ArticlePDF Available

A Framework for Optimizing Process Parameters in Powder Bed Fusion (PBF) Process Using Artificial Neural Network (ANN)

Authors:

Abstract and Figures

Powder bed fusion (PBF) process is a metal additive manufacturing process, which can build parts with any complexity from a wide range of metallic materials. Research in the PBF process predominantly focuses on the impact of a few parameters on the ultimate properties of the printed part. The lack of a systematic approach to optimizing the process parameters for a better performance of given material results in a sub-optimal process. This process needs a comprehensive study of all the influential parameters and their impact on the mechanical and microstructural properties of a fabricated part. Furthermore, there is a need to develop a quantitative system for mapping the material properties and process parameters with the ultimate quality of the fabricated part to achieve improvement in the manufacturing cycle as well as the quality of the final part produced by the PBF process. To address the aforementioned challenges, this research proposes a framework to optimize the process for 316L stainless steel material. This framework characterizes the influence of process parameters on the microstructure and mechanical properties of the fabricated part using a series of experiments. These experiments study the significance of process parameters and their variance as well as study the microstructure and mechanical properties of fabricated parts by conducting tensile, impact, hardness, surface roughness, and densification tests, and ultimately obtain the optimum range of parameters. This would result in a more complete understanding of the correlation between process parameters and part quality. Furthermore, these experiments provide the required data needed to develop an Artificial Neural Network (ANN) model to optimize process parameters (for achieving the desired properties) and estimate fabrication time.
Content may be subject to copyright.
Available online at www.sciencedirect.com
Procedia Manufacturing 00 (2019) 000–000
www.elsevier.com/locate/procedia
47th SME North American Manufacturing Research Conference, NAMRC 47, Pennsylvania, USA
A Framework for Optimizing Process Parameters in Powder Bed Fusion
(PBF) Process Using Artificial Neural Network (ANN)
Mallikharjun Marreya, Ehsan Malekipoura,, Hazim El-Mounayria, Eric J. Faiersonb
aCollaborative Additive Manufacturing Research Initiative (CAMRI), Purdue School of Engineering and Technology, Indianapolis, IN 46202, USA
bQuad City Manufacturing Laboratory-Western Illinois University, Rock Island, IL 61201, USA; efaierson@qcml.org
Abstract
Powder bed fusion (PBF) process is a metal additive manufacturing process, which can build parts with any complexity from a wide range of
metallic materials. Research in the PBF process predominantly focuses on the impact of a few parameters on the ultimate properties of the printed
part. The lack of a systematic approach to optimizing the process parameters for a better performance of given material results in a sub-optimal
process. This process needs a comprehensive study of all the influential parameters and their impact on the mechanical and microstructural
properties of a fabricated part. Furthermore, there is a need to develop a quantitative system for mapping the material properties and process
parameters with the ultimate quality of the fabricated part to achieve improvement in the manufacturing cycle as well as the quality of the final
part produced by the PBF process. To address the aforementioned challenges, this research proposes a framework to optimize the process for 316L
stainless steel material. This framework characterizes the influence of process parameters on the microstructure and mechanical properties of the
fabricated part using a series of experiments. These experiments study the significance of process parameters and their variance as well as study
the microstructure and mechanical properties of fabricated parts by conducting tensile, impact, hardness, surface roughness, and densification
tests, and ultimately obtain the optimum range of parameters. This would result in a more complete understanding of the correlation between
process parameters and part quality. Furthermore, these experiments provide the required data needed to develop an Artificial Neural Network
(ANN) model to optimize process parameters (for achieving the desired properties) and estimate fabrication time.
c
2019 The Authors. Published by Elsevier B.V.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/)
Peer-review under responsibility of the Scientific Committee of NAMRI/SME.
Keywords: Additive manufacturing; selective laser sintering; DMLS; artificial neural network (ANN); optimization framework; parameter optimization;
sensitivity analysis
1. Introduction
PBF process is the most widely used additive manufacturing
technology for metal printing and functional parts [1]. A wide
range of metallic powder can be used as raw material for this
process [2]. As with any other additive technology, PBF fab-
ricates parts directly from 3D CAD data (STL file) and elim-
inates the use of expensive tooling [3, 4]. STL file slices the
overall part into many layers with respect to the layer thickness
and a laser beam sinters/melts each layer. Selective laser melt-
ing (SLM) and selective laser sintering (SLS) are the main two
PBF processes. Unlike the SLM process, where the powder is
completely melted down to form a homogeneous part, the SLS
Corresponding author.
E-mail address: emalekip@purdue.edu (Ehsan Malekipour).
process partially melts the material (sinter the powder) layer-
by-layer at the molecular level [5]. The schematic diagram in
Fig. 1 shows the overall process of the PBF process [6]. The
3D printer machine consists of a supply station for the metal
powder and a sintering/melting unit. A laser selectively sin-
ters/melts the powder with respect to the layer geometry along
a prescribed pattern. After sintering/melting of a layer, the pow-
der dispenser platform moves upward a distance equals to the
thickness of a layer to supply the material required for printing
a new layer and a recoater arm or a roller transfers the material
powder to the sintering/melting zone. The same process contin-
ues until the fabrication of the last layer [7].
Due to the ability of the PBF process to produce homoge-
neous parts with high strength alloys and allowable free-form
fabrication [4], it has found applications in various sectors such
as aerospace, defence, medical etc. [8, 9]. Aerospace industry
widely employs the PBF process because of advantages such
as timesaving and the ability to produce functional assemblies
2351-9789 c
2019 The Authors. Published by Elsevier B.V.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/)
Peer-review under responsibility of the Scientific Committee of NAMRI/SME.
Mallikharjun Marrey /Procedia Manufacturing 00 (2019) 000–000 2
Fig. 1. A schematic diagram of direct metal laser sintering (DMLS) process [6]
[8, 10]. A wide range of metals such as Inconel 625, Inconel
718, 316L stainless steel, cobalt chrome, aluminum, titanium,
and many alloys including Ti-6Al-4V are excellent materials
for the aerospace industry, which are oering a significant cost
and weight reduction [11, 12].
The PBF process has been employed in various industries
however it still suers from some process drawbacks. To over-
come these drawbacks, the research in the PBF process nowa-
days concentrated on the impact detection of few parameters on
the ultimate properties of the printed part [1, 5, 10, 13-18]. The
ultimate goal is to develop a system linking manufacturing pro-
cess, material properties, and the ultimate quality of a fabricated
part to optimize the process parameters. Fulfilling this objective
needs a comprehensive study on all the influential parameters
with their significance on the mechanical and microstructural
properties of a fabricated part. Furthermore, it needs to develop
a quantitative system for mapping material property and process
parameters to achieve improvement in the manufacturing cycle
and quality control of the parts produced by the PBF process.
More than fifty parameters exist and have an influence on
the ultimate quality of the product [19-21]. Scholars classify
the process parameters into dierent groups [20, 22]. In one
approach, Malekipour et al. [21, 22] classified the parameters
into three main categories. The first category is pre-processed
parameters including environmental conditions such as an inert
gas, oxygen level, ambient temperature, powder specifications,
and machine capabilities/limitations. The second category is
the controllable parameters, which include process parameters,
namely, laser specifications and scan strategy, and some few
manufacturing specifications such as layer thickness. The last
category includes the post-processed parameters, which quan-
tify the ultimate quality of the fabricated part such as the yield
strength, fatigue resistance, etc. [22]. Van Elsen named some of
the important parameters in each classification. He mentioned
that the powder specifications and deposition include morphol-
ogy, the surface roughness of the generated grains, particle size
distribution, and the deposition system of powder on to the bed.
The laser specifications include spot size, wavelength, peak
power, mode of the laser, and laser pulse length. The process pa-
rameters include part placement, scan strategy, build direction,
laser power, scan speed, scan strategy, layer thickness, preheat-
ing temperature, hatch distance, and energy density [23].
The aforementioned parameters influence the process and
the fabrication cost [20]. For instance, the process utilizes Ar-
gon instead of Helium as an environment for Ti-6Al-4V be-
cause Helium is 3 to 4 times more expensive than Argon [23].
However, previous literature shows that among all the factors
aecting the PBF part few parameters, namely, laser power,
scan speed, hatch spacing, layer thickness, beam diameter, and
preheating temperature have a tremendous impact on mechani-
cal eciency, economy, and ultimate quality of the entire PBF
process [5, 9, 18, 24].
The objectives of the study are first, to determine the optimal
range of the process parameters and the way that dierent pa-
rameters aect the microstructure, densification, and mechani-
cal properties of the printed part. Second, to specify the sensi-
tivity of dierent parameters and identify which ones aect the
overall performance the most, within the optimal range. Third,
to optimize these parameters to be able to print a part with a
better ultimate quality; and finally, to develop a system for qual-
ity control, and an intelligent network for suggesting optimized
process parameters.
2. The current state of knowledge and gaps
Although PBF technology has significantly developed and
is employed in dierent industries, many challenges and issues
still remain. These challenges hinder the process repeatability,
consistency, and stability of the process. Several research works
have studied the influence of process parameters on quality for
dierent materials and machines; however, it has proven very
dicult to control all aspects of the process or evaluate the col-
lective influence of all the parameters on the properties of a
fabricated part.
