Despite the maturity of laser-based powder bed fusion of metals (PBF-LB/M), some barriers prevent the manufacturing process from fully being established in the industry. One drawback is spatter formation, which is disadvantageous to PBF-LB/M for three main reasons. First, adhering spatter can initiate coater collision, resulting in process failure. Second, large adhering spatter may cause lack-of-fusion defects as they require more energy to remelt sufficiently compared to unprocessed powder. Furthermore, big nonadhering spatter cannot be recycled as powder. The recycling of small spatter particles potentially results in degraded material properties. Ring-shaped beam profiles have been established for deep penetration welding to reduce spatter formation. Investigations on ring-shaped beam profiles in PBF-LB/M focus on improving productivity and process robustness. Qualitative spatter reduction in PBF-LB/M using ring-shaped beam profiles has also been reported. This publication quantitatively examines the influence of ring-shaped beam profiles on spatter formation in PBF-LB/M. Image processing algorithms of on-axis high-speed images are utilized for spatter detection and tracking. A self-developed spatter segmentation is used to determine the spatter size. A Laplacian of Gaussian filter is combined with a Kalman tracker to count and track the spatter. The results show that spatter formation is highly influenced by the beam profile and the chosen process parameters. Considering the melt track width, ring-shaped beam profiles could reduce the number of spatter per fused area. High numbers of spatter are generated when parameter sets result in balling. Moreover, spatter velocity is primarily dependent on the introduced dimensionless enthalpy.
Laser-based powder bed fusion of metals (PBF-LB/M) is the metal additive manufacturing process with the highest industrial maturity. During the process, subsequent powder layers are applied and selectively melted using a laser. Nowadays, single-mode laser beam sources with powers up to 1 kW are typically used.1 The laser beam is scanned across the powder bed. Hereby, the desired areas of the deposited powder layers are fused by hatching individual melt tracks. Due to the high laser intensities and introduced energy, high temperatures, temperature gradients, and partial evaporation of the material occurs during exposure.2 The evaporation and the melt flows caused by temperature gradients lead to high dynamics in the melt pool2 and can result in process errors such as unstable keyhole formation,3 balling,4 or spatter formation.5
In particular, spatter formation represents a major risk for process reliability. On the one hand, spatter adhering to already melted areas represent a risk for a robust coating process.6 On the other hand, incompletely remelted spatter affects solidification behavior,7 provoking lack of fusion,7 and porosity in the final part.6 Furthermore, due to its size, the largest amount of spatter cannot be reused for following processes, which reduces the material recyclability.5 Yet, the reuse of smaller spatter can adversely affect the material properties due to increased oxygen content.8 Spatter can be divided according to two different criteria. The temperature (cold or hot spatter9) and the origin (powder bed or melt pool9). Cold powder spatter are represented by the steam-induced entrainment of surrounding powder particles.5 This effect is also referred to as powder denudation10 and may redistribute powder around the melt track. Whenever entrained powder particles pass the laser beam, they get heated, eventually melted, and become hot powder spatter.9 Multiple powder particles can also fuse and, thus, form larger hot powder spatter.9 In addition, there are spatter originating from melt pool instabilities caused by recoil pressure and Marangoni flows.5,11 This hot melt pool ejecta is often called droplet spatter.5 Ly et al.9 investigated PBF-LB/M processes of AISI 316L and Ti6Al4V with laser powers between 200 and 300 W and scanning speeds between 1.5 and 2 m/s. They concluded that approximately 60 % of all spatter were hot powder spatter, around 25 % were cold powder spatter, and the smallest fraction of approximately 15 % was melt pool ejecta.9
