On the Fabrication of Defect-Free Nickel-Rich Nickel–Titanium Parts Using Laser Powder Bed Fusion

Nickel–titanium (NiTi) is the most common commercially available shape memory alloy (SMA) that exhibits reversible solid phase transformations between martensite and austenite upon temperature or load stimuli [1,2]. On the merits of unique functional properties like shape memory effect (SME) and superelasticity (SE), along with remarkable corrosion resistance and biocompatibility, NiTi SMA has found numerous applications in a variety of industrial sectors such as aerospace [3], automotive [4], biomedical [5,6] in addition to other fields [7,8]. However, fabricating NiTi parts with high degrees of geometric complexity using conventional formative or subtractive manufacturing methods is extremely difficult due to challenges such as accelerated tool wear [9,10], oxidation [11], and poor surface finish [12].

Additive manufacturing (AM) has naturally emerged as a viable alternative to fabricate complex NiTi parts [13,14]. Laser powder bed fusion (L-PBF), as one of the most widely adopted metal AM techniques, has been successfully used to fabricate metallic parts in hard-to-process materials with high degrees of geometric complexity in addition to providing the capability of tailoring microstructural features, mechanical, and functional properties [15,16]. Careful selection of processing parameters such as laser power, scanning speed, hatch spacing, layer thickness, and scanning strategy is essential to enable the fabrication of defect-free parts in L-PBF. Furthermore, through altering these processing parameters, the thermal history within the processed material during fabrication can be modulated to control composition, microstructure, and the resulting mechanical and functional properties [17].

With these capabilities in mind, it is important to point out that L-PBF AM of NiTi does not come without challenges. The material is extremely sensitive to processing conditions and more susceptible to defects and residual stresses than other material systems that were successfully printed in previous studies (e.g., steels and their alloys), particularly for high Ni content compositions [18]. Many research efforts have been pursued to study various aspects associated with L-PBF of NiTi. Process optimization and parameter selection were studied in Refs. [18–21]. The effects of processing parameters on microstructure were reported in Refs. [22–25]. The phase transformation behavior of AM-fabricated NiTi was investigated in Refs. [18,22,25,26]. The relationship between processing parameters and mechanical and/or functional properties was illustrated in Refs. [23,27–32]. The influence of post-heat treatment on final parts’ properties was investigated in Refs. [27,33,34]. References [13,14,35,36] provided reviews of research efforts on AM of NiTi.

One worthwhile observation upon inspecting existing literature is the fact that very few research efforts have investigated AM of NiTi SMAs using a starting powder feedstock with Ni content higher than 50.8 at%, despite the fact that this is desired in order to compensate for the Ni loss during AM and realize the full application potential of final fabricated parts [19]. Indeed, the manufacturability (or printability) of high Ni content NiTi powder (Ni > 50.8 at%) is considerably more challenging compared to lower Ni contents (e.g., equiatomic composition). Xue et al. studied the printability of both Ni_50.1Ti_49.9 (at%) and Ni_50.8Ti_49.2 (at%) powder. Both powders showed good printability, with porosity associated with keyholing and lack of fusion as the main defect mode. Slight warping deformation was observed at the bottom of fabricated specimens using Ni_50.8Ti_49.2 (at%) powder with certain processing parameters [19]. Mahmoudi et al. investigated the printability of Ni_50.9Ti_49.1 (at%) powder via L-PBF and only achieved nine macroscopic defect-free samples out of 47 processing parameter combinations owing to the frequent occurrence of delamination defects [18]. With the increase of Ni content in the starting powder, warping and delamination defects become more prevalent [13]. In addition to the aforementioned defects, when the Ni composition in the starting powder is as high as 51.2 at%, elevated edge/corner or rough surfaces are observed at the early stages of the fabrication process, causing obstruction to the powder feed mechanism (counter-rotating roller or blade) and resulting in early print failure. Some of these defects are also likely to occur simultaneously. Examples of macro defects including warpage, delamination, elevated edge/corner, rough surface, and the combinations of multiple defects in L-PBF of NiTi parts using Ni_51.2Ti_48.8 (at%) powder feedstock are depicted in Fig. 1. Hence, limiting process optimization efforts to focus on common defects such as lack of fusion, keyholing pores, and balling are far from being sufficient. More efforts need to be spared toward eliminating other defects such as warping, delamination, elevated edge/corner, and rough surface in order to truly realize the capability of Ni-rich NiTi parts.