Some previous work studied the eect of dierent process
parameters on the ultimate surface quality. Yasa et al. [2] stud-
ied the staircase eect for nickel-based alloy parts manufac-
tured by the SLS process. This research took the total waviness
as the objective function and developed a predictive model. This
model considered a few process parameters to develop it. Re-
lated to this work, Arasu et al. [1] and Hanzl et al. [3] studied
the surface roughness of a part printed by the SLS process and
conducted Analysis of Variance (ANOVA) to obtain the opti-
mal parameter settings to achieve a better final surface finish.
Similarly, Read et al. [25] used the response surface method to
analyze various process parameters statistically and developed
optimal parameters for surface roughness. This method has the
advantage to consider a greater number of process parameters
and conduct statistical analysis with a smaller number of exper-
iments to print. Furthermore, Fox et al. [14] studied the eect of
the process parameters on the surface roughness of overhanging
structures in a powder-bed fusion process. This work covers a
range of overhang angles and process parameters to determine a
Mallikharjun Marrey /Procedia Manufacturing 00 (2019) 000–000 3
relationship between process parameters, the angle of the over-
hanging surface, and the surface roughness.
There is more research, which focused on other aspects of
the process. Sufiiarov et al. [26] studied the eect of the layer
thickness on the parts printed by the SLM process. This study
found that the microstructure, tensile strength, and elongation
at the break depending on the layer thickness. Asgari et al. [9]
used dierent process parameters available in an SLS system,
such as laser power, scan speed, hatch distance, and laser o-
set distance for three dierent AlSi10Mg samples in 200 C
with dierent surface roughness levels. This work employed
Optical Microscopy (OM) and Scanning Electron Microscopy
(SEM) techniques to study the microstructure of the printed
samples. Moreover, Elsen [23] developed a genetic algorithm
to perform variance analysis to study the complexity of the
SLM process for a limited number of process parameters. The
proposed methodology presented the synergistic possibilities of
the mass-spring-damper system to optimize the density of fab-
ricated parts by varying the selected parameters.
Konecna et al. [17] printed Ti-6Al-4V specimens by the
SLM process in dierent orientations to determine the crack
propagation and presented a stress intensity threshold for the
growth of long cracks. In another study, Zhao et al. [27] evalu-
ated the heat transfer and residual stress evolution in the parts
produced by the SLS process and developed a numerical model
by using COSMOL multi-physics environment for Ti6Al-4V.
This study performed a thermo-mechanical simulation to study
the change of residual stresses of a single layer, and physics and
temperature of melt pool, which give a clear understanding of
the thermo-mechanical evolution of a PBF additive process.
Hofland et al. [28] studied the mechanical properties of PA12
parts printed by the SLS process. In this work, they printed 480
tensile samples with 17 dierent sets of the process parameter.
The part properties selected as output are tensile strength, ten-
sile modulus, elongation at the break, and part density. Monte
Carlo performed a simulation to determine the linear correlation
between the coecients and the sensitivities of the process pa-
rameters. This simulation derived some interesting parameters
properties, which influence the ultimate mechanical properties
of the printed parts. Finally, Munguia et al. [29] employed a
neural network-based model for the estimation of the build time
in the SLS process as well as a MATLAB simulation to validate
the results with the existing cost estimation models.
However, these research studies focused on identifying the
influence of few process parameters, predominantly laser spec-
ifications, on the surface quality or selective mechanical proper-
ties of the printed part; limited research works studied the cor-
relation between the parameters and the ultimate properties of
the printed part. However, there is a lack of a consistent sys-
tem considering/optimizing all the controllable parameters and
mapping the process, material, and parameters onto the ultimate
properties of the fabricated parts. Furthermore, Optimizing the
machine setting by controlling the aforementioned parameters
is a prerequisite for a near flawless fabrication process. The ul-
timate contribution of this work is to examine the eect of a
set of parameters instead of the eect of their individual impact
on the selected properties of a fabricated part. This will help
Fig. 2. Proposed framework and corresponding workflow for developing ANN
model
to fill the existing gap for development of a standardized sys-
tem, which maps all the contributing and controllable process
parameters and material specifications onto the ultimate quality
of fabricated parts. Furthermore, manufacturers can employ this
system to ensure full compliance with the customers demands.
This paper first, provides a two-step framework toward this
ultimate objective. Second, it yields the results, which employ
the microstructure and porosity in order to narrow down the
working energy density for near-full densification (see section
4). We will conduct the experiments in the second step within
the optimized range of the laser power and scan speed obtained
in the first step. The latter step, for the first time, will study
the eects of altering laser power, scan speed, hatch space, and
beam diameter all together on ultimate quality of a printed part
and map them together (see Fig. 2). The significance of this
study versus the previous literature is first, we study the ef-
fects of a set of four parameters on the ultimate quality of a
fabricated part (and not just one parameter or two). Second,
we consider the eects of these parameters on the fabrication
time and various aspects of the ultimate quality of a fabricated
part, including microstructure, densification, mechanical prop-
erties, dimensional accuracy, and surface roughness. Finally,
we employ the results to train a feed-forward back-propagation
(FFBP) neural network (NN). The function will suggest the op-
timized parameters according to the desired ultimate quality of
an ordered part.
3. Research methodology
Researchers have been studying the influence of every con-
tributing parameter on the ultimate properties of a fabricated
product. However, a collective system, which considers all the
controllable parameters, is missed. The framework introduced
in this paper will help in achieving a comprehensive under-
standing of the process and the importance of each parameter
as well as obtaining the optimal range of significant parameters
such as laser power, scan speed, hatch spacing, and beam di-
ameter in the manufacturing process. As Fig. 2 shows, first, we
select the parameters according to the prioritized and influential
order cited in the literature [21, 28] and then, this system con-
ducts the Sensitive Analysis (SA) to figure out the importance
Mallikharjun Marrey /Procedia Manufacturing 00 (2019) 000–000 4
of each individual parameters from the selected ones setting up
the levels for next step. Then, this system conducts two sets of
Design of Experiments (DOE) in the next steps and tests the
consequent properties of the fabricated parts. The system will
finally employ the acquired results to train an intelligent sys-
tem suggesting the optimized process parameters for obtaining
a better ultimate quality as well as estimating fabrication time
of the parts printed by the PBF process with the Ti-6Al-4V ma-
terial. The following sections explain the details of the proposed
framework.
3.1. Material properties
316L stainless steel is the material considered for this re-
search. The alloy composition and actual specifications as sup-
plied are shown in Table 1. 316L SS has widespread application
in additive manufacturing due to its good tensile strength at high
temperatures, low stress to rupture, high hardness, toughness
and corrosion resistance properties [30].
3.2. Design of specimen
This work employs the ASTM E8 standard specimen (Ta-
ble 3) [31] to test the tensile properties. We extend the speci-
men length on both sides to perform more mechanical testing,
namely, hardness, and impact as well as to study the microstruc-
ture more. The quality and precision of the specimens are vital
to get some more accurate metallographic analysis.
Fig. 3 shows the designed specimen, which includes three
sections for the tensile testing specimen (the middle section),
the impact test (the left section), and the hardness test (the right
section) of the specimen. All the dimensions are modified ac-
cording to the ASTM standards.
3.3. Sensitivity analysis
Sensitivity analysis (SA) quantifies the correlation between
the given model and its input parameters [32]. The main ob-
jective of conducting SA are to understand (1) which parame-
ters require additional research for strengthening the knowledge
Table 1. Composition of 316L SS [33]
Grade 316L Min Max Actual
Carbon, C - 0.03% 0.019%
Silicon, Si - 0.75% 0.67%
Manganese,Mn 0.03% <0.1% <0.08%
Phosphorus, P - <0.025% <0.019%
Sulphur, S - <0.01% <0.006%
Chromium, Cr 17.5% <18% <17.9%
Nickel, Ni 12.5% <13% <12.7%
Molybdenum, Mo 2.25% <2.5% <2.36%
Nitrogen, N - <0.1% <0.06%
Copper, Cu - <0.5% <0.2%
Oxygen, O - <0.1% <0.022%
Iron, Fe Balance Balance Balance
Table 2. Tensile testing specimen, ASTM E8/E8M 13a [31]
Dimensions for the subsize specimen (6 mm [0.250 in.] wide) (mm [in.])
G Gauge length 25.0 0.1 [1.000 0.003]
W Width 6.0 0.1 [0.250 0.005]
T Thickness Maximum 6 mm
R Radius of fillet, min 6 [0.250]
L Overall length, min 100 [4]
A Length of reduced section, min 32 [1.25]
B Length of grip section, min 30 [1.25]
C- Width of grip section, approximate 10 [0.375]
Fig. 3. Designed specimen
base, thereby reducing output uncertainty; (2) which parame-
ters are irrelevant and can be eliminated from the final model;
(3) which inputs contribute most to output variability; and (4)
which parameters are most highly correlated with the output
[32].