Spatter characteristics, such as number, velocity, and size, are influenced by part geometry,12 material,13 and process parameters.13–16 Larger amounts of spatter tend to occur in overhanging areas and areas with a high surface-to-volume ratio.12 Gunenthiram et al.13 found that a higher amount of spatter is present for AISI 316L stainless steel than for A404 aluminum for similar process conditions. Furthermore, increased volumetric energy density results in larger amounts of spatter.13,14 Likewise, Andani et al.15 reported that either a decrease in scan speed or an increase in laser power results in a higher number of spatter, with scan speed having a greater effect compared to laser power. In contrast, Young et al.16 found no discernible effect of laser power and scan speed on spatter characteristics in their study. However, a larger spot size results in less spatter formation.14,16 Gunenthiram et al.13 and Sow et al.14 attribute the observed trends to the material evaporation and resulting vapor pressure at the melt pool surface, which promotes spatter formation. Also, in laser welding, the origin of spatter is attributed to local vaporization and the resulting acceleration of the molten metal.17 With PBF-LB/M, spatter formation is often considered an unavoidable process by-product.5 To stabilize the process, an attempt is made to quickly remove any spatter from the process zone via an inert gas flow.18 However, there are also studies on how to prevent spatter formation by process adaptations. For example, adding helium to the shielding gas reduces the number of hot spatter.19 Moreover, transferring the strategy of powder bed presintering from electron beam-based powder bed fusion to PBF-LB/M is possible.20,21 This ensures the powder particles are already bonded together before the melt exposure.21 As a result, hot spatter can be reduced by up to 93 %.20 Koike and Sugiura22,23 analyzed the spatter characteristics with increased gravitational force. Due to the increased gravity, the existing spatter was accelerated more toward the powder bed, resulting in a reduced trajectory height until the suppression of spatter at 10 G.22,23 All these approaches involve increased effort (presintering and high gravity) or cost increases (helium addition).
Another approach is to adjust the melt pool surface temperature and the resulting vapor pressure11 by using alternative beam profiles. In laser deep penetration welding, intensity distributions with a ring surrounding the central intensity peak result in a reduced spatter formation.24,25 Gaussian beam profiles are used as the standard beam profile in PBF-LB/M.1 Several research groups study the influence of different beam profiles on spatter behavior. Slodczyk et al.26 studied the amount of spatter using on-axis high-speed imaging. The PBF-LB/M process was parallelized by a multispot beam shape arranged perpendicular to the scan direction. A similar number of spatter per scan track was counted with a single-spot exposure, a three-spot exposure, and a five-spot exposure, offering enormous potential for reducing the number of spatter per fused area. Due to the distribution of the laser power to several spots, a homogeneous temperature distribution was assumed.26 Tumkur et al.27 compared the spatter characteristics of a Bessel beam shape with a Gaussian beam shape using off-axis high-speed imaging. When using the Bessel beam shape, the amount of spatter was reduced by 50 %.27 Additionally, the velocity of hot spatter was significantly higher when using the Gaussian beam, indicating higher recoil pressures.27 Okunkova et al.28 found a qualitative reduction in the amount of spatter when changing the beam profile from a Gaussian beam profile to a ring-shaped beam profile. Rothfelder et al.29 also found less spatter for a ring-shaped beam. However, Rothfelder et al.29 used tungsten tracer particles and characterized the rearranged particles ex situ on the fabricated sample.
The cited studies26–29 suggest that spatter characteristics strongly depend on the intensity distribution. A change in intensity distribution is typically associated with beam enlargement and loss of resolution in the PBF-LB/M.30 Some modern beam sources offer the ability to change the intensity distribution on the fly.30,31 Therefore, adjusting the beam profile according to the required resolution would be possible. For a deeper insight into spatter behavior with ring-shaped beam profiles, this paper presents an in situ investigation of the spatter characteristics using different ring-shaped beam profiles in different spot sizes and process parameters.
A. Feedstock material
The feedstock material is AISI 316L stainless steel, a widely used material in additive manufacturing. MetcoAdd 316L-A (Oerlikon Metco Europe GmbH, Raunheim, Germany) is selected as powder material. The nominal powder diameters are specified with d10,3 = 19 μm, d50,3 = 30 μm, and d90,3 = 46 μm. The substrate plates are laser-cut, cold-rolled sheets with a corundum-blasted surface (Sa ≈ 3 μm, Sz ≈ 42 μm).
