish eutectoid elements for Ti alloys, suggesting that the eutectoid transformation in the Ti-Fe, Ti-Cr and Ti-Mn alloy systems can be inhibited efficiently by rapid cooling. Nevertheless, these eutectoid elements (Fe, Cr and Mn) can also provide an efficient grain refinement and CET to LAM Ti alloys [83]. Take Fe as an example, 3 wt% Fe addition into LAM Ti-6Al-4V is sufficient for achieving equiaxed grain morphology [83, 93]. The β-grains of LPBF Ti-6Al-4V can be refined efficiently by the rejection of Fe during dendritic growth [94]. However, with increased content of Fe, the microstructure of Ti-6Al-4V alloy transforms from α' into α + ω + β, which deteriorates the tensile properties [83]. Zhang et al [95] fabricated LPBF Ti-6Al-4V with the combined addition of Fe₂O₃ and CP-Ti nanoparticles, and they reported that Fe can promote the in-situ decomposition of α' during LPBF (figures 9(a)–(c)). Therefore, the obtained microstructure and mechanical properties of the alloy were more uniform throughout the building height than that of pure Ti-6Al-4V, as shown in figures 9(d) and (e). In addition, Song et al reported a novel low-cost Ti-(0.35–0.5)O-3Fe alloy for LAM applications using two inexpensive alloying elements (O and Fe) [96]. Fine α–β lamellae microstructure with equiaxed prior-β grains is achieved in these Ti-O-Fe alloys. The prevalent misorientation between α laths in each Ti–O–Fe alloy was [1 120]/60°. Thanks to the high potency of O and Fe in strengthening the α-phase (virtually Fe-free) and β-phase (about 30 vol%, virtually O-free), the novel Ti-O-Fe alloys exhibit superior mechanical performance, with UTS of 1 034 ∼ 1 194 MPa and ductility at fracture of 9% ∼ 21.9%.

Zoom In Zoom Out Reset image size

As discussed above, the strength of LAM Ti alloys can be significantly improved with eutectoid element addition; however, the ductility is generally weakened. The underlying mechanism is that the solubility of these eutectoid elements in Ti is limited; therefore, excessive addition results in the formation of brittle intermetallic phases (e.g. Ti₂Ni, Ti₂Cu). Considering that the solid solubility of the single eutectoid element in Ti alloy is limited, Su et al [84] proposed a novel multi-eutectoid elements alloying approach based on thermodynamic predictions to achieve refined microstructure and high strength-ductility combination in LAM Ti alloys. By adding Co, Cr and Ni elements synergistically into Ti-6Al-4V, the microstructure of the alloy is refined significantly. Meanwhile, Co, Cr and Ni accumulated into the β phase, promoting the in-situ decomposition of α', as shown in figures 10(a) and (b). The multi-eutectoid elements alloying strategy makes full use of the solid-solubility of different eutectoid elements (i.e. Co, Cr, Ni) in Ti-6Al-4V alloy to refine the microstructure and inhibit the formation of brittle eutectoid phases simultaneously. As a result, the strength of the Ti-6Al-4V alloy is increased significantly without significant loss in ductility, as shown in figures 10(c) and (d).

Zoom In Zoom Out Reset image size

The hybrid Ti alloys refer to hybridizing heterostructured microstructures via using LPBF of mixed powders [97]. Thanks to the small size of the melt pool and the ultrahigh cooling rate of LPBF, the local microstructures and compositions can be tailored by the site-specific composition powder feedstock. For instance, Zafari and Xia [97] investigated the LPBF of mixed Ti-6Al-4V (50 wt%) and Ti-5Al-5Mo-5V-3Cr (50 wt%) powders. Due to the fact that powder cluster sizes were larger than that of the melt pool size (width of ∼150 μm and depth of ∼70 μm), the Ti-6Al-4V undergoes a β → α' martensitic transformation, while Ti-5Al-5Mo-5V-3Cr retains as the β phase during solidification. As a result, a unique novel α'+ β dual-phase heterostructured microstructure is achieved in the hybridized alloy (figures 11(a)–(c)). Surprisingly, a superior combination on tensile strength and ductility is achieved in the hybrid alloy. The YS/UTS ratio of the hybridized alloy is only ∼0.6, which is much superior than that of LPBF Ti-6Al-4V and Ti-5Al-5Mo-5V-3Cr alloys. The unique work-hardenability mainly originates from the hetero-deformation between α' and β. In another work, a ternary hybrid Ti alloy which consists of CP-Ti (33.3 wt%), Ti-6Al-4V (33.3 wt%) and Ti-5Al-5Mo-5V-3Cr (33.3 wt%) alloys was also investigated (figures 11(d)–(f)) [98]. Thanks to the highly heterogeneous microstructure, the ternary hybrid alloy also shows an attractive combination of tensile strength (∼1 100 MPa) and ductility (∼20%) compared to various types of conventional Ti alloys (figure 11(g)).

