Spreadability of Metal Powders for Laser-Powder Bed Fusion via Simple Image Processing Steps

This paper investigates the spreadability of the spherical CoCrWMo powder for laser- powder bed fusion (PBF-LB) by using image processing algorithms coded in MATLAB. Besides, it also aims to examine the spreadability dependence with the other characteristics such as powder size distribution, apparent density, angle of repose. Powder blends in four different particle size distributions are prepared, characterized, and spreadability tests are performed with the PBF-LB. The results demonstrate that an increase in fine particle ratio by volume (below 10 µm) enhances the agglomeration and decreases the flowability, causing poor spreadability. These irregularities on the spread layers are quantified with simple illumination invariant analysis. A clear relation between powder spreadability and 3D printed structures properties in terms of residual porosity could not be defined since structural defects in 3D printed parts also depends on other processing parameters such as spatter formation or powder size over layer height ratio.


Introduction
Multimaterial 3D printing is one of the next evolutionary steps in AM that will further improve the degree of freedom in design. Among 3D printing technologies powder bed fusion (PBF) wherein spread powder layers are consolidated by a heat source to obtain three-dimensional (3D) components, seems to be one of the most promising especially for multimetals 3D printing [1]. However, the use of different metals to build 3D parts in one process, has effects on the whole process chain, starting from the pre-process (CAD design and generation of the printing file, setting of printing parameters) to the building process itself where for instance new delivery devices for the powders must be designed [2].
In PBF processes [3][4][5][6][7][8][9][10] that include selective laser sintering (SLS), selective laser melting (SLM), and electron beam melting (EBM), homogeneous powder spreading is important since a well-controlled layer in terms of thickness and roughness may provide satisfying structural and functional properties. For instance, the poor control of surface roughness in selective laser melted embedded channels for cooling systems may negatively affect the cooling efficiency, in contrast with the optimized design. Homogeneity on the powder bed layer depends on powder properties such as shape, size, and roughness, that affect cohesion and apparent/tapped density, so as impurities (such as the presence of a passivating oxidation layer), moisture, etc. [11,12]. Considering combining different types of powders, the control of spreadability becomes one of the first steps necessary to grant high-quality components.
Standard characterization methods of powder metallurgy cannot unfortunately adequately meet the expected properties for AM. In 2014, the ASTM F42 committee created bimodal (fine and coarse), powder mixtures was not successful as well due possibly to the segregation and voids in the layer. While few studies did similar simulations for powder spreadability characterization as well [29][30][31] no other studies have so far been found to propose a simple technique. Such scarce data are partially due to 3D printer manufacturers since they market their instruments with already defined production parameters and raw metal powder sources (same for polymeric and ceramic materials). The companies impose the utilization of such defined parameters and powder sources to guarantee, and otherwise, they do not provide any assurance of the produced parts.
This study aims to add knowledge and develop a simple illumination invariant method by investigating the powder characteristics using digital photos acquired by a digital camera and applying an image processing algorithm in MATLAB. Besides, it will be correlated with other existing characterization methods such as flow rate, apparent density, angle of repose of a model CoCrWMo metallic powder which is commonly used in dental applications such as dental crowns and bridges, for PBF-LB. This research may also open new avenues to provide information on the utilization of outsourced powders that can be used in any available machine, a step further study to render a general tool that is applicable in PBF multimaterials processes.