B. Experimental setup
The experiments are realized on a self-developed test bench. A fiber laser with seven switchable beam profiles and a maximum laser power of 1200 W (AFX-1000, nLIGHT Inc., Vancouver, WA, USA) is used as a beam source. The beam profiles are designed with a central Gaussian peak and a surrounding ring, between which the power can be distributed in discrete steps. In this publication, the shapes are named according to the relative distribution between the Gaussian core and the surrounding ring (see the first column in Table I). Beam deflection and expansion are performed using a four-axis scanning system (AM-MODUL, RAYLASE GmbH, Wessling, Germany). The scanning system is extended by a sensor module (RAYLASE GmbH, Wessling, Germany) that houses a high-speed camera (CP90-3-M-540, Optronis GmbH, Kehl, Germany) oriented coaxially to the processing laser. An additional movable focusing lens system is implemented to keep the camera in focus at different beam expansions and deflection angles. A frame grabber (Coaxlink Quad G3, Euresys SA, Seraing, Belgium) is used for high-speed image acquisition. To ensure an inert PBF-LB/M process environment, all studies are conducted inside an experimental chamber providing an argon atmosphere (maximum 0.2 % oxygen). During the exposure of the single tracks, the substrate is overflowed with a volume flow of 16 l/min to remove any process emissions. The optical part of the experimental setup is shown in Fig. 1.
C. Design of experiments
To investigate spatter characteristics, single tracks with a length of 12 mm are exposed on a 50 μm powder layer. The powder layer is applied manually using a 50 μm gauge strip. The process parameters used are shown in Table I. The listed spot sizes were determined camera-based at 320 W using the second moment method.
D. Determination of spatter characteristics
High-speed camera images are acquired and subsequently evaluated to characterize the spatter characteristics. The images are acquired without additional illumination of the process zone using two exposure times of 2 and 20 μs and a frame rate of 20 000 fps resulting in approximately 300 analyzed images per single track. Due to the high data rates, image processing is performed after acquisition by a separate script. Image processing is done in four steps. The steps are shown in Fig. 2 using an example image.
Initial data are all single images of an individual track taken by the high-speed camera.
In an image preprocessing step, the process zone is blacked to only reveal the spatter in the image. For this purpose, a process mask is calculated for each track, according to Haubold.32 To compute the threshold of the process mask, all images of the image series of a single track are averaged. From this averaged image, an Otsu threshold33 is calculated. To generate a fitting process mask, empirical data have shown that an enlargement of this area is needed.32 Therefore, a morphological dilation filter with a seven times seven kernel size is applied. The pixel values of all pixels included in this area are set equal to 0 for each image of the series. The procedure is illustrated in Fig. 3.
For spatter size determination, image segmentation is performed by binarization with another Otsu threshold. To calculate this additional threshold, the Otsu method is applied to the averaged image with an excluded process zone. This calculation procedure is shown in Fig. 4. For spatter detection, the average image (background model) is subtracted from each image of the respective track with excluded process zone (see Fig. 5). The resulting image foreground is binarized with the calculated global threshold. As only the spatters are shown in these images, the remaining connected components can be analyzed concerning spatial expansion. It should be noted that the detected spatter size represents the projected size of a hot, thermal self-luminescent spatter. Using the Otsu threshold on this projected representation, the spatter are detected as slightly larger than they are. Nevertheless, qualitative trends in spatter size can be determined.
The number and the velocity of the spatter are determined by applying the combination of a Laplacian of Gaussian (LoG) detector followed by a Kalman tracker. This approach has already been implemented by Jäger et al.34 in laser welding and by Slodczyk et al.26 in PBF-LB/M and showed good results. The implementation is done using TrackMate.35
III. RESULTS AND DISCUSSION
A. Influence on the number of spatter
Figure 6 shows the number of tracked spatter per millimeter single-track length. Overall, the results are rather heterogeneous, so no direct correlation between the introduced dimensionless enthalpy and the amount of spatter per millimeter single-track length can be derived. However, some of the selected process parameter sets result in track surfaces that exhibit balling. These process parameters result in significantly higher spatter formation than parameter sets that result in regular smooth surfaces of the single tracks. One reason may be the presence, respectively, the intensity of Marangoni flows in the melt pool. These flows are caused by temperature differences, the corresponding surface tensions in the melt pool, and typically flow from the hot melt pool area within the laser spot backward to the cold tail of the melt pool.11 Depending on the stability and velocity of the melt, the material either solidifies at the melt pool tail, flows forward again toward the laser spot, or is ejected from the melt pool as spatter.11 When single tracks show balling characteristics, the weld is broken due to plateau-Rayleigh instability, and individual melt accumulations (balls) have formed.39 This effect is produced by an unstable melt flow into the melt pool’s rear area without a corresponding backflow.11 The balling effect can be controlled by adjusting the energy input into the melt pool—usually by adjusting the laser power to the scan speed.11 With less energy introduced, the melt pool has a lower temperature and, therefore, solidifies earlier, preventing interruption of the melt track induced by surface tension.40 Along with the balling tendency, material vaporization increases with increased energy introduced into the material. As the metal vapor flow increases, more powder particles are drawn in the laser beam and heated. These particles are also detected as hot spatter, resulting in more tracked spatter. Remarkably, the high spatter numbers and balling tracks in the experiments occur at comparatively low dimensionless enthalpies. The manufacturing of parts with high densities (>99.6 %) using Gaussian intensity distributions is reported in a range of dimensionless enthalpy ΔH/hs between 3.4 (Ref. 41) and 11.42 In the case of having the effect at low dimensionless enthalpy, it can also be caused by insufficient bonding to the substrate and, consequently, insufficient heat dissipation.
