Additive Manufacturing 25 (2019) 239–251

Contents lists available at ScienceDirect

Additive Manufacturing
journal homepage: www.elsevier.com/locate/addma

Powder bed fusion metrology for additive manufacturing design guidance T

⁎
Jared Allison, Conner Sharpe, Carolyn Conner Seepersad
The University of Texas at Austin, Austin, TX 78712, United States

A R T I C LE I N FO A B S T R A C T

Keywords: Design for additive manufacturing (DFAM) guidelines are important for helping designers avoid iterations and
Design guidelines leverage the design freedoms afforded by additive manufacturing (AM). Comprehensive design guidelines should
Powder bed fusion incorporate a variety of features of interest to designers, and given the wide variety of AM processes and their
Selective laser sintering associated capabilities and limitations, those guidelines may need to be process- or even machine-specific. One
Metrology
way to generate detailed DFAM guidelines is to implement a metrology study focused on a strategically designed
test part. This paper describes how quantitative design guidelines are compiled for a polymer selective laser
sintering (SLS) process via a metrology study. As part of the metrology study, a test part is designed to focus
specifically on geometric resolution and accuracy of the polymer SLS process. The test part is compact, allowing
it to be easily inserted into existing SLS builds and therefore eliminating the need for dedicated metrology builds.
To build a statistical foundation upon which design guidelines can be compiled, multiple copies of the test part
are fabricated within existing commercial builds in a factorial study with materials, build orientations, and
locations within the build chamber as control factors. Design guidelines are established by measuring and
analyzing the as-built test parts. The guidelines are summarized in this paper and documented in a publicly
accessible, online web tool.

1. Introduction
Guidelines are important tools for designers selecting the appro-
priate manufacturing process for a specific part and tailoring the part
for functionality and ease of manufacturing. Although Design for
Manufacturing and Assembly guidelines are prevalent for well-estab-
lished processes such as injection molding [1], they are recently
emerging for AM processes. Notable recent work focuses on developing
an ontology and knowledge base for representing Design for AM
knowledge [2] and representing design guidelines as modular compo-
nents that can be implemented and interpreted efficiently across AM
processes and machines [3]. Several benchmarking studies focus on
developing design guidelines that are independent of a specific AM
process or on comparing AM processes to one another utilizing a
standard test part (e.g., [4–6]). However, process-specific investigations
may be required to establish a more complete understanding of a par-
ticular AM process. For example, test parts have been designed for di-
rect metal laser sintering [7,8], material jetting [9], fused deposition
modeling [10], and SLS [11,12]. These test parts can capture the
nuances of specific AM processes more accurately, and the guidelines
created from them provide additional information to designers that
process-independent test parts may overlook.
In recent work, the authors designed a test part specifically for the
polymer SLS process [13]. The test part, shown in Fig. 1, was comprised
of five panels of features attached to a common base. The features re-
presented those of particular interest to designers, and the panels could
be removed from the base for measurement. The panels were arranged
in a cube that measured two inches along each side in an effort to re-
duce the total volume required to build the test part. Unlike metal laser
sintering processes, the SLS powder bed is self-supporting, which means
that parts do not require support structures. The self-supporting powder
bed allowed tight grouping of the test part features and panels and easy
separation of them during post-processing. The resulting arrangement
gave rise to a high density of features in a compact build envelope.
By combining features found in previous test parts (e.g., [4,5,8,11])
but arranging them in a nested form, this design included a compre-
hensive variety of features in a compact cube. The part was built as a
single unit, similar to the parts proposed by Mahesh et al. [4], Castillo
[8], and Moylan et al. [5], but the detachable panels made it easy to
access and measure each feature. The detachable panels were analogous
to a suite of test parts used in the Seepersad et al. study [11], which
utilized a separate test part for each type of feature, but in a much more
compact form. Unlike some previous efforts that focused on char-
acterizing the mechanical properties of SLS parts (e.g., [14]), this test

⁎
Corresponding author.
E-mail address: ccseepersad@mail.utexas.edu (C.C. Seepersad).

https://doi.org/10.1016/j.addma.2018.10.035
Received 19 June 2018; Received in revised form 21 October 2018; Accepted 21 October 2018
Available online 28 October 2018
2214-8604/ © 2018 Elsevier B.V. All rights reserved.
J. Allison et al. Additive Manufacturing 25 (2019) 239–251

Fig. 1. Polymer test cube as-built (left) and disassembled (right) [13].

part focused exclusively on the geometric properties and surface char- Table 1
acteristics of SLS parts. Polymer test cube feature ranges [13].
Furthermore, polymer SLS enabled fabrication of all of the features. Feature Feature ranges Increment
Even complex features such as the thin rods and hinges, as well as the
nested configuration of panels, could be built successfully regardless of Holes 0.8–2.60 mm diameter 0.2 mm
build orientation. The part could not be built in various orientations 1.0–10.0 mm wall thickness 1.5 mm
Thin rods 0.3–0.9 mm diameter 0.1 mm
with a material extrusion process (e.g., FDM) or a metallic powder bed Thin walls 0.2–0.8 mm 0.1 mm
fusion process (e.g., DMLS) without support structures. Gaps 1.4–2.0 mm gaps 0.2 mm
The test cube features and feature sizes are documented in Table 1. 1.0–10.0 mm wall thickness 1.5 mm
Based on the results of the Seepersad et al. study, some feature sizes Cylinders 2.0–8.0 mm diameter 3.0 mm
Hollow cylinders 5.0–25 mm diameter 10.0 mm
were dimensioned such that they range from sizes that are too small to
Domes 6.0 mm diameter -
be resolved to sizes that are large enough to be resolved [11]. This Cones 6.0 mm diameter -
range allowed the resolution of the process to be characterized. 5.2 mm height
The details of the test part's features and procedures for measuring Linear accuracy 5.0–12.5 mm 2.5 mm
the features are described in detail in a previous publication from the Surface roughness 0–90° 15°
Hinges 0.6 mm and 1.0 mm shaft clearances –
authors [13]. As documented in [13], one of the unique characteristics
Lettering 10–18 pt font 2 pt
of the test part was its high feature density, defined as the number of 0.5–2.5 mm embedded/raised depth 0.5 mm
unique features per unit volume, combined with its minimal bounding Snap fits −0.5 to 0.25 mm offsets 0.05 mm
volume. These characteristics allowed it to be placed into unoccupied
volumes in existing SLS builds, enabling characterization of these builds
with respect to a very large number of features without displacing other interior (I) where the cube was built in the center of the chamber and
parts or requiring dedicated builds. In this way, feature density and exterior (E) where the cube was built along the edges of the build
build volume were prioritized over potential heat transfer issues that chamber. Here, exterior is defined as the region bounded by the peri-
may arise from the closely spaced panels. meter of the build area and a second perimeter that is offset from the
In this paper, the focus is on describing how the geometric cap- first by two inches toward the center of the build chamber. Table 2
abilities and surface characteristics of polymer SLS can be characterized shows a summary of the factors and levels that are used in the ex-
using this compact, comprehensive, process-specific test part. A me- perimental design, where each cell represents a specific combination of
trology study was conducted by fabricating the test part repeatedly in material, orientation, and location within the build chamber.
more than 100 commercial polymer SLS builds to investigate and sta- For a statistical characterization of polymer powder bed fusion
tistically analyze the effect of material choice, orientation, and location (PBF), several copies of the test part were built for each combination of
in the build chamber on the accuracy and resolution of a variety of experimental factors. For each unique material, orientation, and loca-
features. Actionable design guidelines were extracted from the results. tion combination, five replicates were built on one machine, and three
replicates on a second machine in order to assess the variability be-
tween PBF machines. In total, 144 test parts were required for the
2. Experimental procedure characterization.
All test parts were built by Stratasys Direct Manufacturing on 3D
A factorial-style study was conducted with the test part to char- Systems Sinterstation 2500 Plus and 3D Systems Sinterstation
acterize the resolution and accuracy of SLS parts under a variety of HiQ + HS machines featuring Integra multizone heater upgrades. The
conditions. Although a wide variety of process parameters can affect the machines were maintained by Stratasys Direct Manufacturing and
resolution and accuracy of features, four specific factors were selected tuned to production level specifications. Each of the test parts was in-
because they are often specified by the designer. The four factors were serted one-at-a-time into a production build filled with other commer-
material choice, orientation of the test part within the build chamber, cial parts. The build parameters were based on the 3D Systems default
location of the test part within the build chamber, and machine iden- settings, but modified by Stratasys Direct Manufacturing in an effort to
tity. Previous studies showed these factors to significantly affect the produce high quality, production parts. Compared to the default set-
mechanical performance of SLS parts [15–21]. However, corresponding tings, the modified parameters sought to balance feature detail with
studies on the geometric effect of these parameters were limited in mechanical strength.1 Nominally, the layer thickness and fill laser
scope [4,11]. By varying each of the four factors, a factorial experiment power were 100 μm and 42 W, respectively. After the build process, the
was conducted. Three different materials were considered: nylon 12 parts were cleaned by Stratasys technicians using a combination of
(PA 12), nylon 12 reinforced with glass beads (GF PA 12), and fire compressed air bead blasting and brush-like tools to remove any re-
retardant nylon 11 (FR PA 11). The test cube was built in three different sidual powder. The cleaning methods were consistent with industry
orientations, denoted by the plane of the build chamber in which the
base was aligned: XY, XZ, and YZ with the z-axis orthogonal to the build
platform and the x-axis parallel to the front window of the machine. 1
The specific values of the process parameters are proprietary to Stratasys
Location within the build chamber was reduced to two specifications: Direct Manufacturing.

