Influence of SLM block support design on geometrical quality of AISI 316 l cantilever features and its impact on post-processing

In Selective Laser Melting, the design of efficient support structures is the key to enable the production of high-quality functional parts exhibiting complex shapes with improved geometrical accuracy. Nonetheless, from a process point of view, supports are waste material that must be minimized to reduce production costs and post-processing. Despite the recent technological advances, support optimization is based on time- and resource-consuming trial-and-error experimental campaigns, while support removal is primarily a manual operation which requires a consistent human effort and consumable consumption. Nowadays, the industry is demanding a tool capable to optimize support design and placement based on part geometry and building orientation, by ensuring high part geometrical accuracy along with reduced timing for post-processing operations. Specifically, the purpose of this experimental campaign, is to evaluate the influence of support thickness and tooth length on the dimensional accuracy of AISI 316 l cantilever specimens in order to form a solid baseline of knowledge for the future realization of an automated algorithm for optimized support structure generation based on both part and process requirements. The experimental results show that the support thickness strongly affects the final part distortion, reducing the as-built geometrical deviation by 72.6% when wall thickness increases up to 0.7 mm, whereas tooth length has a higher impact on post-processing when decreased from 0.7 mm to 0.3 mm, reducing support time removal and consumables usage respectively up to 40.5% and 72.7%. The achieved results highlight that the implementation of optimized support structures ensuring low geometrical deviation and involving reduced resource consumption in post-processing is feasible. These findings provide the starting design rules for the engineering of an empirical methodology, based on thermomechanical modelling, enabling optimized design and implementation of SLM support structures.


Introduction
Additive Manufacturing (AM) unleashes its full potential in the generation of complex, high-value-added geometries unachievable through conventional subtractive manufacturing technologies [1].As AM gains attention in the Aerospace and Medical sectors, Selective Laser Melting (SLM) is an AM technology enabling the fabrication of high-customized functional metal components [2].SLM is a Laser-based Powder Bed Fusion (L-PBF) process that involves a high-density laser source to melt metal powders layer by layer creating near net shape and high densities (up to 99.9%) functional parts [3].Surface roughness up to Ra 2 μm as well as excellent dimensional accuracy (e.g., see [4,5]) are achievable thanks to the implementation of thin layer thickness ranging between 20 μm and 50 μm.In SLM, the fabrication of functionally complex part designs is enabled by the design and implementation of support structures which are defined as any material used to provide structural support to unsupported portions (e.g., overhangs, vertical holes, cantilevers, bridges) or critical features (e.g., thin wall features) of a printed part during the build process [6,7].The use of supports is critical to produce high-quality metal parts in terms of dimensional, geometrical accuracy and structural integrity [8].Miuri et al [9] demonstrated that the staircase effect and consequently the surface roughness of down-facing surfaces is reduced by support optimization, reducing curl formation which compromises the printing process by inducing local detrimental indentation of the recoater blade.Moreover, a proper design and positioning of support structures promotes heat dissipation, alleviating the high thermal gradients generated by melt pool formation [10], and prevents the formation of detrimental residual stresses [11]which undermine the structural integrity of the metal part in terms of mechanical properties and geometrical distortion [12].Mishurova et al [13] demonstrated that the use of supports can reduce up to 15% of the residual stresses within SLM Inconel 718 specimens.
In several industrial applications, support removal remains a manual post-processing operation that impacts process sustainability by increasing product lead time and by requiring consistent consumption of consumables and human resources [14].Different types of support structures are used in SLM part manufacturing (such as block, web, dot, line, and contour structures) [15].Järvinen et al [16] studied the influence of web and tube supports on SS 17-4 PH specimens and concluded that web supports are easier to remove, leading to better surface quality.Support optimization is more critical when high-performing metal alloys are used [17] due to their high mechanical strength and hardness which make support removal more difficult and expensive.Several authors faced this topic for Ti-6Al-4V and Stainless Steel [15][16][17][18][19], but the impact of support design on the economics of the SLM process was poorly discussed.
To the best of the authors' knowledge, the industry still lacks a robust and cost-effective tool that enables efficient support design and placement based on part geometrically complexity and building orientation, ensuring high print quality in conjunction with cost-effective post-processing.This work is the first step of a wide research activity aiming to establish a flexible, scalable, and effective workflow, based on an empirical algorithm, for the design and positioning of support structures, taking into account both part requirements (e.g., geometrical complexity based on functional design, part building orientation) and industrial needs (e.g., reduced product lead time, low consumable consumption, low material waste, increased process sustainability).However, before proceeding with the implementation of this tool, it is necessary to understand and evaluate the influence of support generation parameters on the surface quality and dimensional accuracy of the finished part as well as the impact on the post-processing.Specifically, the purpose of this paper is to analyze the influence of block support structures' design on the dimensional and geometrical quality of AISI 316 lcantilever features to determine minimal support/part contact areas promoting cost-effective support removal and minimizing geometrical deviation of the realized SLM metal parts.The main input variables affecting support design and integrity are identified and analyzed through a full factorial Design of Experiment (DoE) and their impact on the post-processing phase is evaluated in terms of consumable consumption and detachment time.The outputs achieved by the performed research work aim the definition of starting design rules for the design and development of an empirical methodology based on thermomechanical modelling for an optimized positioning and implementation of support structures.
Section 2 deals with the material and experimental method used for sample manufacturing and characterization.Section 3 introduces and discusses the main experimental outputs observed from offline sample characterization.In section 4, the main conclusions are reported.

