1. Introduction
Additive manufacturing (AM), machining (subtractive), and forming are three major classes of manufacturing [
1]. AM is a broad term that encompasses a variety of different technologies and processes for fabricating parts using layer-by-layer addition of materials [
2,
3]. In industry and academia, AM processes are referred to by a variety of terms, including additive fabrication, three-dimensional (3D) printing, and direct digital manufacturing [
4,
5,
6]. There are numerous processes and technologies for AM that follow the same fundamental process but the mechanism for creating layers varies.
Figure 1 illustrates the steps involved in a general AM process. The AM process begins with the creation of a Computer-Aided Design (CAD) file, which can be obtained either by designing a new part or by scanning an existing one. In the second step, the CAD model is converted to the Standard Triangulation Language (STL) file format. This is a critical step because different CAD packages have varying representations of solid objects. To ensure consistency, the STL format has been adopted as the industry’s standard input file format. The STL file is then corrected and sliced into thin multiple cross-sectional layers, followed by the generation of support structures. In step 4, the AM machine’s process parameters such as layer thickness, building speed, laser or electron beam power, and scan pattern are configured. The machine reconstructs the generated slices one layer at a time. After step 5, the built part is removed from the AM machine and subjected to post-processing, which may include the removal of support structures, heat treatment, and surface cleaning. All the steps up until layer slices and tool path are identical for all AM, the only difference comes in step 5 where different AM processes can be implemented. The use of a specific AM process is influenced by several variables, such as the energy source, the layer building mechanism, the intended application, the material used, the cost, and the necessary attributes. Materials are one of the most significant aspects that influence the choice of certain AM methods. The AM technique can be used to produce a variety of materials, including polymers, metals, composites, and ceramics. Polymer-based and composite materials have been extensively used in AM technologies due to their low cost, excellent manufacturability, and high-quality characteristics. Metal-based materials are also essential particularly for medical applications like implants due to their osseointegration and biocompatibility characteristics. To address issues like a high melting point, ceramic materials have also been researched from an AM standpoint.
Every AM process, especially when dealing with complex geometries, requires the use of support material. Indeed, supports are added to the part to prevent the part from curling, distorting, sagging, cracking, shrinking, and/or other deformations caused by thermal stresses, as well as to secure the part to the building platform [
8]. Additionally, it simplifies the removal of the fabricated portion of the platform’s bottom and reinforces thin-and-tall components during the construction process. The support structure is a non-functional frame that is fabricated concurrently with the functional part and removed after the AM is completed [
9]. However, numerous complications, including increased fabrication time and material consumption, additional time and effort spent on support design, removal after the part building is finished, etc., can occur as a result of the support structure’s addition [
8,
9,
10]. Additionally, the EBM process’s support structures are predominantly composed of dense metal, which is difficult to remove, necessitating additional time and expense [
11].
The requirements for support structures and the difficulties associated with their use must be balanced through optimization of the design and manufacturing of support structures. Numerous design rules apply to support structures, which must be specified by designers. For instance, support structures should be designed in such a way that they make minimal contact with the overhang surface components [
8,
9,
12,
13,
14]. Researchers have paid much attention to the design and optimization of support structures for AM. In most cases, optimizing the orientation and implementing self-supporting rules can reduce the amount of support structure required, but cannot eliminate the need for the support structure. Thus, optimizing the design of the support structure is critical to minimizing the amount of support material and making it simple to remove while maintaining performance. Typically, contemporary support generation techniques employ specialized structures to assist the overhang regions. These techniques may result in an overestimation of the quantity of support or the position of an excessive number of supports, which may be redundant and add time to post-processing [
15]. The researchers have paid some attention to the design and optimization of support structures. For instance, M.X. Gan and C.H. Wong evaluated three different designs of Selective Laser Melting (SLM)-fabricated support structures. The majority of the intended 18 samples were successfully constructed, except the ’Y’ support structures. Manufacturing of the ’Y’ support structures was halted due to thermal warpages obstructing the construction process [
16]. The algorithm for 3D printing bridge support structures was developed by Zhen-Hong Shen et al. The algorithm identified regions in need of reinforcement generated a set of support points via adaptive sampling, selected new bridges via a scoring function, and connected the bridges and model to pillars. Experimental comparisons were made between the generated support as well as vertical and branching support structures. During printing, the algorithm conserved both material and time [
17]. Rami and Frederic, on the other hand, proposed a methodology for designing and optimizing EBM support structures. They designed and evaluated novel support structures. The results indicated an improvement in efficiency and a reduction in geometric defects [
18]. Sulaiman et al. focused on developing a support structure that could minimize the deformation of the resulting print part while utilizing the least amount of material and being easily removed during post-processing. The results indicated that modifying the sample orientation, design, and parameter values of the support structure had an effect on the build time, removability, and amount of deformation [
19].