240
J. Allison et al. Additive Manufacturing 25 (2019) 239–251

Table 2 significant, and the multiple comparisons test quantifies the effect of
Summary of factors and levels for experimental design. the factor on the feature of interest. The result of a multiple compar-
isons test provides an indication of the differences in the means be-
tween different levels of a statistically significant factor. The results are
summarized in Table 3. As with the significant effects results, “Average
Mean Difference” is calculated by averaging each of the mean differ-
ences generated from the multiple comparisons test in the ANOVA
study. The results can be interpreted as the average variation associated
with changing between levels for a given process parameter. The results
of the ANOVA study are described in greater depth in the next section in
the context of each of the test cube features.
standard practices so that the test part measurements would reflect
production–quality SLS capabilities. A variety of measurement techni- 3. Results
ques were used to analyze each of the different test cube features. The
measurement techniques are described in detail in [13] and are sum- The measurement results are compiled into a set of guidelines that
marized here alongside the results for each feature. can be used by designers creating parts for polymer SLS. The guidelines
To identify which input parameters have a statistically significant are available publicly through an online web tool that can be accessed
effect on part accuracy, an analysis of variance (ANOVA) test was at http://DesignForAM.me.utexas.edu. A screenshot from the tool is
conducted. The output of the test specifies which factors are significant shown in Fig. 2. The web tool describes the test part, the measurement
at a five percent significance level (α = 0.05). As a designer, these procedure, and the general design guidelines derived from the mea-
factors should be considered when creating a part to be manufactured surements. It also includes graphs and tables summarizing the mea-
using selective laser sintering. A summary of the significant effects for surements of each feature and allows the user to customize those results
all features is provided in Table 3. In the table, “Percentage Significant” for a specific material, orientation, and location in the build chamber.
refers to the fraction of settings for a particular feature for which the In the rest of this section, the general design guidelines are reported,
factor is statistically significant. For example, material choice is statis- along with average measurements across all material choices, orienta-
tically significant for all gap sizes, but location is a significant factor for tions, and build chamber locations for all 144 test cubes in the ex-
only 57% of gap sizes. perimental design. The average values for specific combinations of
After performing the ANOVA test, a multiple comparisons test was material, orientation, and location within the build chamber are tabu-
conducted to quantify the effect of each significant factor on the feature lated in Appendix B. Important insights regarding the significance of
of interest. The ANOVA test tells the designer which factors are the effect of material choice, orientation, or build chamber location on
those measurements are also included in Table 3. Section 3.11 describes
Table 3 the additional information available in the web tool, namely customi-
Significant effects and mean differences for test cube features. “Percentage zation of each of the tables and graphs in this section for a specific
Significant” refers to the fraction of levels within a particular feature in which combination of material choice, orientation, and build chamber loca-
the given factor is statistically significant at a p-value of 0.05. For example, tion.
material choice is statistically significant for all gap sizes, but location is a
significant factor for only 57% of gap sizes. Similarly, “Average Mean 3.1. Surface roughness
Difference” is calculated by averaging the mean differences generated from the
multiple comparisons test in the ANOVA study. For linear accuracy, the average
The test part features for evaluating the surface roughness at various
effect of changing materials causes a 0.23 mm difference in the reported linear
build angles are shown in Fig. 3. Surface roughness was measured using
dimensions.
a Zeta-20 3D optical surface profiler. The Zeta software used internal
Significant effects algorithms to measure the profile of the scan area, and average
roughness (Sa) and root mean square roughness (Sq) values were re-
M O L M*O M*L O*L Machine
ported. The measurement procedure for surface roughness is described
Surface roughness further in Appendix A.
Percentage significant 100% 0% 0% 0% 0% 0% 33% The average surface roughness results from all test cubes are shown
Average mean difference 9.36 – – – – – 8.76 in Fig. 4. The surface roughness is smallest at an angle of 0° relative to
(μm) the build plane. The maximum roughness occurs at 15° then decreases
Linear accuracy at all subsequent angles. Despite the decreasing trend, the roughness
Percentage significant 100% 100% 0% 0% 0% 0% 0% still has a minimum value at 0°. Additional results can be found in
Average mean difference 0.23 0.06 – – – – – Table 13 of Appendix B.
(mm)

Gaps 3.1.1. Design guidance for surface roughness

Percentage significant 100% 14% 57% 50% 0% 11% 50% The results suggest that surface roughness is influenced by the
Average mean difference 0.21 0.09 0.04 – – – 0.09 layering effect that occurs in additively manufactured parts [22,23].
(mm) Surface roughness is smallest at 0° relative to the build plane and is
Holes
largest at 15°. This difference in roughness can be attributed to the stair-
Percentage significant 100% 100% 29% 14% 7% 14% 7% stepping effect between successive layers. As the angle is increased, the
Average mean difference 0.26 0.32 0.13 – – – 0.15
effect becomes less pronounced. The steps disappear at an angle of 90°
(mm) relative to the build plane, but the roughness is still higher than at 0°.
The 90° surface roughness measurement spans multiple layers, and
Thin walls
Percentage significant 100% 100% 21% 21% 7% 0% 21% geometric variation between them causes greater roughness for the 90°
versus the 0° surfaces.
Average mean difference 0.16 0.11 0.05 – – – 0.10
(mm)
3.1.2. ANOVA results
Key: M = material; L = location; O = orientation; * = interaction effect. The analysis of variance results for surface roughness are listed in

241
J. Allison et al. Additive Manufacturing 25 (2019) 239–251

Fig. 2. Online web tool layout.