Zoom In Zoom Out Reset image size

In addition to mixing different types of Ti alloys, the hybrid alloy strategy has also been extended to in-situ alloying Ti alloys with stainless steel [85]. Due to the micrometre-scale concentration modulations and partial homogenization of alloying elements, a unique α' + β heterostructured microstructure is achieved via in-situ alloying Ti-6Al-4V with 4.5 wt% 316L. Besides, the local variation of alloy elements in the Ti-6Al-4V-4.5 wt% 316L alloy also results in the variation of Q, contributing to a significant grain refinement. Due to the unique refined α' + β heterostructured microstructure, the Ti-6Al-4V-4.5 wt% 316 l alloy exhibits high tensile strength (∼1 297 MPa) and good uniform elongation (∼8.8%). More importantly, the alloy also possessed a superior work-hardening capacity (with a YS/UTS ratio of 0.76) owing to the activation of β → α' transformation-induced plasticity (TRIP).

As aforementioned, the benchmark Ti-6Al-4V alloy suffers from the formation of columnar prior-β grains, mechanical anisotropy and insufficient ductility. As a result, some novel LAM-based Ti alloys adapted from Ti-6Al-4V alloy have been reported [61, 99]. Wang et al [99] reduced the Al content in Ti-6Al-4V to design a novel Ti-4Al-4V alloy for LAM and found that the decrease in Al content increased the Q value of the alloy, promoting CET during solidification. In-situ high-energy x-ray diffraction (XRD) results indicated that the lower Al content also reduces the critical resolved shear stress of prismatic slip to basal slip ratio, making dislocation cross-slip and twinning easier. As a result, high ductility (∼13% ∼ 14%) and favourable isotropy mechanical performance were achieved.

Meanwhile, some rare elements (e.g. plumbum (Pd) and rhenium (Re)) have been added to LAM Ti alloy to tailor its microstructure and performance. Pd, a noble element, is also a β-eutectoid element for Ti alloys [100]. Qiu et al [101] reported that trace Pd addition(0.2 wt%) addition could effectively enhance the corrosion resistance of LPBF Ti-6Al-4V. Besides, Re is a β-isomorphous element with a hexagonal close-packed (HCP) lattice structure, which is known for improving the mechanical strength, recrystallization temperature and corrosion resistance of Ti alloys [86, 102]. Chlebus et al [86] showed that 1.5 at% Re addition brings a significant strengthening effect to an LPBF-built Ti-Re alloy. The YS and UTS of Ti-1.5 at% Re alloy are about 1 038 MPa and 1 162 MPa, respectively, which is comparable to the Ti-6Al-4V. Meanwhile, Re addition changes the microstructure of Ti from lath α' into acicular α' due to the lower martensite start temperature [103]. However, Re addition deteriorates the fatigue crack propagation resistance of CP-Ti due to the reduced ductility [103]. Majchrowicz et al [102] investigated the hot corrosion behaviour of LPBF Ti-xRe (x= 0, 2, 4 and 6 wt%) alloys in a mixed Na₂SO₄ and NaCl salts at 600 °C. They reported that complete Re particle dissolution is essential to enhance the hot corrosion resistance of LPBF Ti-Re alloy.

