Content uploaded by Mirko Sinico
Author content
All content in this area was uploaded by Mirko Sinico on Jul 23, 2018
Content may be subject to copyright.
CHARACTERIZATION OF AM METAL POWDER WITH
AN INDUSTRIAL MICROFOCUS CT: POTENTIAL AND LIMITATIONS
Mirko Sinico, Evelina Ametova, Ann Witvrouw, and Wim Dewulf
Department of Mechanical Engineering
KU Leuven
Leuven, Belgium
INTRODUCTION
Final precision of parts produced by Selective
Laser Melting (SLM) and other Powder Bed
Fusion (PBF) technologies relies on material
inputs. Powder characteristics not only influence
part density, mechanical properties and
microstructure but also geometrical parameters,
such as the applicable minimum layer thickness,
the maximum overhang angle and the overall
final surface quality [1]. Effective metrology
techniques are therefore needed to characterize
both particle size distribution and morphology,
and to follow the evolution of these
characteristics during different builds after
powder sieving and reuse.
Cone-beam computed tomography (CT), as
studied for instance in the field of sand and soil
particles characterization [2], has shown a
significant empirical potential as an all-in-one
tool to characterize size and shape at grain
scale. However, compared to these previous
studied applications, SLM powder has much
smaller particles which imposes a new challenge
on CT powder characterization.
CT reconstruction generates a 3D voxelized
attenuation map of the measured object from a
series of 2D projections acquired at different
perspectives. Image segmentation is used to
delineate objects for further analysis. A number
of factors affects the cone-beam CT
measurement performance of micro-objects,
such as spatial resolution loss due to the focal
spot blurring effects and high sensitivity to
inaccuracy in a magnification factor.
This work explores potential and limits of
industrial microfocus CT (µ-CT) as an all-in-one
tool to characterize the morphology of metallic
Additive Manufacturing (AM) powder, the
particles size distribution and, possibly, the
contamination of the feedstock powder.
µ-CT STUDY FOR AM POWDER ANALYSIS
AM powder samples
For the study, we selected three different
powder samples from a specialized AM powders
producer, with D10-D90 of the distribution as
listed in the vendor’s specifications:
- Mar300 15-45: Maraging 300 Alloy Steel
powder 15-45 µm;
- Mar300 5-25: Maraging 300 Alloy Steel
powder 5-25 µm;
- Ti-CP 15-45: Titanium CP powder 15-45 µm.
To confirm the vendor’s specifications and to
validate the subsequent µ-CT analysis, laser
diffraction (LD) measurements were obtained by
means of a Beckman Coulter's LS 13 320. The
underlying mathematics of LD measurements
assume spherical particles to reconstruct the
diffraction pattern. This is a reasonable
assumption for virgin powders, and LD analyses
can be adopted as a reference for the following
µ-CT study.
FIGURE 1. LD distribution curves of Maraging
300 powders.
Particle size distributions from LD are presented
in Figure 1 for the Maraging 300 powders. The
particle size distribution for the Ti-CP 15-45
sample was nearly identical to the Mar300 15-
45, therefore it is not shown in the figure.
Results for all three samples are summarized in
Table 1, with D10, D50, D90 cumulative volume
fractions and the average particle sizes.
TABLE 1. Summary of LD analyses of the three
powder samples.
SAM PL E
D10
[µm]
D50
[µm]
D90
[µm]
AVG
[µm]
Mar300 5-25 5.2 9.6 15.6 9.2
Mar300 15-45 15.9 29.2 44.4 27.6
Ti-CP 15-45 19.8 32.2 44.3 30.5
The average particle size for the Mar300 5-25
sample is out of specifications, presenting a D90
of 15.6 µm instead of 25 µm. This could impede
further investigations of this powder with the
selected µ-CT system due to its finite spatial
resolution.