Table 3. From the ANOVA study, material is a significant factor for

every inclination angle. Furthermore, FR PA 11 is rougher than both PA
12 and GF PA 12, which have similar values of roughness. As shown in
the mean difference results, changing materials influences the mea-
sured surface roughness by 9.36 μm on average. No statistically sig-
nificant correlation is observed between location within the build
chamber and surface roughness.

3.2. Linear accuracy

Linear accuracy indicates how well the machine holds linear di-
mensions with respect to the prescribed dimensions within the CAD
model. Linear accuracy in each build axis was determined by measuring
the gap between successive walls. Fig. 5 shows the linear accuracy
features in the test part. The measurements were taken with iGaging
digital calipers (0.01 mm resolution), and each feature was measured
three times and the average is reported. The measurement uncertainty
was quantified through a gage R&R study, described in Appendix A. Fig. 4. Average surface roughness as a function of build angle for all test cubes.
The error bars in the figure correspond to 95% confidence intervals about the
mean value.
3.2.1. Accuracy
The linear accuracy results are shown in Fig. 6 and detailed in

Fig. 3. Surface roughness features in the test part.

242
J. Allison et al. Additive Manufacturing 25 (2019) 239–251

Fig. 5. Linear accuracy features in the test part. Features of interest are tabs Fig. 7. Features for evaluating gap accuracy as a function of wall thickness in
aligned along the outer edge of the part. the test part.

increase with the linear dimension. The degree of under-sizing does not
appear to scale with feature size, suggesting a greater contribution from
oversintering than shrinkage. Embedded features built on the interior of
the build chamber are usually more undersized than those built along
the perimeter. With respect to orientation, distances along the Z axis of
the build chamber tend to be closer to the defined CAD dimension than
along the X or Y axes. PA 12 has the highest level of accuracy, followed
by GF PA 12 and FR PA 11.

3.3. Gap accuracy as a function of wall thickness

Gap accuracy was determined by varying the gap size and the
thickness of the wall through which the gap was created. The reported
value corresponded to the gap measured along the axis of interest. Four
gaps of varying dimensions were placed within each of the seven walls,
as shown in Fig. 7. Gap accuracy was measured using iGaging digital
calipers (0.01 mm resolution). The measurement uncertainty for gaps
Fig. 6. Linear accuracy trend for all test cube measurements. The red line in- was quantified as described in Appendix A.
dicates the nominal as-designed value, and the error bars correspond to 95%
confidence intervals about the mean. 3.3.1. Accuracy
Similarly to linear accuracy, gaps tend to be undersized by ap-
Table 14 of Appendix B. SLS tends to undersize dimensions by ap- proximately 0.2 mm. As shown in Fig. 8, there does not appear to be a
proximately 0.2 mm. When built, each of the measured distances in significant dependence on wall thickness. Additional results, separated
Fig. 5 are generally 0.2 mm smaller than their designed value. This by input factor, are provided in Table 15 of Appendix B.
trend appears to be independent of size and is consistent across the
entire range of measurements. 3.3.2. ANOVA results
The results generated from an ANOVA study and subsequent mul-
3.2.2. ANOVA results tiple comparisons test are shown in Table 3. Unlike linear accuracy, gap
The complete analysis of variance results are tabulated in Table 3.
The percentages in the table represent the proportion of feature levels
for which the particular parameter is considered statistically significant.
Linear accuracy is assessed at four distinct levels. The ANOVA results
suggest material and orientation are statistically significant for each of
the four levels. Table 3 also displays the differences in mean that can be
attributed to each of the parameters based on the ANOVA study. On
average, material and orientation can contribute to 0.23 mm and
0.06 mm of variation in linear accuracy, respectively. For designers, a
consideration must be made regarding the accuracy associated with
certain materials and orientations. PA 12 produces the highest degree of
linear accuracy, and changing materials can cause linear dimensions to
be undersized by as much as 0.34 mm.

3.2.3. Design guidance for linear spacing

Linear dimensions are undersized due to shrinkage and a phenom-
enon called oversintering. Oversintering occurs when the powder sur-
rounding a part is unintentionally fused during the sintering process.
The additional fused powder reduces the measured linear accuracy Fig. 8. 1.4 mm gap accuracy as a function of wall thickness for all test cubes.
because accuracy is determined by the void space between successive The red line indicates the nominal as-designed value, and the error bars cor-
walls. Shrinkage is a volumetric effect and would be expected to respond to 95% confidence intervals about the mean.

243
J. Allison et al. Additive Manufacturing 25 (2019) 239–251

accuracy appears to depend on the location within the build chamber.

Location is a significant effect for 57% of the gap sizes measured. Ma-
terial choice is significant for all gap sizes. The results also suggest an
interaction effect between material and orientation. Changing materials
can lead to mean differences of 0.21 mm on average. FR PA 11 tends to
undersize gaps the most, followed by GF PA 12 then PA 12. Although
location is statistically significant for more than half of the gap sizes,
the variation associated with location within the build chamber is only
0.04 mm.

3.3.3. Design guidance for gaps

Shrinkage and oversintering appear to be the greatest contributing
factors to the undersizing of gap accuracy. The two effects are con-
founded and cannot be separated in this study, so the accuracy results
reflect a combination of both phenomena. Similar to linear accuracy,
gaps aligned with the Z axis of the build chamber are generally the
closest to the prescribed CAD dimensions. This shows that shrinkage
and oversintering are less severe between layers than within individual Fig. 10. 2.6 mm hole accuracy as a function of wall thickness for all test cubes.
layers. Additionally, PA 12 tends to be more accurate than both GF PA The red line indicates the nominal as-designed value, and the error bars cor-
12 and FR PA 11. Location within the build chamber does not appear to respond to 95% confidence intervals about the mean. (For interpretation of the
significantly affect the accuracy of gaps. references to color in this figure legend, the reader is referred to the web ver-
sion of this article.)
3.4. Hole diameter as a function of wall thickness