Herein, isomorphous β-Ti alloys represent Ti alloys by adding one or more β-isomorphous elements (e.g. Nb, Ta, Mo, V). It is known the β-isomorphous elements can form infinite solid solutions in β-Ti, which can stabilize β-Ti to room temperature. Thanks to the lower density of atoms in the BCC lattice, the elastic modulus of β-Ti is lower than that of α-Ti. As a result, the isomorphous β-Ti alloys are promising low-modulus biomedical structural implant materials without toxic/allergic effects. However, the pre-alloyed powder of isomorphous Ti alloys is still scarce and even commercially unavailable [104]. As a result, numerous studies have been carried out to investigate in-situ alloying LAM of isomorphous β-Ti alloys. The mechanical properties of LAM isomorphous β-Ti alloys are summarized in table 4.

Table 4. Tensile properties of LAM-processed isomorphous β-Ti alloys.

Element	LAM	Composition	Microstructure	Elastic modulus/GPa	YS/MPa	UTS/MPa	/%	References
Nb	LPBF	Ti-15 at% Nb	α'+ β + unmelted Nb	—	501	751	—	[113]
		Ti-25 at% Nb	β + unmelted Nb	—	516	923	—
		Ti-45 at% Nb	β + unmelted Nb	—	583	1 030	—
Mo	LPBF	Ti-15Mo	β + unmelted Mo	—	—	894	2.8	[73]
	LDED	Ti-15Mo	β	73	480	815	17.1	[125]
	LDED	Ti-15Mo	β + athermal ω	87	850	1 099	9.2	[118]
	LPBF	Ti-6Al-4V-10Mo	β + unmelted Mo	73	858	919	20.1	[119]
Ta	LPBF	Ti-6Ta	α'+ unmelted Ta	108	595	697	—	[115]
		Ti-12Ta	α'+ β + unmelted Ta	99	650	783	—
		Ti-18Ta	α'+ β + unmelted Ta	96	668	808	—
		Ti-25Ta	β + unmelted Ta	89	1 029	1 186	—
	LPBF	Ti-50Ta	β + unmelted Ta	76	883	925	11.7	[114]

Overall, two contributions of β-isomorphous element addition can be expected for LAM Ti alloys: (i) promoting grain refinement based on heterogeneous nucleation and (ii) stabilization of β down to room temperature. The effects of β-isomorphous elements on the stability of β-Ti can be evaluated through the Mo-equivalent (Mo Eq.) [105]:

$$\begin{align}Mo{\text{ }}Eq.& = 1.0Mo + 0.28Nb + 0.22Ta + 0.67V + 0.44W\nonumber\\ & \quad + 2.9Fe + 1.6Cr - 1.0Al.\end{align} \tag{ 4 }$$

The unit is in wt%. Note that the β-isomorphous elements (e.g. Nb, Ta, Mo, V) have a higher melting point than Ti. As a result, unmolten β-isomorphous powder particles in Ti alloy are the main challenge for the in-situ alloying of isomorphous β-Ti alloys by LAM. To address this challenge, several solutions have been proposed: (i) adopting ultrafine β-isomorphous powders [106], (ii) using a top-hat profile laser [107], (iii) using a high input laser energy density [104], (iv) using cross-hatching style with an angle of 74° between the layers [108], and (v) using laser re-melting strategy [109].

Ti-Nb alloys are excellent candidates for biomedical applications owing to their superior biocompatibility. Mixing Ti-Nb and pre-alloyed Ti-Nb powder is widely used for LPBF [110–112]. Wang et al [113] studied Nb addition for suppressing the martensitic transformation of LPBF Ti-xNb (x= 0, 15, 25 and 45 at%) alloys. When Nb content increases to 25 at%, the martensitic transformation was completely suppressed, and a full β microstructure could be achieved. Besides, the Nb addition also reduces the elastic modulus, refines the β grains and enhances the apatite-forming capability of the alloy [104, 113]. Nevertheless, the undissolved Nb inclusions are commonly observed in the LPBF Ti-Nb alloys due to their high melting point. To address this issue, Huang et al [107] adopted a top-hat profile laser in combination with high laser power and scan speed for in-situ alloying LPBF of Ti-34Nb mixed powders. They reported that using the specific processing parameters and top-hat profile laser leads to a slow-moving and large melt pool, contributing to the uniform mixing of Ti and Nb powders and suppression of keyhole pores.