The theories described provide an approach to explain why the balling parameter sets also result in an increased number of spatter. Overall, higher power fractions in the ring result in a lower number of spatter per millimeter track length, except the almost Gaussian “95/5” beam profile. The “75/25” and “65/35” beam profiles have the highest number of spatter. The beam profiles “35/65”, “20/80”, and “10/90” have the lowest number of spatter. The “50/50” beam profile lies in between. However, depending on the selected process parameters (laser power, scan speed, beam size, and beam profile), the single tracks have different widths. Depending on the single-track width, larger or smaller hatch distances can be selected in the three-dimensional process. Large hatch spacing can be selected for large single-track widths, reducing the number of single tracks required for an exposed area. Therefore, the number of tracked spatter is related to the fused area in Fig. 7. The fused area refers to the top view and is calculated by multiplying the single-track length of 12 mm and the track width (see Fig. 8). The widths data are taken from cross sections. With a larger beam size and more power in the ring, the track widths get larger. This reduces the number of hatched tracks per area for beam profiles with a high power fraction in the ring. Beam profiles “35/65”, “20/80”, and “10/90” have on average significantly less spatter per fused area than the beam profiles “95/5”, “75/25”, and “65/35”. The beam profile “50/50” with an average tracked spatter of approximately 90 1/mm2 is in a transition range. The dependence of the amount of spatter per fused area on the beam profiles could be attributed to the peak intensities inherent in each profile. The beam profiles “35/65”, “20/80”, and “10/90” show a significantly reduced peak intensity compared to beam profiles “95/5”, “75/25”, “65/35”, and “50/50”. The lower peak intensity in the center, coupled with the higher power in the ring, is expected to result in a decrease in the peak temperature of the melt pool, thereby reducing the temperature gradient.43 Consequently, there are two implications: First, the weakened outflow of metal vapor leads to reduced entrainment of powder particles in the gas jet. Second, the diminished temperature gradient potentially mitigates the intensity of Marangoni flows, resulting in a decrease in the number of melt pool ejecta. However, for a comprehensive assessment of the gas jet, it is crucial to evaluate not only the quantity of the spatter but also its velocity and size.
B. Influence on the spatter size
The mean spatter area as a function of the dimensionless enthalpy is illustrated in Fig. 9. When using the nearly Gaussian beam profile “95/5”, the smallest spatter particles are observed. As the power distribution to the surrounding ring increases, the average spatter sizes increase until a power distribution of “50/50”. With a further increase of the ring power up to “10/90”, the spatter sizes tend to decrease again. Notably, parameter sets resulting in smooth single-track surfaces exhibit significantly smaller spatter compared to those resulting in balling. Furthermore, the magnification, and, thus, the spot size tend to influence the spatter size. Larger magnification values result in larger spatter. Figure 10 presents the spatter sizes as a function of spot sizes. For identical beam profiles, the mean area of the spatter increases with the spot size. In Fig. 10, linear fits for constant beam profiles are drawn. For the beam profiles “95/5”, “75/25”, and “65/35”, the fit functions and their slopes are in a similar range. From beam profile “50/50” on, the slope for beam profiles “35/65” and “20/80” decreases with increasing power fraction in the ring. Due to the small number of parameter sets for beam profile “10/90”, no fit function is available for this profile.
