Comparison of dimensional accuracy and tolerances of powder bed based and nozzle based additive manufacturing processes

Even though the geometrical complexity obtained by additive manufacturing (AM) is seen as the main advantage compared to conventional processes, AM inherent geometrical limitations still exist and are partly addressed in design guidelines.^1–8 These guidelines were mostly developed for laser powder bed fusion (LPBF) using a specific machine and material. The first benchmark artifacts for AM were designed in 1991 by Kruth⁹ for stereolithography, and since then, more than 60 geometrical benchmarks have been designed to evaluate dimensional or geometrical accuracy, repeatability, and minimum feature size¹⁰ of AM parts. These existing benchmarks either concentrate on only a few features (see Fig. 1, Ref. 22) or are very complex and hard to measure accurately due to features being too close together.¹¹ Therefore, the aim of this study was to design, build, and test a scalable demonstrator to quantify the geometrical limitations of different powder based metal AM processes.

The new design developed at Fraunhofer IWS was already used for comparison of electron beam melting (EBM) demonstrators and LPBF demonstrators in Ref. 12. But the analysis was mainly focused on average deviations, not looking further into the relationship between the feature size and achievable accuracy, which is addressed here.

Furthermore, a laser metal deposition (LMD) demonstrator is included in the analysis.

The methodology to investigate the geometrical capability of certain additive manufacturing processes in terms of dimensional accuracy and tolerances is divided into the following four steps:

1. Design of a benchmark artifact incorporating geometric characteristics referenced in Ref. 13;

2. Manufacture benchmark artifacts using three different AM processes: LPBF, EBM, and LMD;

3. Measure geometrical features according to the defined test plan with 3D scanning and CT scanning; and

4. Analyze results and compare manufacturing processes and measuring techniques with regard to the general tolerance capability.

A new benchmark demonstrator (see Fig. 2) with the base dimensions of 40 × 40 × 15 mm³ was designed according to the following specifications:

• different geometrical elements (cylinder, prism, sphere, freeform) according to Ref. 14;

• surfaces facing up and down;

• different feature sizes;

• internal features (cooling channels or holes) with regard to removing excess powder;

• sufficient distance between features for access for 3D scanning;

• feasible and scalable for all AM processes; and

• feature reduction where possible (3–5 feature variations per element) to reduce measurement time and cost.

Overall, the benchmark consists of 64 geometrical elements. For each element, a test plan was created taking into account dimensional and form tolerancing. The test plan is described in detail in Table III in Sec. IV.

For this investigation, three different metal additive manufacturing processes were chosen: LPBF, EBM, and LMD. In this section, the different process principles and manufacturing conditions are shortly described (see Fig. 3). LPBF and EBM are powder bed based processes where a single powder layer is spread and selectively molten and fused by a laser/electron beam that is deflected by a mirror system or electro magnetically. The build plate is then lowered, a new powder layer is spread, and the process continues until the part is fully built.

EBM has the advantage of high scanning speeds and preheating of each powder layer prior to scanning in a vacuum chamber up to 1100 °C, lowering thermally induced stresses, and enabling processing of materials with a high ductile-to-brittle transition temperature such as titanium aluminides.¹⁵ LMD is a nozzle based process where powder delivered through a nozzle onto a substrate is preheated and melted when passing a focused laser beam close to the substrate. The main advantage of LPBF and EBM is the high geometrical complexity, whereas LMD allows for near net shaped scalable geometries ranging from micrometer structures to large components with build rates up to 10 g/min.

Table I gives an overview on the process-material- combinations as well as the machine setup and process parameters used specifically for this study. For the two LPBF demonstrators, the original parameter sets provided by the manufacturer were used without optimization, therefore representing “typical” process conditions. For LMD and EBM, the process parameters for the chosen materials were developed in previous trials.

Two material classes for additive manufacturing were chosen: the titanium alloys Ti-6Al-4 V for LPBF, a newly developed alloy Ti-5Al-5Mo-5V-3Cr (Ti-5553) for EBM, and the nickel-based alloy Inconel 718 (IN718) for LPBF and LMD. This configuration allows the comparison of two materials and one process (LPBF) and one material and two processes (LPBF-EBM and LPBF-LMD). Ti-6Al-4 V is the most commonly used alpha-beta titanium alloy in the aerospace/automotive/dental industry with a density of 4.43 g/cm³ and provides a high specific stiffness. Inconel 718 has a density of 8.19 g/cm³ and is widely used in high temperature applications such as turbomachinery.