Ti-Ta alloys have also received increasing interest in recent years due to their superior biocompatibility and corrosion resistance. However, Ta has a high density of 16.6 g·cm⁻³ (4 times > Ti) and a high melting point (∼3 020 °C), which can result in microstructural inhomogeneities in Ti-Ta alloys. As a result, the fabrication of Ti-Ta alloys is hard using conventional processing methods. Sing et al [114] prepared an LPBF-built Ti-50Ta alloy by using a powder mixture of 50 wt% Ti and 50 wt% Ta. The Ti-50Ta alloy built by LPBF exhibits a single β-phase microstructure with some unmelted Ta particles. Benefiting from the single β microstructure, the elastic modulus of Ti-50Ta alloy is as low as 76 GPa, suggesting a promising potential for biomedical implant applications. Zhao et al [115] investigated the effects of Ta on the mechanical properties and corrosion resistance of LPBF Ti-xTa (x= 0, 6, 12, 18 and 25 wt%) alloys. With increasing Ta content, the microstructure of Ti-Ta alloys changed from lath α grain → lath α' → acicular α' → acicular α' + cellular β grains. Electrochemical testing results in Ringer's solution indicated that the Ta addition enhances the corrosion resistance of Ti-Ta alloys via the formation of stable Ta₂O₅/TiO₂ passive films. Furthermore, Ta alloying can also improve the in vivo biological response of Ti alloys [116]. Through Ta alloying, adding designed porosity and adopting nanoscale surface modification, the in vitro cytocompatibility and early stage in vivo osseointegration of Ti alloy can be further improved significantly.

Similarly, Ti-Mo alloys are typical isomorphous Ti alloys with high corrosion resistance, superior biocompatibility and low modulus [117, 118]. Vrancken et al [119] prepared the Ti-6Al-4V/10Mo alloy by in-situ alloying LPBF, and numerous unmelted Mo particles were present in the alloy. The as-built alloy is similar to a metal matrix composite with β-Ti as the matrix and Mo particles as the reinforcement. Nevertheless, the Ti-6Al-4V/10Mo alloy (∼20.1%) shows significantly larger ductility than the Ti-6Al-4V alloy (∼7.3%). Moreover, the elastic modulus of the Ti-6Al-4V/10Mo alloy is only 73 GPa, which is significantly lower than that of Ti-6Al-4V alloy (∼110 GPa). Shi et al [120] prepared an LPBF Ti-7.5Mo-2.4TiC composite by adding Mo₂C particles into pure Ti. Compared with the Ti-7.5Mo alloy, a higher content of β phase was obtained, and undissolved Mo particles were avoided in the Ti-7.5Mo-2.4TiC composite. Meanwhile, the strength and wear resistance of the Ti-7.5Mo-2.4TiC composite is also higher than that of the Ti-7.5Mo alloy, suggesting great potential in biomedical applications. For the LDED Ti-xMo alloys, the microstructure can be varied along the building height [121, 122]. For instance, Kang et al [122] fabricated a Ti-7.5Mo alloy functionally graded structure by LDED of mixed Ti and Mo powders. Due to the heat accumulation and intrinsic heat treatment during LDED, a graded microstructure is achieved in the as-built state. The fraction of the β phase decreases from 48% to 4% with the increasing build height. As a result, the tensile properties of the Ti-7.5Mo alloy also varied with the build height. The unique functionally graded structure of the alloy could be suitable for specific orthopaedic implant applications.

Vanadium (V) is also a typical β-isomorphous element in Ti alloys. Mantri and Banerjee [121] fabricated a Ti-20V alloy by LDED, and a single β microstructure was obtained. Furthermore, CET occurs gradually with build height. Collins et al [123] prepared a functionally graded structure Ti-V alloy with a composition gradient ranging from pure Ti to Ti-25 at% V. They proposed that a single β phase microstructure could be obtained when the V content reached 17 at%. However, V is cytotoxic to the human body and is unsuitable for biomedical applications [124]. As a result, the research on LAM of Ti-V alloys is less than that of other isomorphous β-Ti alloys (e.g. Ti-Nb, Ti-Ta, and Ti-Mo alloys).