Experimental setup
Spatial resolution is one of the limiting factors in
µ-CT data analysis and, consequently, there
should be a sufficient distance between two
objects to differentiate them. Separating
particles from one another has been reported as
the main issue in CT metrology of fine metallic
powder [3], and a good dispersion of the
particles in the specimen is advised. Specimens’
preparation for this study is realized by dry
spraying powder on a double-sided tape and
subsequently rolling the tape on a 3 mm wooden
stick. This approach allows to relatively
uniformly distribute the powder particles and to
limit sample dimensions to achieve high CT
magnifications.
µ-CT scans were performed on a Nikon XTH
225ST machine at 80 kV, 80 µA, exposure 2000
ms, and tungsten target. In total, 3600
projections were acquired at a magnification of
69 and reconstructed with a voxel size of 2.9
µm. One of the usual concerns associated with
high magnification scans is blurring due to finite
focal spot size. The total power was therefore
limited to constrain the X-ray gun focal spot size
around 3 µm, which is the smallest possible for
the used machine.
Voxel scaling factor
Discrepancy between nominal and actual
positioning of the sample with respect to the
source and the detector due to machine’s
kinematic errors can introduce significant errors
in dimensional measurements [4]. Given the
high magnification, common artifacts [5] were
not suitable; therefore a new object (“fish-eggs”
artifact) consisting of 1 mm high precision
(Grade 10, ISO 3290-1:2014) stainless steel
spheres was designed to calculate the voxel
rescaling factor.
FIGURE 2. (Left) Example of µ-CT projection of
the “fish-eggs” artifact; (Right) after
reconstruction, touching high-precision spheres
are selected for center-to-center measurements.
Spheres were poured in a hollow ~3 mm plastic
cylinder (Figure 2) and the “fish-eggs” artifact
was measured with the µ-CT system at the
same position of the rotation stage and under
the same settings prior to any specimen scan.
Voxel scaling factor was calculated as the mean
value of all center-to-center distances between
spheres in contact; at least 10 pairs of spheres
in contact were considered for each
measurement. Rescaling was consistent
between different scans, with an avg. of 0.9205
± 0.0002 and a standard deviation of 2.6E-4.
Surface determination
In previous studies, authors used a thresholding
binarization of CT data with a subsequent
surface smoothing of extracted voxels clusters
[6], which potentially might introduce
inaccuracies in extracted geometrical
information due to discretization errors and
partial volume effect. In this study, we use
VGSTUDIO MAX 3.1 (abbreviated as “VG”) to
perform surface extraction with sub-voxel
resolution. As a result, we extract the object
surface with no intermediate binarization step.
The standard initial contour value is typically
based on the ISO-50% global thresholding,
selected as the arithmetic mean of the grey
value corresponding to the background and the
grey value corresponding to material peak in the
grey-value histogram. This method was not
applicable to our specimens, since no clear
material peak was detected in the grey value
histograms (top of Figure 3, for Mar300 15-45).
This is caused by the high surface/volume ratio
of the particles, which, for this combination of CT
scan settings and amount of material in the
specimens, induces a smooth transition of grey
values between background and material peak.
Consequently, global thresholding methods are
not suitable for surface extraction of our
datasets.
FIGURE 3. Mar300 15-45 dataset (Top)
Automatic global thresholding ISO-50%;
(Bottom) Local thresholding method, example
for a single particle of the acquired dataset.
Therefore, a local thresholding method was
chosen and its settings were tuned based on the
LD measurements. In general, local thresholding
methods can improve the measurement
accuracy by partial compensation of beam
hardening and other CT artifacts [7]. At the
same time, local thresholding is more time
consuming, settings-sensitive and none of the
methods are suitable for all situations.
Our approach takes in consideration the
development of the grey values local profile for
particles of various size in every data set. The
first derivative of the local profile is analyzed to
find a sharp change in grey level denoting the
edge and, consequently, a particle surface.
Differently from Y. Tan et al. [7] approach, we
always look for the minimum gradient between
the ISO-50% calculated by VG and the ISO
value corresponding to the background peak. An
example of selection of the ISO value for the
Mar300 15-45 is presented in Figure 3 (bottom).
The point of deviation from the linear fit of the
right-hand side gradient corresponds to the min.
of the first derivative of the local profile.
