Content uploaded by Adam Thompson
Author content
All content in this area was uploaded by Adam Thompson on Feb 03, 2021
Content may be subject to copyright.
Joint Special Interest Group meeting between euspen and ASPE
Dimensional Accuracy and Surface Finish in Additive Manufacturing,
KU Leuven, BE, October 2017
www.euspen.eu
Feature-based characterisation of laser powder bed fusion surfaces
Nicola Senin 1,2, Adam Thompson1, Richard K Leach1
1Facul ty of Engineering, University of Nottingh am, UK
2Dipartimento di Ingegneria, Università degli Studi di Perugia, Italy
Abstract
A novel algorithmic pipeline for the automated identification and dimensional/geometric characterisation of topographic formations
of i nte rest (surface features) is proposed, specifically aimed at the investigation of signature features left by laser powder bed fusion
of me tallic su rfaces. Unmelted a nd pa rtially-melted particles, a s well a s s patter formations an d we ld tracks, a re a utomatically
ide ntified and extracted from topography datasets obtained via state -of-the-art areal topography measurement instruments, and
the n characterised in terms of their size and shape properties. Feature -based characterisation approaches, such as the one proposed
in thi s work, allow for development of new solu tions for the study of advanced manufacturing processes through the investigation
of the ir surfa ce fi ngerprin t.
Feature-based topography characte risat ion, lase r powder bed fusion, surface metrology, manufacturing proce ss fingerprin t.
1. Introduction
The i nve stigation of m anufacturing processes throug h the
signature the y leave on the fa bricated surface pla ys an
impo rtant role in process de velo pment and optimisation,
es pecially for those manufacturing technologies that are still at
an ea rly stage of industrialisation, s uch a s ad ditive
ma nu facturing of metals via laser powder bed fusion (LPBF)
[1,2]. Re cent a dvances i n areal top ography me asurement [3]
now allow an unprecedented level of detail in the acquisition of
topogra phic information at mi crometric a nd su b-micrometric
scales. How ever, co nventional top ography da ta a nalysis and
cha ra cterisation meth ods are still strongly ro oted in the
comp utation of a real texture parameters (in particular, the set
of a real pa rameters d efined i n ISO 25178-2), an d th us, are
conceptually oriented towards cap turing the properties of the
en ti re m easured region i nto a s eries of su mmary i ndicators
(texture pa rameters). An opportu nity is , the refore, missed in
ful ly e xploiting the acquired topographic information, pertaining
to i ndi vi dua l topograp hi c fea tures [4].
The focus of this work is o n LPBF of m etals. First attempts at
the i dentification of to pographic featu res in LPBF surfaces are
fou nd in recent works by th e authors [5] a nd elsewhere [6,7]. In
thi s work, a n a pproach is presented for tha t allows for a
comp reh ensive i dentification and ch aracterisation of the most
rel evant sig nature topographic features of LPBF surfaces. An
ori gi nal algorithmic approach to a utomated identification and
cha ra cterisation of the signature features is p roposed, which can
be a pplied to to po graphy datasets n ormally o btainable from
current state- of-the-art topography meas urement instruments.
2. Methods
A Ti 6Al4V sample (40 mm 14 mm 10 m m) fab ricated via
LPBF u s i ng a Renishaw AM250 selective laser melting machine
wi th the manufac build settings is selected
as a test specimen. An Alicona I nfiniteFocus G5 focus variation
(FV) mi cro s cope i s u sed for measurement (20 ma gnification
objective: NA 0.4, s i ngle fi eld of view (FOV) of (0.808
0.808) mm, pixel width (0.439 0.439) µm). Th e top l a yer is
investigated in the a s-b uilt condi tions, i.e . with n o post-
processing, to retain a s ma ny topograp hic forma tions as
possible; us eful to i nves tigate the signature of the p rocess.
Se vera l regions a re mea sured us ing a si ngle FOV wi th no
stitching, located sufficiently far from the s ample b orders to be
considered representative of s teady-state ma nufacturing
process conditions (i.e. avoiding unconventional th ermal e ffects
typical of ed ge regions). The measurement results in a height
ma p of 1840 1840 p oi nts.
The ta rgeted topographic features (attached particles, spatter
formati ons , wel d tracks ) a re su mma ris ed i n fig ure 1.
Figure 1. Main topo graphi c formations visible on the LPBF surface (top
la yer). Focus-stacked image taken with a confocal mi croscope (FOV (1.78
1.78) mm, pixel width (0.6 25 0.625 ) µm).
2.1. Pre-processing of the topography datasets
The fea ture id entification a lgorithmic pi peline con sists of a
fi rst pre-processing step of the entire topography, compri sed of
300 µm
attached par ticles
and spatter formations
Weld tr ack
121
leve lling by le ast-squares mea n pla ne s ub traction, and
rep lacement by weighted interpolation of valid neighbours
of n on -measured p oi nts (voids) a nd localised spike-like
me asurement a rtefacts i dentified via loca l outl ier d etecti on.