The importance of developing the capability to fabricate Ni-rich NiTi stems from the fact that it opens the door for truly tailoring the functional response of fabricated parts. If the transformation temperatures, shape memory, and superelastic behaviors of fabricated parts can be effectively tailored, this will greatly broaden their application prospect. For example, achieving superelasticity under body temperature for biomedical applications (e.g., stents, dental devices) requires the Ni composition in the NiTi matrix to be as high as 50.7 at% to ascertain that the austenite finish temperature is lower than body temperature [37]. Shape memory under room temperature for actuator applications (e.g., robotic arms) requires the Ni composition near equiatomic (i.e., 50.2 at%) [38].

When fabricating Ni-rich NiTi SMAs using AM, higher Ni content enables the formation of precipitates and other microstructural features by controlling the processing parameters and thermal histories during fabrication. Furthermore, since the boiling temperature of Ni (2732 °C) is lower than Ti (3287 °C), the thermal histories can be used to control the extent of Ni differential evaporation and, in turn, vary the Ni–Ti ratio in as-printed parts. In Ni-rich compositions of NiTi, a change of 0.1 at% of Ni composition corresponds to a 10 °C change in phase transformation temperatures [37]. Hence, phase transformation, shape memory, and superelastic behaviors of as-printed parts can be effectively tuned over wide ranges, especially when the Ni content in the raw NiTi feedstock is high enough [39–43]. It has been reported in Ref. [19] that the Ni loss, depending on processing parameters, can reach up to 1% for some levels of input energy. Therefore, using a starting powder material with a Ni composition slightly higher than 51.0 at% (i.e., 51.2 at% in the current study), it can be guaranteed that the fabricated part’s composition still is Ni-rich, which in turn will take full advantage of fabricated parts to meet the needs of different applications.

The current study presents a first investigation into the printability of Ni-rich Ni_51.2Ti_48.8 (at%) powder via L-PBF. A flowchart of the proposed research roadmap is shown in Fig. 2. Specifically, single track experiments are first conducted to classify laser power P and scanning speed v combinations that result in keyholing, lack of fusion, or continuous single tracks. The maximum hatch spacing is then calculated in accordance with the experimentally measured widths and depths of the continuous single tracks. The parameter space determined by P, v, and hatch spacing h was separated into pre-determined non-printable space and parameter space for cuboid specimen fabrication. Next, on the one hand, data augmentation was applied within pre-determined non-printable space. On the other hand, cuboid specimens are fabricated with P, v, h combinations selected within the parameter space for cuboid specimen fabrication. Since printing cuboid specimens is more challenging than printing single tracks, many of the defects and phenomena noted previously are observed and analyzed. The data collected from cuboid fabrication results, together with augmented data, were utilized to train a support vector machine (SVM) classification model in order to classify the processing parameter space (defined by P, v, and h) into printable and non-printable regions. To account for the highly imbalanced dataset—the printable region constitutes a small fraction of the total available process parameter space—the synthetic minority oversampling technique (SMOTE) method was used. Finally, 10 processing parameter combinations are selected for printing test coupons from within the predicted printable region that are applied for experimental validation, and the transformation behavior of these fabricated samples is investigated. The results of the current study could be utilized for further fine-tuning of the processing parameters inside the printable region to achieve targeted phase transformation behaviors or functional properties.

This paper is categorized as follows: Sec. 2 introduces the experimental methods utilized in this study. Results and discussion are illustrated in detail in Sec. 3. Section 4 draws conclusions and points out the future research directions.