As can be seen in Fig. 4, all powders show a spherical shape and low amount of agglomerates. The LPBF powders have a particle size distribution of 15–45 μm. The LMD and EBM powder have a particle size distribution of 50–150 μm.

After conversion to a STL-file, the part is orientated and positioned in the build chamber using the software Magics from Materialise and then sliced and assigned the process parameters. The scan strategy varies depending on the process and software used. As can be seen in Fig. 5, the scan strategy for the Ti-6Al-4 V demonstrator in this study consisted of a borderline (red) with an offset to the edge of the STL boundary due to the beam size, two additional borderlines, and the hatch to fill the element. Vertical walls of the smallest thickness were scanned with two straight borderlines.

It is expected that relative deviations of the width of the smallest wall will, therefore, be high. Small closed outer contours are neglected by the software and will, therefore, not be manufactured. That was the case for the smallest vertical cylinder with a diameter of 0.1 mm as can be seen in Fig. 5 where no borderline is activated. The cylinder with d = 0.5 mm still has a border line and hatch fill line. Consequently, the LPBF process resolution also depends on the software capabilities.

The process parameters and scan strategy for LMD were determined for each geometrical element individually in an iterative manner. The geometry was scaled by factor 1.7 to ensure accessibility and manufacturability since the laser spot size was 1.7 mm and lower distances between features would have led to collisions with the powder nozzle. The geometry of the cooling channels proved to be challenging when closing the circular shaped channels. Therefore, only the three largest outer cylinders, hollow cylinders, and walls were manufactured. For the overhang angle, all five overhangs were realized. The scanning strategies for the individual elements are shown in Table II. The scanning strategy for the base geometry was meander with a double contour to increase dimensional accuracy.

All features were manufactured using three axes. All built demonstrators are depicted in Fig. 6. Visual inspection already shows process inherent differences in resolution and size. The analysis of these demonstrations is presented in Secs. IV–V.

The test plan in Table III lists all features and corresponding characteristics analyzed in this contribution. Besides linear and angular dimensions,¹ form (flatness, cylindricity),^13,16,17 orientation (perpendicularity), and location tolerances (position to plane) were investigated. According to Ref. 13, the flatness is defined as the smallest distance between two parallel planes enveloping a surface. Cylindricity merges straightness and circularity to describe the conformity to a perfect cylinder. It describes the distance of two concentric cylinders in which all points of the surface of the cylindrical feature fall. Angularity is defined as an angle between an inclined surface and a base surface where the inclined surface lies between two planes with a defined distance. Perpendicularity is angularity at an angle of 90°. Position specifies the deviation from specified dimensions of a feature on a part. For all described tolerances, the following applies: the larger the value the larger the deviation from the CAD file.

3D scanning with the GOM ATOS Core 45 (Fig. 7) was used for dimensional and form analysis of outer features with a resolution of 5 μm. Two cameras record a fringe pattern projected onto the surface of the part; thus, a point cloud of the surface is generated, converted to a polygon mesh, and then analyzed using the software Polyworks. When features stand too close together, points of the surface are missing, and therefore, the accuracy of the measurements is negatively impacted (Fig. 8). In this case, the demonstrator was scanned several times in different angles to get better access to critical regions.

For the setup in this investigation, an YXLON FF35 CT equipped with a 250 kV reflection x-ray vacuum tube with a minimum spot size of ≤6 μm was used. Computed tomography produces cross-sectional (tomographic) images of a scanned object by combining numerous high intensity x-ray measurements taken from different angles. The x-rays attenuate during penetration of the object due to absorption and scattering. From the detected signals, an image with different gray values per voxel is reconstructed. The quality of CT scans is problematic at high material density since the penetration depth of x-rays decreases and objects are not properly recognized. Therefore, the large LMD INC718 demonstrator could not be analyzed with the available CT system.

The software “Polyworks” was used to measure dimensions and tolerances from the polygon meshes created with 3D scanning and computed tomography. Auxiliary planes from the surfaces were created using the best-fit method (Fig. 9). The graphs were created with the freeware Gnuplot.

