Download as pdf or txt
Download as pdf or txt
You are on page 1of 14

Available online at www.sciencedirect.

com
Available
Available online
online at
at www.sciencedirect.com
www.sciencedirect.com
Available online at www.sciencedirect.com
ScienceDirect
ScienceDirect
ScienceDirect
ScienceDirect
Available online at www.sciencedirect.com
Procedia Manufacturing 00 (2019) 000–000
Procedia
Procedia Manufacturing
Manufacturing 00
00 (2019)
(2019) 000–000 www.elsevier.com/locate/procedia
ScienceDirect 000–000
Procedia Manufacturing 00 (2019) 000–000 www.elsevier.com/locate/procedia
www.elsevier.com/locate/procedia
www.elsevier.com/locate/procedia
Procedia Manufacturing 48 (2020) 358–371

48th SME North American Manufacturing Research Conference, NAMRC 48, Ohio, USA
48th SME North American Manufacturing Research Conference, NAMRC 48, Ohio, USA
48th SME48th
48th SME
North
SME North
North American
American Manufacturing
Manufacturing
American Research
Research Conference,
Research Conference,
Manufacturing NAMRC 48
Conference, NAMRC 48,
48, Ohio,
(Cancelled
NAMRC due toUSA
Ohio, COVID-19)
USA
Effect
Effect of
of Process
Process Parameters
Parameters on
on Weld
Weld Spatter
Spatter in
in Robotic
Robotic Welding
Welding
Effect of Process Parameters on Weld Spatter in
Effect of Process Parameters a,on Weld Spatter in Robotic
Robotic Welding
Welding
Iqbal Shareefa,a,*, Christopher Martinbb
b
Iqbal
Iqbal Shareef
Shareef a,*,
*, Christopher Martin
Iqbal
a Shareef
Professor, *, Christopher
Christopher
Bradley University,
Martin
Martin
Peoria, Illinois 61625,
b
USA
a
b
Mfg. aProfessor, Bradley
Engineering/Launch University,
Manager, Peoria, Illinois
CNH Industrial, 61625,
Goodfield, USA 61742, USA
Illinois
aProfessor, Bradley University, Peoria, Illinois 61625, USA
b
bMfg. Professor, Bradley University,
Engineering/Launch Peoria, Illinois 61625, USA
Engineering/Launch Manager,
bMfg. Manager, CNH
CNH Industrial,
Industrial, Goodfield,
Goodfield, Illinois
Illinois 61742,
61742, USA
USA
Mfg. Engineering/Launch
* Corresponding author. Tel.: 1-309-677-2981; Manager,
fax: +1-309-677-2953. CNH
E-mail Industrial,
address: Goodfield, Illinois 61742, USA
Shareef@bradley.edu
* Corresponding author. Tel.: 1-309-677-2981; fax: +1-309-677-2953. E-mail address: Shareef@bradley.edu
* Corresponding author. Tel.: 1-309-677-2981; fax: +1-309-677-2953. E-mail address: Shareef@bradley.edu
* Corresponding author. Tel.: 1-309-677-2981; fax: +1-309-677-2953. E-mail address: Shareef@bradley.edu
Abstract
Abstract
Abstract
Abstract
This research deals with the investigation of a real-life industrial problem of excessive weld spatter produced in an automated welding robot
This
cell research
Thisusing
research deals
deals with
gas metal with the
the investigation
arc welding investigation of
of aa real-life
process. Spatter real-life industrial
industrial
generation problem
problem to
was attributed of
ofaexcessive
number ofweld
excessive weld spatter
spatter
sources produced
produced
including in
in an
an automated
incorrect automated welding robot
weldingsurface
machine settings, robot
This
cell research
using gas deals with
metal arc the investigation
welding process. of a real-life
Spatter industrial
generation was problem to
attributed ofaexcessive
number weld
of spatter
sources produced
including in an automated
incorrect machine weldingsurface
settings, robot
cell using gas metal
contaminants, arc welding
and erratic process.Random
wire feeding. Spatter generation
variation ofwas attributed
different to a number
welding process of sources including
parameters incorrectwire
such as voltage, machine settings,
feed speed, surface
and torch
cell using gas metal
contaminants, and arc welding
erratic wire process.Random
feeding. Spatter generation
variation was
of attributed
different to a number
welding process of sources including
parameters such as incorrectwire
voltage, machine
feed settings,
speed, surface
travel angle, affected
contaminants, spatter
and erratic production,
wire which served
feeding. Random as a of
variation motivation
different to embarkprocess
welding on a systematic
parametersstudy
suchinvolving design
as voltage, wireoffeed speed, andAtorch
experiments. and total
torch
contaminants,
travel angle, and erratic
affected spatter wire
spatter feeding.
production, Random
which variation
served of different
as aa motivation
motivation welding
to embark
embark process parameters
on aa systematic
systematic such
study as voltage,
involving wire
design of
offeed speed,
experiments. and
Atorch
total
travel
of ten angle, affected
most significant production,
parameters werewhich served
identified as
through to
a brainstorming on
process involving study
roboticinvolving design
cell operators, experiments.
section A
managers, total
and
travel
of angle, significant
affected spatter production, which served as a motivation to embark on a systematic study involving design of experiments. A total
of ten
ten most
subject most
mattersignificant parameters
parameters
experts. Using were identified
were ten
the selected identified through
through
parameters, aa brainstorming
brainstorming
preliminary process
wereinvolving
process
experiments conductedrobotic
involving robotic cell
cell operators,
to determine operators, section managers,
section operating
the appropriate and
managers,range
and
of ten most
subject matter significant
experts. parameters
Using thehas were ten
selected identified through
parameters, a brainstorming
preliminary process
experiments were involving
conducted robotic
toArray cell operators,
determine the section operating
appropriate managers,range
and
subject
for eachmatter
process experts. Using
parameter the
that selected
the mostten significant
parameters, preliminary
impact on weldexperiments
spatter. Anwere conducted
Orthogonal L16 to determine
fromthe appropriate
Taguchi’s operating
designs range
was chosen
subject
for each matter
process experts. Using
parameter the
that selected
has the ten
most parameters,
significant preliminary
impact on experiments
weld spatter. Anwere conducted
Orthogonal L16 to determine
Array from the appropriate
Taguchi’s operating
designs was range
chosen
for
witheach
fourprocess
parametersparameter that has
at 4-levels, fourthe most significant
parameters impact
at 2-levels, andon twoweld spatter. An
parameters wereOrthogonal L16atArray
held constant from
1-level. Taguchi’s
Following thedesigns was chosen
establishment of a
for
with each
four process parameter that has the most significant impact on weld spatter. An Orthogonal L16 Array from Taguchi’s designs was chosen
with
definedfoursetparameters
parameters
of parametersat
at 4-levels,
4-levels, four
four parameters
parameters
and corresponding rangesat
at 2-levels,
of values,and
2-levels, and two
two parameters
a series of designedwere
parameters were held
held constant
experiments wereat
constant 1-level.
1-level. Following
atconducted to measurethe
Following thetheestablishment
establishment of
of aa
effect of process
with
definedfour parameters
set of at
parameters 4-levels,
and four parameters
corresponding rangesat 2-levels,
of values, anda two
seriesparameters
of designed were held
experimentsconstant
were at 1-level.
conducted Following
to measure thetheestablishment
effect of of a
process
parameters
defined set of onparameters
weld spatter andproduction.
corresponding During eachofdesigned
ranges values, a experiment, six different
series of designed responses
experiments werewere recorded
conducted that included,
to measure the of
the effect weight of
process
defined set of
parameters on parameters
weld spatterandproduction.
corresponding ranges
During each ofdesigned
values, a experiment,
series of designed
six experiments
different werewere
responses conducted to measure
recorded that the effect
included, the of process
weight of
parameters
wire used, theon weight
weld spatter
of the production. During each
weld bead deposited, designed
the mass experiment,
of spatter sixindifferent
collected responses
a tray, audio signalswere recorded
generated, thethat included,
spatter thecount
particle weight of
using
parameters
wire used, on
the weld
weight spatter
of the production.
weld bead During
deposited, each
the designed
mass of experiment,
spatter six
collected in different
a tray, responses
audio signalswere recorded
generated, thethat included,
spatter the
particle weight
count of
using
wire used,
ImageJ, andthetheweight
qualityofof
thethe
weld
weld bead
bead.deposited, the mass
The optimal of spatter
process collected
parameters in a tray,
identified audio
in this signals
study weregenerated,
of practicalthesignificance,
spatter particle
andcount using
as such the
wire used,
ImageJ, and thetheweight
qualityofof
thethe
weld
weld bead
bead.deposited,
The the mass
optimal of spatter
process collected
parameters in a tray,
identified in audio
this signals
study were generated,
of practicalthesignificance,
spatter particle
and count
as using
such the
ImageJ, and theparameters
recommended quality of were
the weld bead. Theinoptimal
implemented process parameters identified in this study were of practical significance, and as such the
the industry.
ImageJ, and the quality of were
recommended the weld bead. Theinoptimal process parameters identified in this study were of practical significance, and as such the
recommended parameters
parameters were implemented
implemented in the the industry.
industry.
recommended
©
© 2019
2020The The parameters
Authors,
TheAuthors,
Authors. were
Published
Publishedimplemented
by Elsevier in
B.V.
by Elsevier the
B.V.industry.
©
© 2019
2019 Theopen
Authors,thePublished
Published by Elsevier B.V.
Peer
This
© 2019
Peer
review
is an under
The under
review access
Authors,the article by
responsibility
Published by
responsibility
Elsevier
underof the
Elsevier
of the
B.V.BY-NC-ND
thescientific
CC committee of NAMRI/SME
B.V. committeelicense
scientific of (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Peer review under
Peer-review under theresponsibility
responsibility of of the
thescientific
Scientific committee