Eutectic β-Ti alloys are characterized by their low melting point and narrow solidification range; therefore, the eutectic β-Ti alloys are promising for LAM applications. Ti-Fe-based eutectic alloys are the most frequently studied eutectic Ti alloy due to their low cost and excellent mechanical strength [126, 127]. According to the Ti-Fe binary phase diagram, the eutectic phase transformation occurs at 1 085 °C with a composition of 32.5 wt% Fe:

$$\begin{align}L \to \beta {\text{-}}Ti + TiFe.\end{align} \tag{ 5 }$$

Gussone et al [126] investigated the LPBF of a Ti-32.5Fe eutectic alloy and reported that a dense and crack-free Ti-32.5Fe alloy could be obtained by preheating the substrate to 600 °C before printing. The LPBF as-built Ti-32.5Fe alloy displays an ultrafine eutectic microstructure composed of TiFe, β-Ti, η-Ti₄Fe₂O_x and a small amount of α-Ti. The formation of primary dendritic η-Ti₄Fe₂O_x is attributed to the pickup of oxygen during LPBF due to the high preheating temperature, and the η-Ti₄Fe₂O_x phase enhances the high-temperature performance.

To further improve the comprehensive performance of Ti-Fe-based eutectic alloys, ternary and quaternary Ti-Fe-based eutectic alloys prepared by LAM were also reported with focuses on Zr, Sn and Y elements addition [128–132]. Wang and Han [131] reported that the Y addition into LDED Ti-Fe eutectic alloy suppresses the Ti₄Fe₂O oxide formation and also leads to significant grain refinement. As such, the Ti-Fe-Y alloy exhibits higher mechanical, tribological and forming properties than the Ti-Fe-base alloy. In another work, the effects of Zr addition on the LDED Ti-Fe-Zr-Y alloy were studied by Han et al [129, 132], and it was reported that 5.86 at% of Zr could enhance hardness and wear resistance as well as corrosion resistance and apatite formation ability. Liang et al [130] investigated the effects of Sn addition on Ti-Fe-Sn-Y alloys built by LDED, and their results indicated that the microstructure of the Ti-Fe-Sn-Y alloys transforms from hypereutectic to triphasic hypoeutectic microstructure with increasing Sn content.

In addition to the isomorphous and eutectic β-Ti alloys mentioned above, customized β-Ti alloys for LAM applications have been reported. Alabort et al [133] reported a novel alloys-by-design approach for designing LAM-based biomedical β-Ti alloy, which considered various aspects. In addition to the martensite start temperature, elastic modulus and passivation, the additive manufacturability was also taken into consideration by using solidification analysis involving solidification range, hot cracking resistance and growth restriction factor. A low-modulus Ti-26.2Nb-8.3Ta-3.3Mo-2.4Zr-4.9Sn metastable β-Ti alloy for additive manufacturing biomedical applications was then derived. The experimental results indicated that the Ti-26.2Nb-8.3Ta-3.3Mo-2.4Zr-4.9Sn alloy exhibits excellent additive manufacturability (i.e. a large process window), and the mechanical properties could be tailored by modifying the laser power. Under the optimized LPBF processing parameters, a low elastic modulus (∼65 GPa) and a high ductility (∼22%) were achieved, suggesting a high potential for biomedical implant applications.

The Bo-Md phase stability diagram map and Q value are also important features for the design of LAM-based β-Ti alloy. Liu et al [106] developed a novel Ti-xFe-xCo-1Mo (1.5 < x < 3.5 at%) metastable β-Ti alloy for LPBF based on the Bo-Md map and Q value. Fe and Co elements promoted the formation of equiaxed grains via promoting constitutional supercooling zone formation. After a super-transus solution treatment, the YS of the alloy reached 1.2 GPa with a decent elongation of 10%. The high YS was mainly attributed to the combined effects of solid solution strengthening of Fe and Co, the fine Mo atomic clusters and ω precipitates. By using the Bo-Md map and the Q value criterion, Chen and Qiu [134] designed a novel Ti-6Mo-5.5Cr-1Co-0.1C alloy for LPBF. The minor alloying of Co and C is to induce CET and grain refinement during LPBF. Overall, the synergic addition of β-isomorphous elements and β-eutectoid elements is favourable for fabricating fine-grained high-strength LAM β-Ti alloys.

