Joint Special Interest Group meeting between euspen and ASPE
Advancing Precision in Additive Manufacturing
Ecole Centrale de Nantes, France, September 2019
Improving the dimensional accuracy of downfacing surfaces
of additively manufactured parts
Umberto Paggi1,2,3, Mirko Sinico1,3, Lore Thijs2, Wim Dewulf1, Brecht van Hooreweder1,3
1 Department of Mechanical Engineering, KU Leuven, 3001 Leuven, Belgium
2 3D Systems Leuven, 3001 Leuven, Belgium
3 Member of Flanders Make - Core lab PMA-P, KU Leuven, 3001 Leuven, Belgium
In this study the accuracy of downfacing surfaces of small LPBF (Laser Powder Bed Fusion) fabricated features is investigated.
Calibration geometries with downfacing angles ranging from 75° to 25° are printed in three different layer thicknesses to quantify
the dimensional error due to dross formation and other melt pool phenomena in the downfacing area. Their dimension is measured
with a digital microscope and compared with the CAD model: all geometries show a negative mismatch which is proportional to the
downfacing angle, therefore a correction parameter is calculated. To validate the effectiveness of this offset, microchannels with
diameter ranging from 1 mm to 0.6 mm were printed both with and without the correction. Their deviations from the nominal
diameter were investigated through the use of micro Computed Tomography. Results show less deviation from the nominal
cylindrical CAD when the offset is applied, while the dross formation remains almost unaffected.
Laser Powder Bed Fusion, Downfacing surfaces, Beam Compensation, Microchannel accuracy, X-ray micro Computed Tomography
The Laser Powder Bed Fusion technology has experienced a
constant increase in popularity in recent years thanks to the
possibility of producing near net shape components with a high
degree of complexity which would be unattainable with
standard manufacturing techniques.
Nonetheless, metal 3D printing still faces challenges in term of
accuracy and surface quality. Downfacing surfaces, for example,
are among the most common and most problematic features in
the LPBF process. Typically, these areas suffer from localised
overheating which produces drosses that negatively affect the
surface roughness [1,2]. Furthermore, their geometrical
accuracy is lower, compared to a middle section, because of
warping and distortions that can happen on unsupported and
overhanging areas [3,4,5]. This can be a major problem in
complex geometries with unreachable downfacing regions
because the dimensional error can be larger than the tolerances
required and can lead to build failure [6,7,8].
To mitigate these problems the most common approach is to
separate the sample in two zones, namely the middle zone
which stands over the bulk metal of the previous layer and the
downfacing zone which is printed directly above lose powder .
For each region the process parameters can be tuned in order to
achieve the required properties (minimum density, low residual
stresses, high surface quality and so on); in particular for the
downfacing area a more stable weld track is generally desired
Another approach which is typically followed in the industrial
environment is to develop offsets and compensation
parameters to balance all the aforementioned distortions. One
main issue with the current procedures is that the same
optimization is applied indiscriminately to downfacing surfaces
with different angles, or at best a distinction is made between
flat or almost flat overhangs (0° to ~20°) and generally sloped
areas. This study investigates the possibility of developing
correction factors which evolve gradually with the downfacing
angle in order to reduce the amount of critical areas and
therefore increase the freedom in the design and production
To do so, calibration geometries are printed with a downfacing
angle that varies from 75° to 25°. The offsets are calculated by
measuring the difference between the samples’ nominal width
and the real width. To validate these results microchannels are
produced both with and without the compensation factor. The
channels are then analysed with an industrial microfocus CT
(µ-CT) and the dimensional error reduction is estimated.
2. Materials and methods
2.1. Benchmarks design and manufacturing
In this study a series of calibration samples are produced to
evaluate the dimensional deviation at different angles due to the
downfacing region. The geometries are first designed with
3DXpert software provided by 3D Systems. This software
Figure 1. a) Calibration geometries with downfacing angle ranging
from 75° to 25°; b) print simulation of the 25° sample; c) cross section
of 1 mm microchannel without compensations (left) and with the
offsets applied (right).
includes a print simulation tool that estimates the distortion
caused by shrinkage phenomena and residual stresses after the
build (Figure 1b, note that it tends to overestimate the deviation
magnitude). In order to account only for the error due to the
localized effects proper of the downfacing region this tool has
been used to create a CAD model which minimize the
aforementioned distortions, typically linked to the sample’s
Samples angles varying from 75° to 25° plus a reference at 90°
(shown in Figure 1a) are printed in 30 μm, 60 μm and 90 μm layer
thickness with a DMP Flex 350 in titanium powder LaserForm
Ti Gr23 (A). To test the effectiveness of the correction parameter
horizontal channels are also produced in 30 μm layer thickness.