Surface determination is finally run on VG with
the selected ISO value in a standard mode and,
for comparison, with the automatic ISO-50%
starting contour in a local adaptive mode (search
distance of 4 voxels). This procedure is repeated
for all three samples, which resulted in a total of
six datasets.
Data analysis
For every dataset, two analysis procedures are
performed: particles pores content estimation
and size/morphology characterization.
A porosity analysis is performed in VG using the
porosity/inclusion analysis module with a
“threshold only” method. For every dataset, the
threshold value was set to the corresponding
ISO value as discussed in the previous section.
For morphology/size analysis, a region of
interest of ~10000 particles is selected in every
dataset, and the particle surface is converted to
a triangular mesh using “manual” mode with a
volumetric sampling interval of 2 μm.
A particle analysis tool developed in-house in
MATLAB was used to process the extracted
particle surface and to calculate characteristics
such as volume, surface area, principal
dimensions, sphericity, bounding sphere, etc.
RESULTS
Particles porosity analysis
As expected, the amount of porosity in the
selected specimens is almost negligible, due to
the small average size of the particles [8] and to
the high quality production processes (generally
gas atomization for Maraging 300 and plasma
atomization for Titanium CP).
FIGURE 4. Normal distribution of the detected
pores in the Mar300 15-45 powder specimen.
Only the Mar300 15-45 sample presents enough
data points to derive a statistical analysis of the
pores content, see Figure 4. The sample
contained 63023 particles, but only 21 pores
were detected with an average dimension of 12
µm. It must be considered that with a roughly 3
µm voxel size resolution, the internal porosity
content might be underestimated, as reported by
F. Bernier et al. [3], and a CT system with a
higher resolution might provide a better
estimation of the total amount of porosity.
Powder distribution analysis
Since particles were distributed on tape using
dry spraying, there were multiple particles either
touching each other or located too close to
resolve them in CT scans. Therefore, prior to
calculation of particle distribution and
geometrical characteristics, we filtered out all
possible artificial agglomerations.
Filtering is implemented in the MATLAB code
based on principal dimensions (with a ≥ b ≥ c):
only spheroids are selected (Figure 5, a & b) for
size distribution estimation and shape analysis,
corresponding to particles with principle
dimensions ratios b/a > 2/3 and c/b > 2/3.
Moreover, particles with a total volume below 8
voxels are also filtered out in 15-45 datasets, to
avoid potential background noise [9] and as no
particles of this size are expected in those
specimens.
FIGURE 5. Examples of a) a particle with
satellites, b) a spherical particle, and c) artificial
particles agglomeration (subsequently filtered)
for the Mar300 15-45 powder.
Even if supported by previous studies [2,9], this
approach will limit further investigations on
powder reuse and powder degradation. The
proposed approach for particle filtration is hence
applicable only for virgin high-quality powder. A
new filtering method, or increased CT scan
resolution is required for more advanced powder
metrology.
Obtained cumulative distributions are presented
in Figure 6. For every specimen we compare the
LD measured distribution against the µ-CT
results.
FIGURE 6. (Top) Mar300 5-25; (Middle) Mar300
15-45; (Bottom) Ti-CP 15-45; calculated
distributions, comparison of LD vs µ-CT.
µ-CT analysis of the Mar300 5-25 was
performed without filtering on size, since a
significant fraction of the particles volume falls
below a total volume of 8 voxels. While
circumscribed sphere measurements can still be
accomplished on those small particles fractions,
erroneous segmentation due to CT noise,
b)
a)
c)
significant blurring and limited data points for
surface extraction hamper the data analysis of
the Mar300 5-25 specimen. For those reasons,
we suggest that specimens with particles below
~8 µm should not be investigated with our
current µ-CT setup, and no shape analysis has
been performed on the Mar300 5-25 dataset.
On the other hand, implementation of the local
thresholding was successful for all the
specimens. Almost a perfect agreement with LD
is found on the average particles size, and total
size spread. Differences in the skewness of the
distributions might be explained by the
differences of the two measuring methods:
- Particle size is obtained as a deconvolution of
the diffraction patterns for LD vs a 3D shape
analysis (circumscribed sphere) for µ-CT;
- Sample measured by LD contained >>10000
vs ~10000 particles measured with µ-CT;
- Sample preparation is obtained with water
dispersion plus sonication for LD vs dry
spraying for µ-CT.