Powders Bed Analysis
Spherical CoCrWMo powders (PURESPHERE 43055; Mn:0.08 wt.%, Cr:26.82 wt.%, Mo:4.66 wt.%, W:5.40 wt.%, Si:0.90 wt.%, kindly supplied by Sentes-BIR A.S., Izmir, Turkey) produced by gas atomization were used as-is in all experiments. Four different powder mixtures with different powder size distribution (PSD) were prepared and coded as M15-45, Mix1, Mix2, and Mix3, where M15-45 was the standard PSD of CoCrWMo powders available in the market and optimized for PBF process, while Mix1, Mix2, and Mix3 were sieved to have increasing average powder size. Powders were sieved by a Retsch Sieve Shaker (AS200 Basic, Retsch, Germany) with 20 µm, 38 µm, 45 µm, and 63 µm sieves. The particle size was measured both by laser diffraction analyzer (Malvern Panalytical Mastersizer 2000, Malvern, UK) and by using the SEM images with the ImageJ software (ImageJ 1.52a, National Institutes of Health, USA) to quantify the average data (from 100 measurements). The morphological characterizations of powders were done by scanning electron microscopy (SEM; Thermo Scientific Apreo S, Waltham, MA, USA). All powder mixtures were dried in an oven at 90 • C prior to any spreadability and 3D printing test.
For spreadability tests, performed at a constant spreading speed (200 mm s −1 ) with a continuous layer thickness of 50 µm, the printer spreading system was modified to install a Bluetooth controlled cell phone fixed with stainless steel apparatus having a 6 mm diameter aperture for the digital camera (8-megapixel, f/2.2, 1.5 µm) of the cell phone, see Figure 1a, for building platform details. A rubber blade on a 30 mm diameter building platform with white LEDs bonded to black cardboard was used.
To characterize the spreadability behavior, preliminary studies were done with the samples having deliberately made irregularities, an example spread layer is shown in Figure 1b.
The camera images of the spreads were processed by Wiener filtering in MATLAB 2019a, (see Figure 2a-f). Initially, an edge detection ("canny") operation was applied to detect the boundary of the circular region. Then, after finding the center and the radius of the boundary circle using the circular Hough transform, thresholding (Otsu method) was conducted inside the extracted circle to calculate black to white ratio simply from the number of black and white pixels, and further average grayscale value in between 0 and 255. It should be noted that here only average grayscale values are given, however, the standard deviation of grayscale values may be useful to properly demonstrate the homogeneity for which further dedicated studies are needed. The camera images of the spreads were processed by Wiener filtering in MATLAB 2019a, (see Figure 2a-f). Initially, an edge detection ("canny") operation was applied to detect the boundary of the circular region. Then, after finding the center and the radius of the boundary circle using the circular Hough transform, thresholding (Otsu method) was conducted inside the extracted circle to calculate black to white ratio simply from the number of black and white pixels, and further average grayscale value in between 0 and 255. It should be noted that here only average grayscale values are given, however, the standard deviation of grayscale values may be useful to properly demonstrate the homogeneity for which further dedicated studies are needed. Flow rate (following ASTM 213), apparent density, and angle of repose measurements were conducted by Hall flowmeter (produced according to ASTM B212 standard by Sentes-BIR, Izmir, Turkey) [12]. Tapped density measurement was performed by Tapped Density instrument (produced according to ASTM B527 standard by Sentes-BIR,  The camera images of the spreads were processed by Wiener filtering in MATLAB 2019a, (see Figure 2a-f). Initially, an edge detection ("canny") operation was applied to detect the boundary of the circular region. Then, after finding the center and the radius of the boundary circle using the circular Hough transform, thresholding (Otsu method) was conducted inside the extracted circle to calculate black to white ratio simply from the number of black and white pixels, and further average grayscale value in between 0 and 255. It should be noted that here only average grayscale values are given, however, the standard deviation of grayscale values may be useful to properly demonstrate the homogeneity for which further dedicated studies are needed. Flow rate (following ASTM 213), apparent density, and angle of repose measurements were conducted by Hall flowmeter (produced according to ASTM B212 standard by Sentes-BIR, Izmir, Turkey) [12]. Tapped density measurement was performed by Tapped Density instrument (produced according to ASTM B527 standard by Sentes-BIR, Flow rate (following ASTM 213), apparent density, and angle of repose measurements were conducted by Hall flowmeter (produced according to ASTM B212 standard by Sentes-BIR, Izmir, Turkey) [12]. Tapped density measurement was performed by Tapped Density instrument (produced according to ASTM B527 standard by Sentes-BIR, Izmir, Turkey). The chemical composition of the used powder was measured by inductively coupled plasma emission spectroscopy (Perkin Elmer Avio 200, Waltham, MA, USA). The oxygen amount was measured with an inert gas fusion principle (Leco TC400, LECO, St. Joseph, MI, USA).

