Pure niobium manufactured by Laser-Based Powder Bed Fusion: influence of process parameters and supports on as-built surface quality

An intense process parameters fine-tuning campaign was performed as the initial step of the research on PBF-LB/M of pure niobium, in order to find a process parameters set that allows obtaining a relative density higher than 99.5%. Several samples were produced with different combinations of laser power, scanning speed, and hatch distance. The power was set in the range of 100–160 W, while the scanning velocity and the hatch spacing values from 250 and 550 mm/s and 0.04–0.06 mm were used, respectively. For all the samples produced, the scanning strategy was the same: a stripes pattern was used, setting the stripes width to 5 mm and the stripes overlap to 0.05 mm, while the scanning rotation angle of 67° was applied between adjacent layers. The layer thickness used in this work was 30 μm. Also, the influence of the scanning strategy on the final density of the parts was investigated by reducing the stripes width to 3 mm. The parts were printed onto a building platform made of copper OFE (Oxygen Free Electrolytic). This choice was dictated by the authors’ experience with other refractory metals [21]: it is known that stainless steel substrates are not suitable for refractory metals because brittle intermetallic phases form at the interface between additive manufactured parts and the platform, or more in general, in the weld zone, so the adhesion to the building substrate is poor and the job failure due to the detachment of the parts under construction is highly likely. As reported by Hajitabar et al. [22], niobium behaves in the same manner as molybdenum and tantalum, for example [21, 23].

Cubic samples were produced with different dimensions: a first trial was performed for evaluating a wider range of laser power values, so the dimensions of the cubes were 7 mm × 7 mm × 7 mm, due to the limited building area.

Relative densities were measured with the Archimedes method, while the top surfaces of the samples with the higher densities were also examined with SEM.

Trials were performed to investigate the influence of scanning speed, laser power, and hatch spacing on the density and find out the optimum process window, basing the choice of the parameters on the volume energy density E_V. It can be estimated according to Eq. (1) [24]: where P is the laser power [W], s is the scanning speed [mm/s], h is the hatch spacing [mm], and t is the layer thickness [mm]. In particular, the combinations of parameters were selected in order to cover a wide range of energy densities. The cubes’ dimensions were increased to 10 mm × 10 mm × 10 mm once the optimal power and the velocity of the laser were detected. In the end, two samples were produced adopting the remelting strategy: for one of them, the first exposure was higher in energy with respect to the second, while for the other sample, the exposure types were the same, but the order was inverted.

Single scan tracks (SSTs) were also produced for evaluating the quality of the melt pools created by the different parameters sets used for printing the cubes. SEM images were taken for each SST in order to evaluate the homogeneity of the single melt pool, then the ImageJ software was employed for measuring the widths.

Contour parameters were also examined, and dimensional verifications were performed on specially designed samples. Ten hollow cubic samples were produced, with a wall thickness of 1 mm, a side of 10 mm, and a height of 8 mm (Fig. 4a). In this test, two contours were applied for each sample. The laser power was kept constant and equal to 110 W, while the scanning speed and the offset parameters varied. The geometrical accuracy was then assessed by using a Zeiss Accura MASS (Zeiss, Oberkochen, Germany) coordinate measuring machine (CMM). A tungsten carbide probe with a diameter of 2 mm was used. The dimensional control was crucial for the further steps of this study, especially for the development of the contactless supporting structures that will be described in the next paragraph. Table 3 summarizes the contour parameters that have been tested. In Fig. 4b are shown the paths of the measurement probe on each sample: 3 different height levels were considered (2 mm, 4 mm, and 6 mm from the platform plane). The integrity of the contour was verified by printing other cubes and acquiring the morphology of the surfaces with a Sensofar S Neox optical profilometer (Sensofar, Barcelona, Spain) since, in previous experiences, the authors observed that the gas flow and the recoater may cause a discontinuity of the contour, especially on the faces on the opposite side with respect to the two different directions, and a comparison of the as-built surface roughness of samples with and without contour was also done.

