Process parameters effect and porosity reduction on AlSi10Mg parts manufactured by selective laser melting

A study was carried out about the processing parameters influence of an AlSi10Mg part manufactured by a selective laser melting process in order to determine its optimal values and improve density and mechanical performance. For this purpose, a design of experiment (DoE) was realized by the Taguchi method selecting a set of appropriate processing parameters (9 combinations), resulting in 45 samples whose relative densities were measured by Archimedes’ principle and metallographic image digital processing. The results showed that a relative density of up to 99.65% was achieved from raw data. While the statistical model by the Taguchi method indicated a comprehensive relationship between process parameters based on relative density, laser power was of major significance among the parameters. The Taguchi model suggested new values of process parameters, achieving AlSi10Mg parts with 99.98% of relative density. To contrast the mechanical properties of both optimized and non-optimized parts, they were characterized by tension testing.


Introduction
Selective laser melting (SLM) is an additive manufacturing (AM) technique that is classi ed as laser powder bed fusion.SLM uses metallic or ceramic powder as supply material and a device called recoater which spreads a bed powder on a substrate/plate (support) where its thickness can vary around 20 µm -1 mm [1].Next, a high-density power laser interacts with the bed powder to consolidate and melt metallic particles whose melting regions obey a CAD model of the piece.The previous process is repeated layer by layer until the piece is nished.
The metal powder plays a signi cant role in an AM process and some commercial alloys are successfully used in the FSL process (over 20 alloys) [2][3].Within appropriate aluminum alloys (Al) for the FSL process, there are Al-Si alloys such as AlSi10Mg, Al12Si, A357, and A356.The AlSi10Mg parts by SLM own higher tensile strength than those manufactured by traditional means due to speci c metallurgy conditions that are inherent to the SLM process [2,[4][5][6].Moreover, the high thermal conductivity of Al allows rapid solidi cation and ne microstructure [4][5][6][7][8].
However, metallurgy defects of the SLM process such as porosities by induced gas and lack of fusion, presence of oxide, delamination between layers, high residual stress level, and deformation, among others [9][10][11][12], are inherent characteristics of the SLM process and it decreases the part performance.
Nevertheless, the usage of suitable process parameters during the SLM process allows for reducing the impact or occurrence of such defects, and for this reason numerous studies have been carried out.
The part performance fabricated via SLM is governed by the main process parameters such as laser power P (W), scan speed v (mm/s), and layer thickness t (mm).These parameters can be normalized to one parameter called the energy density E (J/mm 3 ) which is expressed by Eq. 1 and has received special attention due to the correlation between process parameters and nal part properties.An appropriate E is desired to fabricate full dense parts [2].
Eq. 1 To date, there have been several investigations whose purposes were to correlate the energy density to the formation and behavior of defects (as porosity) [7,, since such defects can adversely alter the mechanical performance of the AM product.Amir et al. [14] examined the dynamic properties of AlSi10Mg parts fabricated via the SLM process.They created a Design of Experiments (DoE) based on the Taguchi method to study the most in uential SLM process parameters and relative densities obtained were in the range 99.5 to 99.25 ± 0.05% with energy density ranging from 49.7 J/mm 3 to 74.7 J/mm 3 .
Praneeth, Venkatesh, and Krishna [13] evaluated the process parameters in uence con AlSi10Mg parts by SLM, being laser power and scan speed the most in uential parameters on hardness and surface roughness.They also attained a maximum density of 2.62 g/cm 3 (98.12%)with energy density between 13.89 J/mm 3 and 125 J/mm 3 .While Hyer et al. [16] proposed a processing window that would produce quasi-dense AlSi10Mg samples via SLM which was established between 32 J/mm 3 to 54 J/mm 3 , accomplishing a maximum relative density (RD) of 99.5% measured by Archimedes' principle when samples were manufactured with 29.14 J/mm 3 .A study was carried out by Bai et al. [15] for optimizing the SLM processing parameters of AlSi10Mg parts by Doehlert DoE whose advantages over other DoE's lies in the response surface methodology and requires fewer experiments.They plotted the response surfaces and contours of the regression models and found a maximum RD of 99.77%.This paper aims to determine the relationship between the main process parameters (namely, laser power, scanning speed, hatch spacing, layer thickness, and laser focus) and the pores formation.A speci c processing window was carefully chosen from literature data in order to fabricate almost entirely dense parts (RD > 98%) according to Fig. 1 where a processing window was proposed between 40 J/mm 3 and 70 J/mm 3 represented by dashed vertical lines.A DoE by the Taguchi methodology was adopted with the purpose of minimizing defects (pores and cracks) and optimizing the processing parameters.Tensile samples were fabricated and tested with optimized parameters to evaluate the tensile performance of virtually pore-free parts manufactured by SLM.