Pre-alloyed Ni_51.2Ti_48.8 (at%) powder prepared via gas atomization method was provided by Nanoval GmbH & Co. The 50th percentile (median) of the powder particle size distribution (D50) was 19.5 μm, while 80% of powder particles possessed a diameter less than 32 μm (D80 = 32 μm). The morphology of the NiTi powder particles is shown in Fig. 3, captured using an FEI Quanta 600 field emission gun (FEG) scanning electron microscope (SEM). Most powder particles show spherical morphology, although some hollow, satellite, and irregularly shaped particles are also visible. It is worth mentioning that the powder particle morphology would also affect the printability and final part properties, while it is not the focus of the current study. The interested reader is directed to Ref. [13] for more details.

About 66 single tracks with a length of 10 mm were printed on a sand-blasted NiTi disk (for better powder coverage) using a ProX 200 DMP commercial L-PBF system from 3D Systems (with a maximum laser power of 260 W and a laser spot size of 80 μm) using different P and v combinations (Fig. 4). The gray dashed lines shown in Fig. 4 represent the boundaries of the processing parameter window investigated in this study. The upper limits of P (260 W) and v (2.5 m/s) were defined by the machine specifications. The lower limit of v was set as 0.05 m/s to mimic the steady yet moving laser status. The lower limit of P was set as 21 W, which was the minimum laser power required to obtain the maximum temperature of the melt pool higher than the melting temperature of Ni_51.2Ti_48.8 (at%) at the lower limit of v, derived using the Eagar–Tsai analytical model [44]. Among these 66 combinations, 60 were chosen using grid sampling with P varying between 40 W and 240 W with an increment of 40 W, and v varying between 0.08 m/s and 2.33 m/s with an increment of 0.25 m/s. To better quantify the region of low P and low v combinations, six additional single tracks were sampled at P of 60 W, 100 W, and 140 W with v of 0.205 m/s and 0.405 m/s. These additional points are important because this region has a more localized change with the change of P and/or v in terms of single track morphologies [45]. The layer thickness was chosen as the D80 of the powder size distribution (32 μm). After single track experiments, the top surfaces of single tracks were investigated under SEM to identify whether a lack of fusion happened and to measure the melt pool widths. A relatively conservative lack of fusion criterion was utilized in this study according to whether the single track was continuous or not. A discontinuous single track indicated that the energy density applied to the powder bed could not penetrate the powder layer, which would result in porosity due to lack of fusion. The linear energy density E_L is defined as P/v, which is a simplified aggregate design parameter quantifying the amount of laser energy delivered per unit length to the powder bed. The single track width (w) was calculated as the average of single track widths measured at five locations.

The single tracks were subsequently cross-sectioned at three equidistant locations using wire electrical discharge machining (EDM) to detect whether the keyholing phenomenon happened and to measure the track depths. The cross-sections were ground by SiC paper up to 1200 grit and polished by water-based diamond suspension and characterized under Keyence VHX digital optical microscope (OM). The single track depth (d) was calculated as the average of single track depths measured at these three locations. A single track was defined as keyholing if 1.2 times single track depth was greater than the single track width (d > w/1.2). A single track with a keyholing defect would exhibit keyhole porosity, which is detrimental to the quality of the fabricated parts. The processing parameter combinations of P and v which resulted in a lack of fusion or keyholing defects were excluded for the next step involving the fabrication of cuboid specimens.

It is worth noting that balling is also a common phenomenon during the L-PBF process, which exhibits a non-uniform ellipsoidal/spherical agglomeration or humping and is caused by the Plateau–Raleigh capillary instability at high laser power and high scanning speed combinations [46]. When the balling phenomenon is excessive, irregular broken tracks would be noticed. The balling phenomenon is closely related to the material properties and processing parameters [47]. In this study, the excessive balling phenomenon was not observed, with all single tracks exhibiting balling phenomenon still being continuous. Therefore, the corresponding processing parameter was considered a candidate for cuboid specimen fabrication as long as the balling single track was continuous.