The instrument was used in the focus variation modality with a 20× objective to scan the surfaces of interest and reconstruct their three-dimensional topography. The acquired data were elaborated in accordance with ISO 25178-2 [25] to compute the Sa areal surface texture parameter, which represents the average absolute ordinate value of the scale-limited surface. In particular, Sa was computed on the S-F surfaces, applying an F-operator based on the least-squares plane to remove the form error, and an S-filter with a cut-off λs equal to 2.5 μm to reduce the instrument noise.

In the PBF-LB/M process, the overhang surfaces that have a critical slope need to be held up by removable structures. This critical angle is influenced by the material (e.g., its thermal properties, the characteristics of the powder), the machine, and the process parameters. In general, supports are produced for those surfaces having overhanging angles lower than 45°. However, the use of traditional supports implies that these surfaces show connection marks once the supports are removed. In this study, the minimum overhanging angle for niobium was researched. Once the fine-tuning of the process parameters has been completed for the core, the attention was focused on the down-skin parameters in order to investigate the minimum self-supporting angle of AM pure Nb. For this test, segmented arcs like that represented in Fig. 5 were printed, with overhanging angles varying in the range of 20–45° by a step of 5°. The samples had a thickness of 2 mm to ensure stiffness, and the down-skin surfaces had an area of 3 mm × 6 mm to perform measurements. Different down-skin parameters were assigned to the samples, and the downward-facing surfaces were then analyzed with the optical profilometer Sensofar S Neox to understand the influence of process parameters on the surface roughness.

Subsequently, the same kind of sample was used for evaluating the effectiveness of contact-free supporting structures. This novel supporting system was designed as reported in Fig. 6. Traditional block supports were also employed, in order to reduce the amount of material used in the job. For simplicity, from now on the parts composing this supporting system will be named simply as “supports,” as reported in the scheme of Fig. 6. The supports were placed under the arcs to cover the entire downward-facing areas. The down-skin parameters were kept constant for all the samples, while the gap between the part and support ranged between 3 and 7 times the layer thickness: 90 μm, 150 μm, and 210 μm. Then, the down-skin surface roughness was measured for evaluating the effectiveness of such innovative supporting structures and establishing the minimum gap value that allows the pieces not to stick and heat to be dissipated properly.

As previously mentioned, the relative density of the cubic samples obtained with different parameters combinations was assessed with the Archimedes method, by measuring first the weight of the parts in air, and then dipping them in distilled water. The formula utilized to obtain the apparent density of each additively manufactured sample is the following [26]: where m_dry is the weight of each cube in air, m_wet is the weight measured in water, ρ_water is the density of water as a function of the temperature at which the measurement is performed, and c is a corrective factor of air, that is equal to 0.0012 g/cm³. The relative density of AMed niobium is then calculated as follows: where ρ_AM Nb is the apparent density computed using Eq. (2), and ρ_bulk Nb is the density of wrought niobium.

The instrument that was used is a Mettler-Toledo ME balance (Mettler-Toledo GmbH, Greifensee, Switzerland). The reference density of wrought niobium was fixed at 8.57 g/cm³ [2, 3]. The densities obtained ranged from 98.17 to 99.87%.

Laser power had the most visible effect on the melting quality during the PBF-LB/M process: the specimens built using the highest power values saw a marked balling effect during the building process. Good quality of the top surface was achieved for those samples manufactured with a laser power equal to 110 W or slightly lower/upper, since the absence of balling or spatters formation was noticed. However, the maximum density was observed for only the cube produced with a power of 115 W. The effect of scanning velocity was less significant. In general, both the hatching distance values tested (0.04 mm and 0.06 mm) were good, because the adjacent melt pools overlapped properly, without leaving Lack of Fusion (LOF) defects in-between or, on the contrary, re-fuse excessively the material of the previous melt pool. Figure 7 shows the top surfaces of several samples printed. It is possible to see that a scanning speed equal to 300 mm/s (Fig. 7b) can be considered adequate for niobium, because the melt pools are more regular, while 500 mm/s (Fig. 7a) is a barely acceptable value: the edges of the melt pools are more irregular; thus, the molten pool is unstable with such a velocity.