Fig. 4. Schematic for global sensitive analysis [34]
Mallikharjun Marrey /Procedia Manufacturing 00 (2019) 000–000 5
The laser power, scan speed, layer thickness, beam diame-
ter, and the hatch spacing are commonly cited in the literature
as the crucial controllable parameters in the DMLS process in-
fluencing the ultimate quality of the fabricated part [21, 28].
SA considers these five parameters to evaluate their correlation
with the volume based energy density (ED) shown in equation
(1) [35]. However,the employment of SA is crucial to demon-
strate the sensitivity of each parameter within the working range
in this work. The sensitivity analysis guides us through select-
ing the levels of parameters and their distribution for designing
the experiments during the next step.
ED =P
SVt(1)
where P is laser power, S is hatch spacing, V is scan speed,
and t is layer thickness, which is set to a constant value of 30 µm
in this research. Fig. 4 shows the schematic process of global
SA employed by MATLAB [36]. The SA results evidently show
the scan speed as the most sensitive parameter, which dras-
tically changes the energy applied per volume and might in-
fluence the ultimate properties of the fabricated part predomi-
nantly [35]. In a similar way, the laser power and hatch spacing
also have a considerable eect. The eect of layer thickness is
not calculated as it is set to a constant value throughout this
research. Fig. 5 shows the values of the total global sensitivity
coecient obtained by SA.
Fig. 5. Total Global Sensitivity (GS) Coecient
Previous literature also confirmed the significant influence of
laser power and scan speed, as two main parameters that aect
the energy transferred to the powder, on the ultimate quality of
the printed part. D. Gu and Y. Shen [8] conducted experiments
with dierent combination values of laser power and scan speed
on stainless steel 316L with the layer thickness of 20 µm. Their
study shows the parameters generates four dierent melting sta-
tus shown in Fig. 6.
In case I, which is called the no-melting zone, the energy
density is insucient to melt the powder leaving the powder
Fig. 6. The dependency of structure on procedural parameters [8]
in its initial state. In case II, a medium laser power scans the
powder with a low scan speed leading to the partially melted
powder. This phase forms coarsened balls after crystallization,
which is the first form of the balling phenomenon. A high laser
power and high scan speed melt the powder with balling phe-
nomenon along the scanned pattern in the form of thin cylin-
drical lines in case III. Complete melting occurs in the case of
IV as the high laser power forms a solid surface by continuous
lines of fully melted powder along the scan paths.
3.4. Design of experiments
In this research, there are two sets of experiments in order
to obtain the optimized process parameters for a PBF machine
with laser power less than 200 W. The first set of experiments
employs full factorial analysis method due to the limited num-
ber of experiments. Table 3 shows the parameters, namely, the
laser power and scan speed, whose values are assigned based
on the literature [13, 26, 35]. The first set of experiments prints
a 10mm ×10mm ×5mm samples considering merely the laser
power and scan speed, while hatch spacing and beam diameter
are kept fix at their machine default values. The layer thickness
is also set to a constant value of 30µm throughout the work. In
this phase, we study the microstructure, porosity, and densifica-
tion of the printed samples to map them onto the energy density.
Previous literature demonstrated that the porosity generated in
the microstructure of a printed part significantly aects the me-
chanical properties of a fabricated part. The porosity change in
low ranges (near full density parts) is seen to alter mechani-
cal properties substantially [37]. Moreover, reducing the poros-
ity enhances the build consistency [38]. Thus in this phase, we
seek to obtain the optimal range of energy density for maximum
densification. It should be noticed that we study only 13 sam-
ples due to very close energy density for four of the samples.
Taguchi method designs the second set of experiments con-
sidering the optimum range of energy density and the number
of levels obtained from the first set. This phase derives the op-
Mallikharjun Marrey /Procedia Manufacturing 00 (2019) 000–000 6
Table 3. Control factors and levels for DOE
Factor Level values Levels
Laser Power, W 100, 125, 150, 175 4
Scan Speed, mm/sec 600, 800, 1000, 1200 4
timum values for the selected set of process parameters, which
have a decisive impact on the ultimate properties of a fabri-
cated part. To do so, we will conduct 64 experiments (L64 OA),
selected by Taguchi method, and study the eect of the laser
power, scan speed, beam diameter, and hatch spacing on the ul-
timate quality of printed samples (see Fig. 2).. Taguchi method
is a statistical method, which designs experiments using Or-
thogonal Array (OA) technique to improve the quality of a man-
ufacturing process [23]. The OA technique converts the param-
eter design values to S/N ratio and calculates the design robust-
ness [15]. To improve the product quality, the quality charac-
teristics must deviate as little as possible from the target value.
OA is a systematic and statistical way of testing interactions
between control factors. It provides a uniformly distributed set
of experiments, which covers all the paired combinations of the
variables [39] instead of the full factorial analysis (256 num-
ber of experiments in this case), which is unnecessary because
it requires a significant amount of material to and a great deal
of time to fabricate numerous specimens. Fig. 3 shows the de-
signed sample for running the second set of experiments. The
results and data acquired from the second set will eventually be
employed in developing ANN.
Fig. 7. Mechanical properties summary
3.5. Mechanical properties
As Fig. 7 shows, a series of tests measure the mechanical
properties and characteristics of the printed samples acquired
by DOE. Various mechanical tests, namely, tensile, hardness,
and impact investigate the mechanical characteristics of the
fabricated product including surface finish, residual stresses,
porosity, microstructure, and densification. The first set of sam-
ples studies the eect of a limited number of process param-
eters, namely, laser power and scan speed on the microstruc-
ture, porosity and density values for the samples printed by Ti-
6Al-4V, using SEM and Archimedes principle. The second set
studies the aforementioned mechanical properties of the printed
samples to optimize a wider range of process parameters.
ANOVA employs the data acquired from the previous sets of
experiments for data analysis, variance analysis, and ultimately
helps in designing the Neural Network (NN) with the capability
of intelligent suggestion of the process parameters. The next
section provides more details about this approach.
3.6. Data analysis
3.6.1. Signal/noise ratio and analysis of variance (ANOVA)
The signal/noise (S/N) is a method of variability measure-
ment of the manufacturing process, which evaluates the process
parameters at all individual levels and ensures the resulting op-
timum process conditions are robust and stable. The following
equations calculate three types of S/N ratios, namely, the lower-
the-better used for surface roughness (Eq. 2), the higher-the-
better used for mechanical properties (Eq. 3), and the nominal
the better used for dimension accuracy (Eq. 4) [15, 25].
S
N=10 log10
1
n
n
X
i=1
Y2
i
(2)
S
N=10 log10
1
n
n
X
i=1
1
Y2
i
(3)
S
N=10 log10(s2)(4)
where n is the number of measurements and Yiis the observed
performance characteristic value and s is the standard deviation
of the responses for the given factor level combination.
S/N value will be calculated in the second set of experi-
ments (64). With the factor of having 64 experiments, even the
slightest variation/error in employing S/N value can be iden-
tified when the resulting S/N values will be used for calculat-
ing the variance. After the calculation of S/N value, a method
called Analysis of Variance (ANOVA) statistically evaluates the
significance of the control factors (i.e. process parameters in
our work) and their influence on the experimental results (me-
chanical properties). ANOVA studies the variance of properties
with the levels of parameters by employing the data available
after material and mechanical testing [39]. The provided graphs
and distribution charts will describe the variance of properties
within the tested range of levels; thus, they will obtain the opti-
mal range of the values for Ti-6Al-4V in the PBF process. This
will complete the correlation between the material and process-
properties for the PBF process, which will guide in the devel-
opment of an ANN system.
Mallikharjun Marrey /Procedia Manufacturing 00 (2019) 000–000 7
4. Results and discussion
We cut each of the samples in the first set of experiments
from the center in both directions, namely, perpendicular to the
build direction and parallel to the build direction (see Fig. 8) by
using electric discharge machining (WEDM) process.
Fig. 8. Cutting planes in the specimens of the first set [40]
We take sixteen micrographs in total from each sample by
using a scanning electron microscope (SEM). Six micrographs
are taken from the horizontal cross-section (each corner plus
two from the center area) and two from the vertical one (Fig.
9), each uses two dierent magnifications i.e. 60X and 300X
(100µm and 10µm scale respectively (see Fig. 4 for an in-
stance)). We employ the MATLAB image processing to mea-
sure the porosity of each sample in two steps (Fig. 3). First,
MATLAB creates black and white (BW) images from the mi-
crographs. In these images, the black pixels represent the poros-
ity and the white pixels represent the solid. In this step, the
threshold level is adjusted by comparing the pore size in the
SEM image (Fig. 10.a) with the image generated by MATLAB
(Fig. 10.b) to increase the accuracy of the method [41]. Then,
we calculate the ratio of the number of black pixels to the total
pixels for the horizontal BW micrographs and the vertical ones
Fig. 9. The position of the images captured from (a) horizontal (b) vertical
cross-section
separately. The overall average ratio for both magnifications on
the BW images of horizontal cross-section and vertical cross-
sections represent the porosity value of the printed sample in
each direction.