The hatch spacing values for cuboid specimens under each P and v combination were selected from the range between 25 μm and the computed maximum hatch spacing h_max. Selecting 25 μm as the lower bound of the hatch spacing value is justified as follows: first, it guarantees that at least one hatch spacing value could be selected for each P, v combination since the calculated h_max for each P, v combination selected for cuboid specimen fabrication was higher than or equal to 25 μm. Second, considering the fact that lowering hatch spacing would increase the total fabrication time, 25 μm is a reasonable lower bound value for hatch spacing in terms of balancing process efficiency. In total, 105 processing parameter combinations with different P, v, and hatch spacing h values were selected for cuboid specimen fabrication. All samples were fabricated under an argon protective atmosphere with oxygen levels controlled under 100 ppm to minimize oxidation. The laser scanning strategy utilized for cuboid specimen fabrication was the hexagonal scanning strategy with a hexagon edge size of 2 mm (R = 2 mm) and overlap between two adjacent hexagons of 200 μm (O = 200 μm) (Fig. 6(b)). Inside each hexagon, the laser scanning angle was set as 45 deg for the first layer in back and forth style and was rotated 90 deg from layer to layer (Fig. 6(a)). The hexagon location was modified from one layer to the other to mitigate the porosity associated with constant overlap zones [51]. Specifically, the center of each hexagon in layer n would become the vertex in layer n + 1. The justification for using the hexagonal scanning strategy is to reduce residual stresses via reduced scanning length. Promoppatum and Yao [52] demonstrated that the residual stress could be reduced by 50% by applying reduced scanning length. All NiTi cuboid specimens were printed on NiTi substrates to eliminate the potential lack of compatibility effect between the melted material and the base substrate [53]. The fabrication outcomes for these specimens were classified and labeled as printable or non-printable. For the non-printable specimens, the corresponding macro defects were also labeled for further analysis.

The dataset utilized to train the SVM model was comprised of three parts. The first part was the data points collected from the cuboid specimen fabrication results (Sec. 2.2). The second part was obtained through augmentation of P and v combinations which resulted in a lack of fusion or keyholing defects during single track experiments. In particular, the data augmentation method was utilized through sampling a series of hatch spacing values for each P and v combination. The third part was obtained by sampling at the high hatch spacing region (h > h_max) for P and v combinations associated with continuous single tracks. It is worth mentioning that all the augmented data were labeled as non-printable since porosity defect would occur if these processing parameter combinations were applied for solid part fabrication. Figure 7 shows an example of how data augmentation works at P = 160 W, in which dot markers indicate printable (y = +1) and cross markers indicate non-printable (y = −1). The black dots and crosses represent the cuboid specimen fabrication results. The crosses are the augmented training data sampled in the lack of fusion or keyholing region with h sampled in H = {10, 25, 50, 100, 150, 200 μm} to sweep over the whole hatch spacing range. The blue crosses are the augmented training data sampled in the high hatch spacing region with h sampled from {h: (h > h_max) ∩(h∈H)}. The purpose of data augmentation is to incorporate more information of what gained in Secs. 2.1 and 2.2 to train the SVM model. Otherwise, the model itself would not know that the lack of fusion region, keyholing region, and the high hatch spacing region are non-printable.

After data augmentation, the final dataset contained 401 observations in total, only 29 of which were labeled as printable. Therefore, the dataset was skewed to the non-printable class with majority-to-minority class ratio higher than 12—in other words, the dataset was heavily imbalanced. To account for this imbalance, synthetic minority oversampling technique (SMOTE) was utilized to synthesize new data points labeled as printable and achieve a balanced distribution with equal amount of data points in printable and non-printable classes. Details of SMOTE method could be found in Ref. [57]. In simple terms, the new data points in the minority class were randomly sampled on a line connecting two adjacent minority data points. It is worth mentioning that during the 5-fold cross-validation mentioned earlier to select the best hyperparameters, for each iteration, SMOTE was only implemented on the subsets used as the training set in which a balanced dataset is desired, while the holdout fold used as the test set was kept as is to ascertain the accuracy of cross-validation. Finally, the cross-validation selected hyperparameters were utilized to train the SVM model on the full dataset oversampled by SMOTE and the resulting predictive printability map was constructed within the P, v, and h space. To validate the printable region predicted by SVM, 10 processing parameter combinations were chosen within the predicted printable region and the 10 mm × 10 mm × 10 mm cubic samples were test fabricated using these processing parameter combinations. Differential scanning calorimetry (DSC, TA Instruments Q2000) was utilized to measure the phase transformation behavior of these test samples. Cuboid DSC specimens with dimensions 3 mm × 3 mm × 1 mm were cut from the middle part of these cubic samples by wire EDM. Two heating/cooling cycles from 150 °C to −150 °C with a temperature changing rate of 10 °C/min were applied.