The samples produced using a smaller stripes width (3 mm) were slightly less dense if compared to the samples produced with the same laser parameters but wider stripes (5 mm), even if the difference was not appreciable, while for the cubes that saw a remelting, it was noticed that the exposure order can also have a little effect on the final properties of the parts. Indeed, the sample produced using first the more powerful exposure had a relative density 0.57% higher with respect to the other cube. However, the authors do not consider the remelting strategy valid for justifying the increase in time that it inevitably leads to. In Table 4, the relative densities, computed using Eq. (2) and Eq. (3), are reported in relation to the printing parameters used for some of the samples produced.

Single Scan Tracks (SSTs) were also produced to see the quality of the melt pools created by using the same exposure parameters of the cubes. The average width, continuity, and homogeneity of the scan tracks were examined. Figure 8 shows the SEM images of the SSTs examined in this work. From this test, it can be seen that a scanning speed higher than 350 mm/s is too elevated independently from the laser power that is employed, and the melt pool appears continuously interrupted because the material did not have enough time to melt completely. Indeed, it is evident that many powder particles are stuck on the edges of these melt pools, while partially melted particles are less numerous on the melt pools generated using lower laser speeds. High power and low velocity combined together are more prone to generate a wider melt pool. This means that the energy supplied to the material is tendentially elevated. In general, the SSTs obtained are mostly homogeneous and there are not-so-evident balling phenomena. The ImageJ software was employed for estimating the width of the single scan tracks. The average widths of the several single scan tracks are plotted in Fig. 9 as a function of laser power and scanning speed. As can be clearly noticed from Fig. 9, a lower scanning speed generates a wider SST, independently from the power employed.

As previously described, the geometrical accuracy of the samples produced was measured at three different heights with a Zeiss Accura CMM machine. This analysis gave preliminary information about the surface quality of the samples. As can be seen from Fig. 10, peaks on the surfaces were detected by the probe. Unfortunately, in all the samples, the contour revealed to be not uniform, and discontinuities were present, especially on the surface of the sample oriented on the opposite side with respect to the gas flow direction. The recoater seemed to contribute to “break” the contour, too, because the peaks were often much higher in the corners where the first contact between the blade and the part occurs (for major clarification, the upper image of Fig. 10 shows the position of the samples onto the platform).

In Table 5, the average values of the displacements acquired with the CMM with respect to the CAD dimensions of the parts are reported. From the test, sample 4 and sample 5 had the most homogeneous surfaces: the contour had much fewer discontinuities with respect to the other samples. Probably, this is due to the central position of samples 4 and 5 on the platform. Samples 7 and 8, however, showed to have a constant average displacement from the CAD dimensions on the three paths considered, though the surfaces had numerous burrs and irregularities.

Cubic samples were then printed defining the contour and were also analyzed with an optical profilometer to measure the as-built surface roughness on the vertical faces of the cubes. The three-dimensional surface topography acquisition and Sa computation were performed as described in Section 2.2. Also here, the surfaces on the opposite side of the gas flow were remarkably rougher with respect to the other sides of the cubes. As previously mentioned, the gas flow might be directly involved in this phenomenon: it is possible that, if the gas flow is vigorous, the contour scan track faces discontinuities or irregularities that may be caused by the local accumulation of the by-products of the process on the laser path. The non-uniform amount of material that must be melted along the track can surely affect the melt pool stability and determines a higher surface roughness. Though, further investigations need to be done to better understand the role of the gas flow on the contour quality. From the measurement, however, it was noticed that the contour helped decrease the roughness of the parts, except for the surfaces indicated above: from a Sa higher than 22 μm measured on the vertical surfaces of samples without contour, the cubes for which the contour was defined had an as-built Sa of 10 μm. Figure 11 shows a qualitative comparison of the as-built surface topography of samples obtained with and without using contours. The vertical surface of the sample produced without contour is visibly more complex and less homogeneous than the surface obtained with contour.

In a first step, samples like that represented in Fig. 5 were produced without any kind of support for evaluating the minimum self-supporting angle, or critical angle, of niobium parts produced with PBF-LB/M. The same core laser parameters were used for all the arcs, while different down-skin parameters were defined, in order to find out the more appropriate combination of laser power and scanning speed. Laser power included between 55 and 70 W and scanning speeds ranging from 450 and 550 mm/s were tested. The down-skin overlap was set equal to 6 μm. As previously mentioned, downward-facing surfaces are generally the worse in terms of as-built surface finish and geometrical accuracy because heat dissipation is strongly hampered by the underlying powder, so the development of a specific set of down-skin parameters would help reduce the energy supplied to the material and so the overheating is avoided [19]. At the same time, the melt pool is smaller than that of the outer skin ones.