3D Printing of the Powder Bed
Following the image analyses, two cubes for each powder mixture (with an edge length of 8 mm) were produced by 3D printing, followed by sample removal from the platform via wire erosion. PBF-LB (Laseral MTL90 Izmir, Turkey) machine, designed for dental restorations, was used for 3D printing by SLM for which the build chamber environment was kept under pure N 2 (>99.9% pure) to ensure the oxygen content was limited to 0.1 wt.%. The printing parameters are given in Table 1. Microstructural investigations, performed by SEM analyses (as reported above), of used powders demonstrated no surface porosity, besides no internal porosity was observed when both fractured bead pieces and Bakelite mounted, ground, and polished samples were examined. The surface porosity of the 3D printed cubes was counted by using a Nikon (Eclipse LV150N, Tokyo, Japan) optical microscope equipped with NIS-Elements BR image analysis. The bulk density measurements were conducted both by Archimedes (following ASTM C20-00 with DI water as buoyancy fluid) and by geometric measurements (for all analyses Radwag AS 220R2 scale was used). The true density was measured by pycnometer (AccuPyc II, Micromeritics, Norcross, GA, USA) according to the ASTM B923-16 standard.

CoCrWMo Powders Characteristics
The characteristics of the CoCrWMo powder mixtures, coded as M15-45, Mix1, Mix2, and Mix3 analyzed by laser diffraction are given in Table 2. While D10, D50, and D90 values of the M15-45 and Mix2 were similar to each other, M15-45 was used since it is a generally recommended powder distribution type for AM, and still there were slight differences in the amount of the particles below 20 µm and above 45 µm. Mix1 had a high ratio of particles below 20 µm being 66.58%, and instead, Mix3 included coarser particles higher in volume.  The particle size distribution (PSD) plots obtained from both laser diffraction and microscopic evaluations (SEM) are given in Figure 3a,b, respectively. For the morphological investigation, the stereological equation D sphere = D circle /0.785 was applied to convert 2D measurements to 3D values [32,33]. The average particle size was assessed as; 31.27 ± 9.46 µm for M15-45, 18.54 ± 7.05 µm for Mix1, 34.30 ± 6.97 µm for Mix2, and 58.72 ± 7.08 µm for Mix3, comparable with the laser diffraction obtained D50 values, as seen also in other works [34]. The particle size distribution (PSD) plots obtained from both laser diffraction and microscopic evaluations (SEM) are given in Figure 3a and 3b, respectively. For the morphological investigation, the stereological equation Dsphere = Dcircle/0.785 was applied to convert 2D measurements to 3D values [32,33]. The average particle size was assessed as; 31.27 ± 9.46 µm for M15-45, 18.54 ± 7.05 µm for Mix1, 34.30 ± 6.97 µm for Mix2, and 58.72 ± 7.08 µm for Mix3, comparable with the laser diffraction obtained D50 values, as seen also in other works [34]. According to visual examinations, M15-45, Mix2, and Mix3 were more flowable, while Mix1 was less flowable due probably to the higher portion of agglomerates, see Figure 4 top. The morphologies of the powder mixtures principally being spherical are shown in Figure 4 bottom. Few non-spherical particles, i.e., irregularities, can be explained with gas atomization process parameters. As known for example the powder morphology is directly related to the cooling rate during atomization that is affected by melting temperature of the metal or alloy, atomization pressure, free-fall distance, and gas to metal ratio.  According to visual examinations, M15-45, Mix2, and Mix3 were more flowable, while Mix1 was less flowable due probably to the higher portion of agglomerates, see Figure 4 top. The morphologies of the powder mixtures principally being spherical are shown in Figure 4 bottom. Few non-spherical particles, i.e., irregularities, can be explained with gas atomization process parameters. As known for example the powder morphology is directly related to the cooling rate during atomization that is affected by melting temperature of the metal or alloy, atomization pressure, free-fall distance, and gas to metal ratio. The particle size distribution (PSD) plots obtained from both laser diffraction and microscopic evaluations (SEM) are given in Figure 3a and 3b, respectively. For the morphological investigation, the stereological equation Dsphere = Dcircle/0.785 was applied to convert 2D measurements to 3D values [32,33]. The average particle size was assessed as; 31.27 ± 9.46 µm for M15-45, 18.54 ± 7.05 µm for Mix1, 34.30 ± 6.97 µm for Mix2, and 58.72 ± 7.08 µm for Mix3, comparable with the laser diffraction obtained D50 values, as seen also in other works [34]. According to visual examinations, M15-45, Mix2, and Mix3 were more flowable, while Mix1 was less flowable due probably to the higher portion of agglomerates, see Figure 4 top. The morphologies of the powder mixtures principally being spherical are shown in Figure 4 bottom. Few non-spherical particles, i.e., irregularities, can be explained with gas atomization process parameters. As known for example the powder morphology is directly related to the cooling rate during atomization that is affected by melting temperature of the metal or alloy, atomization pressure, free-fall distance, and gas to metal ratio.  The angle of repose (AOR, α) can be calculated via (1) in which H is the height, D is the diameter of the formed powder cone as given in (Equation (1)): Figure 5 shows the AOR measurements of all powder mixes. As can be seen, Mix1 was quite cohesive due to the higher fine particle ratio (particles below 20 µm was 66.58 vol%) and was given as 32.85 • . Instead, Mix3 was more flowable (particles below 20 µm is 0 vol%) with an AOR value of 21.33 • , i.e., as expected a lower flow resulted in a lower AOR [20].
The angle of repose (AOR, α) can be calculated via (1) in which H is the height, D is the diameter of the formed powder cone as given in (Equation (1)): (1) Figure 5 shows the AOR measurements of all powder mixes. As can be seen, Mix1 was quite cohesive due to the higher fine particle ratio (particles below 20 µm was 66.58 vol%) and was given as 32.85°. Instead, Mix3 was more flowable (particles below 20 µm is 0 vol%) with an AOR value of 21.33°, i.e., as expected a lower flow resulted in a lower AOR [20]. Density and flow rate results are shown in Table 3. The flow rate of Mix1 was not able to be quantified due to high fine powder volume. The lowest flow rate and AOR values were measured as 14.19 (s/50 g) and 21.33 (s/50 g) for Mix3, respectively. The highest apparent density measured was 4.97 g cm −3 from Mix1, and the lowest apparent density was measured as 4.78 g cm −3 from Mix15-45. As could be seen the highest oxygen content was measured for the Mix1 sample due probably to the higher fine power ratio making the surface area value relatively higher than that of the others.