As has been reported in different studies [1,10],
skewness and shape of the powder distribution
are highly dependent on the instrument, which
must be carefully selected based on powder
characteristics and application.
One clear outcome of this investigation is that
standard global thresholding ISO-50% method
with this µ-CT setup fails to provide a reliable
powder distribution, even with local adaptive
surface determination mode: overestimation of
the ISO value for surface determination will
always lead to a huge underestimation of the
particles size.
Powder shape analysis
Shape is a critical aspect for AM metal powders,
since shape will influence the flowability and its
apparent/tapped density. Different shape
parameters have been developed overtime to
describe powders, mainly taking in consideration
only the 2D shape that might result from an
optical/SEM microscopy or laser beam based
measurement. Qualitative descriptions of
powder particles have been as well standardized
[11], and out of all possible metrics, circularity
and aspect ratio are the most commonly used
shape factors for AM powder. A single 2D shape
factor cannot completely describe a particle
shape, and assumptions must normally be made
for every specimen.
The power of µ-CT analyses relies on the fact
that a full 3D shape is acquired, providing
domain specialists with comprehensive
information which potentially can be integrated
into multi-criteria decision making approaches
and advanced SLM process simulations. As an
example, for every powder specimen, 3D shape
factors like roundness, sphericity and
compactness can be calculated using our in-
house MATLAB code. Sphericity results are
summarized in Figure 7 for both Mar300 15-45
and Ti-CP 15-45 samples.
FIGURE 7. Sphericity index for Mar300 15-45
and Ti-CP 15-45.
As expected, a shape analysis results in high
sphericity for both specimens, with Ti-CP being
overall slightly more spherical.
New 3D shape indicators, such as the possibility
to count and describe particles satellites, as well
as more advanced 3D shape analyses [2] will be
the focus of future work.
PRELIMINARY STUDY ON
MULTIMATERIAL µ-CT POWDER ANALYSIS
A preliminary study is hereby introduced to
explore more advanced applications of CT to
metallic powder characterization. Besides
powder porosity, distribution and shape, µ-CT
has already been used to detect cross-
contamination of powder batches either in the
final printed part [12] or in the raw material state
[13]. Although detection was successful for both
cases, a precise assessment was not possible
because of the difficulties in resolving touching
particles and in the multi-material surface
determination.
To overcome those limitations, a new approach
is developed and implemented in MATLAB. The
main intuition relies on the fact that even if no
well-defined and separated material peaks are
present on the scans, a thresholding ISO value
can be selected looking at the local gray
FIGURE 8. Schematic of the multi-material µ-CT powder analysis to detect cross-contamination of
powder batches; example for an artificial contaminated Mar300 15-45 specimen with 10 vol% Ti-CP.
values profile to iteratively isolate the
contaminant from the main powder batch. To
confirm this assumption, an artificial
contamination is produced on a Mar300 15-45
specimen, with 10 vol% amount of Ti-CP. The
full analysis process is schematized in Figure 8.
A total contamination of 9.7 vol% of Ti-CP is
retrieved with the implemented code, in good
agreement with the prepared sample.
Undergoing efforts are focused on the statistical
validation of the implemented method,
comparing results at different percentages of
artificial contamination for different materials.
CT users must be aware that powders of
materials with similar densities will not be
distinguished with current CT machine setups.
Nonetheless, theoretically, even oxides of the
same material could be distinguished if the
difference in X-ray attenuation is high enough,
and µ-CT analysis could consequently
investigate not only powder cross-
contamination, but also powder degradation
after sieving and reuse.
CONCLUSIONS
After adequate scaling error compensation, the
acquired data set demonstrates how µ-CT could
be a viable metrological technique to derive, with
one measurement, primary characteristics of
PBF starting materials.
For distribution and shape analyses, special
care must be employed on the ISO value
determination prior to surface extraction. The
current main limitation relies on the finite focal
spot size of the µ-CT X-ray source and
consequently on the limited spatial resolution.