Material and experimental methods
The metal powder involved in the experimental campaign is an AISI 316 lpowder (grain size ranging between 15 μm and 45 μm) supplied by Carpenter Additive [20].Powder chemical composition was validated via Inductively Coupled Plasma Atomic Emission Spectroscopy (ICP-OES) and is reported in table 1.
The employed L-PBF system is an SLM Printsharp 250 machine, as shown in figure 4. (system provider: Prima Additive [21]).The system is equipped with a single mode IR fiber laser with a maximum power of 500 W and a laser spot of 70 μm.ISO 14175 Argon 4.6 is used to fill up the printing chamber and reduce the content of Oxygen below 1000 ppm before the printing process.To evaluate the influence of the implemented support structures on part geometrical quality, a cantilever feature is realized on an AISI 316 lsubstrate with the goal to emphasize part distortion at the overhanging surface, specifically on the external point A (see figure 1(a)).The process parameters involved in sample and support fabrication have been optimized in a previous experimental campaign to achieve high density AISI 316 lcubes (i.e., 99.8%) and they are given in table 2. Once the process is finished, the realized samples are detached from the substrate via Electrical Discharge Machining (EDM).To avoid residual stresses relaxation and, consequently, the reduction of part elastic deformation, no stress relief is performed.

Design of experiment (DoE)
An experimental campaign based on two factors and three levels full factorial DoE and an analysis of variance (ANOVA) without replication is implemented.The analysed support structure is a block type, designed through Autodesk ® Netfabb software by setting the default rhomboidal perforation, raster creation with 2 mm of hatching distance, and 2 mm height triangular tooth connection with the part.From a previous analysis performed at the Automation Robotics and Machines laboratory of SUPSI [22], the two main input variables affecting support stiffness, part distortion, and support removal are (see figure 1(b)): • support thickness: defined as the wall thickness along support growth direction; • tooth length: defined as the length of the support tooth at the connection with the part.
For both variables, three experimental levels are chosen: 0.3 mm, 0.5 mm, and 0.7 mm. Figure 2(a) shows the arrangement of the cantilever features on the SLM substrate, while figure 2(b) represents the executed DOE table, where the levels 0, 1, and 2 stand for 0.3 mm, 0.5 mm, and 0.7 mm respectively.The influence of the designed support structures on the final sample quality and post-processing is evaluated based on three outputs: part distortion in terms of maximum geometrical deviation between the nominal CAD and the realized geometry, support integrity in terms of breakage and bonding with the cantilever surface (SB), and resource consumption (RC) involved in support removal.Part distortion analysis is performed based on the comparison between the nominal cantilever CAD and the actual realized feature by considering the displacement (AD) of the external point A, which is defined as the intersection point between the symmetrical plane and a horizontal one defined at Z = 21 mm from the base surface (see figure 1), and the geometrical deviation of the cantilever rounded area (CD) highlighted in figure 3.
Support removal is performed manually by a cutting tool equipped with a rotating disk, while SR is evaluated by measuring both the time required for support separation and the degradation occurring at the cutting disk size (i.e., size variation of the disk diameter before and after support removal).- A GOM ® Atos Core 200 3D scanner [23] with 0.08 mm maximum spatial resolution is used to acquire the actual geometry of the realized features, while the GOM ® Inspect suite is employed to assess part deformation and final dimensional quality.Crack identification and quantification are performed through a Keyence ® VHX 7000 digital microscope [24] with 150X magnification and 1.47 μm x/y pixel resolution.