One possible explanation for the variation in spatter sizes between stable and balling processes may lie in the occurring type of spatter. In a stable process, the detected spatter primarily consists of hot powder spatter.9 These particles retain their original powder particle diameter when they do not agglomerate. Melt pool ejecta typically exhibit larger dimensions compared to the used powder. Wang et al.6 reported that spatter diameters were 3.5 times larger than the original powder particle size. Therefore, the dependence of the spatter size on beam size can be attributed to the scaling of melting phenomena. With larger spot sizes, the corresponding increase in melt pool size scales the flows within the volume, resulting in larger spatter. Beam profiles “20/80” and “10/90” have a significantly reduced peak intensity to mean intensity ratio compared to all other beam profiles. This characteristic can mitigate the temperature gradients, thereby reducing the Marangoni flows and resulting in smaller average spatter sizes due to decreased presence of melt pool ejecta.
C. Influence on the spatter velocity
In Fig. 11, the average velocity of the tracked spatter is presented as a function of dimensionless enthalpy. Remarkably, the spatter velocities are similar across all beam profiles at constant dimensionless enthalpy. Tumkur et al.27 suggest that the high velocity of hot spatter indicates high recoil pressures. In the present context, this implies that the mean vapor pressure depends solely on the dimensionless enthalpy, reflecting the mean energy input during spot crossing. Consequently, the beam profile demonstrates no discernible influence.
Several of the described influences on the spatter characteristics can be exemplified by two parameter sets (black squares in Figs. 6, 7, 9, and 11), which differ only in the laser power and spot size: parameter set 1 (750 W, 365 μm) and parameter set 2 (600 W, 277 μm). Parameter set 1 results in a lower number of spatter due to less energy introduced, resulting in less metal vapor and less melt pool ejecta (see Figs. 6 and 7). For the same reason, the average spatter velocity of parameter set 1 is slightly lower (see Fig. 11). Because parameter set 2 produces balling effects, the average spatter size is larger than for parameter set 1 (see Fig. 9) due to a higher portion of melt pool ejecta compared to powder spatter.
In this study, an investigation of various hot spatter characteristics during PBF-LB/M of AISI 316L was conducted. The experimental setup involved the utilization of seven beam profiles with different spot sizes and a variation in process parameters. High-speed on-axis camera images were recorded and evaluated using various image processing algorithms for spatter evaluation. The key findings can be summarized as follows.
The number of hot spatter relative to the fused area primarily depends on the beam profile. Beam profiles with high peak intensities result in a significantly higher amount of spatter per fused area than beam profiles with lower peak intensities. In this investigation, a factor of 3–4 is observed for constant dimensionless enthalpies.
The spatter size is influenced by two main factors: the spot size and the beam profile. Larger spot sizes generally result in larger spatter sizes, while beam profiles with lower peak intensities tend to generate smaller spatter sizes. The hypothesis is that the origin of spatter changes with the beam profile. Due to the lower peak intensity, beam profiles characterized by less power in the center and more power in the ring exhibit a significant reduction in ejections from the melt pool.
The spatter velocity is primarily influenced by the dimensionless enthalpy introduced. The beam profile and its corresponding peak intensity have a minor influence. High dimensionless enthalpies result in faster spatter, which indicates high recoil pressure and high process dynamics, limiting the process robustness.
Across all spatter characteristics examined (number, size, and velocity), single tracks with balling surfaces consistently yield higher characteristic values than those resulting in smooth surfaces.
In future work, validating the hypotheses made, e.g., using a Smooth Particle Hydrodynamics (SPH) simulation, will be necessary.
Conflict of Interest
The authors have no conflicts to disclose.
Jonas Grünewald: Conceptualization (lead); Data curation (lead); Formal analysis (lead); Funding acquisition (supporting); Investigation (lead); Methodology (lead); Software (lead); Validation (lead); Visualization (lead); Writing – original draft (lead); Writing – review & editing (lead). Jan Reimann: Conceptualization (supporting); Data curation (supporting); Investigation (supporting); Software (supporting); Writing – review & editing (supporting). Katrin Wudy: Conceptualization (supporting); Resources (lead); Supervision (lead); Writing – review & editing (supporting).