The critical angle is defined as the minimum angle between the inclined downward-facing surface and the building platform plane at which the use of supporting structures can be avoided. It depends on the material, of course, but also on the process parameters [17], in terms of energy supplied by the thermal source. When the building process is taking place, the smaller the inclined angle of the downward surface, the harder the heat dissipation. Because of this, warping may occur, and edges of the part may start arising layer after layer. If this happens, the powder layer will not have a homogeneous thickness across the building area because the recoater starts jumping (see Fig. 12) and, in the worst case, the job fails after the collision of the blade on the warped part under construction.

It was found that the critical angle of niobium lies between 20 and 25°. The down-skin parameters with which overheating was limited only for angles lower than 25° are 70 W of power and a scanning velocity of 450 mm/s. However, for angles lower than 30°, the surface still appeared with dross formation and irregularities.

The use of contactless supports allowed for reaching smaller angles and, in some cases, completing the formation of the arc. It was noticed that the down-skin parameters and the gap work synergically. So, it is extremely important to experiment with both two variables at the same time for optimizing as much as possible the production of inclined surfaces by using contact-free supporting structures. Indeed, it was seen that using a scanning velocity of 450 mm/s for the down-skin, the parts were in general characterized by a lower as-built surface roughness with respect to those obtained with a slightly higher speed (500 mm/s, lines with square marks in the graph). The parts that went to completion had generally the lower of the two velocities considered. In Fig. 13 are reported the Sa values obtained on the samples with and without contact-less supports, for each inclination angle. As can be seen, Sa is lower for the samples obtained with a down-skin velocity of 450 mm/s (lines with rounded marks in the graph). Moreover, it should be noted that all the samples have comparable Sa values for the 45° surfaces, probably because the down-facing surfaces are self-supporting for angles equal to or wider than 45°; thus, the effects of the supporting system are missed. The use of contactless supports helps reduce the down-skin roughness when the inclination angle is smaller, but the effectiveness of such supports is not so evident for the biggest gaps examined. The best results have been obtained for the supported samples with a gap of only 90 μm between the part and the support. If compared to the unsupported arcs, Sa of such samples is more or less reduced by 50%. No weld points were detected. On the contrary, a gap of 7 times the layer thickness was revealed to be inefficient for reducing the surface roughness but helped reduce the critical angle achieved anyway. A gap of 150 μm did not show improvements in terms of down-skin roughness, but also in this case, it helped further reduce the limiting angle, with respect to the unsupported parts.

One of the most challenging applications of additively manufactured niobium is the production of superconducting radiofrequency (SRF) cavities to be used in particle accelerators. The main difficulties concerning this are the assurance of the purity of the material and the surface finish of the inner surfaces, which has to be excellent (absence of any defect on the surface, Sa lower than 1 μm, and mirror-like). These cavities are generally obtained by coupling two halves together and welding is done in the equatorial region of the component. However, the possibility that impurities and defects are introduced in the welded zone is highly likely, and the performances of the cavities would be greatly influenced. Thus, new manufacturing processes that allow the creation of seamless cavities are under investigation. Studies on additive manufacturing of SRF cavities are reported in the literature [27], but difficulties in obtaining one-piece cavities are stated since the most limiting factor is the extremely poor down-skin surface quality, which makes the as-built surface texture unacceptable for the final application. As mentioned above in this paper (see Section 1), the use of traditional supports is also deleterious. So, a one-piece 6 GHz SRF cavity was printed by using the contact-free support strategy. The upper half of the cavity is the most critical region because the inclination angle is just 18°. Using the contactless supporting system, the inner downward-facing surface was remarkably improved and the job went to completion successfully (Fig. 14a). Figure 14 shows the inner up-skin and down-skin of the cavity printed. As can be seen, the as-built surface finish is similar between the two regions. Thus, the efficacy of contactless supports was demonstrated. Due to the small dimensions of the object, it is not possible to carry out surface roughness measurements without cutting the cavity. Further investigations on the part are ongoing to test its performance.