Part distorsion charcterization
The methodology followed for part distortion analysis is schematically represented in figure 3.
Firstly, the nominal CAD geometry is aligned with the acquired mesh by defining three reference geometric elements (step 1).To emphasize part distortion on the overhanging surface, CAD/mesh alignment is performed by using three reference planes which are constructed (both in the CAD and mesh) according to the method of fitting points.A plane on the base surface of the component (the green one), one on the back vertical surface (the red one), and one on the symmetry plane (the blue one) are defined.Table 3 summarizes the resulting alignment error for each realized cantilever sample.After sample alignment, the deformation experienced by the realized cantilever structure is assessed by evaluating the displacement of the point A along the X direction (see figure 3distortion of point A).The aim of this analysis is to quantify the deformation experienced by the cantilever structure against the original CAD.Displacement measurements are performed at the point A defined as the intersection point between the symmetrical plane and a horizontal plane defined at Z = 21 mm from the base surface.
Finally, the characterization of the geometrical deviation of the cantilever rounded surface is performed (see figure 3-cylinder deviation analysis).To perform this analysis, both in the nominal CAD and in the actual mesh a cylinder is constructed by fitting point method and the deviation between their mutual axes is measured.The results are used to characterise the inclination of the lower surface of the cantilever with respect to the nominal XZ plane and YZ plane, as shown in figure 3. The distortion measurement coming from AD and CD analysis provide important information about the effectiveness of the wall thickness in limiting part distortion.

Analysis and discussion of the results
Table 4 summarizes the experimental results for the four evaluated outputs (i.e., AD, CD, BS, and ER).To assess statistical significance, the resulting p-values are analysed based on a 95% confidence interval (i.e., 0.05 a value).In the next paragraphs the experimental results are introduced and discussed.