The horizontal cross-section micrographs of the samples
from the first set of experiments show the energy densities ap-
plied to the samples generate three dierent types of porosity
Fig. 10. (a) SEM image and (b) MATLAB image
according to low, medium, or high value of volumetric energy
density (VED). Low volumetric energy density leads to incom-
plete melting of the powder particles and formation of irregular
pores due to lack of fusion such as sample 3 (the low VED with
LP 100 W and SS 900 mm/s) (Fig. 11.a). While, the exertion
of high volumetric energy vaporizes the material and hence, it
leads to the formation of circular gas pores such as sample 13
(the highest VED with LP 175 W and SS 700 mm/s) (Fig. 11.b).
These circles can be a cross-section for a keyhole porosity.
Samples with medium VED such as sample 16 (medium VED
with LP 175 W and SS 1000 mm/s) possess microscale holes
with a nearly uniform distribution throughout the cross-section,
which is an evidence in better mechanical properties compared
with the other types [42]. Table 4 illustrates the porosity val-
ues of the three aforementioned samples. The results are confi-
dently in agreement with the results from the previous literature
Mallikharjun Marrey /Procedia Manufacturing 00 (2019) 000–000 8
Fig. 11. Dierent pores formed during the process (a) Lack-of-fusion pores (b)
Gas pores
Fig. 12. Horizontal vs vertical cross-section SEM images
such as cherry et. Al. [33]. Dierent types of porosity are visi-
ble in horizontal and vertical cross sections, which lead to dif-
ferent porosity percentage in each cross section. Vertical cross
sections illustrate the less frequent but bigger size porosity usu-
ally progressing through layers. Whereas, horizontal cross sec-
tions illustrate the widespread porosity in dierent size ranges,
which scatters throughout the entire section (Fig. 12). The aver-
age porosity will be used in the future phase of this framework.
Fig. 13. Porosity vs VED for first set of experiments
Table 4. Porosity of the samples
Sample no. Horizontal (%) Vertical (%) Average porosity
3, VED =61.7 J/mm35.453 3.5 4.48
16, VED =97.2 J/mm30.8793 0.502 0.69
13, VED =138 J/mm31.8736 1.308 1.59
The porosities in horizontal and vertical cross-sections for
the samples according to the applied VED is shown in Fig.12.
As Fig. 13 shows, the energy density alters between 55 and 138
J/mm3. This energy density creates a part with the density be-
tween 95.52% and 99.31% with a maximum of 99.31%. The
maximum density is achieved with the VED of 99.2 and 104.17
J/mm3. Considering the densification percentage, we can nar-
row down the range of optimum VED to 85 J/mm3and 110
J/mm3. This range of VED suggests the optimized range be-
tween 150 W to 200 W for the laser power and 800 mm/s to
1000 mm/s for scan speed to obtain the maximum densifica-
tion. These ranges will be employed as the inputs for the sec-
ond phase (second set of experiments) in the framework. The
second set will map the set of four process parameters into the
ultimate characteristics of a printed part.
5. Future Work: Neural Network method
There are two main methods for modeling a manufacturing
process. First, a physics-based and second, a data-driven mod-
eling. The physics-based modeling technique analyzes a manu-
facturing process from a physical point of view. However, this
traditional analytical modeling method is not always suitable to
model some modern complex manufacturing processes, such as
AM, due to the number of process variables and the non-linear
complex nature of the process.
Another modeling method is empirical modeling, which em-
ploys experimental data and statistical theory [39]. Many ap-
plications in manufacturing engineering successfully imple-
mented the ANN methodology as a good empirical modeling
method. This work employs the acquired data from the exper-
imental sets, explained in the previous sections, to model the
process by creating a correlation function between the process
Mallikharjun Marrey /Procedia Manufacturing 00 (2019) 000–000 9
parameters and ultimate properties of the fabricated parts. This
function helps in adjusting the process parameters in order to
fabricate parts in accordance with the users desired require-
ments, namely, mechanical properties, microstructure, fabrica-
tion time, dimensional accuracy, and surface roughness. The
aforementioned qualities may have an adversarial nature or ef-
fects on each other. Thus, providing a guideline to control them
through the selection of the process parameters seems crucial.
Numerous literature such as [1-4, 21-25] demonstrates the con-
tributing factors to the ultimate qualities and studied the cor-
relation between them. For instance, Malekipour et al. [22]
classified the process parameters according with their correla-
tion with the defects generated during the powder-bed fusion
(PBF) process. The results demonstrated that laser specifica-
tions, namely, laser power, scan speed, hatch space, and beam
diameter are the most eective parameters contributed to the
defects during the PBF processes. Moreover, this current paper
employs SA to demonstrate the influence of every single param-
eter within a set of parameters on a selected ultimate quality.
However, no literature neither has investigated the influence of
the set of these parameters simultaneously on the ultimate qual-
ities of a fabricated part nor has considered all the aforemen-
tioned ultimate qualities for developing an optimization model.
These two gaps are highly in demand and can be achieved by
the employment of ANN for the PBF process.
This study is employing a feed-forward NN, which adopts
the back-propagation (BP) algorithm to develop a multiple in-
put/output optimization models. The inputs include the process
parameters (laser power, scan speed, hatch space, and beam
diameter) while the outputs include ultimate properties of a
fabricated part (densification, mechanical properties, surface
roughness, dimensional accuracy, and fabrication time) (see
Fig. 14). Development of this multi-input/output ANN function
will eventually lead to an intelligent system capable of control-
ling multiple parameters simultaneously.
Fig. 14. The schematic ANN architecture of this research
The framework presented in this work employs two sets of
experiments to first, narrow down the process parameters (the
inputs) to their optimized ranges and second, study how a set of
parameters contribute to the ultimate properties (the outputs).
This data is essential for building the necessary database to de-
velop this multiple input/output optimization models. Employ-
ment of plentiful data to train an NN function with high accu-
racy is crucial. First, this work considers 16 full factorial ex-
periments to find the optimized range of laser power and scan
speed to fabricate parts with a nearly full density during the first
set of experiments (see subsection 3.4). This set does not use
ANN for optimization, but rather it employs a MATLAB code
to measure the density of the printed samples one-by-one ac-
cording to the images taken by scanning electron microscopes
(SEM). However, the second set is employing ANN to acquire
the optimum ranges of the input parameters. In this case, we
profit from Taguchi method to select and print 64 samples with
a various range of parameters. Furthermore, we are studying
the accuracy of employment of generative adversary network
(GAN) in order to generate a sucient number of inputs for
training the function by ANN. The trained function can be cus-
tomized even further according with the necessities and unique
product of each company.
We will start designing the feed-forward NN with one hid-
den layer and traingdx training algorithm with by using MAT-
LAB deep learning toolbox. The number of iterations is set to
a maximum of 10000, the performance goal is set to 0 accord-
ing with the mean square error (MSE), and the other training
parameters are set to their default values. The best approxima-
tion with the least MSE specifies the best network topology. BP
algorithm calculates and adjusts the weights, which gradually
brings the output closer to the required output.
This system will go towards a comprehensive control sys-
tem in the PBF process within two steps. First, we develop a
NN intelligent system, which makes users/manufacturers capa-
ble of selecting the process parameters according with the re-
quired/desired ultimate properties of a fabricated part (the cur-
rent project) (Fig. 15). Second, we integrate this system with
an online monitoring and control (OMC) system to fabricate
nearly flawless parts with the desired ultimate qualities (the
long-term objective). Plentiful research nowadays has focused
on the development of OMC systems [43-48] to avoid/diminish
the defects and abnormalities generated during the fabrication
process [21, 22, 49-52]. Monitoring and control of the thermal
specifications and thermal evolution of any inherently thermal
AM process has been recognized as a crucial step towards im-
proving the microstructure and ultimate mechanical properties
of a fabricated part [53-55]. Nowadays, most vendors try to han-
dle the frequent thermal anomalies of the fabricated parts such
as distortion by designing some temporary support structures.
These supports facilitate the conduction during the fabrication
process and strengthen the structure.
Designing the topology-optimized support structures re-
duces the fabrication time and material [56] however the fab-
ricated parts still need a significant work for post-processing.
Optimization of process parameters by using an ANN model in
this project integrated with an OMC system can considerably
improve the mechanical properties and surface quality of fabri-
cated parts, increase the repeatability, reduce fabrication time,
and significantly decrease the need for the post-processing op-
erations.
Mallikharjun Marrey /Procedia Manufacturing 00 (2019) 000–000 10
Fig. 15. The algorithm of selection process parameters [57]
6. Conclusion
This study presents a framework for optimization of the PBF
process by mapping the ultimate quality of a fabricated part into
the process parameters and material properties. The first phase
studies the eects of laser power and scan speed on the mi-
crostructure and porosity of SS 316L fabricated parts. To do
so, it studies the energy density between 55 and 138 J/mm3
and demonstrates that the porosity alters between 0.69% to
4.48%. The near-fully densification of 99.31% with a uniform
microstructure is achieved for energy density equals to 97.2
J/mm3. The results suggest that the optimum range of energy
density is between 85 J/mm3 and 110 J/mm3 to achieve maxi-
mum densification. Furthermore, this phase measures the influ-
ence and sensitiveness of laser power, scan speed, hatch spac-
ing, and beam diameter within the optimal range. The results
show that the energy density is sensitive to scan speed more
than other parameters within the optimal range.