Zhang et al. provided a method for polar angles to be supported and shrunk in SLM. The viability of producing an overhang without support structures with various overhang angles was first investigated, followed by the proposal of a method for generating reduced support structures using varying polar angle requirements. The results revealed an average reduction of 35% in support infrastructure [
20]. Vanek et al. proposed a novel method for reducing the number of support structures required for overhanging surfaces created with Fused Deposition Modeling (FDM) technology. The described approach began by orienting the input 3D model into a position that resulted in the smallest possible area requiring support. To keep the total length of the employed support structure points to a minimum, the supporting structure for the decanted point of the area support was constructed incrementally (tree-like shape). Utilizing this approach resulted in a tree-like form structure that efficiently supported the overhanging surfaces while decreasing the material and processing requirements for the support structure [
21]. Zhu et al. employed the tree-supports to support the FDM-created overhang structures. The authors provided a set of formulas for developing tree supports sustainably and reduced the support volume using a combination of particle swarm optimization and a greedy approach. The combination was proven to be beneficial in reducing the volume of tree supports [
22]. The literature has indicated that designing and optimizing support structures is a critical factor in improving the performance of AM parts. Additionally, the literature review has revealed a significant dearth of research on the accurate production of Ti6Al4V overhanging surfaces in EBM. The majority of published research has been on the Selective Laser Sintering (SLS) and SLM overhang surface. The current work proposes tree-like support structures for EBM components and evaluates and optimizes their design and process characteristics.
2. Experimental Procedures
Tree-like support resembles a tree with its structure including branches, and trunk to assist the overhang regions during the AM building process. The support merely contacts the overhang at a few spots with this style of support. The tree-like support is composed of two components: the trunk and the branches. The trunk portion symbolizes the support’s main body, while the branch part serves as a connector between the support and the part. Tree-like support structures have many design parameters such as trunk diameters top (D1) and bottom (D2), branch diameters top (d1) and bottom (d2), the height of trunks (h), minimum distance between rows, branch length, branch angle, and number of branches in each trunk. In this study, only two main design parameters branch top diameters and minimum distance between rows (as shown in
Figure 2) have been considered in addition to AM processing parameters.
The Solidworks software is used to create the ledge overhangs specimen, while the Materialise software is used to support the tree-like support structures. These tree-like structures are generated using the semi-automated method by the Magic’s software and fabricated using the EBM machine. The intended overhang with tree-like support systems is displayed in
Figure 3.
The Design of Experiments (DOE) method is employed in this study to plan the experiments and examine the results. The regulated parameters and their values are chosen in the current study based on the possibility of varying the parameters level and the results of the screening experiments. The selected parameters and their corresponding tree-like support levels are summarized in
Table 1. The remaining tree-like support parameters are kept at their default values. For instance, trunk diameters top and bottom (1 mm), branch diameters bottom (1 mm), maximum branches per trunks (3), branch diameters bottom (10 mm), and minimum distance between points (1.5 mm) as well as the focus offset (0 mA). It is true that the number of branches and the length of the branches will affect how well the support performs; for instance, the support volume will increase with more branches and longer branches, whereas the support removal time will increase with more branches and decrease with greater branch length, and the overhang deformation will decrease with more branches. It is currently not possible to change these factors due to software restrictions, hence they have been kept constant in this work.
The experiments are designed using a response surface methodology based on central composite design, and
Table 2 details the DOE runs. A total of 31 runs are generated using the factors and their respective levels (alpha = 1, 2 factorial point replications, and 6 center point replications).