Actual displacement of point a (AD) and cylinder deviation (CD) analysis
Based on p-value analysis, table 4 shows that both support thickness and tooth length have a significant influence on AD, with p-values of 5e-4 and 0.032, respectively.The experimental results also highlight the decrease of part distortion with support thickness.Moving from a wall thickness of 0.3 mm to 0.7 mm, the maximum displacement experiences a 72.6% reduction, decreasing from an average value of 1.72 mm to 0.49 mm (figure 5(a)).By increasing the tooth length from 0.3 mm to 0.7 mm, AD reduces from 1.17 mm to of 0.99 mm, observing a 15.4% distortion reduction.This trend is also highlighted by the contour plot in figure 5(b), where the support thickness is plotted in the Y-axis and the tooth length in the X-axis.The banded pattern along the Yaxis confirms the influence of support thickness on the geometric part distortion is more significant than tooth length, showing a strong reduction in displacement as thickness increases.The observed results agree with [25] where the inherent strain is computed by assuming a Gaussian laser source.Specifically, when the deposition of a new layer begins, the total inherent strain vector ò_tot is computed as the sum among the elastic strain (ε e ), the plastic strain (ε p ), and the thermal strain (ε t ): In a simple welding process, since the deformations are analysed at the end of the process, the thermal strains can be neglected as they only have an effect during the process.Moreover, if a stress-relief heat treatment is performed, the elastic strains can also be ignored since the residual stresses are released by the performed heat treatment.
Based on these assumptions, equation (1) can be rewritten as follows [26]: where the total inherent strain is equal to the plastic one.
In general, when metal AM processes are considered, the previous assumptions become invalid since the elastic deformation of the previous layer affects the total distortions of the subsequent layers [26].Moreover, in  the specific case of this research paper, the influence of the elastic deformation due to the residual stresses cannot be ignored since a stress-relieving is not performed.Based on these considerations, equation (1) should be considered as valid.
Assuming an ideally plastic-elastic material, the plastic strain vector component can be obtained through the Prandtl-Reuss equation: where , x s ¢ , y s ¢ z s ¢ are the instant deviator stress in the x, y, and z direction, and dl is the instant positive constant of proportionality [25].Despite most residual stresses are released as a result of the detachment of the component from the baseplate, the elastic strain experienced by each single layer during part manufacturing cannot be neglected since it affects the resulting distortions of the subsequent layers.By using the linear relation between stress and strain defined in [26] and assuming an isotropic behaviour of the material, the elastic strain along the three spatial directions can be expressed as: where D is the elasticity matrix and , , x y s s and z s are the stress vectors in the three directions x, y, and z.Finally, the thermal strains generated during the deposition of one layer can be obtained with the following thermal equation: where e a is the coefficient of thermal expansion of the AISI 316 lmetal powders.
By combining equations (3)-( 5) into the equation (1)for the calculus of the total strain, it can be found that ò_tot can be stated as: T D of the equation ( 7)can be expressed by using Fourie's law of heat conduction: where Q is the amount of heat delivered by the laser source, t is the exposition time of the laser, A is the area where the heat is dissipated, and k is the thermal conductivity of the material.Heat accumulations and high thermal gradients generated during the SLM process have been demonstrated to promote the formation of high residual stresses, which, by exceeding the elastic limit of the material, increase the plastic component of total deformation ò tot .[25,27].If Q, t, and k are assumed as constant, thermal gradients are mainly affected by the size of the block support thickness.Indeed, by increasing support thickness, the support/part contact areas also increase, reducing the heat flux in the cantilevered section and leading to smaller thermal gradients.The observed experimental results tabulated in table 4 agree with the theoretical considerations expressed by the equations equations ( 6)-( 7), showing a reduced geometrical deviation of point A with increasing support thickness and highlighting a strong dependence of the total deformation on the amount of heat dissipated during the SLM process.Because of the implemented block support structure, high thermal dissipation is enhanced by thicker support structures rather than longer tooth lengths, resulting in a reduced total deformation thanks to limited residual stresses' formation.
Contrary to what observed in AD, in CD analysis ANOVA highlights a low statistical significance of the tooth length compared to the support width (i.e., p-value of 0.011 and 0.386 for wall thickness and tooth length, respectively, see table 4).From the two measured angles, respectively the XZ and the YZ angles, the YZ exhibits low deviation from the nominal (i.e., ranging between 0.28°to 0.01°), confirming a negligible deformation occurs along the longitudinal direction of the specimen.On the contrary, the geometrical accuracy of XZ strongly depends on the support stiffness, exhibiting a deviation of the measured angle ranging from 1.63°to 0.03°for wall thickness of 0.3 mm and 0.7 mm respectively (see figure 5(c)).Since CD is mainly focused on the XZ plane, a strong correlation exists between the previous detected output (i.e., AD) and the geometrical deviation of the cantilever rounded surface (i.e., CD).Indeed, the greater the deviation of point A from the CAD surface, the greater is also the deviation angle between the actual and the CAD cylinder.The contour plot confirms the observed behaviour, highlighting that the geometrical distortion of the cantilever structure and consequently the measured cylinder axis deviation from the nominal mainly depends on support thickness (see figure 5(d)).Moreover, for each realized feature, the deviation angle between the realized support structure and the vertical is measured by means of image analysis (see figure 6(a)).The strong deformation of the support structure observed when 0.3 mm support thickness is implemented confirms the strong influence of the wall thickness in reducing the geometrical distortion of the realized features.