The second phase of this framework leverages the acquired
data from the first phase to correlate the parameters with the
mechanical properties, residual stresses, surface roughness, di-
mensional accuracy, and microstructure development. This cor-
relation will help in the development of an intelligent neural
network for parameter suggestion and build-time estimations.
Further steps introduced in the second phase have to be taken
to eventually create a mapping between process, parameters,
and ultimate quality properties of a fabricated part.
7. Acknowledgement
We appreciate CAMRI Research Laboratory that supported
this research. We would also like to show our gratitude to our
colleagues from the Quad City Manufacturing Laboratory, who
provided insight, expertise, and facilities for the experimental
studies that greatly assisted the research, although they may not
agree with all the interpretations/conclusions of this paper. We
also appreciate Nojan Aliahmad for his valuable help in this
work.
References
[1] Arasu, I., et al., Optimization of surface roughness in selective laser sin-
tered stainless steel parts. Vol. 6. 2014. 2993-2999.
[2] Yasa, E., et al., A Study on the Stair Stepping Eect in Direct Metal Laser
Sintering of a Nickel-based Superalloy. Procedia CIRP, 2016. 45: p. 175-
178.
[3] Hanzl, P., et al., The Influence of Processing Parameters on the Mechanical
Properties of SLM Parts. Vol. 100. 2015.
[4] Dingal, S., et al., The application of Taguchis method in the experimental
investigation of the laser sintering process. Vol. 38. 2008. 904-914.
[5] Hardik Varia, B.G., A Review on Eect of Process Parameters on Surface
Quality and Properties of Parts Realized by Selective Laser Sintering Pro-
cess. International Journal on Recent and Innovation Trends in Computing
and Communication (IJRITCC), 2017: p. 45-51.
[6] Custompart.net: ’Direct Metal Laser Sintering’. [cited 2018 June
25]; Available from: http://www.custompartnet.com/wu/direct-metal-laser-
sintering.
[7] Verma, A., P.S. Tyagi, and K. Yang, Modeling and optimization of direct
metal laser sintering process. Vol. 77. 2014.
[8] Gu, D. and Y. Shen, Processing conditions and microstructural features of
porous 316L stainless steel components by DMLS. Applied Surface Sci-
ence, 2008. 255(5, Part 1): p. 1880-1887.
[9] Mohammadi, M. and H. Asgari, Achieving low surface roughness
AlS i10Mg200Cparts using direct metal laser sintering. Additive Manu-
facturing, 2018. 20: p. 23-32.
[10] Calignano, F., et al., Influence of process parameters on surface roughness
of aluminum parts produced by DMLS. The International Journal of Ad-
vanced Manufacturing Technology, 2013. 67(9): p. 2743-2751.
[11] DMLS.com. DMLS applications. [cited 2018 10/01]; Available from:
http://dmls.com/dmls-applications.
[12] Stratasys. Direct metal laser sintering materials. [cited 2018 10/01]; Avail-
able from: https://www.stratasysdirect.com/materials/direct-metal-laser-
sintering.
[13] Gong, H., et al., Defect Morphology of Ti-6Al-4V Parts Fabricated by Se-
lective Laser Melting and Electron Beam Melting. 2013.
[14] Fox, J.C., S.P. Moylan, and B.M. Lane, Eect of Process Parameters on the
Surface Roughness of Overhanging Structures in Laser Powder Bed Fusion
Additive Manufacturing. Procedia CIRP, 2016. 45: p. 131-134.
[15] Kumar, N., H. Kumar, and J.S. Khurmi, Experimental Investigation of Pro-
cess Parameters for Rapid Prototyping Technique (Selective Laser Sinter-
ing) to Enhance the Part Quality of Prototype by Taguchi Method. Procedia
Technology, 2016. 23: p. 352-360.
Mallikharjun Marrey /Procedia Manufacturing 00 (2019) 000–000 11
[16] Delgado, J., J. Ciurana, and C.A. Rodrguez, Influence of process param-
eters on part quality and mechanical properties for DMLS and SLM with
iron-based materials. The International Journal of Advanced Manufactur-
ing Technology, 2012. 60(5): p. 601-610.
[17] Konen, R., et al., Long Fatigue Crack Growth in Ti6Al4V Produced by
Direct Metal Laser Sintering. Vol. 160. 2016. 69-76.
[18] Pal, S., et al., The Eect of Post-processing and Machining Process Pa-
rameters on Properties of Stainless Steel PH1 Product Produced by Direct
Metal Laser Sintering. Vol. 149. 2016. 359-365.
[19] Yadroitsev, I., P. Bertrand, and I. Smurov, Parametric analysis of the se-
lective laser melting process. Applied Surface Science, 2007. 253(19): p.
8064-8069.
[20] Spears, T.G. and S.A. Gold, In-process sensing in selective laser melting
(SLM) additive manufacturing. Integrating Materials and Manufacturing
Innovation, 2016. 5(1): p. 2.
[21] Malekipour, E. and H. El-Mounayri, Defects, Process Parameters and Sig-
natures for Online Monitoring and Control in Powder-Based Additive Man-
ufacturing, in Mechanics of Additive and Advanced Manufacturing, Vol-
ume 9. 2018, Springer. p. 83-90.
[22] Malekipour, E. and H. El-Mounayri, Common defects and contributing
parameters in powder bed fusion AM process and their classification for
online monitoring and control: a review. The International Journal of Ad-
vanced Manufacturing Technology, 2018. 95(1): p. 527-550.
[23] Van Elsen, M., Complexity of selective laser melting : a new optimisation
approach. 2007.
[24] Ruban, W., et al., Eective process parameters in selective laser sintering.
International Journal of Rapid Manufacturing, 2014. 4(2-4): p. 148-164.
[25] Read, N., et al., Selective laser melting of AlSi10Mg alloy: Process opti-
misation and mechanical properties development. Vol. 65. 2015. 417-424.
[26] Sufiiarov, V.S., et al., The Eect of Layer Thickness at Selective Laser
Melting. Procedia Engineering, 2017. 174: p. 126-134.
[27] Zhao, X., et al., Numerical modeling of the thermal behavior and residual
stress in the direct metal laser sintering process of titanium alloy products.
Additive Manufacturing, 2017. 14: p. 126-136.
[28] C. Hofland, E., I. Baran, and D. Wismeijer, Correlation of Process Param-
eters with Mechanical Properties of Laser Sintered PA12 Parts. Vol. 2017.
2017. 1-11.
[29] Mungua, J., J. Ciurana, and C. Riba, Neural-network-based model for
build-time estimation in selective laser sintering. Proceedings of the Institu-
tion of Mechanical Engineers, Part B: Journal of Engineering Manufacture,
2009. 223(8): p. 995-1003.
[30] DebRoy, T., et al., Additive manufacturing of metallic components Process,
structure and properties. Progress in Materials Science, 2018. 92: p. 112-
224.
[31] ASTM, Standard Test Methods for Tension Testing of Metallic Materials,
in E8 /E8M - 13. 2013: ASTM International, West Conshohocken, PA.
[32] Hamby, D.M., A review of techniques for parameter sensitivity analysis of
environmental models. Environmental Monitoring and Assessment, 1994.
32(2): p. 135-154.
[33] Cherry, J.A., et al., Investigation into the eect of process parameters on
microstructural and physical properties of 316L stainless steel parts by se-
lective laser melting. The International Journal of Advanced Manufacturing
Technology, 2015. 76(5): p. 869-879.
[34] Nasir, B., Implementation of Sobols Method of Global Sensitivity Analy-
sis to a Compressor Simulation Model, in 22nd International Compressor
Engineering Conference. 2014, Purdue e-Pubs: Purdue.
[35] Thijs, L., et al., A study of the microstructural evolution during selective
laser melting of Ti6Al4V. Acta Materialia, 2010. 58(9): p. 3303-3312.
[36] Pianosi, F., F. Sarrazin, and T. Wagener, A Matlab toolbox for Global Sen-
sitivity Analysis. Environmental Modelling Software, 2015. 70: p. 80-85.
[37] Gangireddy, S., Gwalani, B., Liu, K., Faierson, E. J., Mishra, R. S. (2019).
Microstructure and mechanical behavior of an additive manufactured (AM)
WE43-Mg alloy. Additive Manufacturing, 26, 53-64.
[38] Palanivel, S., Dutt, A. K., Faierson, E. J., Mishra, R. S. (2016). Spatially
dependent properties in a laser additive manufactured Ti6Al4V component.
Materials Science and Engineering: A, 654, 39-52.
[39] Chen, H., A process modelling and parameters optimization and recom-
mendation system for binder jetting additive manufacturing process, in De-
partment of Mechanical Engineering 2016, McGill University: Montreal
[40] Spierings, A.B., M. Schneider, and R. Eggenberger, Comparison of density
measurement techniques for additive manufactured metallic parts. 2011.