To date, the research regarding LAM of Ti alloys mainly focuses on conventional commercial Ti alloys (e.g. CP-Ti, Ti-6Al-4V and Ti-5553 alloys). Notably, these commercial Ti alloys are designed for traditional processing methods (e.g. cast and forging), in which the unique metallurgy characteristics of the melt pool, ultra-high cooling rate and intrinsic thermal cycling of LAM were left unaccounted for. To this end, it is of great importance to develop novel Ti alloy customized for LAM applications, thereby enlarging the application area of LAM Ti alloys. In turn, both LPBF and LDED can be used to accelerate the development of novel Ti alloys based on the high throughput method. Differing from conventional alloy design, the design of LAM-based Ti alloy should also consider the following factors: (i) solidification temperature range, (ii) hot cracking susceptibility, (iii) growth restriction factor, (iv) intrinsic thermal cycling effect, (v) cost and energy efficiency. Due to the fact that LAM is a typical non-equilibrium process, out-of-equilibrium methods such as the simple Scheil model should be used for solidification simulations, hot cracking susceptibility and growth restriction factor. These physical parameters can be obtained from CALPHAD-based software (e.g. Pandat and Thermo-Calc).

(i) Solidification temperature range . Generally, a narrow solidification range may reduce the remelting of underneath layers during LAM, increasing the risk of de-cohesion. However, a wide solidification range could increase the risk of hot cracking and the appearance of defects upon solidification. As a result, the solidification range needs to be optimized to ensure a high quality of LAM parts [133]. According to the criterion of generating constitutional undercooling [84, 156]:

$$\begin{align}\frac{G}{R} < \frac{{m{c_o}\left(1 - k\right)}}{{{D_L}k}} = \frac{{\Delta T}}{{{D_L}}}.\end{align} \tag{ 6 }$$

Where G, R, m, c₀, k, D_L , and ΔT are the temperature gradient, the growth rate of the solid/liquid interface, the slope of the liquidus, the initial content of an alloying element, the solute partition coefficient, the diffusion coefficient in liquid, and the solidification temperature range, respectively. The solidification temperature should be high enough to promote sufficient constitutional undercooling, facilitating nucleation and grain refinement during LAM.

(ii) Hot cracking susceptibility . Hot cracking susceptibility (HCS) is also an essential factor for the additive manufacturability of LAM-based Ti alloys. HCS can be calculated by the ratio of the time required for solidification of the last 10% liquid to the time required for solidification from 40% solid to 90% solid, as shown below:

$$\begin{align}HCS = \frac{{{t_{{f_S} \to 0.9}} - {t_{{f_S} \to 1.0}}}}{{{t_{{f_S} \to 0.4}} - {t_{{f_S} \to 0.9}}}}.\end{align} \tag{ 7 }$$

Where t is the time required to reach the particular solid volume fraction. Note that the low HCS is generally contradicted to achieve a high growth restriction factor and solidification range, so the balance between these factors should be considered.

(iii) Growth restriction factor. The growth restriction factor (Q) controls the rate of development of a constitutional supercooling zone, which plays a decisive role in grain refinement. The Q value can be calculated by the derivative of the fraction of solid concerning undercooling as follows [157]:

$$\begin{align}Q = {\left(\frac{{\partial \left(\Delta {T_S}\right)}}{{\partial {f_S}}}\right)_{{f_S} \to 0}}.\end{align} \tag{ 8 }$$

Where f_s and ΔT_S are the fraction of solid and the solute undercooling, respectively. This can be estimated through the Scheil analysis by calculating the slope of a linear regression fit to the temperature profile as a function of the solid fraction for solid fractions below 0.05. Generally, a large Q value is beneficial for achieving a large constitutional supercooling zone, contributing to CET and grain refinement.

(iv) Intrinsic thermal cycling effect. Utilising thermal history and composite material modulation to control the microstructure of Ti alloys is also an ideal research direction. The intrinsic thermal cycling is one of the most important metallurgical features of LAM, which brings about an intrinsic heat treatment effect to the printed alloy parts [135]. The composition design of LAM-based Ti alloys should also take the intrinsic heat treatment into consideration. Differing from conventional PHT, the rapid intrinsic heat treatment is beneficial to preserving fine structures obtained during LAM. The intrinsic heat treatment can be employed elaborately to tailor the microstructure, which can sometimes be used to replace the PHT. For LAM Ti alloys, the intrinsic heat treatment effects can be utilized to promote the in-situ decomposition of metastable phases and to achieve a fine equilibrium α/α + β microstructure. By using alloying elements with a high diffusion rate (e.g. Cu, Co, and Ni) in Ti [84], it can be inferred that intrinsic heat treatment effects of LAM (α + β)-Ti alloy can be promoted. Besides, the intrinsic heat treatment can also be used to replace post-heat treatment processes, thereby reducing the manufacturing life-cycle and improving energy efficiency.