Figure 10 gives an overview of the manufactured demonstrators and presents the 3D scan fitted to the top surface and side surfaces A and B of the CAD file. Green marks very low deviation from the CAD (±150 μm for the LPBF demonstrators and ±250 μm for EBM and LMD), orange/red marks areas with positive deviation, blue marks areas with negative deviations, and purple is a deviation larger than ±1.5 mm for EBM/LMD and ±0.5 mm for LPBF. Visual inspection of these images shows the tendency that smallest features are larger and overall dimensions of the base plate are smaller than the CAD file. Besides the smallest diameter of 0.1 mm, all other features were successfully built and could be analyzed according to the test plan (Table III).

For the EBM demonstrator, the CT and 3D scan results of the outer features were compared (Fig. 11). Overall, the smallest deviations were found for the dimensional values and form tolerances such as flatness and cylindricity, see Table IV. Relative deviations remained large (>20%) given the small absolute values of some tolerances. For example, the perpendicularity of the widest vertical wall (VW1) was measured by 3D scan to be 0.062 mm. The corresponding CT value was 0.079 mm that is a relative deviation of 27%.

Overall, the CT scans and 3D scans follow the same trends and lead to the same conclusions regarding the geometrical accuracy. For outer features, 3D scanning is a faster inspection technique and material-independent compared to CT scanning and therefore advantageous. For inner features, CT scanning presents a nondestructive inspection technique and is especially suited to measure the accuracy of additively manufactured parts with internal complex structures. From the CT scans in Fig. 12, it is clearly visible that the pockets and meander cooling channels of the EBM demonstrator are sealed with sintered powder that could not be removed after the build process. Additionally, the pockets demonstrate an increasing shift in the upper region to the right generated during the build process. It is unclear, why this shift occurred during the build process, but it could be attributed to the rake mechanism and different cooling rates of the part after finishing the base geometry. Similar EBM build defects were investigated in Ref. 18, classified as a loss of edge due to different cooling rates at free ends of the part. In contrast, the LPBF IN718 demonstrator presents straight pockets with no remaining powder.

The measured linear dimensions length (base dimension), width (prisms), diameter (cylinder), and angular dimension (overhang angle) were compared to the CAD file, and the resulting relative deviations are presented in Fig. 13. For the base geometry, all processes show low deviations below 5%. The dimensional deviations for the smallest vertical walls rose to more than 300% for LPBF and EBM. The smallest achievable wall thickness was 303 μm. In the literature, it has been shown that an optimization of process parameters can lead to smaller wall thicknesses of 140 μm,^19–23 but this was not the target of this study. The LPBF Ti-6Al-4 V demonstrator is consistently more accurate by factor 10 than the LPBF IN718 demonstrator. This proves that the choice of process parameters and materials can lead to significant differing dimensional accuracy within the same process. For vertical walls, the LMD deviations were in the range of the EBM deviations and more accurate than the IN718 LPBF demonstrator. The accuracy of all overhang angles was below 5% for the LPBF and EBM processes. The LMD process had large deviations above 10% for overhang angles below 60°. This deviation may be improved with further optimization of process parameters. For the diameters of cylinders and holes, the dimensional accuracy also increased with smaller diameters similar to the findings of walls and pockets. The orientation of cylinders had no impact on the accuracy for all processes. The largest differences in the accuracy of the LPBF demonstrators were found for the inner feature pockets and holes.

In Fig. 14, all measured absolute linear dimensional deviations are sorted by the nominal length and classified by tolerance classes according to Ref. 1. For nominal lengths below 6 mm, the majority of dimensional deviations of the LPBF Ti-6Al-4 V demonstrator are in the “fine” tolerance class. The IN718 LPBF demonstrator had larger deviations in the class “medium” and “coarse” for nominal lengths between 0.5 and 3 mm. The accuracy of EBM and LMD is comparable and mostly found in class “coarse.” This is mainly contributed to the higher powder particle size, focus diameter, and layer height compared to LPBF (see Table I).

In summary, the smallest achievable structure size for LPBF is around 300 mm, for EBM is 500 μm, and for LMD with the tested nozzle is 1.2 mm, which was expected due to different particle sizes and layer thicknesses of the processes. Smaller structures are possible with LMD when the laser spot diameter, the powder nozzle, and powder particle size are adjusted.