Hole accuracy and resolution were determined by varying the hole

diameter as well as the thickness of the wall through which the hole
was created. Hole resolution denoted the smallest hole that is reliably
formed at each wall thickness. To be considered reliably formed, the
hole must have resolved in at least 75% of the test cubes in this study.
Accuracy values provided quantitative comparisons of the mean hole
diameter measured at each wall thickness to the as-designed diameter
for the two largest holes (two rightmost columns of holes in Fig. 9).
Accuracy was measured using a flatbed scanner and a Matlab digital
image processing script that was calibrated against pin gages, as de-
scribed in Appendix A. Hole resolution was determined through visual
inspection.
3.4.1. Accuracy
Accuracy was measured for hole diameters of 2.4 mm and 2.6 mm
across all wall thicknesses. Hole diameters have a tendency to be un-
dersized by approximately 0.4 mm, as shown in Fig. 10. The trend does
appear to depend on the thickness of the wall, with thicker walls
yielding smaller holes than thinner walls. Additional accuracy results
are documented in Table 16 of Appendix B.
3.4.2. Resolution
Hole resolution was determined by visual inspection. Holes that
were fully formed and allowed light to pass through them received a
“Pass” rating, while those that did not allow light to pass were given a
“Fail” rating. A description of this convention is shown in Fig. 11. It is
Fig. 11. Convention used for hole resolution. The two leftmost holes (circled)
failed while the remaining seven holes passed. The only criterion for passing
holes is the ability for light to pass through the hole, regardless of the shape of
the resulting hole.
important to note that all parts were post-processed with compressed
air bead blasting and brush-like tools prior to making these observa-
tions.
Hole resolution appears to depend greatly on wall thickness. As
shown in Table 4, thinner walls resolve smaller holes, and the smallest
resolvable hole diameter tends to increase with wall thickness. In
general, 1.0 mm thick walls resolve 1.0 mm holes in at least 75% of the
test cubes. Conversely, 8.5 mm thick walls resolve only 2.6 mm holes in
at least 75% of the test cubes.
3.4.3. ANOVA results
Table 3 outlines the significant effects generated from the ANOVA
test for hole accuracy. At every wall thickness for both the 2.4 mm and
2.6 mm holes, material and orientation are both statistically significant
factors. The table also shows that location within the build chamber is
significant for nearly one third of the holes measured. Table 3 shows
interaction effects, but they are present only for holes built in wall
Table 4
Hole diameter resolution as a function of wall thickness for all test cubes. The
values within each cell indicate the proportion of observed instances that cor-
rectly resolved. The color scheme corresponds to the proportion of resolved
instances, where green cells represent 75% or greater, yellow cells represent
25–75%, and red cells denote less than 25% resolved. (For interpretation of the
references to color in this table legend, the reader is referred to the web version
of this article.).

Fig. 9. Features for evaluating hole accuracy and resolution as a function of

wall thickness in the test part.

244
J. Allison et al. Additive Manufacturing 25 (2019) 239–251

thicknesses of 4.0 mm and smaller. Both material choice and hole or-
ientation contribute to large observed differences in means. Orientation
has the greatest impact, with mean differences of 0.32 mm on average.
PA 12 holes tend to be larger than both GF PA 12 and FR PA 11 holes.
Holes built in the Z-direction are the most undersized because the entire
cross section of the hole lies in the plane of the laser path. The spot size
of the laser is larger than the layer thickness, so holes tend to be more
undersized when they are built in-plane.

3.4.4. Design guidance for small holes

Holes tend to be the most undersized when the central axis is or-
thogonal to the build chamber, further supporting the hypothesis that
oversintering and shrinkage are more severe within individual layers.
This trend is consistent with the gap results that show more under-
Fig. 13. Thin wall accuracy for all test cubes. Measurements were taken in the
sizing for gaps built within the build plane. As with the previous fea-
middle of the wall. The red line indicates the nominal as-designed value, and
tures, PA 12 produces holes that are the most accurate and is able to the error bars correspond to 95% confidence intervals about the mean.
resolve the smallest holes. Location within the build chamber does not
appear to influence the resolution or accuracy of holes.
Table 5
Thin wall resolution for all test cubes. The values within each cell indicate the
3.5. Thin walls
proportion of observed instances that correctly resolved. The color scheme
corresponds to the proportion of resolved instances, where green cells represent
Both accuracy and resolution are evaluated for thin walls. Two 75% or greater, yellow cells represent 25–75%, and red cells denote less than
measurements are taken for thin walls–one at the base, and one at the 25% resolved. (For interpretation of the references to color in this table legend,
middle. Fig. 12 shows the thin wall features in the test part as well as the reader is referred to the web version of this article.).
the two points where measurements are taken. The accuracy of thin
walls was measured using iGaging digital calipers (0.01 mm resolution).
The measurement system for thin walls was verified using a gage R&R
study as described in Appendix A. Resolution was dictated by the
thinnest wall that could be built reliably. Thin walls were built with
thicknesses ranging from 0.2 mm to 0.8 mm in increments of 0.1 mm.

3.5.1. Accuracy
As shown in Fig. 13, thin walls are usually oversized by approxi-
mately 0.2 mm. This trend does not depend significantly on the thick-
ness of the wall. The results are further stratified by input factor in are the most accurate. XZ and YZ walls are generally 0.10 to 0.12 mm
Table 17 of Appendix B. thicker than the nominal dimension by comparison.

3.5.2. Resolution 3.5.4. Design guidance for thin walls

In general, the thinnest walls to resolve reliably (in more than 75%
of test cubes) are 0.5 mm. At thicknesses below 0.5 mm, the proportion
of correctly resolved walls drops substantially. All walls of 0.6 mm
thickness or greater resolved successfully in 100% of test cubes, as
documented in Table 5.
3.5.3. ANOVA results
Table 3 lists the significant factors as determined by the ANOVA
test. Measurements taken at one edge of the wall (outer) and at the
center of the wall (middle) are shown. Interestingly, the two mea-
surement locations do not produce identical significant effects. How-
ever, material and orientation are significant factors for all of the walls
measured. From the multiple comparisons test, changes in material
appear to have a smaller effect on thin walls than the other features. PA
12 walls tend to be the most dimensionally accurate. Conversely, GF PA
12 walls can be up to 0.19 mm thicker than the nominal value while FR
PA 11 walls are as much as 0.26 mm thicker. For orientation, XY walls
Thin walls are the most accurate when they are built parallel to the
build platform. The accuracy of thin walls parallel to the build platform
is governed primarily by layer thickness, whereas the accuracy of thin
walls orthogonal to the build platform is affected more strongly by laser
spot size. Since SLS layers are 100 μm thick while the spot size is ty-
pically 450 μm in diameter, greater accuracy is achievable for thin walls
that are parallel to the build platform. Again, PA 12 produces the most
accurate walls; GF PA 12 and FR PA 11 have similar accuracy. Location
within the build chamber does not usually affect thin wall accuracy.
When the effect is observed, however, walls built in the interior portion
of the build chamber are generally less oversized than in the exterior.
3.6. Thin rods
Three types of rods were tested for resolution, as depicted in Fig. 14.
Supported rods were connected to side walls at both ends, while

Fig. 12. Thin wall features in the test part. Fig. 14. Thin rod features in the test part.