Committee of NAMRI/SME
NAMRI/SME
of the NAMRI/SME.
Peer review under the responsibility of the scientific committee of NAMRI/SME
Keywords: Weld Spatter; Weld Parameers; Gas Metal Arc Welding; Taguchi Orthogonal Array, Weld Accoustic Signals; Weld Optics
Keywords:
Keywords: Weld
Weld Spatter;
Spatter; Weld
Weld Parameers;
Parameers; Gas
Gas Metal
Metal Arc
Arc Welding;
Welding; Taguchi
Taguchi Orthogonal
Orthogonal Array,
Array, Weld
Weld Accoustic
Accoustic Signals;
Signals; Weld
Weld Optics
Optics
Keywords: Weld Spatter; Weld Parameers; Gas Metal Arc Welding; Taguchi Orthogonal Array, Weld Accoustic Signals; Weld Optics
With increased welding speed, it is essential to control
1. Introduction With
With increased
increased welding speed, it is
is essential
essential[10].
to control
control
1. Introduction voltage,
With shorten welding
increased the arc, speed,
welding and reduce
speed, it
it is spatter to
essential[10]. Spatter
to control
1.
1. Introduction
Introduction voltage, shorten the arc, and reduce spatter Spatter
voltage,
severity
voltage, shorten
is dependent
shorten the arc,
on
the arc, and
the reduce
composition
andcomposition spatter
reduce spatter of [10].
the Spatter
electrodes
[10]. Spatter
Spatter is defined as droplets of molten material that are severity is dependent on the of the electrodes
Spatter
Spatter is
is defined
defined as
as droplets of molten material that are severity
[11], is dependent
andis the on the
type ofoningredientscomposition of
used inof the the electrodes
the flux [12].
formed
Spatter
formed
near
near is the
defined
the as droplets
welding
welding
arc during
droplets
arc during
of molten
of the welding
molten
the welding
material
material that
process,
process,
are
that and
are
and
severity
[11],
[11],
Welding and
and
dependent
the
the
polarity type
typealsoof
of
the composition
ingredients
ingredients
has used
an effectused on in in the
the
weld
electrodes
flux
flux
spatter [12].
[12].
and
formed
ejected
formed near
from
near the
the welding
weld
welding arc
puddle.
arc during
The
during the
ejection
the welding
of
welding process,
molten and
material
process, and [11],
Welding and the
polarity typealsoof ingredients
has an effect used
on in the
weld flux
spatter [12].
and
ejected from the weld puddle. The ejection Welding polarity also has an effect on weld spatter and
ejected
has from
from the
numerous
ejected weld
causes.
the weld A puddle. The
prominent
puddle. cause ofof
The ejection
ejection of molten
molten
ofejection
moltenis
material
material
the arc
material
welding
Welding
welding
arc
arc
stability,
polarity
stability,
when
alsowhen
has an direct current
effect
direct
electrode
on electrode
current weld spatter positive
and
positive
has
has numerous
numerous causes.
causes. A
A prominent
prominent cause
cause of
of ejection
ejection is
is the
the arc welding
welding arc
(DCEP) andstability,
direct whencurrentdirect current
electrode electrode
current negative positive
electrode (DCEN)
failing to
has numerous start [1]. In
causes. this case,
A prominent the electrode comes
cause of ejection in the arc
contact
is contact
arc (DCEP)
(DCEP)
arc
and
and
stability,
direct when
current direct
electrode negative positive
(DCEN)
failing
failing
with
failing
to
to
the start
start [1].
[1].
workpiece,
to workpiece,
In
In this
this
and
start [1]. In and
case,
case,
joule
this joule
the
the
case, the
electrode
electrode
heating triggers
electrode
comes
comes in
in contact
disintegration
comes in contact
welding
welding and direct
(DCEP) operations
operationsdirectare
current
current electrode
are performed
performed
[13].
electrode
[13].
negative
negative (DCEN)
(DCEN)
with
with the heating triggers disintegration welding
CO operations
gas bubbles are performed
that explode [13].
during the welding process
of
withthethe
the workpiece,
wire. The wireand
workpiece, and joule
piece heating
jouleejects away
heating triggers
from the
triggers disintegration
tip of the
disintegration welding
CO
2 operations are performed [13].
2 gas bubbles that explode during the welding process
of
of the
the wire.
wire. The
The wire
wire piece
piece ejects
ejects away
away from the tip of the canCO 2 gas
also be bubbles
the cause that explode
thatof excess during
spatterthe welding
[14-15]. process
Important
electrode
of the wire.
electrode
in
in
the form
Theform
the
of spatter
wireof piece
spatter
[2].
ejects
[2]. away from the tip
from the tip of
of the
the can
can
CO 2 gas
also
also
parameters be
be
bubbles
the
the
that cause
cause
need
explode
toof
ofbeexcess
excess
during
spatter
spatter
controlled
the welding
in [14-15].
process
the gas Important
[14-15]. Important
metal arc
electrode
electrode in
Spatterinisthethe form
often of
formaa of spatter
result [2].
of [2].
spatter parameters that have not been can also
parameters be the
that cause
need toofbe excess spatter
controlled in [14-15].
the gas Important
metal arc
Spatter
Spatter is
is often result of parameters that have not been parameters
welding that
that need
(GMAW) to
to be
be controlled
needprocess, are welding in
in the gas
gas metal
current, arc
welding
properly
properly is often
often aa result
Spatter optimized
optimized
[3-4]. of
result
[3-4]. ofForparameters
example,
parameters
For example,
that
thatif
if
have
there
have
there
not been
not is
beenaa
is
parameters
welding
welding
travel (GMAW)
(GMAW)
speed, arc process,
process,
voltage, type
controlled
are
are
of welding
welding
shielding
the
current,
current,
gas, wire
metal arc
welding
welding
feed rate,
properly
momentary optimized
short [3-4].
circuit duringFor the example,
process if
or there
voltage is
there isisnot
is aa welding
travel (GMAW)
speed, arc process,
voltage, are welding current, welding
properly optimized
momentary
momentary
tuned properly,short
short
[3-4].
circuit
circuit
spatter during
during
occurs
Fortheexample,
the
when process
process
ifvoltage
or
or
the breakage voltage is
in current not
not travel
and speed,
torch
travel travel
speed, arc
arcand work type
voltage,
voltage, type
angles
type
of
of shielding
shielding
shielding gas,
of [16-17].
gas, wire
gas, wire
feed rate,
wire feed
feed rate,
rate,
momentary
tuned properly,short circuit
spatter during the process or voltage is not and
and torch
torch travel
travel and
and work
work angles
angles [16-17].
[16-17].
tuned
occurs properly, spatter
[5]. Spatter is occurs
occurs
generallywhen
when the
the breakage
the breakage
whenknown
in
in current
in current
to increase with The
andThe
torchgoal of
travel
goal of
this study
and study
this
is
work angles
is
to
to
optimize
[16-17]. these parameters and
optimize
these parameters
tuned
occurs
occurs
increase
properly,
[5].
[5].
in
spatter
Spatter
Spatter
voltage is
is
[6-7].
occurs
generally
generally
Welding known
known
power
breakage
to
to increase
increase
devices have
current
with
with
also The
The goal
eliminate of
of this
this study
or minimize
goal weld
study is to
to optimize
is spatter. In thethese
optimize GMAW
these processand
parameters
parameters and
an
and
occurs [5].
increase in Spatter [6-7].
voltage is generally
Welding known
power to increase
devices have with
also eliminate
eliminate
electrical or
or
arc minimize
minimize
is weld
weld
established spatter.
spatter.
between In
In athe
the GMAW
GMAW
base process
process
material and an
an
increase
been
increase in
improved voltage
to
in voltage be[6-7]. Welding
digitally
[6-7]. power
controlled
Welding power so devices
that
devices morehave
have also
precise
also eliminate
electrical or
arc minimize
is weld spatter.
established between In athebase
GMAW process
material and anaa
been
been
controlimproved
improved
of weldingto
to be
be digitally
digitally
current and controlled
controlled
voltage so
so that
that
waveform more
more
can precise
precise
be used electrical
continuously
electrical arc
arc is
fed
is established
consumable
established between
electrode.
between a
a base
The
base material
electrode
material and
is
and the aa
been improved
control of weldingto be digitally
current and controlled
voltage so that more
waveform can precise
be used continuously
continuously
filler metal, andfed
fed isconsumable
consumable
protected electrode.
electrode.
from atmosphericThe
The electrode
electrode
conditions is
is the
the
control
control of
of welding
to stabilize welding current
the arc and and
reduce
current voltage
and the waveform
occurrence
voltage waveform can
can be
of spatter used
be[8-9].
used continuously
filler metal, andfedisconsumable
protected electrode.
from atmosphericThe electrode
conditions isbytheaa
by
to
to stabilize
stabilize the
the arc
arc and
and reduce
reduce the
the occurrence
occurrence of
of spatter
spatter [8-9].
[8-9]. filler metal,
shielding and
gas
filler metal,gas is
that
andthat protected
ensures
is protected from
the atmospheric
integrity ofconditions
the weld by
byis a
to stabilize the arc and reduce the occurrence of spatter [8-9]. shielding
shielding ensuresfrom the atmospheric
integrity of conditions
the weld isa
2351-9789 © 2020 The Authors. Published by Elsevier B.V. shielding gas gas thatthat ensures
ensures the the integrity
integrity of of the
the weld
weld is is
2351-9789 © 2019 The Authors, Published by Elsevier B.V.
This is an open
2351-9789 access article under by the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Peer review©
2351-9789 2019
©under
2019 The
the Authors,
Authors, Published
Theresponsibility by Elsevier
of the scientific
Published B.V.
committee
Elsevier of NAMRI/SME
B.V.Committee
Peer-review
2351-9789
Peer review under
©under
2019 responsibility
Theresponsibility
the of scientific
Authors, Published
of the the Scientific
by Elsevier B.V. of NAMRI/SME
committee of the NAMRI/SME.
Peer review under the responsibility of the scientific committee of NAMRI/SME
10.1016/j.promfg.2020.05.058
Peer review under the responsibility of the scientific committee of NAMRI/SME
Iqbal Shareef et al. / Procedia Manufacturing 48 (2020) 358–371 359
2 Iqbal Shareef and Christopher Martin/ Procedia Manufacturing 00 (2019) 000–000