The form tolerances for flatness and cylindricity are presented in Fig. 15. The flatness values of the parallel planes E1 and E3 (see Fig. 9) representing the long vertical sides were averaged for the base geometry, vertical and horizontal walls, pockets, and overhangs. The LPBF process shows lower form and cylindricity values throughout all feature elements and sizes than the EBM and LMD processes. The form tolerances for the two LPBF demonstrators are very similar opposed to the high dimensional results (Fig. 13). Best overall flatness results were achieved for vertical walls. There was no large dependency of flatness on feature size except for the overhang angle and the smallest vertical wall. The flatness increased consistently with the overhang angle for the LPBF process. This could be explained by increased waviness at high overhang angles due to decreased overhang thickness and resulting remelting of up skin surfaces and borderlines. The high flatness of the smallest vertical wall is also caused by borderline remelting and results in a wavy surface. The EBM demonstrator had the largest flatness values for all geometrical features except for the overhang structures where LMD values are larger. Specifically, the base geometry and pockets have high maximum flatness values caused by the high distortion of the EBM demonstrator on the top surface, which was already anticipated through the CT scan in Fig. 12. No significant difference between cylindricity of outer or inner features nor feature size dependency was found for all processes. All processes show similar results leading to the conclusion that cylindric elements are well suited for the layerwise additive manufacturing processes. When summarizing all flatness values and sorting them by the nominal length, it can be seen in Fig. 16 that the LPBF processes were not able to reach the “fine” tolerance class for features smaller than 10 mm. Instead, the flatness ranges evenly between tolerance classes “medium” and “coarse.” For the LMD and EBM processes, the flatness values are hardly in the “coarse” tolerance class and mostly even exceed the specified tolerance classes. This leads to the conclusion that the EBM and LMD processes still need machining as a postprocess to meet tolerance specifications, whereas for the LPBF process, general tolerances can be met.

The perpendicularity of walls and cylinders to the top base surface is depicted in Fig. 17. For features smaller than 100 mm, the classification in Ref. 5 lists the following: “H” for 0.2, “K” for 0.4, and “L” for 0.6 mm. For base geometry, the EBM process showed highest values between 0.5 and 2 mm and is, therefore, mostly in class L and above. The results for the vertical walls and vertical cylinders were below 0.1 mm for all processes (except the LMD process for higher feature size) and therefore in class “H.” The perpendicularity of internal features such as pockets and holes was particularly high (above 1 mm) for the LPBF IN718 demonstrator well exceeding values of the EBM and other LPBF processes.

At last, the position to side A for prisms and to side A and B for cylinders was analyzed (see Fig. 18). For vertical walls and overhangs, the best position tolerances with values around 0.1 mm were achieved by the Ti-6Al-4 V LPBF demonstrator followed by the IN718 LPBF (around 0.3 mm), LMD (around 1 mm), and finally EBM demonstrator (over 2 mm). The large position offset of the EBM demonstrator is caused by the aforementioned shift during buildup (see Fig. 12). For the vertical cylinder and hollow cylinder, the position tolerance for the two LPBF demonstrators was around 0.3 and 0.5 mm opposed to around 2 mm for the EBM and LMD demonstrators.

For inner features such as pockets and holes, the position tolerances of the EBM demonstrator were significantly lower than for corresponding outer features. The IN718 LPBF demonstrator had the opposite effect, and inner features had higher position tolerance than outer features.

In this study, 3D scanning and CT scanning were used to analyze the geometrical accuracy and tolerances of a special designed demonstrator part. Not only dimensional tolerancing but also form deviations such as flatness and cylindricity as well as perpendicularity and position tolerances were considered. This is important since the end user has certain specifications for the final part and needs knowledge on the geometrical capabilities of different AM processes. The comparison of outer feature results obtained by 3D scanning and CT scanning showed high consistency in the range of ±100 μm, and therefore, both techniques are suited to determine the part quality. For inner structures, CT scanning is advantageous compared to 3D scanning.

It was found that the dimensional accuracy highly varies within one demonstrator and therefore not only depends on the process but also on the feature element and size. For example, within the LPBF demonstrator, deviations can range between 0.01 and 0.2 mm for the same element size. The following tolerance classes for linear dimensions were reached:

• LPBF: mainly f (fine) for features 0.5–6 mm,

• EBM: mainly m (medium) and c (coarse), and

• LMD: large range of deviations, all tolerance classes possible.

There were significant differences of the dimensional accuracy between the two LPBF demonstrators leading to the conclusion that the parameter set developed for Ti-6Al-4 V was much more accurate. The difference between LPBF and EBM/LMD is mainly attributed to the optimized parameter set and the smaller powder particle size, laser beam spot diameter, and layer thickness. Inner structures were best represented by LPBF. The EBM process had sintered powder remnants in the cooling channels, and the channel geometry was extremely challenging for the LMD process.