17(5): p. 380-386.
[41] Wol, S. J., Lin, S., Faierson, E. J., Liu, W. K., Wagner, G. J., Cao,
J. (2017). A framework to link localized cooling and properties of di-
rected energy deposition (DED)-processed Ti-6Al-4V. Acta Materialia,
132, 106117.
[42] Yan, F., Xiong, W., Faierson, E. (2017). Grain structure control of addi-
tively manufactured metallic materials. Materials, 10(11), 1260.
[43] Amini, M. and S. Chang, A review of machine learning approaches for
high dimensional process monitoring. 2018.
[44] Amini, M. and S. Chang, Process Monitoring of 3D Metal Printing in In-
dustrial Scale. 2018(51357): p. V001T01A035.
[45] Amini, M. and S.I. Chang, MLCPM: A process monitoring framework for
3D metal printing in industrial scale. Computers Industrial Engineering,
2018. 124: p. 322-330.
[46] Imani, F., et al., Layerwise In-Process Quality Monitoring in Laser Powder
Bed Fusion. 2018.
[47] Yao, B., et al., Multifractal Analysis of Image Profiles for the Characteriza-
tion and Detection of Defects in Additive Manufacturing. Journal of Man-
ufacturing Science and Engineering, 2018. 140(3): p. 031014-031014-13.
[48] Imani, F., et al., Process Mapping and In-Process Monitoring of Porosity
in Laser Powder Bed Fusion Using Layerwise Optical Imaging. Journal of
Manufacturing Science and Engineering, 2018. 140(10): p. 101009101009-
14.
[49] Montazeri, M., et al., In-Process Monitoring of Material CrossContamina-
tion Defects in Laser Powder Bed Fusion. Journal of Manufacturing Sci-
ence and Engineering, 2018. 140(11): p. 111001-111001-19.
[50] Sedaghati, A. and H. Bouzary, A study on the eect of cooling on mi-
crostructure and mechanical properties of friction stir-welded AA5086 alu-
minum butt and lap joints. Proceedings of the Institution of Mechanical
Engineers, Part L: Journal of Materials: Design and Applications, 2017: p.
1464420717726562.
[51] Igbelina, C., The Synergistic Integration of Mass Customization, Paramet-
ric Design and Additive Manufacturing: A Case of Personalized Footwear.
2018 (Doctoral dissertation, The University of Texas at San Antonio).
[52] Taheri, H., Koester, L. W., Bigelow, T. A., Faierson, E. J., Bond, L. J.
(2019). In-situ additive manufacturing process monitoring with an acous-
tic technique: clustering performance evaluation using K-means algorithm.
Journal of Manufacturing Science and Engineering, 1-27.
[53] Attoye, S., E. Malekipour, and H. El-Mounayri, Correlation Between Pro-
cess Parameters and Mechanical Properties in Parts Printed by the Fused
Deposition Modeling Process. in Mechanics of Additive and Advanced
Manufacturing, Volume 8. 2019. Cham: Springer International Publishing.
[54] Malekipour, E., S. Attoye, and H. El-Mounayri, Investigation of Layer
Based Thermal Behavior in Fused Deposition Modeling Process by In-
frared Thermography. Procedia Manufacturing, 2018. 26: p. 1014-1022.
[55] Ebrahimpour, A., et al., Experimental Measurement and Numerical Sim-
ulation of Temperature Distribution and Thermal History During Friction
Stir Welding of Pure Copper Joints. Journal of Mechatronics, 2014. 2(2):
p. 113-118.
[56] Malekipour, E., A. Tovar, and H. El-Mounayri. Heat Conduction and Ge-
ometry Topology Optimization of Support Structure in Laser-Based Addi-
tive Manufacturing. in Mechanics of Additive and Advanced Manufactur-
ing, Volume 9. 2018. Cham: Springer International Publishing.
[57] Yu, Ning. Process parameter optimization for direct metal laser sintering
(DMLS). PhD diss., 2005.
... They developed process maps for 316L and 17-4 PH stainless steel using that model. Marrey et al. (2019), Mehrpouya et al. (2019), and Nikam et al. (2020) developed ANN models to optimize process parameters. Marrey et al. (2019) proposed an ANN framework to find optimum process parameters for desired mechanical properties in LBPF fabricated 316L stainless steel. ...
... Marrey et al. (2019), Mehrpouya et al. (2019), and Nikam et al. (2020) developed ANN models to optimize process parameters. Marrey et al. (2019) proposed an ANN framework to find optimum process parameters for desired mechanical properties in LBPF fabricated 316L stainless steel. Mehrpouya et al. (2019) developed an MLP neural networks-based model to predict optimal process parameters for additively manufactured NiTi shape memory alloys. ...
Article
Full-text available
Additive manufacturing (AM) offers the advantage of producing complex parts more efficiently and in a lesser production cycle time as compared to conventional subtractive manufacturing processes. It also provides higher flexibility for diverse applications by facilitating the use of a variety of materials and different processing technologies. With the exceptional growth of computing capability, researchers are extensively using machine learning (ML) techniques to control the performance of every phase of AM processes, such as design, process parameters modeling, process monitoring and control, quality inspection, and validation. Also, ML methods have made it possible to develop cybermanufacturing for AM systems and thus revolutionized Industry 4.0. This paper presents the state-of-the-art applications of ML in solving numerous problems related to AM processes. We give an overview of the research trends in this domain through a systematic literature review of relevant journal articles and conference papers. We summarize recent development and existing challenges to point out the direction of future research scope. This paper can provide AM researchers and practitioners with the latest information consequential for further development.
... Properly trained ANNs can model the correlations between the given input and output data and accordingly predict the responses based on unseen input values. ANNs have been employed in the SLM process in various ways; some of the applications are the design for AM, including topology and material design [48], in situ process monitoring, including melt pool or powder bed monitoring using optical or acoustic techniques [49][50][51][52], and process-property correlation [53,54]. Recently, the process-property correlation application of the ANN models has been extended to the optimization of this process for different materials. ...
Article
Full-text available
In this study, the influence of selective laser melting (SLM) parameters on the surface characteristics of “single-layer” raster scanning deposits, with Ti6Al4V powder, is thoroughly investigated through experimental, computational, and data-driven approaches. In layer-wise manufacturing processes, such as SLM, the evaluation and understanding of the surface characteristics of individual layers of materials are important; however, there is limited work in the literature on the “single-layer” surface characteristics in SLM. Single-layer samples are designed on a smooth surface on top of semi-cylinders and fabricated using different levels of SLM processing parameters, i.e., laser power, scan speed, and hatch spacing, in different numbers of scan lines, i.e., 5, 10, 20, and 40. The fabricated sample surface data are acquired using non-contact optical interferometry. The computational approach utilizes the discrete element method (DEM) for the powder bed distribution, as well as computational fluid dynamics (CFD) for simulating the laser-powder interaction. One of the main contributions of this study is to investigate and provide predictive models for the correlations between the surface roughness of raster scans and SLM process parameters through experiments and computational work. In addition, a method for comparing the surface characteristics of the thermo-fluid simulation and experimental results is presented. A technique is employed to extract the surface roughness from the simulated results. The discussion on the experiment-simulation comparison challenges provides a deep insight into the criteria selection for experiment-simulation comparison in the SLM process. Moreover, an artificial neural network was trained using the back-propagation method and K-fold cross-validation, as well as regularization techniques to avoid overfitting. The laser power proved to be the most significant parameter, having a contribution of 48.3%, in determining the surface roughness with a unique trend affected by the spattering of powder particles. The simulation results showed noticeable deviations from the experiments in cases with extreme parameters, e.g., too high or too low values of power, speed, or hatch spacing. The machine learning algorithm achieves reasonable predictability of the single-layer surface roughness, showing a root mean square error and a coefficient of determination of 0.9589 µm and 98.78%, respectively, for a separate testing dataset.
... ANNs have been employed in the SLM process in various ways; some of the applications are the design for AM, including topology and material design [21], in-situ process monitoring, including melt pool or powder bed monitoring using optical or acoustic techniques [22][23][24][25], and processproperty correlation [26,27]. Recently, the process-property correlation application of the ANN models has been extended to the optimization of this process for different materials. ...
Article
Full-text available
Selective laser melting (SLM) is a widely used powder-bed fusion additive manufacturing (AM) process for the fabrication of parts from metal powders in a variety of industries such as aerospace, medical, automotive, etc. Despite significant improvements in the design flexibility and mechanical performance, the poor predictability in surface finish, and yet oftentimes with large variability, remains a major challenge in the SLM use. Numerous factors affect the surface roughness of SLM-manufactured parts, which have been reported in the literature, but mostly for bulk samples composed of several layers. In this work, single-layer raster scanning of Ti6Al4V samples are designed and fabricated. The influence of the four most dominant SLM process parameters, i.e., laser power, scanning speed, hatch spacing, and layer thickness on sample surface roughness is thoroughly investigated using a fractional factorial design. Surface roughness data, acquired by white-light interferometry, from 216 data sets are then used to train a machine learning model with the back-propagation method and predict the surface roughness based on the input process parameters. The results show that the laser power is the most significant parameter in determining the top surface roughness of samples. Interestingly, although the investigated samples are single layer raster scanning areas on a solid SLM-built sample with the same parameter set, the layer thickness has a contribution of 10% to 15% in the variations of the surface roughness of the single layers. Furthermore, the machine learning algorithm achieves reasonable predictability, showing a coefficient of determination of 98.8% for a separate 32 testing data set.