The results of single track experiments are shown in Fig. 8. Different colors correspond to different classes of single tracks. The processing parameter combinations resulting in keyholing single tracks are shown as diamonds and representative top-view SEM image and corresponding cross-sectional OM image are shown in Fig. 9(a). The gas pore shown in the cross-sectional OM image was caused by the evaporation of melted material due to the high energy input applied to the powder bed and subsequent collapse of the vapor cavity [58]. Keyholing porosity is detrimental to the fabricated part's properties. The processing parameter combinations resulting in a lack of fusion single tracks are shown as blue crosses in Fig. 8. Almost no track or a highly discontinuous track are classic forms of this class. Figure 9(d) shows the representative top and cross-section of a lack of fusion single track. The processing parameter combinations utilized in the lack of fusion tracks did not cause sufficient penetration and melting of the powder layer and bonding between deposition material and substrate. This ultimately resulted in a lack of fusion porosity which also compromised the fabricated part's properties. The processing parameter combinations resulting in continuous single tracks are shown as black dots in Fig. 8. Two classes of single tracks were considered as continuous: good single tracks (Fig. 9(b)), which were uniform and continuous, and balling single tracks (Fig. 9(c)), which were non-uniform but continuous. The balling phenomenon would affect the surface roughness of printed samples [59]. However, in virtue of its continuity, it was still possible to use these processing parameter combinations to produce defect-free part if the hatch spacing was tuned to a smaller value. Therefore, the processing parameter combinations resulting in both good and balling tracks were considered continuous and chosen for cuboid specimens fabrication.

The P and v combinations that resulted in continuous single tracks were selected to print the cuboid specimens using a series of hatch spacing values. The maximum hatch spacing values for P and v combinations resulting in continuous single tracks were calculated and are shown in Table 1. The hatch spacing values chosen for cuboid specimens fabrication are located in the range below the maximum hatch spacing values as illustrated in Sec. 2.2. In total, 105 processing parameter combinations of P, v, and h were selected for cuboid specimen fabrication. All other processing parameters were kept constant throughout the fabrication process. There were six macro-defect modes observed (Fig. 1), and the corresponding processing parameters at six different laser power levels are shown in Fig. 10, in which green dots represent printable processing parameters, and all crosses represent processing parameters resulting in macro-defects. Details regarding each mode are summarized as follows:

1. Warpage: Warpage was the distortion phenomenon that happened during L-PBF where the bottom portion of the specimen was tilted (or warped) due to the accumulation and relief of thermal-induced residual stress [60], as shown in Fig. 1(b). Compared to the geometry of the defect-free specimen (Fig. 1(a)), the contact area between the warping specimen and the substrate was reduced, causing it to be easily removed by the powder feed roller, which resulted in build failure. The processing parameters resulting in warpage defect are shown as orange crosses in Figs. 10(b), 10(c) and 10(e). It could be noticed that individual warpage defect was rarely observed and at slightly higher hatch spacing values compared with the printable processing parameters with the same P and v combinations. For example, when hatch spacing increased from 50 μm to 75 μm at a fixed laser power of 80 W and scanning speed of 0.33 m/s, a noticeable warpage defect would occur. This aligns well with the results derived in Ref. [61]. At lower hatch spacing values, the amount of material subject to reheating/remelting increased with the increase in the overlap between adjacent tracks, which provided an annealing effect to reduce the residual stress and in return mitigate the warpage.

2. Delamination: Delamination was a macro defect that was also caused by the relief of accumulated residual stress generated during the fabrication process [62]. Unlike warpage defect, delamination was manifested as the interlayer cracks perpendicular to the building direction, as depicted in Fig. 1(c), which was caused due to the low tensile strength along the building direction resulting from the incomplete fusion and adhesion between successive layers. The processing parameters resulting in delamination are shown as blue crosses in Figs. 10(a) and 10(b). Individual delamination defect was inclined to generate at lower laser power (P = 40, 80 W) and higher hatch spacing (h ≥ 100 μm for P = 80 W and v = 0.08 m/s), within which region E_V was relatively low.