245
J. Allison et al.
Fig. 15. Example of feature resolution for thin rods. The intended geometry from the CAD model is shown on the left. The as-built geometry is on the right, with
arrows to show the pass/fail criteria.
unsupported rods were cantilevered. Aspect ratios (length/diameter) of
5 and 10 were used for short and long unsupported rods, respectively,
with rod diameters varying from 0.3 mm to 0.9 mm in 0.1 mm incre-
ments. Thin rod resolution was evaluated through visual inspection.
3.6.1. Resolution
Resolution was determined through visual inspection. Features that
did not build successfully were given a “Fail” rating. Unsuccessful
features included those that built poorly or failed to build at all. An
example of this convention is shown in Fig. 15, where the bottom four
rods received a “Pass” rating and the top three missing rods were given
a “Fail” rating.
As shown in Table 6, no significant relationship is observed between
the support conditions and the ability to resolve thin rods. In order to
ensure reliable resolution of thin rods, they should generally be at least
0.6 mm in diameter. Smaller rods tend to fail or leave an “empty” spot
in their place.
3.6.2. Design guidance for thin rods
Thin rods tend to have the best resolution when the primary axis is
along the Z axis of the build chamber. For this orientation, the laser spot
traces the cross section of the rod at each layer. The self-supporting SLS
powder bed secures the rods in place during the build process, elim-
inating the effect of unsupported length. Concerning material choice,
GF PA 12 resolves thin rods better than the other two materials. Thin
rod resolution does not depend on location within the build chamber.
3.7. Hinges
Two hinge designs were tested by varying the gap between the shaft
of the hinge and the knuckle. Fig. 16 shows the two designs. The hinges
were tested by checking whether they can move freely and assigning a
“pass” or “fail” rating.
3.7.1. Resolution
As shown in Table 7, the results of the study indicate that a 1.0 mm
Table 6
Thin rod resolution for short (aspect ratio = 5), long (aspect ratio = 10), and
fully supported rods. The values within each cell indicate the proportion of
observed instances that correctly resolved. The color scheme corresponds to the
proportion of resolved instances, where green cells represent 75% or greater,
yellow cells represent 25–75%, and red cells denote less than 25% resolved.
(For interpretation of the references to color in this table legend, the reader is
referred to the web version of this article.).
246
Additive Manufacturing 25 (2019) 239–251
Fig. 16. Hinge features in the test part.
Table 7
Hinge resolution at shaft clearances of 0.6 and 1.0 mm for all test cubes. The
values within each cell indicate the proportion of observed instances that cor-
rectly resolved. (For interpretation of the references to color in this table le-
gend, the reader is referred to the web version of this article.).
shaft clearance is generally sufficient for creating mechanical hinges.
Conversely, it is not recommended to design hinges with only a 0.6 mm
clearance because the knuckle will most likely fuse to the shaft, ren-
dering it immobile.
3.7.2. Design guidance for hinges
Hinges with a shaft clearance of 1.0 mm generally resolve success-
fully while those with a shaft clearance of 0.6 mm generally do not. As
seen in the linear accuracy and gap accuracy results, shrinkage and
oversintering tend to affect measurement results by about 0.2 mm.
Using this rule, all hinges with a shaft clearance greater than 0.6 mm
should be able to move freely. However, the results suggest that a shaft
clearance closer to 1.0 mm is necessary. The design of the hinge ac-
counts for this discrepancy. For a hinge to move freely, there needs to
be a small amount of unfused powder between the shaft and the
knuckle. However, the small shaft clearance allows additional heat to
be trapped. The resulting effect is that more oversintering occurs than
would otherwise be expected. Consistent with the hole results, hinges
built with the primary axis along the Z axis of the build chamber do not
resolve as reliably. Also similar to previous features, the PA 12 hinges
generally resolve much better than both GF PA 12 and FR PA 11. Hinges
built in the exterior portion of the build chamber have slightly better
resolution results than those built in the interior. The perimeter of the
build chamber is susceptible to lower part bed temperatures as heat
conducts away from the build volume. This slight variation in thermal
environments may improve the resolution of hinges by limiting the
amount of oversintering that occurs in the exterior region.
3.8. Lettering
Both embedded and raised lettering was tested at letter depths of
0.5–2.5 mm and font sizes ranging from 10 pt to 18 pt (Fig. 17). The
font was Arial (Regular), a Sans-Serif font. Both uppercase and
J. Allison et al.
Fig. 17. Lettering features for embedded (left) and raised (right) letters.
Table 8
Convention used to evaluate lettering resolution. (For interpretation of the re-
ferences to color in this table legend, the reader is referred to the web version of
this article.).
lowercase forms of the letter “A” were used as the test font. Lettering
resolution was determined using the criteria outlined in Table 8.
3.8.1. Resolution
Tables 9 and 10 show the resolution results for embedded and raised
lettering, respectively. For each cell in the tables, the total number of
instances of “pass,” “intermediate,” and “fail” ratings according to the
criteria in Table 8 are tallied. The cell color is determined by the cri-
terion with the highest number of occurrences. For instance, green cells
indicate more “passing” features than “intermediate” or “failing” ones,
but not necessarily a majority. Lettering resolution does not appear to
depend significantly on the depth of the embed/extrude, but depends
substantially on the size of the font. Embedded lettering resolved for all
font sizes at all depths. For this reason, a designer should consider
embedded lettering as opposed to raised lettering when adding text to a
part that is additively manufactured using selective laser sintering. If
raised lettering is desired, fonts larger than 16 pt should be used.
3.8.2. Design guidance for lettering
Embedded lettering is preferred to raised lettering because raised
lettering requires many fine “positive” features. At font sizes below
16 pt the features are too small to be resolved by the laser spot and
result in significant defects. Embedded lettering is less susceptible to
these defects because only the feature “negative” is scanned by the
laser. Lettering oriented along the Z axis of the build chamber performs
worse than the other directions. This trend is attributed to the effects of
Table 9
Resolution of embedded lettering for all test cubes. The cell color follows the
pass/intermediate/fail convention outlined in Table 8 corresponding to green,
yellow, and red, respectively. The color is determined by the highest count of
either pass, intermediate, or fail observations. (For interpretation of the refer-
ences to color in this table legend, the reader is referred to the web version of
this article.).
247
Additive Manufacturing 25 (2019) 239–251
Table 10
Resolution of raised lettering for all test cubes. The cell color follows the pass/
intermediate/fail convention outlined in Table 8 corresponding to green,
yellow, and red, respectively. The color is determined by the highest count of
either pass, intermediate, or fail observations. (For interpretation of the refer-
ences to color in this table legend, the reader is referred to the web version of
this article.).
spot size versus layer thickness mentioned in the thin walls section.
Consistent with many of the other features, the best results are achieved
with PA 12, followed by GF PA 12 then FR PA 11. Location within the
build chamber does not produce a discernable effect on lettering re-
solution.
3.9. Snap fits
The offset required for mechanical snap fits was determined by
testing an extruded square peg against slots of varying size (Fig. 18).
Square pegs of 3.0 mm nominal width were tested against slots of
2.9 mm to 3.5 mm nominal width in increments of 0.1 mm. Snap fit
resolution was defined by the offset distance between the peg and slot
that provides a snug fit.
3.9.1. Resolution
In general, as documented in Table 11, when incorporating me-
chanical snap fits into a design, an offset of 0.15 mm to 0.2 mm should
be used between the peg and slot. An example design and a depiction of
the offset distance are shown in Fig. 19.
3.9.2. Design guidance for snap fits
Snap fits that are built with the primary axis orthogonal to the build
platform require a larger offset for the best performance compared with
the other orientations. Attributed to the poorer resolution of the ma-
terial, FR PA 11 snap fits require the largest offset compared with GF PA
12 and PA 12. The results do not suggest a significant dependence on
location within the build chamber.
3.10. Machine variation
Five test cube replicates were built on one machine, and three re-
plicates were built on a second machine for each combination of ma-
terial, orientation, and build chamber location in an effort to assess
variability between machines. However, due to manufacturing
Fig. 18. Snap fit features in the test part.
J. Allison et al. Additive Manufacturing 25 (2019) 239–251