maintained. The arc formed between the consumable wire and Nomenclature
workpiece heats the workpiece metal resulting in melting and
joining. The process input parameters must be adjusted to 𝜌𝜌" Density of the weld wire (lb/in3).
minimize noise deviations in output, while the robustness and 𝑑𝑑" Diameter of the weld wire consumed in each test (inch)
parameter fluctuations are maximized [18-19]. Some 𝑓𝑓" Weld wire feed rate used in the test (inch/min)
parameters are programmable including, voltage, and wire 𝐼𝐼& Intensity of sound during welding (dB)
feed rate, weld arm traveling speed, whereas others like 𝑁𝑁&( Number of weld spatter particles (#)
surface condition, welding process type and wire stick out are 𝑡𝑡" Welding time for each test (sec)
not programmable. 𝑉𝑉" Volume of the weld wire consumed in a test (in3)
One reason improperly set machines can produce excess 𝑊𝑊,- Exact weight of metal deposited (lb)
spatter is that the welding process uses either a globular 𝑊𝑊. Quality of the weld produced in test (#)
transfer or a short circuit transfer. A globular transfer is when 𝑊𝑊& Weight of the spatter collected (lb)
the weld metal travels across the arc in globs or large droplets. 𝑊𝑊/&0 Weight of the test specimen before spatter test
A short circuit transfer is when the wire comes in contact with 𝑊𝑊/&1 Weight of the test specimen after spatter test
metal which creates a short circuit prior to detaching from the 𝑊𝑊" Input weight of the weld wire (lb)
electrode. Both interrupt the molten weld puddle production 𝑊𝑊"2 Weight of the wire lost due to spatter and vaporization (lb)
and cause some of the puddle to leave the weld puddle in the 𝑊𝑊"2% Percent loss of weld wire during welding (%)
form of spatter. A way to reduce spatter is to use a spray
transfer. Spray transfer uses a transfer method similar to the
spray paint process. It deposits metal into the weld puddle 2. Test Materials, Input Variables, and DOE
from an arc length. In spray transfer the wire electrode does
not come into contact with the weld puddle. Another way to 2.1. Materials
reduce spatter is to use pulse gas metal arc (MIG) welding, Test specimen material: The test pieces used were cut
which generally produces less spatter than standard constant from ¼” thick ASTM A572 Grade 50 structural steel that is
voltage MIG welding [20-21]. Both constant voltage and pulse most commonly used in the industry, where these tests were
MIG welding processes were used for testing. conducted. As such, ASTM A572 Grade 50 became the
Investigation into some of these methods to reduce weld obvious choice for selection of this material to be used for
spatter can be done through experimentation, which can be spatter reduction tests. The chemical composition of the test
optimized using orthogonal arrays based on a technique specimen material is given in Table 1.
developed by Dr. Genichi Taguchi. Use of orthogonal arrays
Table 1. Chemical composition of test specimens
reduces the number of experiments and serves as an excellent
tool for screening a large set of parameters and identifying the Carbon Iron Manganese Phosphorous Silicon Sulfur
C Fe Mn P Si S
most significant parameters that affects the process
performance [22]. Furthermore, Taguchi technique involves 0.21% 98% 1.35% 0.03% 0.35 0.05%
reducing the variation of a process through a robust design of
experiments (DOE). The objective of the technique is to Weld wire material: The weld wire used was Lincoln
produce high-quality products at a lower cost to the ED032863 0.045” (1.1 mm) SuperArc L-59 MIG Wire. This
manufacturer [23]. wire has an AWS classification of ER70S-6, which conforms
A cell phone camera can be used to take high speed video to the AWS A5.18/A5.18M:2017 specification for carbon
of each experiment. The videos can then be used to analyze steel electrodes and rods for gas shielded arc welding standard
the amount and pattern of spatter emerging from the weld by the American Welding Society. The chemical composition
[24]. The videos can also be used to analyze the weld puddle of the weld wire is given in Table 2.
to see how much it is affected by the wire electrode, if at all. Table 2. Chemical Composition of weld wire
Still frame captures of each video can also be used to quantify Element Weight %
spatter by counting the number of particles coming off the Carbon C 0.06 - 0.15
weld [25]. Manganese Mn 1.40 - 1.85
ImageJ is an open sources image processing program that Silicon Si 0.80 - 1.15
can be used to count those particles. These spatter particles Phosphorous P 0.025 max
can be quantified by converting the images to black and white Sulphur S 0.035 max
and filtering bit sizes and thresholds within the picture to Nickel Ni 0.15 max
Chromium Cr 0.15 max
prevent the weld arc flash from skewing the data. Particle Molybdenum Mo 0.15 max
counts will be consistent so long as the still frame images are Vanadium V 0.03 max
taken at the same intervals. Copper Cu 0.50 max
The audio coming out of the welding process can also be
analyzed to optimize the welding machine parameters. The The weld wire’s salient features include an engineered
welding spatter loss is proportional to the arc sound energy alloy system which enhances silicon island management and
produced [26]. A poorly optimized machine will sound facilitates minimal spatter. It is coated with copper for long
choppy and uneven, whereas a well optimized machine will contact tip life, increased electrical conduction, and offers fast
sound like a constant hiss or a buzz [27]. travel speeds. MicroGuard Ultra coating provides superior
360 Iqbal Shareef et al. / Procedia Manufacturing 48 (2020) 358–371
Iqbal Shareef and Christopher Martin/ Procedia Manufacturing 00 (2019) 000–000 3