The developed models are imported into the Magics software and arranged in rows; tree-like support structures for the models are generated in accordance with
Table 2, as illustrated in
Figure 4. The models are then exported along with their supporting structures to the build assembler program for slicing and creating constructed files for the EBM machine.
Using the default preheating and melting parameters, the ARCAM A2 machine is utilized to produce the overhang structures specimens from the Ti6Al4V alloy [
23], where the support structure is constructed using a variety of process parameters (
Table 2). The various EBM components are shown in
Figure 5 [
24]. It utilizes a heated tungsten filament inside a grid cup to generate electrons in the shape of a beam. The electrons emitted by the tungsten filament are heated to a temperature in excess of 2500 °C. Through the anode cup, these high-energy electrons are accelerated. Magnetic lenses focus the electron beam, which is then used to scan the powder bed electromagnetically. The drift tube contains three magnetic lenses: an astigmatism lens, a focus lens, and a deflection coil that controls the direction of the electron beam. The astigmatism lens generates a circular e-beam with a Gaussian energy distribution, while the focus lens concentrates the beam to the required diameter and the deflection coil directs the concentrated beam to the desired location (or scans the e-beam across the building area) [
25]. When the high-energy electrons collide with the titanium powder, the kinetic energy of the electrons is converted to heat energy, melting the powder. Two hoppers hold the stock material, while the titanium powder is gravity fed from cassettes and spread over the build area by the raking blade. The constructed specimen is lowered into the build tank, and as each layer melts away, a new layer of powder is fed on top of the previous melt layer until the build is complete. The process is carried out under a vacuum of 10
−4 to 10
−6 mbar, which eliminates impurities and results in a material with high-strength properties. Throughout the melting process, inert gas (helium gas) is used to reduce the vacuum pressure, allowing the part to cool and the beam to remain stable. Once the build is complete, the parts are cooled to room temperature using helium gas within the machine chamber [
26,
27,
28].
Following specimen fabrication, the unmelted powder is removed using a powder recovery system.
Figure 6 illustrates the fabricated specimens after the unmelted powder has been removed.
Following completion of the fabrication step, the support design and process parameters are evaluated for performance by measuring support volume (SV), support removal time (SRT), and deformation (deviation). The objective is to minimize the volume of support and the time required to remove it while monitoring their effect on the deviation. The STL viewer is used to calculate the volume of the supporting structure. The STL Viewer is a powerful online application that allows you to view CAD models created or exported in the STL file format. The support structures’ STL files are exported from the Magics software and imported into an STL viewer application. The volumes of the structures are calculated automatically in the STL viewer following the removal of the unmelted powder, and the fabricated specimens undergo the support structure removal process. To find the impact of support structure design and process factors on the removability of support structures, specimens are manually removed using simple pliers, and the time required to remove each specimen is recorded. To ensure consistency, all specimens have the same pattern of support structure removal, and the operator is given sufficient rest time (approximately 10–15 min) between specimens to avoid fatigue.
Following the removal of the support structures, specimens are inspected for deformation using 3D comparison analysis in the final step. The 3D comparison approach is employed to appropriately quantify the deformation of the specimens produced by varying the support structure variables. It is regarded among the most robust and complete methods to graphically portray the surface discrepancies between the test surfaces and the reference CAD model [
29]. This deformation analysis using a 3D comparison approach is carried out in Geomagics Control
® (Geomagics Control 2014, 3D System, Valencia, CA, USA) [
30]. The test model is positioned on the reference CAD model using the best fit alignment at the beginning of this procedure. Subsequently, the analysis software automatically computes the best fit between the test and reference object. This best fit alignment guarantees that the test and reference entities are in the same coordinate system. Additionally, the average maximum deviation in the outward direction is chosen as the statistic to measure the deformation. This statistic is used because it reflects the deviation in the specimens, allowing the approximation of the gap between the test specimens and the reference CAD model. The test specimens (specimens manufactured at various tree-like support parameters) are captured as a point cloud set using the laser scanner mounted on the Faro platinum arm (FARO, Lake Mary, FL, USA) as illustrated in
Figure 7.