Table 11
Snap fit resolution for all test cubes. Green boxes indicate the offsets that cor-
respond to the best fit between the peg and slots. The values within each cell
indicate the proportion of observed instances that correctly resolved. (For in-
terpretation of the references to color in this table legend, the reader is referred
to the web version of this article.).

constraints, each machine was dedicated to a particular material and

was not used to process other materials; therefore, machine variability
must be considered on a per-material basis. Machine variability was
evaluated by implementing an analysis of variance test with multiple
comparisons, just as with the other input factors. Fig. 20. Interactive web tool filtered to display the results for all cubes built in
The rightmost column in Table 3 shows the machine variability for GF PA 12 and in the XY orientation.
each feature. The values in Table 3 represent the average variability
across each of the three materials to show the general trend. Mean
differences for machine variability signify the average difference in
measured dimensions between the first machine (five replicates built
for each factor combination) and the second machine (three replicates
built for each factor combination). The material-specific results for
machine variability can be found in the online web tool (http://
DesignForAM.me.utexas.edu). As with the other input parameters,
machine variability changes with feature type. Notably, linear accuracy
does not show a machine dependence based on the results of this study,
but half of the gap sizes do show statistically significant machine de-
pendence. This result can have implications for a designer creating a
part for polymer SLS. Depending on the part design, changing machines
may have a stronger effect on certain features over others. The results
also show that machine variability typically influences accuracy by
0.09–0.15 mm. Although the results suggest machine variability can be
inconsistent across materials and features, in many cases statistically Fig. 21. Interactive chart tool filtered to display results for all cubes built in GF
significant differences in feature accuracy are observed. The effects of PA 12 and in the XZ orientation.
switching machines therefore cannot be ignored.
to “XZ.” It can be seen that GF PA 12 walls built along the “XY” axes of
3.11. Additional results the build chamber (Fig. 20) successfully resolved for all thicknesses
measured in the study. Conversely, when the orientation is changed to
The online web tool (http://DesignForAM.me.utexas.edu) provides “XZ” (Fig. 21), only walls greater than 0.5 mm resolved successfully.
a platform in which designers can view additional results from the SLS This same method of comparison can be used to observe the effects of
metrology study. In some cases, designers may want to observe the any combination of input parameters that were included in the me-
effects of different process parameters on certain features. The web tool trology study.
includes an interactive component that enables the designer to display The level of detail offered by this tool can be achieved only in the
trends based on selected input parameters (http://DesignForAM.me. form of an interactive capability, as each combination of inputs gen-
utexas.edu/graphs). An example of the interactive chart tool is shown erates a unique set of associated graphs and charts. In fact, in order to
in Fig. 20. In this case, the designer has selected GF PA 12 as the ma- account for all input combinations in the polymer study, more than 800
terial and “XY” as the orientation across both locations within the build graphs and charts are needed. It is difficult for a designer to interpret
chamber. By selecting different options from the dropdown menu, the such a large number of figures, but the interactive tool supports quick
designer can compare the relative effects of any input parameter comparisons of parameters of interest. The interactive component of the
combination. For instance, if a particular design relies on the ability to web tool places the power in the hands of the designer, showing how
resolve thin walls in different orientations, the designer can quickly each processing parameter or combination of parameters can affect part
identify the thinnest observable walls in each orientation. Fig. 21 shows quality.
the updated graph selection when the orientation is changed from “XY”

Fig. 19. Example snap fit design. The image on the right shows the peg and slot dimensions, representing an offset distance of 0.15 mm.

248
J. Allison et al. Additive Manufacturing 25 (2019) 239–251

4. Discussion metrology study is enabled by a compact and comprehensive test part

The test cube features can be categorized into two regimes: “ad-
ditive” features and “subtractive” features. An “additive” feature is one
in which the laser scans the feature geometry as in thin walls and rods.
Gaps and holes are “subtractive” features in which the feature geometry
is determined by voids in the laser scan area. The results indicate that
“additive” features tend to be oversized compared to the nominal di-
mensions, while “subtractive” features tend to be undersized. The
driving factor behind this phenomenon is oversintering. Oversintering
occurs when the powder surrounding a part is unintentionally fused
during the sintering process. The additional fused powder causes the
dimensions of “additive” features to be larger than the as-designed
values and “subtractive” features to be smaller. The oversintering effect
appears to maintain a relatively constant value of 0.2 mm for both
feature regimes and does not appear to depend on feature size. The
degree of oversintering, however, does appear to depend on feature
orientation. Gaps and holes tend to be most undersized when the fea-
ture geometry lies within the build plane. Similarly, thin walls built
orthogonally to the build plane are typically oversized. Part of this over-
and under-sizing could also be attributed to laser scan patterns in-
corporated into the process planning software, especially since the
observed offset value of 0.2 mm is similar in magnitude to the laser spot
size (0.25 mm). Currently, SLS process planning software accounts for
shrinkage through part scaling; in future applications, it could be
adapted to compensate for the oversintering effects observed in this
study, as well.
The results of the ANOVA test with multiple comparisons suggest
that material and orientation are the most common statistically sig-
nificant factors. The variation attributed to changing materials or or-
ientations can lead to mean differences in dimensions as great as
0.4 mm. Location within the build chamber is statistically significant for
certain features, but the effect is generally minor with mean differences
in dimensions of approximately 0.05 mm. Machine variation is also
statistically significant for nearly one third of the features measured,
but the effect appears to be relatively small. The online web tool con-
tains much more detailed information on the effects of the various input
parameters, including an interactive component that allows the user to
filter the results for each parameter of interest.
5. Conclusions
The metrology study presented here investigates the effect of ma-
terial choice, orientation within the build chamber, location within the
build chamber, and machine identity on the accuracy and resolution of
features of interest for the polymer powder bed fusion process. The
that can be inserted into voids in existing builds to quantify design
constraints. The results are compiled into statistically meaningful de-
sign guidelines and integrated into an online web tool. The web tool can
be used by designers to reduce design iterations and aid in making
design decisions.
Future studies should be conducted to investigate new features or to
build additional copies of the test part in order to increase the statistical
fidelity of the results. It would be particularly interesting to build copies
of the test part on other types of machines at other industrial facilities
and compare the results to those reported here. The metrology strategy
used for polymer PBF can also be applied to different AM process and
added to the online web tool. The parts used in such studies should be
process-specific so that the capabilities and limitations of each AM
process are accurately captured. Finally, it could be enlightening to
develop and publish case studies that investigate the potential benefits
of applying the design guidelines presented in this paper to various
examples of interest to AM part designers. Of particular interest would
be the potential ability to increase the accuracy of the parts and prevent
defects without iterative redesign.
Disclaimer
The views and conclusion contained in this document are those of
the authors and should not be interpreted as necessarily representing
the official policies, either expressed or implied of the Government.
Distribution authorized to U.S. Government Agencies and America
Makes Members; Critical Technology. Other request for this document
shall be referred to AFRL/RXMS, Wright-Patterson Air Force Base, OH.
Acknowledgements
The authors gratefully acknowledge Stratasys Direct Manufacturing,
and especially Mr. Steven Kubiak, for supporting this study by fabri-
cating the test cubes. The authors also recognize Jay Pi, Prapti Ghiya,
and Connor Dilgren for their extensive contribution in collecting the
measurement data for each of the test cubes; David Rosen for helpful
guidance and support throughout the project; and Desiderio Kovar for
his assistance with metrology. This effort was performed through the
National Center for Defense Manufacturing and Machining under the
America Makes Program entitled “A Design Guidance System for
Additive Manufacturing (Project 4053)” and is based on research
sponsored by Air Force Research Laboratory under agreement number
FA8650-12-2-7230. The U.S. Government is authorized to reproduce
and distribute reprints for Governmental purpose notwithstanding any
copyright notation thereon.