... The optimum parameters play a crucial role in enhancing the quality of the building part and its mechanical properties. Most of the researchers utilized different ML techniques to optimize process parameters [44][45][46]. ...
Article
Full-text available
Machine learning (ML) is one of the artificial intelligence tools which uses past data to learn the relationship between input and output and helps to predict future trends. Powder bed fusion additive manufacturing (PBF-AM) is extensively used for a wide range of applications in the industry. The AM process establishment for new material is a crucial task with trial-and-error approaches. In this work, ML techniques have been applied for the prediction of the density of PBF-AM. Density is the most vital property in evaluating the overall quality of the AM building part. The ML techniques, namely, artificial neural network (ANN), K-nearest neighbor (KNN), support vector machine (SVM), and linear regression (LR), are used to develop a model for predicting the density of the stainless steel (SS) 316L build part. These four methods are validated using R-squared values and different error functions to compare the predicted result. The ANN and SVM model performed well with the R-square value of 0.95 and 0.923, respectively, for the density prediction. The ML models would be beneficial for the prediction of the process parameters. Further, the developed ML model would also be helpful for the future application of ML in additive manufacturing .
... Zuowei et al. [17] used ANN in order to predict the geometric deviations for fused deposition modelling, classifying the process images as well. Marrey et al. [18] presented a framework for the PBF process parameters optimization by mapping the ultimate quality of a fabricated part into the process parameters and material properties. The effects of laser power and scan speed on the microstructure and porosity of SS 316L fabricated parts were investigated in their first phase studies. ...
Article
Full-text available
Directed energy deposition (DED) is an additive manufacturing process used in manufacturing free form geometries, repair applications, coating and surface modification, and fabrication of functionally graded materials. It is a process in which focused thermal energy is used to fuse materials by melting. Thermal effects can cause distortions and defects on the parts during the DED process, therefore they should be evaluated and taken into account during the manufacturing of products. Melting pool control and DED bead geometries should be defined properly as well. In this work, an Artificial Neural Network model has been applied considering the DED process parameters in order to predict the geometrical patterns and create a local reinforced product as a hybrid manufacturing technology. Although lots of studies are available on topology optimization for manufacturing methods such as casting, extrusion, and powder bed fusion, topology optimization for the DED process is not widely taken into consideration to predict the design geometrical patterns. DOE RSM and ANN approaches were applied in this study to predict convenient dimensions, topology based geometrical patterns of local stiffeners and heat source power optimizing the energy, total mass, and peak force results of the hybrid part. A single bead track deposition is simulated in terms of validation of the numerical heat source model, and cross-sections of the beads are analysed. A cross-member structure is manufactured using the DED device and the structure is correlated under the three point bending physical conditions on test bench. It has been investigated that locally reinforced cross beam has much more energy absorption and peak force values than plain model. The results showed that the proposed NN-GA is a promising approach to generate the topology based geometrical patterns and process parameters which can be used to create a local reinforced product as hybrid manufacturing technologies.
... The downward-facing surface condition can be optimized by minimizing supports using different methods such as numerical simulations, statistical analyses of LPBF build parameter, [4] topologically optimized supports [212] and artificial neural networks. [213] Contactless supports were explored by researchers via building a thin blade parallel to the down-facing surface. [212,214] The key point behind this design is to facilitate efficient heat transfer from the melt pool to the thin blade through the powder bed. ...
Article
The design freedom offered by additive manufacturing (AM) enables the fabrication of components with internal surfaces that are challenging to access post-manufacture. This is of concern, as the surface condition can markedly deteriorate fatigue performance. Additionally, the adaptation of surface finishing methods for AM components with topologically optimised designs can be a costly practice. It is therefore desirable to consider deploying AM parts with no or minimal surface processing for targeted applications. This requires an in-depth understanding of the formation of various types of AM surfaces, including the variation in surface condition and controlling factors, and their influence on mechanical performance. The last few years have seen significant research advances in these aspects. Ti-6Al-4V is the most extensively studied alloy for AM. The research data available now allows an informative treatment of this topic for both practical applications and future research. Using laser powder bed fusion (LPBF) of Ti-6Al-4V as a model AM−alloy system, this article examines (i) the characteristics of various types of LPBF surfaces including horizontal, vertical, inclined, upward, downward, internal isolated, and slotted surfaces; (ii) the design features and LPBF variables that affect the surface topography; (iii) the capabilities of existing post-AM surface processing methods; and (iv) the influence of AM surface topography on mechanical properties by focusing on the fatigue performance. On this basis, design considerations are recommended for AM of consistent surfaces, and priority surface-related research issues are identified. The purpose is to establish an essential knowledge base for improved commercial designs for LPBF for suitable dynamically loaded applications, with no or minimal surface processing. While centring on LPBF of Ti-6Al-4V, the insights derived are expected to be applicable to other AM processes or metallic materials.
... Direct metal laser sintering process(Marrey 2019). For a color version of this figure, seewww.iste.co.uk/kumar/materials.zip ...
Chapter
This chapter explains an appraisal of the methodologies and suitable materials used in additive manufacturing (AM) for advancing the aerospace industry with the performances of topological and material optimization, design integration and part consolidation that promotes profitable quantity and quality products. It discusses several 3D printing methods and the use of different materials and their applications in the aerospace industry. The prominent process of AM is classified into the following seven types, namely laser beam melting, electron beam melting, selective laser melting, direct metal laser sintering, laser metal fusion, and direct metal deposition. The materials play a predominant role in AM, which are considered for engineered operational applications. AM contributes its foremost capability to serving the aerospace industry, because this industry faced a great challenge to reduce the weight and size of components in order to avoid dangers and promote safety and durability.
Chapter
Additive Manufacturing (AM) of metallic objects is the process of adding layer-upon-layer of material to produce a hard-compact physical object. During the deposition, many process parameters, such as thermal history, power, material feeding rate, microstructure, etc., are involved and need to be regulated. The soar of the 3D printing process in various sectors has recently added a set of challenges, creating a massive demand for an automated process monitor to ensure a good deposition. In this paper, a literature review of vision-based techniques in additive manufacturing process is presented. A molten pool profile extraction methodology is performed. The image processing results can be used to create a visual monitoring technique and inspire future research in the use of vision sensing in 3D printing process.
Article
Additive manufacturing (AM) is rapidly transitioning to an accepted production technology. This transition has led to increasing demands on data analysis and software tools. Advances in data acquisition and analysis are being propelled by an increase in new types of in-situ sensors and ex-situ measurement devices. Measurements taken with these sensors and devices rapidly increasing the volume, variety, and value of AM data but decreasing the veracity of that data simultaneously. The number of new, data-driven software tools capable of analyzing, modeling, simulating, integrating, and managing that data is also increasing; however, the capabilities and accessibility of these tools vary greatly. Issues associated with these software tools are impacting the ability to manage and control AM processes and qualify the resulting parts. This paper investigates and summarizes the available software tools and their capabilities. Findings are then used to help derive a set of functional requirements for tools that are mapped to AM lifecycle activities. The activities include product design, design analysis, process planning, process monitoring, process modeling, process simulation, and production management. AM users can benefit from tools implementing these functional requirements implemented by (1) shortening the lead time of developing these capabilities, (2) adopting emerging, state-of-the-art, AM data and data analytics methods, and (3) enhancing the previously mentioned AM-product-lifecycle activities.
Article
Full-text available
Laser powder bed fusion (LPBF) is receiving widespread attention for its capability to build components with complex geometries. Post-processing can address the adverse effects of various imperfections exhibited in LPBF parts in their as-built state, including inhomogeneous microstructure, tensile residual stresses and poor surface quality. In a recent experimental study, we investigated the influences of different post-processing techniques of heat treatment and shot peening as well as their combination on rotating bending fatigue behavior of V-notched LPBF AlSi10Mg samples. Herein, we further examined those samples regarding the specific parameters that directly influence fatigue performance with the aim to develop a deep learning based approach by means of artificial neural network. The effect of yield stress, ultimate tensile strength, elongation, porosity, microhardness, compressive residual stresses, and surface roughness and morphology were assessed and implemented in the model. Fatigue behavior of the samples was predicted and analyzed using sensitivity and parametric analyses. The obtained results reveal the high potential of deeply learned neural network for unlocking the role of post-processing on the fatigue performance of LPBF AlSi10Mg specimens.