Characterization of metallic powder
Gas-atomized AlSi10Mg powders from SLM Solutions Group were acquired.The chemical elements of the metallic powder in wt.% (Si − 9.8, Mg − 1.04, and Al: Balance) were measured by the energy dispersive x-ray spectroscopy technique.Besides, the powder morphology and distribution were examined by scanning electron microscopy (Jeol, JSM-6510 LV).

Design of experiment and manufacturing processes E = P vht
A DoE was proposed whose processing parameters were properly selected, as mentioned in Fig. 1, and designed with three factors: laser power (P), scanning speed (V), and laser focal plane (F) with three levels (minimum, medium, and maximum) per factor.During experiments, the powder bed thickness and hatch space were kept constant (30 µm and 130 µm, respectively).The Taguchi methodology was applied, resulting in an orthogonal array L9 of 9 combinations of process parameters as shown in Table 1.

Total samples 45
The samples were manufactured in SLM 280HL of SLM Solutions® which is equipped with a continuouswave Ytterbium IPG ber laser (1064 nm of wavelength); nominal power and spot diameter laser of 400 W and 70 µm, respectively.The substrate was heated to 200°C during the samples manufacture and the atmosphere in the chamber was controlled via argon with less than 0.1% oxygen.A stripe partition scanning was chosen as strategy scanning where the angle orientation change was 67° among adjacent layers and 45 samples were fabricated as mentioned in Table 1.

Archimedes' method
The RD was measured in accordance with the ASTM B311 Standard Test Methods which is based on Archimedes' method.An OHAUS Discovery balance was used and a resolution of 1×10-9 kg was employed to apply Archimedes' method.Moreover, deionized water 0.1 wt.% and 0.9976 g/cm 3 of density (ρ w ) was utilized as uid during mass measurements.This measurement process was carried out under normal conditions.The RD was calculated according to Eq. 2.
Eq. 2 Where ρ r is a reference density which was taken from the pore-free alloy density (2.67 g/cm 3 ), M s_air is the sample mass in air, M s_w is the sample mass and support immersed in water, and M s is the support mass immersed in water.

Micrographs
One sample was randomly selected per each experiment subset to expose and analyze pores.Each sample were cut in cross-section and normal to the printing direction.Both cuttings were encapsulated by phenolic resin Multifast Black (Struer®) and metallographically prepared with SiC papers from 350 grits up to 4000 grits before being polished using 0.24 µm colloidal silica.Samples were examined in random areas by confocal microscopy ZEISS LMS 700.Micrographs were analyzed by digital image processing via ImageJ software.Pore size and morphology were evaluated by equivalent diameter D (µm) and circularity C, respectively, according to the following expressions: Eq. 3 Eq. 4 where A (µm 2 ) is the area of pores and L (µm) is the pore perimeter.A value of 1 (Eq.4) indicates a perfect circle and when the value is close to 0, it means that the polygon is getting longer and longer.

Tensile tests
The tensile tests were conducted using an MTS 647 testing machine with a rate of straining of 2 mm/min and an extensometer of 25.4 mm was used.Tensile samples were dimensioned according to ASTM E8 (Standard Test Methods for Tension Testing of Metallic Materials) and fabricated using optimized and non-optimized process parameters.The test orientations were 0°, 45°, and 90° about manufacturing direction.

Powder morphology
The AlSi10Mg powder exhibited a spherical shape with agglomerated satellites and some few irregular and elongated particles as shown in Fig. 2 (a).The particle size distribution and average were analyzed by digital image processing via ImageJ software, resulting in a distribution size of D10 = 10.92 µm, D50 = 29.39µm, and D90 = 9.6 µm as shown in Fig. 2 (b) and average particle size of 24,66 µm.