Appendix A. Measurement procedure

Surface roughness was measured using a Zeta-20 3D optical profiler. The profiler created 3D surface scans generated from a series of images. The
images were taken at 0.3 μm steps in the Z-direction and the reported X–Y resolution was 0.9 μm. Each scan had a field of view of 1646 × 1235 μm,
and surface roughness was measured after stitching two surface scans together with a 10% overlap for a total field of view of 3128 × 1232 μm.
A gage repeatability and reproducibility (R&R) study was performed to quantify the uncertainty for the caliper measurements of linear accuracy,
gap size, and wall thickness. The R&R was conducted using 12 parts, 3 trials, and 2 operators. The R&R percentages listed in Table 12 correspond to
the fraction of total part variation attributed to the measurement system. While the R&R values for linear accuracy and wall thickness were within
the range of what is generally acceptable, the R&R for gap size was slightly higher. The observed part-to-part variation for gaps was smaller than the
other features, so a larger percentage of variation was associated with the measurement system.

Table 12
Gage R&R results for caliper measurements.
Feature R&R (%)

Linear accuracy 20.21

Gap size 40.49
Wall thickness 20.16

249
J. Allison et al. Additive Manufacturing 25 (2019) 239–251

Fig. 22. Hole size as a function of wall thickness with sample flatbed scan image.

Table 13
Average surface roughness for each combination of input parameters. The values in the table represent the average surface roughness across materials, locations, and
inclination angles relative to the build platform in microns.
Inclination angle Surface roughness

PA 12 GF PA 12 FR PA 11

Interior Exterior Interior Exterior Interior Exterior

0° 19.18 19.58 22.67 29.37 25.23 24.14

15° 30.82 29.94 31.98 31.15 34.63 36.82
30° 28.46 29.23 29.56 30.53 34.10 36.77
45° 26.34 27.14 28.16 28.07 35.86 37.05
60° 24.94 26.46 26.20 26.94 38.50 39.85
75° 24.28 24.25 24.46 26.86 34.59 39.21
90° 23.51 23.26 23.24 24.04 34.08 37.68

Table 14
Average linear accuracy for each combination of input parameters. Accuracy is defined as the measured dimension minus the nominal dimension. The values in the
table represent the average accuracy for a specific material, location, and orientation.
Linear accuracy

PA 12 GF PA 12 FR PA 11

Interior Exterior Interior Exterior Interior Exterior

XY XZ YZ XY XZ YZ XY XZ YZ XY XZ YZ XY XZ YZ XY XZ YZ

−0.02 −0.07 −0.09 −0.02 0.03 −0.02 −0.28 −0.23 −0.17 −0.36 −0.25 −0.18 −0.40 −0.34 −0.35 −0.36 −0.31 −0.30

Table 15
Average gap accuracy for each combination of input parameters. Accuracy is defined as the measured dimension minus the nominal dimension. The values in the
table represent the average accuracy for a specific material, location, orientation, and wall thickness.
Wall thickness Gap Accuracy
(mm)
PA 12 GF PA 12 FR PA 11

Interior Exterior Interior Exterior Interior Exterior

XY XZ YZ XY XZ YZ XY XZ YZ XY XZ YZ XY XZ YZ XY XZ YZ

1.0 −0.21 −0.22 −0.01 −0.12 −0.11 0.01 −0.26 −0.25 −0.14 −0.32 −0.22 −0.19 −0.26 −0.27 −0.23 −0.26 −0.27 −0.26
2.5 −0.13 −0.17 −0.08 −0.08 −0.08 −0.05 −0.23 −0.25 −0.23 −0.33 −0.28 −0.26 −0.35 −0.40 −0.34 −0.34 −0.36 −0.52
4.0 −0.11 −0.16 −0.09 −0.09 −0.06 −0.06 −0.24 −0.26 −0.25 −0.33 −0.29 −0.27 −0.36 −0.42 −0.40 −0.34 −0.39 −0.60
5.5 −0.12 −0.18 −0.10 −0.09 −0.09 −0.06 −0.25 −0.27 −0.27 −0.32 −0.29 −0.27 −0.34 −0.44 −0.46 −0.34 −0.40 −0.50
7.0 −0.11 −0.18 −0.11 −0.12 −0.09 −0.06 −0.25 −0.26 −0.27 −0.32 −0.28 −0.26 −0.34 −0.43 −0.45 −0.38 −0.39 −0.52
8.5 −0.14 −0.19 −0.12 −0.09 −0.10 −0.06 −0.23 −0.26 −0.27 −0.31 −0.26 −0.25 −0.35 −0.42 −0.42 −0.35 −0.38 −0.51
10 −0.10 −0.19 −0.11 −0.10 −0.11 −0.05 −0.20 −0.24 −0.24 −0.29 −0.23 −0.22 −0.33 −0.36 −0.35 −0.36 −0.32 −0.45

250
J. Allison et al. Additive Manufacturing 25 (2019) 239–251

Table 16
Average hole accuracy for each combination of input parameters. Accuracy is defined as the measured diameter minus the nominal diameter. The values in the table
represent the average accuracy for a specific material, location, orientation, and wall thickness.
Wall thickness Hole accuracy
(mm)
PA 12 GF PA 12 FR PA 11