Powders Spreadability by Image Processing
Spreadability was analyzed via MATLAB on the first three spread layers. Simply, in the beginning, the raw image taken was cropped as a circle, then edge detection and Density and flow rate results are shown in Table 3. The flow rate of Mix1 was not able to be quantified due to high fine powder volume. The lowest flow rate and AOR values were measured as 14.19 (s/50 g) and 21.33 (s/50 g) for Mix3, respectively. The highest apparent density measured was 4.97 g cm −3 from Mix1, and the lowest apparent density was measured as 4.78 g cm −3 from Mix15-45. As could be seen the highest oxygen content was measured for the Mix1 sample due probably to the higher fine power ratio making the surface area value relatively higher than that of the others.

Powders Spreadability by Image Processing
Spreadability was analyzed via MATLAB on the first three spread layers. Simply, in the beginning, the raw image taken was cropped as a circle, then edge detection and threshold were applied on the cropped circle. Followed by the determination of the black to white ratio of the threshold image which was used to identify the defects on the spread layer. The obtained and processed images, and the extracted data are given in Figures 6 and 7 and Table 4, respectively. The latter demonstrates the average grayscale values and black to white ratios that are used to evaluate the spread quality of each layer. Mix2 was also a flowable powder blend akin to M15-45 but the flowability was higher (with a flow rate of 14.35 s/50 g and AOR value of 22.01°) due probably to coarser average particle size, compared to that of M15-45. While raw images demonstrate a homogeneous outlook (see Figure 7a), some extent of noise can be visible from the edge detection applied layer images (see Figure 7b) but not from the threshold applied ones given in Figure 7c   Mix2 was also a flowable powder blend akin to M15-45 but the flowability was higher (with a flow rate of 14.35 s/50 g and AOR value of 22.01°) due probably to coarser average particle size, compared to that of M15-45. While raw images demonstrate a homogeneous outlook (see Figure 7a), some extent of noise can be visible from the edge detection applied layer images (see Figure 7b) but not from the threshold applied ones given in Figure 7c  As seen in Figure 6a-c, no irregularity (such as precipitates, clusters, waves) was observed from layers 1 to 3 of the spreads for M15-45. It was previously shown that M15-45 was a flowable powder (with a flow rate of 15.08 s/50 g and AOR value of 22.39 • ) and predictably homogeneous powder layers were observed. For M15-45 powder mixture layer 1, layer 2, layer 3, average grayscale values and black to white ratios were quantified as 119.84, 115.96, 135.05, and 0.31, 0.30, 0.21, respectively.
The flow rate of Mix1 was not able to be assessed, and parallel to that, clear irregularities were observed from the raw and edge detection applied images for layers 1, 2, and 3. However, these irregularities were not noticed in threshold images (see Figure 6d-f). The average grayscale values of layers 1, 2, 3 were quantified as 94.49, 91.98, 92.66, while average black to white ratios were 0.37, 0.40, 0.46. It is important to note that Mix1 was a very cohesive powder with the highest AOR (32.85 • ) in all powder mixtures. Accordingly, different than all other samples an average grayscale value and black to the white ratio for all layers were quantified lower than 95 and higher than 0.32, respectively. Mix2 was also a flowable powder blend akin to M15-45 but the flowability was higher (with a flow rate of 14.35 s/50 g and AOR value of 22.01 • ) due probably to coarser average particle size, compared to that of M15-45. While raw images demonstrate a homogeneous outlook (see Figure 7a), some extent of noise can be visible from the edge detection applied layer images (see Figure 7b) but not from the threshold applied ones given in Figure 7c. Accordingly, for Mix2 layer 1, 2, and 3, average grayscale values of 107.04, 107.38, 107.33, and black to white ratios of 0.31, 0.32, 0.31 were found, respectively.
The most flowable blend was the Mix3 with a flow rate of 14.19 s/50 g, and AOR being 21.33 • . High flowability caused inadequate filling in layer 1, as shown in Figure 7(d-I). However, irregularities were not seen in the spread layers of 2 and 3, see Figure 7(d-II,III). The average grayscale values of layers 1, 2, and 3 were quantified as 82.38, 113.67, and 115.09, while average black to white ratios were found as 0.60, 0.30, and 0.30, respectively (see Figure 7e,f).