Article
Full-text available
This paper presents the survey of effect of different process parameters on surface quality and properties of parts fabricated by selective laser sintering (SLS) process. SLS is one of the Additive Manufacturing (AM) technique used for 3D printing of metal powders. AM is a layer based material additive process and can produce 3D parts directly from Computer Aided Design .stl data automatically without using any traditional tooling. Additive Manufacturing is an umbrella term given to all technologies that manufacture parts by adding material in layers, as opposed removing material in more traditional subtractive processes like CNC milling, turning, WEDM etc. As one of the advanced Rapid Prototyping and Manufacturing (RPM) processes, the SLS process not only save a lot of inventory but realize the components of almost any complexity in a wide range of metallic materials in quickest possible time. It offers rapid, cost-effective and low-volume manufacturing of physical parts. The various process parameters that affect the quality of final fabricated product are laser power, hatching distance, build orientation, layer thickness, scan speed, laser beam diameter, hatc/h length, scanning pattern and power particles size. From this the effect of various parameters on properties of final part were discussed below.
Conference Paper
Full-text available
Metal 3D printing is one of the fastest growing additive manufacturing (AM) technologies in recent years. Despite the improvements and capabilities, reliable metal printing is still not well understood. One of the barriers of industrialization of metal AM is process monitoring and quality assurance of the printed product. These barriers are especially much highlighted in aerospace and medical device manufacturing industries where the high reliability and quality is needed. Selective Laser Melting (SLM) is one of the main metal 3D printing methods where it is known that more than 50 parameters are affecting the quality of the print. However, the current SLM printing process barely utilizes a fraction of the collected data during production. Up to this point, no study to the best of our knowledge examines the correlation of factors affecting the quality of the print. After reviewing the current state of the art of process monitoring for metal AM involving SLM, we propose a method to control the process of the print in each layer and prevent the defects using data-driven techniques. A numerical study using simulated numbers is provided to demonstrate how the proposed method can be implemented.
Conference Paper
Full-text available
The goal of this work is to understand the effect of process conditions on part porosity in laser powder bed fusion (LPBF) Additive Manufacturing (AM) process, and subsequently, detect the onset of process conditions that lead to porosity from in-process sensor data. In pursuit of this goal, the objectives of this work are two-fold: (1) Quantify the count (number), size and location of pores as a function of three LPBF process parameters, namely, the hatch spacing (H), laser velocity (V), and laser power (P). (2) Monitor and identify process conditions that are liable to cause porosity through analysis of in-process layer-by-layer optical images of the build invoking multifractal and spectral graph theoretic features. This is important because porosity has a significant impact on the functional integrity of LPBF parts, such as fatigue life. Furthermore, linking process conditions to sensor signatures and defects is the first-step towards in-process quality assurance in LPBF. To achieve the first objective, titanium alloy (Ti-6Al-4V) test cylinders of 10 mm diameter × 25 mm height were built under differing H, V, and P settings on a commercial LPBF machine (EOS M280). The effect of these parameters on count, size and location of pores was quantified based on X-ray computed tomography (XCT) images. To achieve the second objective, layerwise optical images of the powder bed were acquired as the parts were being built. Spectral graph theoretic and multifractal features were extracted from the layer-by-layer images for each test part. Subsequently, these features were linked to the process parameters using machine learning approaches. Through these image-based features, process conditions under which the parts were built was identified with the statistical fidelity over 80% (F-score).
Article
Additive manufacturing (AM) is based on layer-by-layer addition of materials. It gives design flexibility and potential to decrease costs and manufacturing lead time. Because the AM process involves incremental deposition of materials, it provides unique opportunities to investigate the material quality as it is deposited. Development of in situ monitoring methodologies is a vital part of the assessment of process performance and understanding of defects formation. In situ process monitoring provides the capability for early detection of process faults and defects. Due to the sensitivity of AM processes to different factors such as laser and material properties, any changes in aspects of the process can potentially have an impact on the part quality. As a result, in-process monitoring of AM is crucial to assure the quality, integrity, and safety of AM parts. There are various sensors and techniques that have been used for in situ process monitoring. In this work, acoustic signatures were used for in situ monitoring of the metal direct energy deposition (DED) AM process operating under different process conditions. Correlations were demonstrated between metrics and various process conditions. Demonstrated correlation between the acoustic signatures and the manufacturing process conditions shows the capability of acoustic technique for in situ monitoring of the additive manufacturing process. To identify the different process conditions, a new approach of K-means statistical clustering algorithm is used for the classification of different process conditions, and quantitative evaluation of the classification performance in terms of cohesion and isolation of the clusters. The identified acoustic signatures, quantitative clustering approach, and the achieved classification efficiency demonstrate potential for use in in situ acoustic monitoring and quality control for the additive manufacturing process.
Article
Magnesium alloys are highly attractive in aerospace and auto industries due to their high strength-to-weight ratio. Additive manufacturing of Mg alloys can further save cost from materials and machining time. This paper investigates the microstructure and dynamic mechanical behavior of WE-43 Mg alloy built through the powder bed fusion process. Samples from four different combinations of processing parameters were built. These builds were studied in both as-built and hot isostatically pressed conditions. The resultant complex microstructures were studied under scanning and transmission electron microscopes while their dynamic mechanical behavior was evaluated using a split-Hopkinson pressure bar testing system. Effects of initial porosity and microstructural evolution during HIP treatment on mechanical response are discussed.
Article
Metal 3D printing is one of the fastest growing additive manufacturing (AM) technologies in recent years. Despite much improvements in its technical capabilities, reliable metal printing is still not well understood. One of the barriers of industrialization of metal AM is process monitoring and quality assurance of the printed product. These barriers are especially much highlighted in aerospace and medical device manufacturing industries where the high reliable and quality products are needed. Selective Laser Melting (SLM) is one of the main metal 3D printing methods where more than 50 parameters may affect the quality of the print. However, current SLM printing processes only utilize a fraction of the collected data for quality related tasks. This study proposes a process monitoring framework named MLCPM (Multi-Layer Classifier for Process Monitoring) to predict the likelihood of successful printing at critical printing stages based on collective data provided by identical 3D printing machines producing the same part. The proposed framework provides a blueprint for control strategies during a printing process and aims to prevent defects using data-driven techniques. A numerical study using simulated data is provided to demonstrate how the proposed method can be implemented.
Article
The goal of this work is to understand the effect of process conditions on part porosity in laser powder bed fusion (LPBF) Additive Manufacturing (AM) process, and subsequently, detect the onset of process conditions that lead to porosity from in-process sensor data. In pursuit of this goal, the objectives of this work are two-fold: (1) Quantify the frequency (number), size and location of pores as a function of three LPBF process parameters, namely, the hatch spacing (H), laser velocity (V), and laser power (P). (2) Monitor and identify process conditions that are liable to cause porosity using in-process optical images of the powder bed invoking multifractal and spectral graph theoretic analysis. To achieve the first objective, titanium alloy (Ti-6Al-4V) test cylinders owere built under differing H, V, and P settings on a commercial LPBF machine (EOS M280). The effect of these parameters on frequency, size and location of pores was quantified based on X-ray computed tomography (XCT) images. To achieve the second objective, layerwise optical images of the powder bed were acquired as the parts were being built. Spectral graph theoretic and multifractal features were extracted from the layer-by-layer images for each test part. Subsequently, these features were linked to the process parameters using machine learning approaches. Through these image-based features process conditions under which the parts were built was identified with the statistical fidelity over 80% (F-score).
Article
The goal of this work is to detect the onset of material cross-contamination in laser powder bed fusion (L-PBF) additive manufacturing (AM) process using data from in situ sensors. Material cross-contamination refers to trace foreign materials that may be introduced in the powder feedstock used in the process due to such reasons as, poor cleaning of the machine after previous builds, or inadequate quality control during production and storage of the powder. Material cross-contamination may lead to deleterious changes in the microstructure of the AM part and consequently affect its functional properties. Accordingly, the objective of this work is to develop and apply a spectral graph theoretic approach to detect the occurrence of material cross-contamination in real-time as the part is being built using in-process sensors. The central hypothesis is that transforming the process signals in the spectral graph domain leads to early and more accurate detection of material cross-contamination in L-PBF compared to the traditional delay-embedded Bon-Jenkins stochastic time series analysis techniques, such as autoregressive (AR) and autoregressive moving average (ARMA) modeling. To test this hypothesis, Inconel alloy 625 (UNS alloy 06625) test parts were made at Edison Welding Institute (EWI) on a custom-built L-PBF apparatus integrated with multiple sensors, including a silicon photodetector (with 300 nm to 1100 nm optical wavelength). During the process two types of foreign contaminant materials, namely, tungsten and aluminum particulates under varying degrees of severity were introduced. To detect cross-contamination in the part, the photodetector sensor signatures were monitored hatch-by-hatch in the form of spectral graph transform coefficients. These spectral graph coefficients are subsequently tracked on a Hotelling T 2 statistical control chart. Instances of Type II statistical error, i.e., probability of failing to detect the onset of material cross-contamination, was verified against X-ray computed tomography (XCT) scans of the part to be within 5% in the case of aluminum contaminant particles. In contrast, traditional stochastic time series modeling approaches, e.g., ARMA had corresponding Type II error exceeding 15%. Furthermore, the computation time for the spectral graph approach was found to be less than one millisecond, compared to nearly 100 milliseconds for the traditional time series models tested.