Interior Exterior Interior Exterior Interior Exterior

XY XZ YZ XY XZ YZ XY XZ YZ XY XZ YZ XY XZ YZ XY XZ YZ

1.0 −0.14 −0.23 −0.34 −0.01 −0.03 −0.21 −0.18 −0.24 −0.48 −0.29 −0.24 −0.55 −0.16 −0.18 −0.44 −0.16 −0.16 −0.62
2.5 −0.14 −0.24 −0.53 −0.02 −0.15 −0.30 −0.22 −0.30 −0.62 −0.41 −0.35 −0.75 −0.34 −0.35 −0.73 −0.36 −0.33 −0.81
4.0 −0.21 −0.29 −0.63 −0.16 −0.23 −0.43 −0.27 −0.38 −0.74 −0.50 −0.46 −0.76 −0.52 −0.45 −0.88 −0.46 −0.42 −0.94
5.5 −0.26 −0.34 −0.64 −0.14 −0.25 −0.52 −0.31 −0.41 −0.88 −0.60 −0.55 −0.83 −0.45 −0.50 −0.72 −0.45 −0.49 −0.92
7.0 −0.34 −0.40 −0.68 −0.23 −0.25 −0.53 −0.32 −0.53 −0.90 −0.54 −0.55 −0.91 −0.52 −0.59 −0.75 −0.52 −0.53 −0.99
8.5 −0.26 −0.50 −0.72 −0.23 −0.19 −0.59 −0.36 −0.55 −0.75 −0.49 −0.52 −0.88 −0.56 −0.60 −1.09 −0.59 −0.58 −1.06
10 −0.27 −0.48 −0.64 −0.30 −0.21 −0.61 −0.34 −0.50 −0.81 −0.54 −0.44 −0.78 −0.63 −0.57 −0.98 −0.62 −0.54 −0.99

Table 17
Average thin wall accuracy for each combination of input parameters. Accuracy is defined as the measured dimension minus the nominal dimension. The values in
the table represent the average accuracy for a specific material, location, and orientation.
Thin wall accuracy

PA 12 GF PA 12 FR PA 11

Interior Exterior Interior Exterior Interior Exterior

XY XZ YZ XY XZ YZ XY XZ YZ XY XZ YZ XY XZ YZ XY XZ YZ

0.04 0.26 0.26 0.04 0.10 0.10 0.10 0.19 0.20 0.21 0.12 0.20 0.25 0.29 0.36 0.22 0.31 0.42

Hole diameter was measured using an optical flatbed scanner and processed in Matlab. As shown in Fig. 22, the part was scanned at a resolution
of 600 dpi and converted to a black and white image. The area of the white space occupied by each hole in the image was used to calculate the
diameter. It should be noted that eccentricity of the holes was not taken into account, so average diameter was reported. This measurement system
was verified against pin gages in increments of 0.001 in. Furthermore, each image was individually calibrated to the dimensions of a square with a
known size (red square in Fig. 22). Additional information on the measurement methodology is provided in [13] and in the accompanying web tool
(http://DesignForAM.me.utexas.edu).

Appendix B. Summary of measurement results by feature type

See Tables 13–17.

References design rules, EOS e-Manufac. Solut. (2008).

[1] G. Boothroyd, P. Dewhurst, W. Knight, Product Design for Manufacture and
Assembly, CRC Press, Boca Raton, FL, 2011.
[2] M. Dinar, D. Rosen, A design for additive manufacturing ontology, J. Comput.
Inform. Sci. Eng. 17 (2017) 021013 (1–9).
[3] H. Jee, P. Witherell, A method for modularity in design rules for additive manu-
facturing, Rapid Prototyp. J. 23 (6) (2017) 1107–1118.
[4] M. Mahesh, Y.S. Wong, J.Y.T. Fug, H.T. Loh, Benchmarking for comparative eva-
luation of RP systems and processes, Rapid Prototyp. J. 10 (2) (2004) 123–135.
[5] S. Moylan, J. Slotwinski, A. Cooke, K. Jurrens, M.A. Donmez, An additive manu-
facturing test artifact, J. Res. Natl. Instit. Stand. Technol. 119 (2014) 429–459.
[6] G. Adam, D. Zimmer, Design for additive manufacturing – element transitions and
aggregated structures, CIRP J. Manfac. Sci. Technol. 7 (2014) 20–28.
[7] J. Kranz, D. Herzog, C. Emmelmann, Design guidelines for laser additive manu-
facturing of lightweight structures in TiAl6V4, J. Laser Appl. 27 (S1) (2015)
S14001.
[8] L. Castillo, Study about the rapid manufacturing of complex parts of stainless steel
and titanium, TNO Ind. Technol. (2005).
[9] N. Meisel, C. Williams, An investigation of key design for additive manufacturing
constraints in multimaterial three-dimensional printing, J. Mech. Des. 137 (2015).
[10] F. Fernandez-Vicente, M. Canyada, A. Conejero, Identifiying limitations for design
for manufacturing with desktop FF 3D printers, Int. J. Rapid Manufac. vol. 5, (1)
(2015) 116–128.
[11] C. Seepersad, T. Govett, K. Kim, M. Lundin, D. Pinero, Design rules for selective
laser sintering, SFF Symposium, Austin, TX, 2012.
[12] D. Sippel, Investigation of detail resolution on basic shapes and development of
[13] J. Allison, C. Sharpe, C. Seepersad, A test part for evaluating the accuracy and
resolution of a polymer powder bed fusion process, J. Mech. Des. 139 (10) (2017)
100902.
[14] A. Wegner, G. Witt, Correlation of process parameters and part properties in laser
sintering using response surface modeling, Phys. Procedia 39 (2012) 480–490.
[15] I. Gibson, S. Dongping, Material properties and fabrication parameters in selective
laser sintering process, Rapid Prototyp. J. 3 (4) (1997) 129–135.
[16] D.L. Bourell, M.C. Leu, D.W. Rosen, Roadmap for Additive Manufacturing, (2009)
Washington, DC.
[17] J.P. Kruth, X. Wang, T. Laoui, L. Froyen, Lasers and materials in selective laser
sintering, Assemb. Autom. 23 (4) (2003) 357–371.
[18] U. Ajoku, N. Saleh, N. Hopkinson, R. Hague, P. Erasenthiran, Investigating me-
chanical anisotropy and end-of-vector effect in laser-sintered nylon parts, Proc.
Instit. Mech. Eng. B: J. Eng. Manufac. 220 (7) (2006) 1077–1086.
[19] W. Cooke, R.A. Tomlinson, R. Burguete, D. Johns, G. Vanard, Anisotropy, homo-
geneity and ageing in an SLS polymer, Rapid Prototyp. J. 17 (4) (2011) 269–279.
[20] B. Caulfield, P.E. McHugh, S. Lohfeld, Dependence of mechanical properties of
polyamide components on build parameters in the SLS process, J. Mater. Process.
Technol. 182 (2007) 477–488.
[21] D.L. Bourell, T.J. Watt, D.K. Leigh, B. Fulcher, Performance limitations in polymer
laser sintering, Phys. Procedia 56 (2014) 147–156.
[22] D.T. Pham, S. Dimov, F. Lacan, Selective laser sintering: applications and techno-
logical capabilities, Proc. Instit. Mech. Eng. 213 (5) (1999) 435–449.
[23] G.D. Kim, Y.T. Oh, A benchmark study on rapid prototyping processes and ma-
chines: quantitative comparisons of mechanical properties, accuracy, roughness,
speed, and material cost, J. Eng. Manufac. 222 (2) (2008) 201–215.

251