RD and analysis of variance (ANOVA)
A study of data control was realized to eliminate outliers (unusual observations) whose in uences were disproportionate during the results analysis.Table 2 shows RD results and E employed in each experiment.The minimum and maximum RD values were 99.40% for combination number 6 (P = 400 W, V = 1800 mm/s, and F = 0 mm) and 99.648% for combination number 9 (P = 400 W, V = 1800 mm/s, and F = 0.1 mm), respectively.An ANOVA was carried out from obtained measurements of RD (Table 2) using a statistical software and set up to 95% con dence level (α = 0.05).The ANOVA response (Table 3) suggests that the null hypothesis is not rejected because of the F-value for each factor was less than the critical value F (F = 19).Whilst the t general linear model showed an acceptable adjustment (R 2 = 0.88) between process parameters and RD results.Besides, the ANOVA pointed out that the RD was mainly affected by the laser power (72.73% contribution) for the proposed parameters window; while the laser focus laser and scanning speed contributions were low (8.64% and 6.79%, respectively).
The ANOVA exhibited in Table 3 determines if the association (relationship) between the response and factors in the model would be statistically signi cant.This ANOVA took a risk of 5% in order to conclude that exists a relationship when there is not a true association.The p-value for each factor is less than α = 0.05, i.e., a signi cance value of 0.05 could indicate that factors do not have any effect on the statistical model output response.However, the null hypothesis was not rejected because the f-value was less than the critical value (Fischer rejection).The Fischer rejection criterion considers degrees of freedom of each factor whose critical value was 19 for a con dence level of 95%.

Pores morphology
The Fig. 3 shows images taken from confocal microscopy which were digitally processed, resulting in circular (induced by gas trapping) and irregular (induced by the process) pores due to lack of fusion.Pores diameters were measured in each combination, as longitudinal as cross section, resulting in pores diameter among 4.95 µm − 8.5 µm in cross section and 4.97 µm − 7.9 µm in longitudinal section.The behavior of average pore diameter is shown in Fig. 4 This difference of the pore diameter may be related to a non-homogenous powder bed distribution and particles were scattered during the consolidation process, yielding voids on the surface.Further, the measured pores size can be classi ed as micro-pores (0.1 µm y 100 µm) according to [28].While other studies measured circular pores several tens of microns in size [19,29].

Optimization and validation of the response
To establish a suitable optimization, the objective function is xed to maximize the RD response from DOE.The multiple linear regression model was employed to estimate the relationship between process parameters (independent variables) and RD (dependent variable) according to Eq. 5 which was solved to calculate (predict) RDs for different factors and levels (see Table 1) and its behavior can see in Fig. 5 when F = 0 mm; resulting in a maximum RD = 99.62% when the process parameters are P = 400 W, V = 1800 mm/s, and F = 0 mm.Besides, the relative error was calculated between experimental and predicted RD also displayed in Table 1.
Eq. 5 Based on the nding described above (RD = 99.62%),new samples were manufactured with such optimized process parameters (see Fig. 6).RD measurements were again realized by Archimedes' method and image digital processing, resulting in a RD = 99.98% with relative error of 0.37% compared to the prediction of multiple general linear regression.The Fig. 7 is displayed the longitudinal and cross section from optimized samples where average pores diameter were 6.45 µm and 7.26 µm, respectively; achieving better RD than predicted one.

Microstructure characteristics
The typical microstructure of AlSi10Mg parts manufactured by SLM and viewed in cross section (perpendicular to printing direction) is displayed in Fig. 6a where the shape melt pool shows an overlap among them due to a complete melting and good consolidation among layers.Moreover, in Fig. 7 (a) can be observed continuous lines (melt pool boundary) with a sh scale morphology.The Fig. 7 (b) presents the melt pool morphology in longitudinal section (parallel to printing direction) whose shape was as ellipse elongated in direction to laser scanning and rotation angle of 67°.The different sizes of ellipses in Fig. 7 (b) were due to melt pools affected several previous layers during bed powder consolidation.
In Fig. 8 is shown in detail the boundary melt pool which presented three different microstructures in AlSi10Mg parts fabricated by SLM.The central region of the melt pool (CMP) displayed ne cellular grains due to laser exposure time was relatively low, yielding rapid cooling and solidi cation and therefore hindering the formation of homogeneous crystalline nuclei.Furthermore, the cellular growth direction was parallel to local heat ux, resulting in a ne cellular structure.The boundary melt pool (BMP) had a coarser cellular grain because of the melt pool deepened several previous layers.And lastly, the heat affected zone (HAZ) was identi ed in adjacent melt pools whose grains were undergone to re nement.Such regions were reported in previous studies.
In Fig. 7 is made evident that most pores were located in regions near the BMP because of there were interstitial spaces among metal particles occupied by inert gas and when the laser scanned the powder bed, these interstitial spaces were trapped as bubbles by liquid metal and driven to the melt pool bottom due to the phenomenon known as recoil pressure.Hence, pore formation mechanisms are doubtful to coexistence of liquid and vapor phases during the SLM process.
A mapping of the parts manufactured with optimized process parameters is displayed in Fig. 9 both longitudinal and cross section.A higher concentration and larger pores were found at part edges, exhibiting a pore diameter between 135 µm and 140.4 µm and being larger than pores located at the part core (6.45 µm and 7.26 µm).The high E used in edge process parameters (99.72 J/mm 3 ) was away from the processing window (32 J/mm 3 to 54 J/mm 3 ) proposed in this research and the edge process parameters were P = 350 W, V = 900 mm/s, Hs = 0.13 mm, and h = 30 µm.Besides, some researchers have pointed out that parts manufactured with E > 70 J/mm3 showed most presence of pores due to a phenomenon known as keyhole [17,[30][31].