feeding and arc stability. Type ER70S-6 is a wire with higher while keeping the rest constant was ruled out in the interest of
levels of deoxidizers and is suitable for welding of steels with optimizing time and resources needed to conduction the
moderate amounts of scale and/or rust. For MIG welding, the experiment. Consequently, alternative design of experiments
wire manufacturers recommendation is to use + CO2 or Argon was explored. With four parameters at 4-levels, four
+ CO2 or Argon + 2% Oxygen as shielding gases. In this additional parameters at 2-levels, and last two parameters at
study 98% Argon and 2% CO2 mix was used for shielding. 1-level listed in Table 3, the most feasible and appropriate
design suitable for determining the effect of 10 parameters on
2.2 Test Variables - Input Parameters
weld spatter was determined to be Taguchi’s L16 orthogonal
The goal of the experimental work was to conduct a series array. Table 4 lists Taguchi’s L16 orthogonal array that was
of well-designed experiments incorporating key process run in randomized order to eliminate any biases that may arise
parameters that affect the weld spatter. This was in line with in the experiments. Unlike in the MIG constant voltage (CV)
the overall goal to reduce weld spatter without sacrificing the weld process, where the voltage is held constant, in Pulse
quality and cost aspects of production. To identify the most welding process when the trim is decreased from 1 to 0.85 the
significant parameters that influence production of weld weld arc length reduces and in doing so reduces the welding
spatter, a well-diversified team of experts with knowledge, voltage. Similarly, when the trim is increased from 1 to 1.15,
experience, and economic stake in weld spatter reduction was the weld arc length increases and therefore increases welding
formed. The team consisted of robotic cell operators, sections voltage. Accordingly, in Table 4 two numbers in the column
managers, and subject matter experts both from industry as for Voltage/Trim are for Pulse welding process and just one
well as from academia. With the team in place the first number for the CV process.
brainstorming session was held only to collect as many ideas
as possible. During the second brainstorming session all ideas Table 4. Design matrix using L16 orthogonal array
collected and the parameters listed were shared with the team Weld For Constant Weld
Electrical Process Size (6mm x 6mm)
and the team was asked to identify the most significant Test Wire Stick Surface (MIG Weld Wire Arm
Voltage Torch
parameters affecting weld spatter in each members’ view. The /Trim Travel
No. Out Condition Axial Direction Feed Travel
(Volts) Angle
ideas generated in the 2nd brainstorming session were (inches) Spray Speed Speed
Transfer) (in/min), (in/min)
prioritized in the order of importance assigned by the team
1 0.5 Clean CV Push 325 14.7 24 10
members and the team was asked to finalize the number of 2 0.5 Clean Pulse Drag 375 17.0 26/1.06, 10
parameters to be tested. A final brainstorming session was 3 1 Oily CV Push 425 19.3 28 10
held that resulted in 10 input parameters for testing. These 4 1 Oily Pulse Drag 475 21.5 30/1.21 10
5 1 Oily Pulse Push 325 14.7 26/1.07 15
included: i) Gas Mixture, ii) Grounding Wire Length, iii) 6 1 Oily CV Drag 375 17.0 24 15
Electrical Wire Stick-out, iv) Surface Condition, v) Weld 7 0.5 Clean Pulse Push 425 19.3 30/1.22 15
Process Type, vi) Weld Direction, vii) Wire Feed Speed, viii) 8 0.5 Clean CV Drag 475 21.5 28 15
9 0.5 Oily Pulse Drag 325 14.7 28/1.16 20
Arm Travel Speed, ix) Voltage, and x) Torch Travel Angle. 10 0.5 Oily CV Push 375 17.0 30 20
Once the 10 most significant parameters to be tested were 11 1 Clean Pulse Drag 425 19.3 24/0.95 20
identified, each parameter was further examined from a 12 1 Clean CV Push 475 21.5 26 20
practical feasibility point of view and the number of levels 13 1 Clean CV Drag 325 14.7 30 25
14 1 Clean Pulse Push 375 17.0 28/1.15 25
needed to investigate each parameter was selected through 15 0.5 Oily CV Drag 425 19.3 26 25
team consensus, keeping in mind the best approach to study 16 0.5 Oily Pulse Push 475 21.5 24/0.93 25
weld spatter. Even after the selection of input parameters and
the corresponding number of levels to be tested for each 3. Experimental Techniques
parameter was finalized, there were three additional revisions
made before the very final design that was actually 3.1 Design and fabrication of test samples
implemented. The ten input parameters and the corresponding
The test samples were made from ¼” thick ASTM A572
levels of investigation are listed in Table 3.
Grade 50 steel. Two plates of dimensions 10”x3”x1/4” and
Table 3. Input parameters for weld spatter tests 10”x2”x1/4” were tag welded at the ends to form a “T”
Grounding Electrical
Weld
Process
For Constant Weld
Size (6mm x 6mm)
junction as shown in the schematic in Figure 1. A total of 80
Gas Wire Wire Stick Surface (MIG Weld Wire Arm Voltage
Torch
Travel
such test samples were made for weld spatter testing to cover
Mixture Length
(meters)
Out
(inches)
Condition Axial
Spray
Direction Feed
Speed
Travel (Volts)
Speed
Angle 64 test pieces for DOE, and the rest for use in preliminary trial
Transfer) (in/min), (in/min) experiments.
92% Ar, Constant
3.5 0.5 + delta Clean Push 325 14.7 24 10
8% CO2 Voltage
Oily wiped
1.0+delta Pulse Drag 375 17.0 26 15
with cloth
425 19.3 28 20
475 21.5 30 25

2.3 Taguchi L16 orthogonal design


Having arrived at ten different input variables with each
designated to be tested a specific level, the investigation team
carefully examined various test options. After initial review of
the experiment the option of testing one variable at a time Fig. 1. Weld spatter test sample geometry
4 Iqbal Shareef and Christopher Martin/etProcedia
Iqbal Shareef Manufacturing
al. / Procedia 00 (2019)48
Manufacturing 000–000
(2020) 358–371 361

3.2 Design and fabrication of spatter collection pans 3.3 Tools & Software
Welding Machine: Lincoln Electric Power Wave 455M
Initially three spatter collection trays were designed and
welding machines were used during sample preparation and
fabricated from a 1/16” sheet metal. A ¼” diameter steel rod
test parameter validation. The Power Wave® platform is
bent in the shape of a shallow U, and welded at the ends of the
designed around three principles of process control, which
trays, served as handles for easy placement and handling of
allows for optimized performance and reduced waste:
spatter collected in the trays. During preliminary trial
Implement – Optimized arc performance with Waveform
experiments spatter was collected by placing three trays
Control Technology®, Control – Process standardization with
directly under the welding fixture as shown in Figure 2.
Power Wave® Manager and Weld Sequencer® Software, and
Verify – Productivity and quality monitoring with
CheckPoint® Production Monitoring.
Welding Robot: A FANUC ARC Mate 100iC Industrial
Robot was used in this study. The FANUC ARC Mate Series
that are built specifically for Arc welding includes a variety of
models with payloads of up to 20 kg. It has an arm reach of up
to 2.0m, is suitable for a wide range of arc welding, laser
welding, soldering and cutting applications. The robot used
had a variety of customized tools to increase its versatility,
and to assist in making the weld processes more efficient and
profitable.
Robot Programs: In this study, FANUC’s specialized
robot software products were used that included dedicated
Fig. 2. Three spatter collection trays placed below the weld location functions, simple to use interfaces, and exclusive features to
simplify and standardize robot programming. These features
This method of using three trays for spatter collection was
enabled the robot to run more effectively during the
discarded for two reasons: a) it did not catch the weld spatter
experimentation, and also allowed program adjustments to arc
in its entirety, and b) the process was too cumbersome. The
start and end parameters. A sample of one of the programs
improved method of collecting spatter consisted of designing,
manipulations is given below. Manipulating the highlighted
and fabricating an enclosure in the shape of a five sided cone
program schedule creates adjustments to the arm speed,
with a narrow opening at the bottom, and a wide opening at
voltage, and wire feed speed. In this example “{36}” in the
the top. The enclosure was specifically designed to fit the
program listed below represents the adjustments which are
envelope of the test robots, while at the same time not
detailed in the program scheduled settings given at the end of
interfering with robot operations. The primary aim was to
the program
prevent weld spatter from escaping the catchment area. The
enclosure extended all the way from the bottom of the floor to
CREATE = DATE 19-02-13 TIME 11:05:48;
the base of the robot arm. A small spatter collection tray was
MODIFIED = DATE 19-04-26 TIME 13:28:42;
also designed to fit the narrow opening of the enclosure at the
1 : J P{1} 100% CNT100
bottom. The spatter collection enclosure with the small tray at
2 : J P{2} 100% CNT100
the bottom is shown in Figure 3. All components were
3 : J P{3} 100% CNT100
designed, welded, and fabricated at the industrial test facility.
4 : L P{5} 250mm/sec FINE Arc Start E2{36}
Anti-spatter was sprayed on trays and pans in order to prevent
5 : L P{6} WELD_SPEED FINE Arc End E2{36}
spatter from sticking to the sides. This also facilitated easier
6 : L P{7} 500mm/sec FINE
collection and weighing during the experiments.
7 : J P{8} 40% FINE
8 : J P{18} 40% FINE
9 : J P{19} 40% FINE
10 : L P{20} 250mm/sec FINE Arc Start E2{36}
11 : L P{21} WELD_SPEED FINE Arc End E2{36}
12 : L P{22} 500mm/sec FINE
13 : J P{23} 40% FINE
14 : CALL ROBOT_B_HOME
END