A cube of 10x10x10 mm was printed with two series of
cylindrical holes of 1 mm, 0.8 mm and 0.6 mm diameter, one
compensated and one not compensated. These features are
selected because their dimension is in the same order of
magnitude of the offsets, and therefore will benefit the most
from this correction procedure. Likewise, the layer thickness is
set at 30 μm because it grants the higher accuracy for small
geometries and reduce the noise in the subsequent measures
due to the surface roughness.
The initial calibration samples are lightly sandblasted to
remove the sintered powder on the surface which would
otherwise introduce high noise rending the data unreliable.
2.2. Samples analysis
The top surface of the calibration geometries is measured with
a Keyence VHX-6000 digital microscope, using a 100x
magnification. To avoid any operator dependency in the final
value the width of the sample is calculated by interpolating the
sample’s sides with two parallel lines using the edge detection
tool available in the microscope software. For each point four
samples are measured and the standard deviation of these data
is used as an estimation of the error.
To calculate the correction factors for different angles we
compare the measured and nominal width of the calibration
geometries’ top surface. From these measurements two
different offsets are derived: the first one is proper of the middle
areas and is correlated to the error due the melt pool width. As
shown in Figure 2a, if no correction is applied then the laser will
scan the CAD contour and because of the melt pool dimension
the final cross section will be bigger than the CAD one.
Therefore, a middle offset is calculated as shown below and a
new scanning contour is created (Figure 2b) so that the final
dimension matches the 3D model.
Where Off is the correction factor for a middle region,
is the nominal width of the sample and W
real width of the sample.
When a downfacing re gion is printed the compensation for the
melt pool width is still applied, but other phenomena (explained
below) affect the final cross section dimension (Figure 2d).
Therefore, a new downfacing offset is calculated (Figure 2e),
which is zero for a 90° sample since no downfacing area is
present, and calculated as follow.
is the correction factor for a general downfacing
region at an angle of α°, (
is the nominal width of
the sample with the applied middle offset and W
real width of the sample.
Regarding the microchannels benchmark, this was analysed
trough a µ-CT scan performed on a Nikon XTH 225ST machine at
195 kV, 62 µA, exposure 2000 ms, and tungsten target. A copper
filter of 0.5 mm was used during the scan to reduce the beam
hardening artefact. In total, 3142 projections were acquired at a
magnification of ~20 and reconstructed with a voxel size of
Figure 3. a) Downfacing geometry graph that shows the trend of the
negative mismatch with the printed angle. b) 30° 30 μm layer thickness
sample with detail of maximum deviation on the edge. c) 30° 90 μm
layer thickness sample with detail of maximum deviation on the edge.
Figure 2. a) Sample printed without any correction; b) sample printed with middle offset; c) schematic of a general downfacing calibration sample;
d) downfacing sample printed only with middle offset; e) downfacing sample printed with middle and downfacing offset.
10 µm. The collected CT projections after reconstruction were
imported in VGSTUDIO MAX 3.2.3. The mentioned software was
used to first determine the surface of the acquired dataset, using
an advanced ISO50% surface determination with a set searching
distance of 40 µm. Subsequently, nominal cylinders of 1 mm,
0.8 mm and 0.6 mm were fitted against the compensated and
non-compensated channels present in the benchmark. To avoid
fitting errors, a first raw fit was performed with a manual
alignment of the nominal cylinders followed by a refined best-fit
executed with the advance refinement VGSTUDIO algorithm. A
nominal/actual comparison was extrapolated comparing the
nominal cylinders to the printed micro-channels, and a total of 6
deviation plots were exported (3 compensated and 3
3. Results and discussion
3.1. Downfacing offset
The results for 30 μm, 60 μm and 90 μm layer thicknesses are
shown in Figure 3a. A clear trend is shared between the three
series which links larger downfacing angles to larger negative
This could be explained with the mechanism of fluid flow that
occurs on melt pools supported mostly by the metal particles, as
shown by Chen et al. . The liquid metal is subj ect to a number
of factors (capillarity forces, gravity, Marangoni convection flow
and others) which create an unstable melt pool that sinks in the
powder bed, reducing the final width of the scan track.
Moreover, as reported by Wang et al. , the unsupported
edges of the scan tracks are prone to bending upwards due to
residual stresses, which leads to a negative deviation of the final
shape in the x-y plane.
It is also clear that a thicker layer is correlated to a larger error.