(v) Cost and energy efficiency. High cost is one of the most important thrusts of LAM Ti alloys. The high cost is due to the feedstock material as well as the expensive equipment and consumables (e.g. raw powder/wire and argon gas). For the large-format component and mass production, the cost reduction of the raw material (either powder or wire) is important to consider. To this end, the low-cost hydrogenated–dehydrogenated (HDH) Ti powder could be a feasible alternative feedstock material [158, 159]. The high oxygen content in the HDH-Ti powder can be scavenged through trace rare earth element (e.g. Y) addition [160, 161]. Moreover, low-cost alloying elements (e.g. Fe and O) can also be considered to reduce the overall cost [96]. More importantly, challenges remain in improving the productivity and energy efficiency of LAM processes and lowering the equipment cost, which deserves further work in the future.

(vi) High-performance Ti-based composites. In addition to Ti alloys, LAM of Ti-based matrix composites (TMCs) is also a promising research domain. TMCs can enhance the properties of conventional Ti alloys, including stiffness, strength, and wear resistance, through minor addition of reinforcements like TiB₂, TiC, B₄C, and hydroxyapatite (HA) [162]. However, it is noteworthy that the mechanical performance of TMCs hinges on the precise control of reinforcement dispersion and interfacial reactions between the Ti matrix and reinforcements. The pre-alloyed composite powders rather than mechanically mixed composite powders could alleviate the segregation or aggregation of reinforcing particles, which is more feasible to achieve better mechanical performance. The intrinsic low ductility and suboptimal fatigue performance of LAM-built TMCs necessitate further attention and may require the implementation of advanced post-processing techniques.

As previously discussed, traditional trial-and-error methods for parameter optimization and alloy design are both costly and time-consuming. Machine learning presents a promising auxiliary tool for the LAM of Ti alloys, offering numerous advantages across various aspects:

(i) Rapid processing parameters optimization. By employing suitable machine learning models, the costly and time-consuming trial-and-error LAM optimization experiments can be significantly reduced. Machine learning also demonstrates immense potential in predicting LAM-produced Ti alloys' mechanical and fatigue properties, enabling more accurate and efficient process optimization. However, the current research in machine learning-assisted LAM parameter optimization predominantly focuses on limited processing parameters (e.g. laser power, scanning speed). It lacks comprehensive consideration of factors affecting the process (e.g. thermal history, part geometry, and speed inconsistencies.). Future research should explore developing more sophisticated models that incorporate a wider range of parameters and their interactions, as well as integrating multi-scale and multi-physics simulations to improve the accuracy and efficiency of parameter optimization.

(ii) Customized alloy design. Recent years have borne witness to a burgeoning academic interest in the realm of high-throughput alloy design facilitated by LPBF and LDED [163–166]. In contrast to conventional high-throughput alloy synthesis methodologies, such as diffusion multiples, combinatorial thin films, combustion synthesis, and in situ nanotip melting, in-situ alloying AM exhibits the capacity to fabricate 3D bulk samples of freely chosen compositions, achieved through adjustments of the powder-feed rates within the hoppers [164]. Integrating CALPHAD, high-throughput alloy synthesis and machine learning to assist alloy design for LAM could be a promising research direction. The merits of high-throughput alloy synthesis mainly include: (i) rapid development of new materials with tailored properties, (ii) reduced material waste and cost compared to traditional trial-and-error methods, (iii) ability to explore a wide range of alloy compositions and microstructures, (iv) greater control over the design and production process. Future research should emphasize the development of advanced machine learning algorithms for LAM-based Ti alloy design that can capture the complex relationships between alloy composition, processing parameters, and resultant material properties. Additionally, research should explore novel alloy systems and microstructural features to create alloys with enhanced performance characteristics tailored for specific applications.