Although other factors such as spreading speed and blade type are highly possible to affect the spread layer characteristics [20], setting these constant while altering only the powder characteristics by changing the particle size distribution, it is possible to comment that for a homogenous layer covering, it is better to have a powder with average flowability value. Accordingly, for the tested powder blends the measured average grayscale value should not be lower than 95 on the cropped images, and black to a white ratio higher than 0.32 on threshold applied images. These results are in accordance with [27], where the spreadability was linked to the flowability by the AOR (lower AOR means higher flowability) and to powder specific energy (the lowest the specific energy the highest the flowability).

Production of 3D Printed Cubes
To correlate the spreadability results with the quality of 3D printed parts, all blends were used to manufacture 3D-printed cubes. As seen in Figure 8, the printed samples have a surface finish which is generally seen in additively manufactured parts and might not be only due to particles with different PSD but other processing parameters [35]. flowability).

Production of 3D Printed Cubes
To correlate the spreadability results with the quality of 3D printed parts, all blends were used to manufacture 3D-printed cubes. As seen in Figure 8, the printed samples have a surface finish which is generally seen in additively manufactured parts and might not be only due to particles with different PSD but other processing parameters [35]. As seen in the results reported in Table 5, all samples, including the M15-45 that was produced by using commercially recommended powder, show residual porosity. The calculated total porosity values were around 7 vol% from geometric, and 3-7 vol% from Archimedes technique, implying probable measurement errors. From Archimedes using also the assessed true density, it is possible to state that all samples contained both open and closed porosity for which the former remained ~2 vol% (see Table 5). In addition, image processing was applied on three defined points on both top (XY plane) and left side (XZ plane) of the 3D printed cubes, see representative images given in Figure 9, and data in Table 6. It is clear from these analyses that for all produced samples a higher amount of surface porosity was observed on the XZ plane compared to that of the XY plane, being the XZ plane representative of a higher number of layers. As seen in the results reported in Table 5, all samples, including the M15-45 that was produced by using commercially recommended powder, show residual porosity. The calculated total porosity values were around 7 vol% from geometric, and 3-7 vol% from Archimedes technique, implying probable measurement errors. From Archimedes using also the assessed true density, it is possible to state that all samples contained both open and closed porosity for which the former remained~2 vol% (see Table 5). In addition, image processing was applied on three defined points on both top (XY plane) and left side (XZ plane) of the 3D printed cubes, see representative images given in Figure 9, and data in Table 6. It is clear from these analyses that for all produced samples a higher amount of surface porosity was observed on the XZ plane compared to that of the XY plane, being the XZ plane representative of a higher number of layers.  Among all samples, the lowest surface porosity together with high relative density (minimum total porosity) were obtained from the sample made by using Mix3 blend. This may be related to the layer thickness since during the production, layer thickness was always kept at 50 µm parallel to the D50 of Mix3, measured as ~50 µm. However, no other clear trend to observe the effects of spreadability on the final 3D printed parts was observed. It should be underlined that other SLM process parameters (laser power, speed, layer thickness, moisture, and oxygen level during the production, etc.,) might also affect the final properties, for example, during the printing process spatter formation was noted and this probably altered the oxygen content, and the final pore formation [36][37][38], therefore certainly further studies with variations on the above-mentioned parameters are needed.