This is because larger melt pools are more instable and capillary
forces make the melt track prone to break into spherical
agglomerates . The resulting rough edge has a maximum
deviation comparable with the offset calculated for the
corresponding angle, as shown in Figure 3c.
In order to have a more consistent offset which smoothly
transitions from zero to the maximum measured deviation, the
experimental data is fitted with a polynomial fit of grade 2, fixing
a value of zero at 90° as explained above. This choice is arbitrary
and does not represent any theoretical prediction of the melt
pool behaviour, but it’s still selected because it’s quick to
perform and because the resulting curves are within the error
bars of almost all the data point. A linear fit was also tested but
in this case the R-squared values are higher compared to the
3.2. Compensation for microchannels
In microchannels two types of dimensional error can be
typically identified : at angles close to flat overhangs
overheating and dross formation occurs, while from 90° to ~30°
we see a straightening of the side of the cylinder which leads to
a more squared cross section. As shown in Figure 4c, in the first
case the cross section of the printed cylinder will be smaller than
the desired dimension, therefore dross defects will be referred
to as negative deviation. On the contrary, the squared shape
resulting from the straightening is bigger than the cylindrical one
and will be referred to as positive deviation.
In Figure 4d-e it’s possible to identify these features both in
the non-compensated and compensated chan nels. The negative
deviations (in purple) are mainly due to dross and in some cases
high surface roughness, whereas the positive deviations (in red)
consist of the straightening effect on the downfacing region and
the staircase effect on the bottom of the cylinder. Qualitatively,
it’s already possible to say that the downfacing offset plays a
major role in smaller channels while the 1 mm cylinder still
shows a relevant amount of straightening in the compensated
In Figure 4a a more detailed representation of the surface is
given in the deviation plot .The main peak in the centre of the
graph corresponds to the cylindrical surface, shown in green in
Figure 4d-e. If the microchannel were perfectly round, smooth
and equal to the CAD dimension, then on the graph there would
be only a single vertical line placed at x = 0. In reality, the sample
Figure 4. a) Deviation plot of the non-compensated 1 mm channel; b) deviation plot of the compensated 1 mm channel; c) schematic
representation of the different deviations in a microchannel. CT scan of microchannels of d) 1 mm and e) 0.6 mm , compensated one on the rig ht.
has a rough surface which deviates both in positive and negative
direction and causes the values to spread. Following Schmähling
and Hamprecht , this spread is assumed to be randomly
distributed and therefore the peak can be interpolated with a
Gaussian fit. Moreover, the centre of the peak is not at zero but
is always shifted between -20 μm and -40 μm, which means that
the diameter of the printed channel is smaller than the nominal
one. These numbers are consistent with the thickness of the
sintered powder layer that cannot be removed from internal
features via sandblasting, but shrinkage phenomena could also
be contributing to the cross section reduction.
After removing the central peak from the data three main
regions can be defined: on the left we have the negative
deviation caused by the dross, in the middle we have the noise
and residue from the Gaussian subtraction which will not be
considered further, and on the right there is the positive
deviation due to the straightening and staircase effects. It’s
important to mention that several fitting intervals were tested
to assess the interpolation influence on the final data and the
results showed very little variation. To evaluate the
effectiveness of the offsets the total areas affected by the
positive and negative deviations are calculated and normalized
with the nominal channel surface from the CAD model. Results
are shown in Table 1.
All samples show a reduction in the straightening effect
although the compensation strategy seems to work best for
small feature which dimension is comparable to the offset value.
It’s interesting to notice that the dross effect is also slightly
reduced because by printing a rounder shape the overhanging
area close to 0° is less compared to the squared cross section
shown in Figure 4c.
In this paper, LPBF calibration geometries are printed in 30
μm, 60 μm and 90 μm layer thickness with a downfacing angle
that varies from 75° to 25°. The middle and downfacing offsets
are calculated by measuring the difference between the
samples’ nominal width and the real width. The experimental
data are then interpolated with a parabolic fit to obtain a
smooth transition between zero and the maximal deviation,
which equals ~70 μm for the 30 μm layer thickness. To validate
these results microchannels of 1 mm, 0.8 mm and 0.6 mm
diameter are produced both with and w ithout the compensation
factor. The channels are then analysed with an industrial
microfocus CT and a reduction in the straightening effect of,
respectively, 1.6 %., 11.7 %, and 8.2 % is achieved. A small
improvement is also obtained on the sag defect with a reduction
of 7.1 %, 5.3 %, and 4.7 %. This is because the squared profile
caused by the straightening has a larger overhanging area close
to 0° compared to the cylindrical shape. Therefore, by printing a
cross section closer to the CAD model this critical overhanging
zone is reduced.