Program Schedule Settings adjusted during testing


PULSE Sched. {36}:
475 in/min. wire speed,
1.0 Trim (voltage),
21 in/min. arm travel speed
Fig. 3. Spatter enclosure on the left, enclosure in use on the right
362 Iqbal Shareef et al. / Procedia Manufacturing 48 (2020) 358–371
Iqbal Shareef and Christopher Martin/ Procedia Manufacturing 00 (2019) 000–000 5

MIG Weld Sched. {35}: quantifying spatter in production. Consequently, a high-


475 in/min. wire speed, volume robot was identified running normal settings and three
28.5 volts, trays shown in Figure 1 were placed at the base of the welding
21 in/min. arm travel speed fixture. These remained for an entire shift after which they
Sched. {19} : 420 ipm, 26.0 volts, 20 ipm arm speed were removed, and the trays were emptied, and spatter
Sched. {14} : 400 ipm, 23.0 volts, 20 ipm arm speed accumulated in the trays was collected. Subsequently the
Sched. {06} : 350 ipm, 22.5 volts, 15 ipm arm speed spatter was transferred into a Ziploc bag and weighed. The
costs associated with weld spatter were also explored and
summarized. A summary of the standard settings are listed in
ImageJ Vision Program: ImageJ, an open source Java
Table 5 below, and the weight in pounds of spatter collected
image processing program with tremendous plugin options for
on each day from the collection trays is shown in Table 6. The
extended functionality, was used for vision parameter data
total spatter collected each day is the sum of the amount of
collection. ImageJ can read many image formats including
spatter collected from the trays and the spatter collected off
TIFF, GIF, JPEG, etc., and display, edit, analyse, process and
the floor surrounding the trays because the trays could not
print images. In this work, ImageJ was used to calculate area
cover all the areas where spatter could fall each time.
and pixel value statistics of user-defined selections, density or
grey scale calibration, measurement of distances, and angles Table 5. Standard settings for high volume parts from industry
to create density histograms and line profile plots.
Wire Arm
Audacity Acoustics Program: Audacity is an open source Wire Torch
Schedule Surface Weld Weld Feed Travel Voltage
user friendly multi-track audio editor and recorder that can be Stick Angle
Number Condition Direction Process Speed Speed (Volts)
used in various operating systems. Audacity can be used to Out (in) (deg)
(in/min) (in/min)
record live audio, and convert records into digital recordings,
edit sound file formats, cut, copy, splice change the speed or 12 0.75 - 1 Clean Push CV 475 22 27.5 15
pitch of a recording. During the experiments we used this
program to monitor weld sound and convert these to data Table 6 Weight of the spatter collected from trays
points for spatter analysis.
Weld pieces per
Date Spatter Weight (lb.)
shift
3.4 Test procedure
5-Feb 150 2.35
Each of the 16 experiments listed in Table 4 were repeated 6-Feb 150 1.76
4 times and the average results used for final analysis. Two 7-Feb 150 1.88
special jigs were designed and fabricated, one to hold the test 8-Feb 150 2.41
pieces for tack welding and one to hold the tack welded test 11-Feb 140 2.07
pieces to conduct the designed experiments in the welding 12-Feb 150 2.09
robot cell. The weld cell fixture contained two stops and a 13-Feb 150 2.01
magnet to secure the part. Average 2.08142857
Std. D 0.23426684
3.5 Data collection % Difference 11.3%

In determining how to collect data, importance was placed


on identifying a data collection plan, identifying data sources, The annual costs of weld spatter were broken down into
and finally the data itself. Key elements of this study included annual costs per robot cell and annual costs from all robotic
frequency of occurrence, and resultant weight of excess welders. The annual costs from each robotic cell include
spatter. To that end, multiple responses were measured to cleaning, scraping, and weld wire loss. The annual costs per
ensure full capture of the phenomena with comparative data robotic cell are shown in Table 7. The annual costs from all
streams leading to more definitive findings. Measurement robotic welders include quality issues at the paint line,
focused on weight of the test piece before and after weld, tool/fixture repair hours, and consumables. The annual cost
weight of the spatter collected, video recording of the weld for all robotic weld cells is also shown in Table 7.
process via multiple cameras, and audio recordings of each
Table 7. Annual cost for all robotic weld cells
weld test.
Type of Cost Classification Annual Cost/cell Cost for 12 cells
3.6 Weld spatter baseline
Cleaning 100 min/week $3,625 43,500
Various methods were employed for collection and Scrapping 40 min/week $1,450 17,400
analysis of weld spatter data, to express the loss due to spatter
Weld wire loss 2.5 pounds/day $675 8,100
in terms of cost. Several materials and tools were used, first to
establish a baseline, and subsequently to test and analyse the Quality Issues 40 hours/year All robot cells 1,740
responses Tool/Fixture
10 hours/year All robot cells 435
Degradation
Even before the start of experiments, the current state of
Consumables Nozzles, tips All robot cells 753
weld spatter generation had to be quantified and a baseline
established. The process of determining the baseline included Total Annual Costs all robotic cells $71,928
Iqbal Shareef et al. / Procedia Manufacturing 48 (2020) 358–371 363
6 Iqbal Shareef and Christopher Martin/ Procedia Manufacturing 00 (2019) 000–000