Form deviations such as flatness and cylindricity were lowest for the LPBF process, but the tolerance class “H” was not reached. EBM and LMD values were mostly higher than tolerance class “L.” The results for perpendicularity and position were similar and showed clearly the advantage of the LPBF process.

Overall, the LPBF provides highest accuracy for dimensions, as well as form, perpendicularity, and position tolerance. The EBM/LMD process for now is not suitable for complex internal structures due to sintered powder in the case of EBM and complex programming in the case of LMD. The results are partially attributed to the difference in powder particle size and difference in spot size used for the different processes. LMD accuracy is comparable to the EBM process but still has high potential for increased accuracy at high buildup rates in the future through optimized process parameters and automated tool path generation. In future studies, the effect of the powder particle size on roughness quality could be investigated as well as how much dimensional accuracy and tolerancing could be improved by adjusting and optimizing the process parameters. The results of this study underline the need for mechanical reworking of functional surfaces of AM parts to meet specifications and tolerances as well as the need to further improve process parameter development.

This work was funded by the Program Zwanzig20 of the German Federal Ministry of Education and Research (Consortium AGENT-3D). The authors would like to thank Fraunhofer IPK for the cooperation with the fabrication of the SLM Ti-6Al-4 V demonstrator.

Samira Gruber studied mechanical and process engineering at the Technical University Darmstadt. In November 2015, she finished her master's thesis about producing defined porous structures with Laser Powder Bed Fusion. Her work at the Fraunhofer IWS focuses on improving the geometrical accuracy of additive manufactured parts.

Christian Grunert completed an apprenticeship as a mechatronic technician at Koenig & Bauer AG in 2011 and is working in the group of 3D Manufacturing at the Additive Manufacturing Center Dresden (AMCD). His tasks include maintenance, servicing, and project-specific modifications of machines and equipment. He is involved in several projects, in particular dealing with optical 3D-measurement systems.

Mirko Riede studied mechatronics at the Technische Universität Dresden. In 2011, he finished his master thesis about high precision laser cladding at the Fraunhofer IWS Dresden. For the last 5 years, he has been working on several research projects related to additive manufacturing and structuring. Today he is the group manager of 3D Manufacturing at the Additive Manufacturing Center Dresden (AMCD) at Fraunhofer IWS.

Dr. Elena Lopez studied chemical engineering at the Universidad de Valladolid and Friedrich-Alexander-Universität Erlangen-Nuernberg. She finished her Ph.D. thesis about the topic of plasmachemical etching of silicon solar wafers at the Technische Universität Dresden. After focusing on CVD technologies, she moved to Printing and Additive Manufacturing technologies in 2014. She is the department manager of Additive Manufacturing and Printing at the Additive Manufacturing Center Dresden (AMCD) at Fraunhofer IWS. Nowadays, she leads a big consortium named Agent-3D with more than 120 companies involved.

Dr. Axel Marquardt finished his diploma in 2009 at the Technische Universitaet Freiberg and his Ph.D. in 2015 at the Technische Universtaet Dresden. He is currently postdoctoral research fellow for materials science at the Technische Universitaet Dresden and member of the “Additive Manufacturing” working group at the Fraunhofer IWS Dresden. His research work covered a wide range of topics with a focus on high temperature/lightweight materials and nondestructive testing

Prof. Dr. Frank Brueckner studied automation and control engineering as well as business administration at the Technische Universität Dresden. He finished his Ph.D. about theoretical aspects of laser cladding. Now he leads the business unit of Additive Manufacturing and Printing at the Fraunhofer IWS Dresden. Together with his team working at the Additive Manufacturing Center Dresden AMCD, he mainly focuses on nozzlebased, powder-bed based processes as well as printing technologies. In addition, he supervises Ph.D. candidates at the Luleå University of Technology.

Prof. Dr. Christoph Leyens studied physical metallurgy and materials technology at RWTH Aachen, Germany, where he earned his diploma in 1993 and his Ph.D. in 1997. He is currently a full professor for materials science and engineering at the Technische Universität Dresden, Germany, and director of the Fraunhofer Institute for Material and Beam Technology, Dresden. Dr. Leyens has covered a wide range of research topics with a focus on high temperature and lightweight materials, surface technology, and additive manufacturing. He has published more than 200 papers and seven books and holds 11 patents.