It’s important to note that these final numbers should be
considered more as qualitative results rather than absolute
ones. A series of approximations has been applied to the data,
starting from the CT measurement itself which has a limited
achievable resolution and yet unknown uncertainty. Moreover,
the Gaussian fitting of the analysed deviations for the roughness
contribution is not a perfect representation of the actual surface
and some residual deviation from the model is always present.
Nonetheless, a reduction in both positive and negative
deviations is clearly seen in the compensated channels,
therefore it can be concluded that applying a localized
correction specific for every downfacing angle definitely
improves the accuracy of small and fine features.
In future work a larger set of calibration samples should be
measured to decrease the standard deviation on the single data
points. Small geometries in 60 μm and 90 μm should also be
printed to check the effectiveness of the downfacing offset for
these parameters, but the accuracy of these two layer
thicknesses is generally low and often leads to occlusions which
render the microchannels non-functional. Therefore, some
other types of benchmark should be designed. Finally, more
attention should be given also to shrinkage and other
phenomena that affect the build accuracy and that scale with
the feature dimension, to understand why the offset strategy
seems less effective on the 1 mm cylinder.
This research was funded by The EU Framework Programme
for Research and Innovation - Horizon 2020 - Grant Agreement
No 721383 within the PAM2 research project.
 Jamshidinia, M. & Kovacevic, R. T he influence of heat accumulation
on the surface roughness in powder-bed additive manufacturing. Surf.
Topogr.: Metrol. Prop. 3, 014003 (2015).
 Xiang, Z., Wang, L., Yang, C., Yin, M. & Yin, G. Analysis of the quality
of slope surface in selective laser melting process by simulation and
experiments. Optik 176, 68–77 (2019).
 Cheng, B. & Chou, K. Deformation Evaluation of Part Overhang
Configurations in Electron Beam Additive Manufacturing. Processing
V001T02A072, ASME, Volume 1 (2015).
 Calignano, F. Design optimization of supports for overhanging
structures in aluminum and titanium alloys by selective laser melting.
Materials & Design 64, 203–213 (2014).
 Wang, D., Yang, Y., Yi, Z. & Su, X. Research on the fabricating quality
optimization of the overhanging surface in SLM process. The
International Journal of A dvanced Manufacturing Technology 65, (2013).
 Pakkanen, J. et al. Study of Internal Channel Surface Roughnesses
Manufactured by Selective Laser Melting in Aluminum and Titanium
Alloys. Metall and Mat Trans A 47, 3837–3844 (2016).
 Snyder, J. C., Stimpson, C. K., Thole, K. A. & Mongillo, D. J. Build
Direction Effects on Microchannel Tolerance and Surface Roughness. J.
Mech. Des 137, 111411-111411–7 (2015).
 Atzeni, E. & Salmi, A. Study on unsupported overhangs of AlSi10Mg
parts processed by Direct Metal Laser Sintering (DMLS). Journal of
Manufacturing Processes 20, 500–506 (2015).
 Cloots, M., Zumofen, L., Spierings, A. B., Kirchheim, A. & Wegener, K.
Approaches to minimize overhang angles of SLM parts. Rapid
Prototyping Journal 23, 362–369 (2017).
 Charles, A. et al. Effect of process Parameters on Dimensional
accuracy of Down-facing surfaces in Selective Laser Melting of Ti6Al4V.
ASPE and euspen Summer Topical Meeting - Lawrence Berkeley National
Laboratory, Berkeley, California, USA Volume: 69, (2018)
 Fox, J. C., Moylan, S. P. & Lane, B. M. Effect of Process Parameters
on the Surface Roughness of Overhanging Structures in Laser Powder
Bed Fusion Additive Manufacturing. Procedia CIRP 45, 131–134 (2016).
 Mertens, R., Clijsters, S., Kempen, K. & Kruth, J.-P. Optimization of
Scan Strategies in Selective Laser Melting of Aluminum Parts With
Downfacing Areas. Journal of Manufacturing Science and Engineering
 Chen, H., Gu, D., Xiong, J. & Xia, M. Improving additive
manufacturing processability of hard-to-process overhanging structure
by selective laser melting. Journal of Materials Processing Technology
250, 99–108 (2017).
 Schmähling, J. & Hamprecht, F. A. Generalizing the Abbott–
Firestone curve by two new surface descriptors. Wear 262, 1360–1371
Table 1. Normalized area affected by positive and negative
deviations. The nominal area employed for 1 mm, 0.8 mm and 0.6
diameter is respectively 31.4 mm2, 25.12 mm2, and 18.84 mm2.