4. Output Responses Measured metal deposited in each test (Wmd), the weight of the wire lost
in the form of spatter and vaporization (Wwl)is calculated.
A total of 6 test responses were measured during each test Once knowing the weight of the wire lost (Wwl) during each
listed in Table 4. The responses measured were: i) Input test, the percent loss in weight of the wire is calculated.
weight of the weld wire, ii) Weight of the spatter collected,
𝑊𝑊𝑊𝑊𝑊𝑊𝑊𝑊ℎ𝑡𝑡 𝑜𝑜𝑜𝑜 𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤 𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙 𝑊𝑊"2 = 𝑊𝑊" − 𝑊𝑊,- (4)
iii) Percent loss of weld wire during welding, iv) Number of
weld spatter particles, v) Intensity of sound during welding, TF U TVW
𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙 𝑜𝑜𝑜𝑜 𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤 𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤 𝑊𝑊"2% = S X 100 (5)
and vi) Weld quality TF
Input weight of the weld wire (𝑾𝑾𝒘𝒘 ): The actual weight of
the weld wire consumed during each spatter collection test
was expressed in pounds, and it is calculated as follows. The
weight of the input wire consumed in a test (Ww lb) is a
function of the diameter of the weld wire consumed in each
test (dw inch), wire feed rate used in the test (fw inch/min), the
time of weld test (tw min), and the density of the weld wire (rw
lb/in3). The volume of the weld wire consumed in a test (Vw
in3) was obtained by multiplying the cross-sectional area of
the weld wire with wire feed rate (fw in/min), and weld time
(tw). Then the weight of the weld wire was obtained by
multiplying the volume with the density of the weld wire. The
diameter of the wire was measured using a micrometre from
which the cross-sectional area was calculated. The feed rate of
Fig. 4. Scale for measuring test specimen weight
the weld wire was obtained from the welding machine.
During each test, five video cameras including a GoPro Number of weld spatter particles (Nsp): A GoPro camera
camera was placed in the cell at a strategically located and four stationary video cameras were installed in the robotic
predetermined position to record the entire welding operation. cell shown in Figure 5. These cameras were used to monitor
The robot was programmed in such a way that the starting of welding robots and manual welding processes. During each
the robot was the same for each experiment and was within test videos were recorded and transferred to a video editing
the observation range of the GoPro camera. This ensured that software for analysis.
the cameras were synchronized to the process to see spatter
development throughout the cycle from start to end. The exact
amount of time taken in each test was calculated by importing
videos from the GoPro camera paced in the weld cell into a
video editing software, and finding the frame in which the
weld started and the frame in which the weld ended. The
difference in time stamps of the weld start and weld end
picture frames was calculated to determine the exact time of
the weld in each test.
𝑊𝑊𝑊𝑊𝑊𝑊𝑊𝑊ℎ𝑡𝑡 𝑜𝑜𝑜𝑜 𝑡𝑡ℎ𝑒𝑒 𝑊𝑊𝑊𝑊𝑊𝑊 𝑊𝑊𝑖𝑖𝑖𝑖𝑖𝑖. 𝑊𝑊" = 𝑓𝑓(𝑑𝑑" , 𝑓𝑓" , 𝑡𝑡" , 𝜌𝜌" ) (1)
G ×I ×/
E-F F F
𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉 𝑜𝑜𝑜𝑜 𝑡𝑡ℎ𝑒𝑒 𝑊𝑊𝑊𝑊𝑊𝑊𝑊𝑊 𝑊𝑊𝑊𝑊𝑊𝑊𝑊𝑊 𝑉𝑉" = (2)
J
G ×I ×/ ×K
E-F F F F
𝑊𝑊𝑊𝑊𝑊𝑊𝑊𝑊ℎ𝑡𝑡 𝑜𝑜𝑜𝑜 𝑡𝑡ℎ𝑒𝑒 𝑊𝑊𝑊𝑊𝑊𝑊𝑊𝑊 𝑊𝑊𝑊𝑊𝑊𝑊𝑊𝑊 𝑊𝑊" = (3)
J
Weight of the spatter collected (Ws): Actual weight of the Fig. 5. Robotic cell with multiple video cameras
spatter ejected during the welding process (Ws) was the
Each video from the welding cell was analyzed and
second response. Before each test the spatter collection tray
screenshots for each part were taken. While welding was
was weighed using a scale correct to a gram, and was placed
performed on both sides of the 10 inch long inverted-T
at the narrow opening of the enclosure at the bottom as shown
section test specimen shown in Figure 1, data from only one
in Figure 3. After the welding experiment was complete, the
side of each test piece was processed due to the robotic
tray was weighed again. The difference in weight of the
welding arm covering some of the spatter on the other side,
spatter collection tray before and after the weld, was recorded
leaving only one side of each test piece to be analyzed. Using
as the actual weight of the spatter collected in pounds.
the weld on the back side of each test piece to capture spatter
Percent loss of weld wire during welding (𝑾𝑾𝒘𝒘𝒘𝒘%): Each
optically, screenshots were taken at regular intervals from the
test specimen was weighed before (Wtsb) and after each
videos. Those images from each piece were ran through a
experiment (Wtsa). The scale used is shown in Figure 4. The
particle counting software called ImageJ. The ImageJ is an
difference in weight of the sample before and after the test
open source image processing program that was used to count
was the exact weight of metal deposited (Wmd). Knowing
the spatter particles coming off of the weld. These spatter
input weight of the weld wire (Ww), and the weight of the
particles were quantified by converting the images to black
364 Iqbal Shareef et al. / Procedia Manufacturing 48 (2020) 358–371
Iqbal Shareef and Christopher Martin/ Procedia Manufacturing 00 (2019) 000–000 7

and white and filtering bit sizes and thresholds within the software, each test listed in Table 4 was repeated four times.
picture to prevent the weld arc flash from skewing data. During each experiment, four screen shots at regular intervals
Particle count was found to be consistent as long as the still were taken and the number of particles in each screenshot
frame images were taken at the same intervals. A screenshot determined. The average of the four screenshots was taken to
of the ImageJ software navigation tool bar along with still reduce variability in measurement. Then taking the average of
frame of the spatter image imported in ImageJ is shown in each of the four repeat experiments, an overall average (Nsp)
Figure 6. of each test listed in Table 4 was determined.

Fig. 6. Still frame of weld spatter imported in Image J Fig. 8. Screenshot used by ImageJ to count spatter particles

After each image is imported into ImageJ, the image can Intensity of sound during welding (Is): The GoPro
then be inverted and the dark scale threshold changed in order camera installed in the robotic welding cell recorded both
to get the best focus on the weld spatter itself. The same audio and video information during each test. The audio data
threshold was used for every image to keep the spatter was imported from the GoPro camera into the open source
quantification process consistent. An example of an image audio software called Audacity. In Audacity the stereo data
that has been inverted and focused by changing the threshold was converted to mono at a sample rate of 48,000 Hz. From
is shown in Figure 7. each weld five samples each of two second duration were
exported into text files. Each test specimen shown in Figure 1
had two welds. Each two second data packet contained 96,000
data points. The text file contained the normalized decibel
waveform data, and a decibel scale range from -1 to 1. A
value of zero is quietest while -1 or 1 indicates the loudest
relative sounds. The exported text file is inserted into Excel
where the relative minimum and maximum values were
determined. The maximum values are averaged. The larger
the average of maximum values, the louder the weld is. A
louder weld is an indication of spatter being produced.
Weld quality (Wq): After each experiment, the test
samples were allowed to reach the room temperature before
evaluating the quality of the weld. The quality of each weld
was determined from visual inspection, and rating of the
Fig. 7. Image threshold changed to focus on weld spatter and weld arc
quality on a scale of 1 to 5, where a test specimen of best
As can be seen from Figure 7, the arc flash from the weld quality was rated as 1, while a test sample of unacceptable
causes reflections off of the robot arm, but ImageJ software quality was rated as 5. Based on the prior experience and
allows for this kind of interference to be cropped out. Figure 8 input from subject matter experts in DOE of the weld quality,
shows the cropped image that is devoid of any extraneous a rating of 1 or 2 was given to only those that were deemed to
reflections and shows only the region that covers weld spatter. be acceptable.
Then this cleaned image is used by ImageJ to count the spatter An example of 5 rated welds would include heavy
particles detected. The pixel size that is detected by ImageJ undercuts and porosity as shown in Figure 9a. Test specimens
can be changed to prevent the arc flash from skewing the data. rated 4 had minor undercuts that were still visible to the naked
By filtering the size, the arc flash is not counted as a particle eye as seen in Figure 9b. Test specimens rated 3 are the ones
while all the other smaller particles are. To test the accuracy that do not have enough heat to melt the weld wire and
of the ImageJ software known number of triangles, square, produce a very rounded weld bead. A quality rating of 3 is
rectangles and circles, of different sizes were drawn with few shown in Figure 9c. Test specimens rated 2 are the ones with
hundred dark areas. Each time ImageJ provided exactly the weld beads that are slightly convex and in need of a little bit
same number of the images that were placed in the calibration more voltage to get the weld to a level one quality. A quality
still image. Having established the accuracy of the ImageJ rating of 2 is shown in Figure 9d.
Iqbal Shareef et al. / Procedia Manufacturing 48 (2020) 358–371 365
8 Iqbal Shareef and Christopher Martin/ Procedia Manufacturing 00 (2019) 000–000

Test specimen rated 1 would be a six millimeter flat to 5. Results


slightly concave weld bead with no undercut or porosity. A
quality rating of 1 is shown in Figure 9e. Each experiment listed in Table 4 was repeated four times,
and the average of the four values was taken as the mean
response. Six output responses described in Section 4 were
measured and the measured responses are given in Table 8.