(iii) In-situ defect detection and closed-loop process control. Although progress has been made in employing machine learning for in-situ defect detection in LAM of Ti alloys, several challenges and research gaps still need to be addressed. The most common sensing techniques employed are thermal imaging and in-situ high-speed x-ray imaging. However, these methods are expensive and challenging to integrate into commercial AM machines. Recent literature has proposed more cost-effective, flexible solutions, such as in-situ acoustic monitoring, demonstrating promising potential for defect detection tasks [167–169]. Another important future direction is the development of multi-sensor monitoring systems, which combine the strengths of various sensing techniques to enhance defect detection outcomes [170]. This approach can help overcome the limitations of individual sensors and provide more comprehensive and reliable information for in-situ monitoring. Furthermore, future research should focus on developing and implementing closed-loop process control systems that can dynamically respond to detected defects. This will enable better part quality and reduced waste, leading to more efficient and reliable LAM processes for Ti alloys.

(iv) Process-structure-property (PSP) causal analytics. Current research in machine learning for understanding process-microstructure-property relationships in LAM-produced Ti alloys is still in its early stages. Future efforts are needed to develop more advanced machine learning models that can decipher the complex interdependencies among processing parameters, microstructural features, and material properties, as well as uncover causal relationships for improved process control and alloy performance [170]. In addition, constructing and sharing Ti-6Al-4V material PSP relationship datasets, such as the one presented by Luo et al [171], is of particular importance in stimulating research progress in this area. By facilitating collaboration and data sharing among researchers, the scientific community can accelerate advancements in machine learning-assisted modelling applications for LAM of Ti alloys.

In summary, machine learning is poised to revolutionize the LAM of Ti alloys by rapid processing parameter optimization, facilitating customized alloy design, enabling real-time defect detection and process control, and enhancing our understanding of process-structure-property relationships. Machine learning can significantly enhance the LAM of Ti alloys, leading to more efficient, reliable, and innovative manufacturing solutions. The integration of machine learning techniques will enable the development of next-generation LAM processes and materials tailored to specific applications and industries.

To improve the comprehensive performance of LAM Ti alloys, integrating auxiliary fields has been proven as a feasible solution. In terms of the auxiliary process, the effects of auxiliary energy fields on the laser-material interaction and their effects on microstructure and properties still need to be further examined. Multi-field coupling and other novel energy fields (e.g. electric field) could also be used to assist the LAM of Ti alloys. More importantly, proper auxiliary fields could also assist in developing novel advanced LAM-based Ti alloys. For instance, ultrasonic vibration can promote the dispersion of alloying elements during LAM, which could be beneficial for fabricating low-cost Ti alloys with high content of Fe addition.

Thanks to the high forming freedom of LAM, hybrid processing is also a promising trend for LAM of Ti alloys [172–174]. Herein, hybrid processing refers to the involvement of various processing methods in the fabrication of a single part. The LAM techniques combined with other processing methods (e.g. forging-LAM hybrid manufacturing [175]) could be favourable for fabricating large-format and/or complex-shaped components [176]. Besides, different LAM techniques can also be combined (e.g. LPBF combined with LDED [173], wire arc additive manufacturing (WAAM) combined with LDED [172]) for the Ti alloy fabrication, which can take advantage of different LAM techniques. To make full use of advantages of different processing methods, the interfacial bonding behaviour between Ti parts fabricated by different processing methods should be investigated intensively.

In addition to different processing methods, different materials can also be consolidated in a single part via LAM [177]. Specifically, the high forming freedom of LAM enables the fabrication of multi-materials [178–180]. LAM of Ti-based functionally graded materials is promising for industrial applications, which allows the direct fabrication of complex components with less post-machining work than conventional processing methods. Moreover, it is known that the Ti alloys fabricated by LAM generally exhibit poor work-hardening capacity and low uniform ductility [85]. To address this challenge, manufacturing heterostructured materials could be a feasible solution [181, 182]. Especially for the LDED, it has a larger flexibility in fabricating heterostructured materials than conventional processing methods. As a result, fabricating heterostructure multi-materials (e.g. lamellar heterostructure, voxelized heterostructure) could be adopted for achieving high-strength yet ductile mechanical performance and improving work-hardening capacity via introducing hetero-deformation strengthening.