The surface of the test specimens is scanned and imported as an STL model into Geomagics control
® (Geomagics Control 2014, 3D System, Valencia, CA, USA) in order to assess them with the reference CAD. The results are represented by the 3D comparison analysis software as an error scale (color coded) based on the shortest distance between the test model and the surface of the reference model. The color cod demonstrates the location, magnitude and direction of the discrepancies between the reference and test models using a color-coded scale. Positive deviations represented by yellow-red bars and negative deviations represented by green-blue bars.
Figure 8 illustrates the results of the 3D comparison analysis graphically. According to the findings, there is a significant distortion at the free end as compared to the fixed end. This distortion rises as the location shifts from the fixed to the free end. The positive deviation in the figures represents deformation in the outer direction, which is much greater at the free end as compared to the fixed end.
In order to establish the importance of parameter impacts, statistical analysis using the ANOVA test is also conducted following the 3D comparison analysis. Additionally, the optimization search is carried out in the Mode Frontier program utilizing a multi-objective genetic algorithm (MOGA-II) to create workflows for discovering ideal parameters.
3. Results Analysis and Discussions
As illustrated in
Table 3, three performance measures are investigated, and the ANOVA test is used to determine the significance of parameter effects.
The relative strength of input variables for SV is illustrated in
Figure 9, which is calculated using smoothing spline analysis of variance in the Mode Frontier
® software (Esteco S.p.A., Trieste, Italy). Minimum distance between rows (MDBR) is found to be the most significant factor. It shows that choosing the right value of MDBR is more crucial for minimizing the SV than choosing the Branch top diameters (BTD). The support volume usage will be lower if the MDBR value is low. MDBR may have greater significance because it governs the density of tree structures in a given space. As a result, the SV is greatly increased by decreasing the MDBR. However, since BTD is kept at a low level to preserve the other branches’ spatial positions, it has little of an impact on SV.
The effect of the interactions of the tree-like support parameters on the SV is shown in
Figure 10. The contour map for the SV shows that the SV is greatest when BTD and MDBR are at their lowest values. It points to the fact that the SV would rise greatly when the BTD and MDBR values are 0.5 and 1 mm, respectively. The fact that only MDBR can achieve the lowest SV illustrates the importance of MDBR in lowering the SV. Moreover, the SV can be greatly reduced if we can raise the MDBR while maintaining a constant BTD. For example, the SV is about between 79 and 88 mm
3 at BTD = 0.5 mm and MDBR = 2 mm.
Figure 11 shows the relative strength of input variables for the SRT. Again, MDBR is the main cause of variation in the SRT in comparison to the other variables. It should be obvious that as we reduce the MDBR, we must also remove more support structures. It will also result in reduced accessibility of the cutting tool during the removal process because supports are removed in smaller segments thus raising the SRT. The BTD is only a small fraction of the overall tree; hence, its contribution is obviously insignificant in contrast to MDBR, where the entire tree contributes to the supports.
According to the interaction between MDBR and BTD in
Figure 12, SRT is at its lowest point (about 9 min) when MDBR is at its highest level (approximately 3 mm) and BTD is actually at its lowest level (0.5 mm). This highlights once again how important it is to optimize the MDBR in order to lower SRT.
The effect of the interactions of the tree-like support parameters on the SRT is shown in
Figure 12,
Figure 13 and
Figure 14.
Figure 13 shows that varying either beam current (BC) or beam scanning speed (SS) is not influencing the SRT significantly as it is only varying between 14 and 29 min. Moreover, the SRT is quite high at higher values of BC. For example, at 1000 mm/s (SS) and 1.6 mA (BC), the SRT is quite high approximately 29 min. This is because when the BC grows, the energy input into the supports increases as well, increasing the support strength and lengthening the time needed to remove the supports as a result.
The relative strength of input variables for the overhang portion deviation is shown in
Figure 14. It indicates that BC, SS, and MDBR have an effect on the overhang deviation. BC and SS are the greatest contributors to variation in the overhang deviation. The effect of the interactions of the tree-like support parameters on the deviation is shown in
Figure 15. It shows that reducing the BC or the beam energy density and increasing the BSS resulted in a significant reduction in deformation. Thus, it is critical to find a balance between BC, SS, and MDBR that can give the lowest SV, lowest SRT, and minimal deviation.