3. Rough surface: Bumpy top surfaces were detected (Fig. 1(d)) at high laser power (P = 240 W) and high scanning speed (v ≥ 2.08 m/s) scenarios, shown as purple crosses in Fig. 10(f). The peaks of these bumpy surfaces could project from the powder bed and cause collision with the movement of the powder feeding mechanism. Furthermore, the rough surface would cause the non-uniform powder spreading and in turn degrade the fabricated part properties. This defect type was caused due to the combined action of surface tension and flow inertia, which promoted swelling formation at the end of the melt pool [63].

4. Elevated edge/corner: The elevated edge/corner phenomenon during the L-PBF process (Fig. 1(e)), also called the edge/corner effect, was caused by the heat accumulation near the edge/corner of the printing area, where the surrounding powder bed possessing low thermal conductivity slowed the thermal dissipation process [64,65]. Similar to the rough surface defect, the elevated edge/corner was harmful to the printability of Ni_51.2Ti_48.8 (at%) powder investigated in the current study given that it collides with the powder recoating roller as it deposits powder for the next layer. This generates a significant torque that could stop the movement of the recoating system and consequently stop the fabrication process. The elevated edge/corner defect often happened at an early stage and specimens with such defect often failed at a very early age. The processing parameters leading to elevated edge/corner are represented as rufous crosses in Figs. 10(c)–10(f). It could be noticed that this defect form was associated with lower scanning speed values at higher laser power (P ≥ 120 W), in which higher thermal accumulation was generated due to higher E_L.

5. Combined warpage + delamination: In addition to the aforementioned four distinctive defects, certain processing parameters led to more than one macro defect. The combination of warpage and delamination is illustrated in Fig. 1(f), and the corresponding processing parameters are marked as rose-crosses in Figs. 10(c)–10(f). Similar to individual warpage defects, the hatch spacing resulting in this defect mode was relatively higher than the printable points with the same P and v combinations. Higher hatch spacing is associated with lower E_V, which facilitated the generation of delamination defect.

6. Warpage + elevated edge/corner: Figure 1(g) represents the typical profile of this defect mode. The specimen with such combinations of defects would fail at a premature age since the powder feeding mechanism would easily peel it off from the substrate. The processing parameters generating such defects are shown as olive crosses in Figs. 10(c)–10(f). All crosses were located at higher hatch spacing and lower scanning speed region, which aligned well with the locations for individual warpage and individual elevated edge/corner defects as discussed before.

SVM was utilized to classify the processing parameter space of P, v, and h into printable and non-printable regions as described in Sec. 2.3. The final SVM model with hyperparameters tuned using 5-fold cross-validation possessed the RBF kernel with γ = 10⁻¹ and C = 10⁷. This hyperparameter combination resulted in an accuracy of 0.988 in the cross-validation step.

Figure 11 shows the classification results at six different values of laser power P, with horizontal and vertical axes representing scanning speed v and hatch spacing h, respectively. In the circumstance when P was too low (P = 40 W), no printable processing parameter was detected as shown in Fig. 11(a). When P was reasonably low (P = 80 W), the printable hatch spacing range was comparatively wider (Fig. 11(b)). With the increase of P, the printable values of v increased while the printable hatch spacing was narrowed to lower values. The occurrence of rough surface defect at P = 240 W further narrowed the printable region. This pattern could be explained by macro defect modes. In the lower power situation (P ≤ 80 W), the major defect mode was delamination. With the increase of P (P ≥ 120 W), multiple defects were generated at different v and h combinations, which significantly narrowed the printable region.