Tensile properties
The results of tensile tests are shown in Fig. 10.The ultimate tensile strength (UTS) on optimized samples was higher than non-optimized ones, e.g., The UTS of samples manufactured to 0° direction resulted in 448.85 MPa, being 36.46%more resistant than no-optimized samples with same direction (328.92MPa).Besides, the yield strength (σ y ) of optimized samples also exhibited better mechanical behavior in 0°, 45°, and 90° directions whose increases were 36.10%,13.11%, and 21.14%, respectively.Both stiffness and strain of optimized samples signi cantly increased in 0° and 90° directions with regard to non-optimized ones, whose increases were 36.10% and 21.14% in elasticity modulus, respectively; while in strain the increases were 53.04% and 10.06%, respectively.In Fig. 11 is shown strains in non-optimized and optimized samples, achieving a homogeneous strain with regard to the manufacturing direction in optimized samples.
The mechanical response of optimized samples, regardless of the manufacturing direction, improved the tensile properties in all aspects.Therefore, the optimized RD enhanced the tensile properties of AlSi10Mg samples due to pores reduction (0.02%) and pore size range reduction (6.45 µm -7.26 µm) compared to non-optimized samples.

Conclusions
The process parameters effects was investigated about porosity reduction, characteristics microstructure and tensile properties of AlSi10Mg parts fabricated via selective laser melting.The main conclusions were the following: -The measured relative densities, according to the proposed processing window, were greater than 99.62%, no apparent cracks, and few pores in longitudinal and cross section.
-The analysis of variance was performed via a multiple linear regression model and found that the laser power was most signi cant process parameter among studied factors whose contribution to relative density response was 72%.
-An optimization process has been carried out for maximizing the relative density response as objective function and resulted in optimum process parameters: 400 W laser power, 1800 mm/s scanning speed, and 0 mm laser focus calculated via a multiple linear regression model.According to statical model, the relative density was predicated 99.6%, while measured density relative resulted in 99.98%.
-A microstructure characterization was realized in AlSi10Mg samples which exhibited melt pools.In central and heat affected zones of melt pools showed ne cellular grains, while melt pool boundaries displayed coarser grain.
-Tensile tests were performed both optimized and non-optimized samples.The tensile properties of optimized samples presented better performance than non-optimized ones in all aspects due to pores optimization and reduction realized in this study.
Funding The received nancial support provided by the Institutional Fund for Regional Scienti c,   Analysis of porosity and pore shape in longitudinal area.
Average pore diameter behavior in each combination.
Optimization contour plot to predicate RD according to Eq. 5 when F = 0 mm.
New samples manufactured with optimized process parameters.P = 400 W, V = 1800 mm/s, and F = 0 mm.
Microstructure of optimized AlSi10Mg samples in a) cross and b) longitudinal section.Identi cation of large pore concentration at the part edges.
Bar chart of strain in 0°, 45°, and 90° directions for non-optimized and optimized samples of AlSi10Mg.

Figure 2 A
Figure 2

Table 1
Design of experiments by Taguchi method.

Table 2
Experimental measurements of RD according to DoE where P is the laser power, V is the scanning speed, F is laser focus, E is the energy density, RD is the relative density, and S is the standard deviation.