Table 8. Response table of Taguchi L16 design

Output Responses
Actual vs Audio
Wire Spatter ImageJ
Theoretical (Avg. Weld
Input Collection (Particle
Weight (% Max Quality
(lb.) (lb.) Count)
Loss) Value)
Test No Ww Ws Wwl% Nsp Is Wq
Fig. 9. (a) Example of a 5 rated weld with large undercuts and porosity 1 0.1831 0.0042 5.67% 56 0.1324 1
2 0.1817 0.0006 1.96% 20 0.1032 2
3 0.186 0.0098 10.22% 82 0.1365 1
4 0.1872 0.0018 2.88% 19 0.0854 5
5 0.1855 0.0006 4.41% 7 0.0810 5
6 0.1858 0.0078 6.65% 65 0.1080 3
7 0.1837 0.0024 4.07% 3 0.0907 5
8 0.1836 0.0184 6.43% 9 0.0644 5
9 0.1832 0.0002 4.35% 8 0.0481 4
10 0.1762 0.0022 5.23% 13 0.0827 5
11 0.1859 0.0004 1.96% 26 0.1151 3
Fig. 9. (b) Example of a 4 rated weld with minor undercuts 12 0.192 0.0054 6.23% 108 0.1421 1
13 0.1855 0.0002 2.64% 11 0.0544 3
14 0.1886 0.0012 6.59% 14 0.0898 5
15 0.1822 0.0058 6.77% 28 0.1193 2
16 0.1851 0.0092 7.58% 35 0.1377 1

Minitab was used for statistical analysis of Taguchi L16


orthogonal design listed in Table 4. Taguchi Analysis was
performed with output responses 𝑊𝑊" , 𝑊𝑊& , 𝑊𝑊"2% , 𝑁𝑁&( , 𝐼𝐼& , and
𝑊𝑊. versus Torch Travel Angle, WfsAts, Voltage, Stick Out,
Surface Condition, Weld Direction, and Weld Process. Linear
Fig. 9. (c) Example of a 3 rated weld without rounded weld bead Model Analysis of S/N ratios versus Torch Travel Angle,
WfsAts, Voltage, Stick Out, Surface Condition, Weld
Direction, and Weld Process was completed. The estimated
Model Coefficients for S/N ratios are given in Table 9a and
the corresponding ANOVA table for S/N is given in Table 9b.

Table 9a. Estimated Model Coefficients for S/N ratios

Term Coef SE Coef T P


Constant -19.1288 1.332 -14.359 0.005
Torch Angle 1 -4.3164 2.307 -1.871 0.202
Torch Angle 2 4.0916 2.307 1.773 0.218
Fig. 9. (d) Example of a 2 rated weld with slightly convex shape beads Torch Angle 3 -0.7474 2.307 -0.324 0.777
Wfs Ats 1 3.5368 2.307 1.533 0.265
Wfs Ats 2 -0.1374 2.307 -0.06 0.958
Wfs Ats 3 -0.8946 2.307 -0.388 0.736
Voltage 1 -5.7503 2.307 -2.492 0.13
Voltage 2 -1.7029 2.307 -0.738 0.537
Voltage 3 1.7017 2.307 0.738 0.538
Stick Out 1 2.1238 1.332 1.594 0.252
Surface Cond 1 0.6134 1.332 0.46 0.69
Weld Direction 1 -1.2306 1.332 -0.924 0.453
Weld Process 1 -3.4572 1.332 -2.595 0.122
Fig. 9. (e) Example of a 1 rated weld free of undercut and porosity S = 5.329 R-Sq = 93.4% R-Sq(adj) = 50.7%
366 Iqbal Shareef et al. / Procedia Manufacturing 48 (2020) 358–371
Iqbal Shareef and Christopher Martin/ Procedia Manufacturing 00 (2019) 000–000 9

Table 9b. Analysis of Variance for S/N ratios is the largest in the table, and is nearly twice that of the next
largest value of 3.38, which is for voltage. The next largest
Source DF Seq SS Adj SS Adj MS F P
being 2.54 and 1.73 for stick out and torch travel angle
Torch Angle 3 147.503 147.503 49.168 1.73 0.386
respectively. The ANOVA Table 9b for S/N ratios shows that
WfsAts 3 78.41 78.41 26.137 0.92 0.558
the weld process, voltage, stick out, and torch travel angle are
Voltage 3 287.769 287.769 95.923 3.38 0.237
cause for most of the variation.
Stick Out 1 72.172 72.172 72.172 2.54 0.252
From Table 10a for estimated Model Coefficients for
Surface Cond 1 6.019 6.019 6.019 0.21 0.69
Means, it is clear that the T-values are again spread equally
Weld Direction 1 24.232 24.232 24.232 0.85 0.453
around both sides of zero, and the R2 values of 84.5% is again
Weld Process 1 191.237 191.237 191.237 6.74 0.122
relatively good, given the fact that it is the real life data
Residual Error 2 56.789 56.789 28.394
obtained from industry. In Table 10b the F-value of 3.31 for
Total 15 864.131
weld process is more than double the next largest value of 1.6
for stick out, and more than three times for weld direction.
Linear Model Analysis of Means versus Torch Travel
The response table for signal to noise ratio is shown in
Angle, WfsAts, Voltage, Stick Out, Surface Condition, Weld
Table 11a, and the corresponding response tables for means
Direction, and Weld Process was completed. The estimated
and standard deviations are shown in Tables 11b and 11c.
Model Coefficients for Means are given in Table 10a, and the
corresponding ANOVA table for Means in Table 10b.
Table 11a. Response table for Signal to Noise ratios
Table 10a. Estimated Model Coefficients for Means Torch Surface Weld Weld
Level Wfs Ats Voltage Stick Out
Angle Cond Direction Process
Term Coef SE Coef T P 1 -23.45 -15.59 -24.88 -17 -18.52 -20.36 -22.59
Constant 5.82972 1.323 4.408 0.048 2 -15.04 -19.27 -20.83 -21.25 -19.74 -17.9 -15.67
Torch Angle 1 1.94608 2.291 0.849 0.485 3 -19.88 -20.02 -17.43
Torch Angle 2 -1.4744 2.291 -0.644 0.586 4 -18.16 -21.63 -13.38
Torch Angle 3 1.22908 2.291 0.537 0.645 Delta 8.41 6.04 11.5 4.25 1.23 2.46 6.91
Wfs Ats 1 -1.8285 2.291 -0.798 0.508 Rank 2 4 1 5 7 6 3
Wfs Ats 2 -0.50754 2.291 -0.222 0.845
Wfs Ats 3 0.48062 2.291 0.21 0.853 Table 11b. Response table for means
Voltage 1 2.18167 2.291 0.952 0.441
Torch Surface Weld Weld
Voltage 2 1.445 2.291 0.631 0.593 Level Wfs Ats Voltage Stick Out
Angle Cond Direction Process
Voltage 3 -0.46781 2.291 -0.204 0.857
1 7.776 4.001 8.011 4.154 5.741 7.181 8.237
Stick Out 1 -1.67553 1.323 -1.267 0.333
2 4.355 5.322 7.275 7.505 5.919 4.478 3.423
Surface Cond 1 -0.08879 1.323 -0.067 0.953
3 7.059 6.31 5.362
Weld Direction 1 1.3517 1.323 1.022 0.414
4 4.129 7.685 2.671
Weld Process 1 2.40708 1.323 1.82 0.21
Delta 3.647 3.684 5.341 3.351 0.178 2.703 4.814
S = 5.291 R-Sq = 84.5% R-Sq(adj) = 0.0%
Rank 4 3 1 5 7 6 2

Table 10b. Analysis of Variance for Means


Table 11c. Response table for standard deviations
Source DF Seq SS Adj SS Adj MS F P Torch Surface Weld Weld
Level Wfs Ats Voltage Stick Out
Torch Angle 3 41.457 41.4574 13.8191 0.49 0.722 Angle Cond Direction Process
WfsAts 3 29.099 29.0986 9.6995 0.35 0.8 1 17.848 8.373 18.488 8.81 12.669 16.279 18.859
Voltage 3 68.18 68.1797 22.7266 0.81 0.593 2 8.92 11.242 16.611 16.82 12.961 9.351 6.771
Stick Out 1 44.918 44.9184 44.9184 1.6 0.333 3 15.756 14.264 11.453
Surface Cond 1 0.126 0.1261 0.1261 0 0.953 4 8.736 17.38 4.708
Weld Direction 1 29.234 29.2336 29.2336 1.04 0.414 Delta 9.112 9.007 13.78 8.01 0.292 6.928 12.089
Weld Process 1 92.705 92.7049 92.7049 3.31 0.21 Rank 3 4 1 5 7 6 2
Residual Error 2 55.981 55.9813 27.9907
Total 15 361.7 It is clear from the main effects plot of S/N ratios in Figure
10a, that the voltage and WfsAts are almost linear. The S/N
From the examination of the estimated model coefficient ratio increases with an increase in voltage and decreases with
for Signal to Noise ratios in Table 9a, it is clear that the data is an increase in WfsAts. The response Table 11a for S/N ranks
symmetric about the mid-point with both positive and voltage as 1, torch travel angle as 2, weld process as 3, and
negative T-values spread almost evenly on both sides of the WfsAs as 4.
zero. The coefficient of determination R2 is the percentage of The main effects plot for Means in Figure 10b shows a
variation in the response that is explained by the model. It is a huge difference between weld processes, and insignificant
statistical measure of how close the data are to the fitted difference between surface conditions. Although, torch travel
regression line, and a R2 value of 93.4% is relatively good for angle is somewhat significant, it is not clear why the response
the real data obtained from industry. Examination of F-values fluctuates with linear increase in torch travel angle. One
in Table 9b shows that the F-value of 6.4 for the Weld Process reason may be due to the sensitivity of the response to
Iqbal Shareef et al. / Procedia Manufacturing 48 (2020) 358–371 367
10 Iqbal Shareef and Christopher Martin/ Procedia Manufacturing 00 (2019) 000–000