Breakage of supports (BS) analysis
The optical analysis performed with the digital microscope highlights the presence of structural cracks along the support/part connection interface for all the cantilever features implementing 0.3 mm support thickness (see figure 6(b)), confirming that the lowest level of support thickness is not suitable to withstand the high heat fluxes generated during the SLM process.By decreasing the contact area between the supports and the part, the thermal gradients along the building direction increase following equation 7, resulting in increased thermal stresses and causing the part to fail when these stresses exceed the Ultimate Tensile Strength (UTS) of the material.
In addition, according to the formula: for the same amount of force F acting on the part, the point of contact between the support and the workpiece results in an area of stress concentration because of the smaller cross section A, which is why failure occurs at the point observed in figure 6(b).Nevertheless, thermal analysis validation will be performed to corroborate this statement.
In addition, cracks are also detected in the support structure of specimen 2 (0.3 mm support thickness and 0.5 mm tooth length), and 5 (0.5 mm support thickness and 0.5 mm tooth length).Since the observed cracks do not compromise the integrity of the contact surface between the supports and the realized feature, they are considered negligible.

Support removal (ER) analysis
ANOVA analysis highlights the significance of support thickness on the post-processing time and consumable consumption in the support removal phase.Table 5 and figure 6(d) show the detected wear of the cutting tool disk as difference between the starting and the final measured disk diameter, and the total time required for support removal.For thinner support structure (i.e., 0.3 mm), support removal requires low post-processing time (i.e., 9 min in average) and low disk wear (i.e., 1.05 mm in average).Removing supports with 0.3 mm wall thickness and 0.3 mm tooth length takes approximately 8 min and the cutting disk diameter reduces of 0.49 mm, starting from Ø38.2 mm and ending to Ø37.71 mm.An increased of the tooth length increases the post-processing time and the tool wear up to 12 min and 1.97 mm, respectively.For 0.5 mm of support thickness, post-processing time range between 10 min and 14 min, showing an average tool wear of 1.81 mm.Compared to 0.3 mm and 0.7 mm support thickness, these support structures are not sensitive to tooth length as shown in table 5. On the contrary, a further thickening the supports wall increases the time and the deviation of the cutting disk diameter, reaching 30 min for removal, and 9.01 mm diameter reduction for the support structures consisting of 0.7 mm support thickness and 0.7 tooth length.

Optimization of the parameter combination
The Minitab Response Optimizer (RO) tool is used to optimize response variable to identify the optimum combination of process parameters (i.e., support thickness and tooth length) capable to minimize part  distortion and resource consumption in the post-processing phase.The generated Composite Desirability parameter allows to evaluate the quality of a solution generated by the RO, assessing the best combination of process parameters ensuring the desired response.The Composite Desirability ranges from 0 to 1, where 1 indicates that all goals and constraints are met.The input variables for RO are the DOE output variables (i.e., AD, CD, BS, and ER).A 3D scatter plot, showed in figure 7, is used to evaluate the output of the RO.On the x-axis and y-axis, respectively the tooth length and the support thickness are represented, and on the z-axis the Composite Desirability is shown.
To enable RO analysis, each input variable needs to be weighted arbitrarily choosing a value ranging from 0.5 to 2. In this research work, the most critical parameter, which is assigned an importance value of 2, is BS (i.e., the breakage of supports).Indeed, from the application point of view, supports with cracks or breaks cannot be implemented, as they do not enable defect-free part manufacturing.The variables support's manual removal and the observed geometrical deviation of the realized cantilever feature are assigned to an importance value of 1.These parameters do not prevent part production but affect both product final quality and post-processing efficiency.Finally, the cylinder axes parameter was assigned an importance of 0.5.The reason is due to the strong correlation between AD and CD (see section 3.1).The goal set in the RO for parameter optimization is the minimization of all variables, this means looking for the combination of parameters that minimizes all input variables.As highlighted from figure 7, the optimized combination which best satisfies the output variables, is 0.7 mm of support thickness and 0.3 mm of tooth length.This combination, in fact, reaches a Composite Desirability of 0.8318 which is the highest off all the available combination of parameters.
The points in the graph show that support structures implementing a thickness of 0.3 mm are non-optimal due to the crack formation which make the part unusable in an industry point of view.The resulted graphic strengthens the analysis performed in the previous chapter, which states that the support thickness was the major responsible for heat dissipation and thus to minimum deviation, whereas the tooth length had an influence in the removal from the part since they modified the connection area between specimen and support structure.For this reason, the optimal combination of parameters for supports generation are a structure thickness of 0.7 mm and a tooth length of 0.3 mm.In fact, such supports achieve the best trade-off between lower part distortion (0.65 mm) and easier support removal (2.46 mm).