Statistical validation of the proposed local
thresholding method for surface determination
must be done at different µ-CT scan settings
and with different powder specimens.
The potential for detecting powder cross-
contamination and eventually powder
degradation has been introduced and will be the
main effort in future studies.
ACKNOWLEDGEMENTS
This research was funded by The EU
Framework Programme for Research and
Innovation - Horizon 2020 - Grant Agreement No
721383 within the PAM2 (Precision Additive
Metal Manufacturing) research project.
REFERENCES
[1] Sutton AT, Kriewall CS, Leu MC, Newkirk JW. Powder characterisation
techniques and effects of powder characteristics on part properties in
powder-bed fusion processes. Virtual and Physical Prototyping. 2017
Jan 2;12(1):3–29.
[2] Su, D., and W. M. Yan. “3D Characterization of General-Shape Sand
Particles Using Microfocus X-Ray Computed Tomography and Spherical
Harmonic Functions, and Particle Regeneration Using Multivariate
Random Vector.” Powder Technology 323 (January 1, 2018): 8–23.
[3] Bernier, Fabrice, Rui Tahara, and Mathieu Gendron. “Additive
Manufacturing Powder Feedstock Characterization Using X-Ray
Tomography.” Metal Powder Report, February 2018.
[4] Stolfi, A., and L. De Chiffre. “3D Artefact for Concurrent Scale Calibration
in Computed Tomography.” CIRP Annals 65, no. 1 (2016): 499–502.
[5] Müller P, Hiller J, Dai Y, Andreasen JL, Hansen HN, De Chiffre L.
Quantitative analysis of scaling error compensation methods in
dimensional X-ray computed tomography. CIRP Journal of
Manufacturing Science and Technology. 2015 Aug;10:68–76.
[6] Slotwinski, JA, EJ Garboczi, PE Stutzman, CF Ferraris, SS Watson, and
MA Peltz. “Characterization of Metal Powders Used for Additive
Manufacturing.” Journal of Research of the National Institute of
Standards and Technology 119 (September 16, 2014): 460–93.
[7] Tan Y, Kiekens K, Kruth J P, Voet A and Dewulf W. Material dependent
thresholding for dimensional x-ray computed tomography. Int. Symp. on
Digital Industrial Radiology and Computed Tomography (Berlin,
Germany, 2011).
[8] Guo, Rui-Peng, Lei Xu, Bernie Ya-Ping Zong, and Rui Yang.
“Characterization of Prealloyed Ti–6Al–4V Powders from EIGA and
PREP Process and Mechanical Properties of HIPed Powder Compacts.”
Acta Metallurgica Sinica (English Letters) 30, no. 8 (August 2017): 735–
44.
[9] Pavan, Michele, Tom Craeghs, Raf Verhelst, Olivier Ducatteeuw, Jean-
Pierre Kruth, and Wim Dewulf. “CT-Based Quality Control of Laser
Sintering of Polymers.” Case Studies in Nondestructive Testing and
Evaluation 6 (November 2016): 62–68.
[10] Iacocca, R. G., and R. M. German. "A comparison of powder particle
size measuring instruments." International journal of powder metallurgy
33.8 (1997): 35-48.
[11] ASTM B243-17, Standard Terminology of Powder Metallurgy, ASTM
International, West Conshohocken, PA, 2017.
[12] Jamshidinia, M, P Boulware, J Marchal, H Mendoza, L Cronley, S Kelly,
and S Newhouse. “In-Process Monitoring of Cross Contamination in
Laser Powder Bed Fusion (L-PBF) Additive Manufacturing (AM).” In
Solid Freeform Fabrication 2016, 15, 2016.
[13] Brandão, Ana D., Romain Gerard, Johannes Gumpinger, Stefano
Beretta, Advenit Makaya, Laurent Pambaguian, and Tommaso Ghidini.
“Challenges in Additive Manufacturing of Space Parts: Powder
Feedstock Cross-Contamination and Its Impact on End Products.”
Materials 10, no. 5 (May 12, 2017).