The 3D representation of the predicted printable region is shown in Fig. 12. The blue surface represents the iso-probability contour plane with a printable probability of 0.9. Below that surface, the printable probability is higher than 0.9 and the corresponding processing parameters were considered printable. The green squares and crosses with different colors indicate the printable processing parameters and parameters leading to macro defects resulting from cuboid specimen fabrication, respectively. For detailed information on the processing parameters and corresponding fabrication results and/or defect modes, please refer to Table 3 in  Appendix. It could be seen that SVM classification results performed well in splitting the printable processing parameters from non-printable parameters. With the laser power P ranging from 50 W to 100 W, the printable region narrowed down drastically with feasible hatch spacing value decreasing from 80 to 40 μm. With further increase of the laser power, the printable region was found to be confined to a narrow “tube,” with a hatch spacing value of less than 30 μm. This was in accordance with what was detected in Fig. 11.

Ten processing parameter combinations inside the predicted printable region were selected and test fabricated to validate the proposed classification model. The points are shown as black dots in Fig. 12. The detailed processing parameters are listed in Table 2. All ten samples were built successfully, free of macroscopic defects. Figure 13 shows the transformation behavior (obtained by DSC characterization) of four representative as-fabricated samples, showing an increasing trend in transformation temperature with increasing linear energy density E_L. At a smaller P value (P = 50 W), a high E_L value could be successfully applied, which resulted in higher Ni evaporation and in turn led to higher transformation temperatures. With the increase of P, the printable E_L gradually decreased and the transformation temperatures were consequently reduced. When P exceeded 100 W, the transformation behavior didn't show a noticeable difference due to the relatively constant E_L and E_V settings. It could be seen that although the printable region for Ni_51.2Ti_48.8 (at%) was quite narrow compared with relatively lower Ni compositions investigated in previous studies [18,19], the range over which transformation temperatures can be tuned was still larger than NiTi compositions with lower Ni content [18]. This demonstrates the potential of Ni-rich NiTi compositions in tailoring transformation temperatures and further emphasizes the impact of the current study.

In the current study, the printability of high Ni-content Ni_51.2Ti_48.8 (at%) powder via L-PBF was investigated. A variety of macro defect modes were detected during the fabrication process. The cause of these defects was systematically investigated, and the relationship with key processing parameters was analyzed. The printable region within the space defined by laser power P, scanning speed v, and hatch spacing h was predicted via the SVM classification model and validated by test printing. The conclusions are summarized as follows:

1. Compared to NiTi powder feedstock with lower Ni compositions investigated in previous studies, the printability of Ni_51.2Ti_48.8 (at%) is significantly degraded and the printable region within the 3D space defined by laser power P, scanning speed v, and hatch spacing h is narrowed. These processing parameter combinations should be fine-tuned in order to achieve parts free of macropscopic defects.

2. In addition to the lack of fusion and keyholing defects, several different types of macroscopic defects, namely delamination, warpage, rough surface, and elevated edge/corner were observed during the fabrication process which highly compromised the printability of Ni_51.2Ti_48.8 (at%) powder. The combination of more than one defect (warpage + delamination, warpage + elevated edge/corner) also emerged. Elevated edge/corner defect was associated with a lower scanning speed at higher laser power (P ≥ 120 W). Warpage happened along with relatively higher hatch spacing values. Delamination defect was generated when the energy input was relatively low. Rough surface was caused by the combination of high laser power (P = 240 W) and high scanning speed (v ≥ 2.08 m/s). The occurrence of more than one defect was detected when the processing parameters that favored the generation of multiple defects were chosen.

3. At lower laser power (P < 100 W) settings, the printable region was relatively large. On the contrary, at higher laser power (P ≥ 100 W) settings, the printable region was narrowed to a low hatch spacing range.

4. The transformation behavior of as-printed samples showed a clear trend relative to linear energy density. In order to achieve higher transformation temperatures, a lower P value (P < 100 W) should be considered due to the higher likelihood of successful prints at lower scanning speed.

Since the functional properties of NiTi SMAs, e.g., SME and SE, are the most attractive properties of this class of alloys, the relationships between processing parameters and the transformation temperatures as well as these functional properties for NiTi specimens fabricated via L-PBF will be studied systematically in future work. The effects of other processing parameters, e.g., scanning strategy, and post-heat treatment will also be studied in detail in future work.

The co-authors thank the support from NSF through Grant No. 1846676.

The datasets generated and supporting the findings of this article are obtainable from the corresponding author upon reasonable request.