Conclusion
A 3 2 full factorial DoE and an ANOVA without replication are implemented to analyse the influence of support thickness and tooth length on AISI 316 lpart geometrical distortion, estimating the impact of support design on the post-processing phase in terms of consumable consumption and removal time.The main findings are as follows: Support thickness is the main support design variable affecting the final geometrical accuracy of the realized AISI 316 lcantilever structure.Increasing the support thickness reduces the temperature gradient within the layer and results in lower geometrical deviations by limiting the total strain vectors and the resulting residual stresses.The tooth length has a small statistical influence on ER and AD.On the contrary, by reducing tooth length up to 0.3 mm, the total time for support removal reduces up to 40.5% along with a consumable saving of 72.7%.Specimen with 0.3 mm thickness show structural cracks along the connection zone between supports and the realized feature.
The best support structure design ensuring low geometrical deviation and reduced resource consumption in the support removal phase consists of a support thickness of 0.7 mm and a tooth length of 0.3 mm, corresponding to 0.65 mm AD, and a 50% and 33% reduction in cutting disc consumption and removal time, respectively, compared with the support structure implementing the equal wall thickness and 0.7 mm tooth length.
These experimental outputs will provide the basis on which an empirical algorithm enabling automated and optimized support design and implementation will be designed and developed.

Figure 1 .
Figure 1.(a) cantilever feature realized in the experimental campaign [mm]; (b) support thickness and tooth length definition.

Figure 2 .
Figure 2. (a) positioning of the specimens on the substrate; (b) design table of the 3 2 full-factorial DOE and sample coding.

Figure 3 .
Figure 3. Flow chart of part distortion and ease of removal analysis.Step 1, alignment by geometric elements of the mesh obtained from the three-dimensional scan with the CAD geometry; Step 2, analysis of the maximum deviation due to thermal shrinkage of the component; Step 3, analysis of the variation between cylinder axes; Step 4, manual support removal.

Figure 5 .
Figure 5. (a)-(b) main effect and contour plot of support thickness and tooth length on AD; (c)-(d) main effect and contour plot of support thickness and tooth length on CD.

Figure 6 .
Figure 6.(a) specimen '00' with support thickness 0.3 mm, and specimen '30' with support thickness 0.7 mm; both specimen with tooth length of 0.3 mm.(b) crack present on every 0.3 mm support thickness specimens; (c) crack formation on specimen '02', and on specimen '21'; (d) Behavior of the diameter deviation of the Dremel disk on the sample number.

Figure 7 .
Figure 7. Response optimizer graphic for choosing the best parameter combination.

Table 3 .
General alignment error coming from CAD/mesh comparison and angle actual deviation.

Table 4 .
Experimental data observed of the different outputs and statistical significance of the input variables through p-value.

Table 5 .
Results of the consumable usage and time for support removal.