Conclusions
Powder spreadability was analyzed quantitatively via image processing algorithms using MATLAB. According to the applied algorithm on the tested powder blends, average grayscale values (0 to 255) lower than 95 on cropped images, and black to white ratio higher than 0.32 on threshold applied images, irregularities were observed. It was noted Figure 9. NIS-Elements BR image analysis processed optical microscope images taken from the top (XY plane: (a-d)) and left side (XZ plane: (e-h)) of the 3D printed cubes. Among all samples, the lowest surface porosity together with high relative density (minimum total porosity) were obtained from the sample made by using Mix3 blend. This may be related to the layer thickness since during the production, layer thickness was always kept at 50 µm parallel to the D50 of Mix3, measured as~50 µm. However, no other clear trend to observe the effects of spreadability on the final 3D printed parts was observed. It should be underlined that other SLM process parameters (laser power, speed, layer thickness, moisture, and oxygen level during the production, etc.) might also affect the final properties, for example, during the printing process spatter formation was noted and this probably altered the oxygen content, and the final pore formation [36][37][38], therefore certainly further studies with variations on the above-mentioned parameters are needed.

Conclusions
Powder spreadability was analyzed quantitatively via image processing algorithms using MATLAB. According to the applied algorithm on the tested powder blends, average grayscale values (0 to 255) lower than 95 on cropped images, and black to white ratio higher than 0.32 on threshold applied images, irregularities were observed. It was noted that when the powder mixture had a high flowability, it caused a reduction in the angle of repose (AOR), as seen for Mix3, demonstrating poor spreadability. In addition, when the volume fraction of fine particles (e.g., below 10 µm) increased, enhanced agglomeration decreased the flowability, and caused a similar poor spreadability effect. For example, Mix1 had the highest volume of particles sized below 10 µm (28.06%) and it demonstrated to have an average grayscale value lower than 95, and a black to a white ratio higher than 0.32. Therefore, it could be stated that when flowability is neither too high nor too low, but the one allowing an optimum coating, high-quality spreadability can be obtained. While the spread layer irregularities may cause unwanted microstructural developments, they can also be deliberately used to form porous components depending on the user's requests.
3D printed samples were found to be porous which cannot be merely related to the poor spreadability but instead, is due probably to the combination of various factors including spreadability, spatter formation, scanning speed, laser speed, and other PBF-LB processing parameters [39,40]. As shown that it is currently not an easy task to investigate the correlation between spreadability and standardized characterization techniques. To propose a fully accepted method to quantify spreadability, there is still a need for more detailed and statistical studies. For example, while in the current work, spreadability was examined from the layers distributed at constant spreading speed (200 mm·s −1 ) with continuous layer thickness (50 µm); the alteration of these parameters and their impact with varying powder characteristics on the final component morphology together with further image enhancement techniques (e.g., contrast enhancement) will be conducted in future studies.
Author Contributions: Conceptualization, supervision, project administration, and writing, review, and editing, C.V.; formal analysis and investigation B.H.; data curation, writing-original draft preparation C.V. and B.H. Supervision, writing-review, and editing L.B. All authors have read and agreed to the published version of the manuscript.