Figure 16 makes it clear that the deformation will be at its minimum when MDBR is at its lowest. However, BTD is not particularly significant from the perspective of deformation. By limiting the MDBR, the overhang portion is supported at the maximal number of locations, reducing the likelihood that the manufactured component would deform.
To minimize the volume of supports, the time required to remove them, and warping deformation, multi-response optimization problems are formulated as illustrated in
Figure 17. To maintain a very low level of warping deformation while minimizing support volume and removal time, an additional constraint (Deviation < 0.1 mm) is added to the optimization problem. Additionally, a constraint (Deviation ≥ 0) is added to ensure that the optimal solution is feasible. The Mode Frontier software ((Esteco S.p.A., Trieste, Italy)) is used to develop workflows for solving optimization problems. The optimization search is conducted using a MOGA-II. It is a robust algorithm that intelligently exploits multi-search elitism. The algorithm performs as many evaluations as the number of points in the design of the experiment table (the preliminary population) multiplied by the generation count.
Figure 18a,b illustrate the results of using bubble charts to optimize the tree-like support for multiple responses. By plotting the design points against two objective functions, the three-dimensional bubble chart is constructed, SV and SRT, one of which is represented by the bubble diameter. The design points in the initial DOE matrix are real, whereas the design points predicted by RSM are virtual; additionally, feasible design points are gray, while infeasible design points are yellow.
Since the optimization problem’s objective is to minimize the volume of support and the time required to remove it, feasible design points corresponding to the bubble chart’s lower-left corner will be considered candidates for the optimal solution, as illustrated in
Figure 19a,b. The optimal design points and details are summarized in
Table 4.
A parallel coordinate chart, as illustrated in
Figure 20, is another technique for analyzing the design points. A parallel coordinate chart can be used to display all of the design points associated with the study’s parameters. In summary, optimal results are obtained when the current beam is low, the beam scan speed is high, the branch top diameters are small, and the "Min distance between rows" is set to a moderate value. It shows that by combining a beam current of 0.8 mA, a scan speed of 1800 mm/s, a branch top diameter of 0.5 mm, and a minimum distance between rows of 2.1 mm, a tree-like support volume of 82.29 mm
3, a tree-like support removal time of 15.89 s, and an overhang deformation of 0.099 mm can be attained.
4. Conclusion
The purpose of this study was to determine the feasibility of using tree-like support structures in metal additive manufacturing and to optimize their design and process parameters for the minimum possible support volume, support removal time, and deviation. The outcome of this study reveals that the right value of MDBR is crucial for minimizing the SV. It is because MDBR controls how many tree structures there are in a specific area. When compared to the other variables, MDBR is also the primary reason for variation in the SRT. It should be clear that we need to remove more support structures as we lower the MDBR. It will also result in limited accessibility to the cutting tool during the removal operation as well as supports are removed in smaller pieces thereby boosting the SRT. Moreover, BC, SS, and MDBR have an effect on the overhang deviation. The results also show that increasing the SS and decreasing the BC or beam energy density significantly reduced deformation. Therefore, finding a balance between BC, SS, and MDBR that can result in the lowest SV, lowest SRT, and the least amount of deformation is crucial. Thus, the following conclusions are drawn from the analysis of the data:
The results demonstrated the feasibility of using tree-like support structures in metal additive manufacturing;
In general, the design and manufacturing parameters of tree-like support structures have a significant impact on their performance;
The minimum distance between rows has the greatest effect on the volume of tree-like support and the support removal time;
The beam current, and beam scan speed, have the most significant effect on the overhang surface deviation;
By combining a low beam current (0.8 mA), a high scan speed (1800 mm/s), a small branch top diameter (0.5 mm), and a moderate minimum distance between rows (2.1 mm), a minimum tree-like support volume (82.29 mm3), a minimum tree-like support removal time (15.89 s), and a minimal overhang deformation were achieved (0.099 mm).
There are many opportunities for research in various aspects because the tree-like supports are still in their infancy and are still being developed. Future research on optimizing the number of branches, the branch length, and the branch angle is recommended with regard to the design of tree-like support.