2.2. Identification and characterisation of attached particles
and spatter formations
Attached p articles and spa tter fo rma tions are process ed
throu gh the same algorithmic procedures as th ey are both seen
as protruded singularities; the main discriminating factor being
size (the s pa tter formations a re l a rger, resultin g from
coa lescence of mul tiple me lted pa rticles). The s hape o f the
protrude d singularities is approximately spherical, except when
mul tiple particles a re clustered together, so shape/size-related
considerations ca n be u s ed for further dis crimination, once
gene ric pro truded s ingularities have b een isolated. To isolate
the fe atures, the to pography is fi rs t filtered using a high-pass
Ga ussian filter (ISO 16610-61) to remove th e underlying, larger-
scale wa viness, th en h ei ght-based s egmentation is pe rformed
via k-means clustering [8] to obtain a coarse identification of the
features. Shape/size post-processing on the two-dimensional
feature footprint (blob analysis vi a i mage moments) leads to
furth er discrimination between individual particles (unmelted or
pa rtially-melted p a rticles), spatter formations (a pproximately
circular footprint but projected area larger than a single particle)
and particle clusters (large projected area, no n-circular footprint
and lo w is operimetric q uotient). Attributes such a s pa rticle
numerosity, asp ect ratio, size a nd l ocalisation wi thin th e FOV
can be de termined on ce the fe atures ha ve been i ndividually
extra cted .
2.3. Identification and characterisation of weld tracks
Once deprived of particles an d spatter (masked out a s non-
me asured points), the top ograp hy i s fitte d to a smoothed
appro ximation via l ocal non-parametric regression ( Lowess
loca lly weighted scatterplot smo othing [9]). Morp hological
segmentation i nto hills with Wolf and area pruning (ISO 25178-
2 and [10]) is then applied to isolate the weld tracks. Once the
individual tra cks have b een isolated, their shape/size properties
(thi ckness, width and cross-sectional shape) can be investigated
by s l icing each track along their resp ective axes, found via blob
ana l ysi s o f th e weld tra ck footp rint.
3. Results
The i dentification results for the attached particles are shown
in figure 2, as obtained on one of the test datasets . The identified
features are rend ere d in fal se col our.
Figure 2. Results of the algorithmic identificat ion of attached particles
for one of the test data set s.
In fig ure 3, feature post-processing vi a blo b-analysis is shown
as a means to further discriminating the protruded singularities
into individual particles, spatter formations and particle cl usters.
Figure 3. Feature pos t-proce ss ing via blob-ana lysis .
The ide ntification results for th e we ld tra cks a re s hown in
fi gure 4, for the same test dataset shown in figure 2 and figure
3. In figure 5, an extracted weld track is sh own, along with the
res ults of we ld track s licin g a nd l ocal we ld track width
comp uta tion on th e cross -se ctions.
Figure 4. Results of the algorithmic identifi cation of the weld tracks for
the sa me tes t dataset shown i n fi gure 2 and figure 3. Each track is
rende red in different colour.
Figure 5. Extract ion, cross-sectioning a nd l ocal width computation for
one of the identified weld tracks .
4. Conclusions
An a lgo rithmic pipeline has been implemented which allows
for bo th the automated identification an d the dimensional/
geome tric characterisation of localised topograph ic features of
inte rest, starting f rom areal topography datasets. The pipeline
122
ha s been designed to target signature features left by the LPBF
process on the top s urface of metal pa rts, an d a llows the
inves tiga tion o f th e manufacturi ng proce ss fin gerpri nt.
The resu lts promote the a pp roach of fea ture-based
cha ra cterisation a s a vi able a lternative to the s ummary
de s cription of topographic properties via computation of areal
fi eld texture parameters (e.g. rough ness parameters), and
allows a mo re di rect ta rgeting of th e ge ometric and size
prope rties of top ographic fe atures l eft by th e ma nu facturing
proces s under i nves tiga tion .
Acknowledgments
AT and RKL would like to thank EPSRC (Grants EP/ M008983/1
and EP/L01534X/1) and 3TRPD Ltd. for funding this work. NS and
RKL wou ld a lso l ike to tha nk the EC for sup porting this work
throu gh the grant FP7-PEOPLE-MC 624770 METROSURF. The
authors would l ike to tha nk Pro f. Chris Tu ck (Uni ve rsity of
Notti ngha m) for he lpfu l di scuss io ns on th e LPBF proces s.
References
[1] Sames W J, Li st F A, Pannala S, Dehoff R R a nd Babu S S 20 16 The
meta llurgy and processing science o f metal additive
manufacturing. In t. Mat. Rev. 61 315-360
[2] Leach R K 2016 Me trology for additive manufacturing. Meas. +
Control 49 132-135
[3] Leach R K 2011 Op tical measurement of surface to pography
(Springer: Berlin)
[4] Senin N and Blunt L A 2013 Characterisation of i ndividual a real
features. In: Leach R K Characterisation of areal surface texture
(Springer: Heidel berg)
[5] Leach R K 2016 An Introduction to the UK Strategy in Metrology for
Additive Manufac turing Proc. ASPE/Euspen 2016 Summer Topical
Meeting on Dimensional Ac curacy and Surface Finish in Additive
Manufactu ring, Raleigh, NC, June 27-30
[6] Lou S, Sun W, Zeng W, Abdul-Rahman H S, Jiang X and Scott P J
201 7 Extraction of process signature fe atures from additive
manu factured me tal surfaces Proc. Met rology and Properties of
Engineerin g Surfaces, Gothenb urg, J une 27-29
[7] Reese Z, Eva ns C J, Fox J C and Taylor J S 2017 Observations on the
surfa ce morphology of laser powder bed fusion metal surfaces
Proc. Metrology and Properties of Engineering Surfaces,
Gothenb urg, June 27-29
[8] Senin N, Ziliotti M a nd Groppetti R 2007 Three-di mensional surface
topogra phy segmentat ion through clustering Wear 262 395-410
[9] Clevel and W S 1 979 Robust locally weighted regression and
smoothing scatterplots J. Am. Stat. Assoc. 74 829-836
[10] Scott P J 2004 Pattern analysis and ,etrology: the extraction of
sta ble features from observable measurements Proc. R. Soc. L ond.
A 460 2845-2864.
123