changes in torch travel angle, or the data may be corrupted variations of parameters close to the mean were used for
resulting in a nonlinear change in response due to linear optimization testing in search of the sweet spot. A torch travel
change in angle. As such, this may require further angle of 15, Wire feed and Arm travel speed of 425, voltage
investigation. The corresponding response Table 11b ranks between 26 and 28 volts, average stick out of 0.75, and a
voltage as 1, weld process as 2, WfsAts as 3, and torch travel clean surface condition were used. There was huge noise
angle as 4. variation due to weld process so variations of both CV and
The main effects plot for standard deviations given in Pulse were tested. The results of these tests are detailed in
Figure 10c, and the corresponding response table 11c for Table 12a.
standard deviations shows very much the same trends as that
of the main effects plot of Means. Table 12a. Response table for standard deviations
Verification Tests (1 & 2) / Optimization Testing (3 - 8) and Results
Experiment
1 2 3 4 5 6 7 8
Number
Wire Feed Speed 325 375 425 425 425 425 475 475
(in/min)
Travel Speed 15 17 19 19 19 19 21 21
(In/min)

Input Parameters
Torch Angle 15 15 15 15 15 15 15 15
Voltage/ Trim 26/ 24 26.5 25.2/ 27 27.3/ 27.5 25.5/
(Volts) 1.1 1 1.1 1.05
Stick out (in.) 1.00 1.00 0.75 0.75 0.75 0.75 0.75 0.75
Surface Oily Oily Clean Clean Clean Clean Clean Clean
Conditions
Weld Direction Push Drag Push Push Push Push Push Push

Weld Process Pulse CV CV Pulse CV Pulse CV Pulse

Wire Input (lb.) 0.172 0.035 0.032 0.114 0.098 0.031 0.076 0.141
Output Responses

Fig. 10. (a) Main effects plot for S/N ratios Actual vs 2.58% 4.28% 4.25% 3.27% 3.98% 3.35% 4.23% 3.74%
Theoretical Weight
Audio
(% Loss) 0.073 0.0995 0.148 0.131 0.147 0.121 0.148 0.137
(Avg. Max Value)
ImageJ (Particle 10 70 154 90 147 82 144 94
Count)
Quality 5 3 1 1 1 1 1 1

Better quality was observed for tests 3 to 8 with large


variations in particle count and nominal variations in noise
and weight loss. The only significant variant was weld
process so further comparison was done to understand which
weld process presented more optimal results. Table 12b shows
the comparison of pulse and constant voltage welding at
similar wire feed speeds, and voltages.

Table 12b. Direct comparison of optimization of Pulse vs CV


Fig. 10. (b) Main effects plot for Means Experiment
3 4 5 6 7 8
Number
Wire Feed Speed
425 425 425 425 475 475
(in/min)
Voltage/ Trim 25.17/ 27.3/ 25.5/
27 27 28
(Volts) 1 1.1 1.05
Weld Process CV Pulse CV Pulse CV Pulse
Actual vs
Theoretical 4.25% 3.27% 3.98% 3.35% 4.23% 3.74%
Weight (% Loss)
% Improvement
13.03% 8.59% 6.15%
with Pulse
Audio
0.1477 0.1307 0.1471 0.1208 0.1479 0.1373
(Avg. Max Value)
% Improvement
11.57% 9.80% 3.72%
with Pulse
ImageJ (Particle
154 90 147 82 144 94
Count)
Fig. 10. (c) Main effects plot for Standard Deviations % Improvement
25.92% 28.53% 20.91%
with Pulse
Optimization of the wed process: DOE attributes were
based on the Taguchi signal responses shown in Figure 10. Further tests were done to refine the results and establish
Settings were based on parameters with the main effects that final recommendations. The results from the recommendation
resulted in a smaller signal to noise ratio due to our desire to verification experiments are shown in Table 13a. Table 13b
reduce spatter. Two verification tests were developed, and 6 shows a comparison of schedule 12 (current settings) to the
368 Iqbal Shareef et al. / Procedia Manufacturing 48 (2020) 358–371
Iqbal Shareef and Christopher Martin/ Procedia Manufacturing 00 (2019) 000–000 11

three recommended settings which resulted in the lowest


particle count and highest quality.
Table 13a. Final optimized test parameters and responses
Experiment
1 2 3 4 5 6
Number
Wire Feed Speed
425 425 450 450 475 475
(in/min)
Travel Speed Fig. 11. (c) Weld penetration Left and Right for Expt. 3 in Table in 13a
19 19 20 20 21 21
(in/min)
Torch Angle 15 15 15 15 15 15
Parameters

Voltage/ Trim 1.05 1.05 1.05


27.0 28.0 28.5
(Volts) (26.9) (27.5) (28.1)
stick out (inches) 0.75 0.75 0.75 0.75 0.75 0.75
Surface Condition Clean Clean Clean Clean Clean Clean
Weld Direction Push Push Push Push Push Push
Fig. 11. (d) Weld penetration Left and Right for Expt. 4 in Table in 13a
Weld Process CV Pulse CV Pulse CV Pulse
Wire Input (lb.) 0.1803 0.1937 0.1953 0.1951 0.1968 0.1966
Actual vs
Theoretical Weight4.10% 3.50% 3.76% 4.19% 4.11% 4.10%
Responses

(% Loss)
Audio (Avg. Max
0.1426 0.1263 0.1361 0.136 0.1313 0.1369
Value)
ImageJ Particle
143 82 120 86 130 88
Count)
Fig. 11. (e) Weld penetration Left and Right for Expt. 5 in Table in 13a
Quality 1 1 1 1 1 1

Table 13b. Direct comparison of pulse vs CV recommendations


Current Rec. 1 Rec. 2 Rec. 3
Wire Feed Speed
475 425 450 475
(in/min)
Voltage/ Trim
27.5 26.9/ 1.05 27.5/ 1.05 28.1/ 1.05
(Volts)
Weld Process CV Pulse Pulse Pulse Fig. 11. (f) Weld penetration Left and Right for Expt. 6 in Table in 13a
Actual vs
Theoretical Weight 4.23% 3.50% 4.19% 4.10%
(% Loss)
Percent Change 9.44% 0.48% 1.56% 6. Discussion
Audio
0.1479 0.1263 0.1360 0.1369
(Avg. Max Value)
Percent Change 7.91% 4.22% 3.87% 6.1 Taguchi Designed Experiments
ImageJ (Particle After running the L16 Orthogonal Array, the Minitab
144 82 86 88
Count)
Percent Change 27.43% 25.22% 24.14% analysis showed what parameters impact spatter the most. The
most impactful parameter was voltage. At the highest voltage,
The test pieces with the best-looking welds from each set there was very little spatter produced while the weld quality
of four experiments were cut into thirds with a band saw. The was reduced, which led to less aggressive voltages being
middle section of each piece was grinded, polished and etched recommended. The second most impactful parameter was
on both sides to view the weld penetration. The penetration torch travel angle. Torch travel angle was a four-level
shows as triangular darker areas at the welds. Photos for each parameter and the torch travel angle of 15 degrees produced
weld can be seen in Figures 11a to 11f. the best results. The third most influential parameter was weld
process. Pulse welding produced less spatter. The ranking of
the remaining parameters was wire feed speed, stick out, weld
direction, and surface condition respectively.
6.2 Responses from Experiments
Input of weld wire deposit rate: The calculated weight of
Fig. 11. (a) Weld penetration Left and Right for Expt. 1 in Table in 13a the wire deposited into the weld was found to have, by nature,
a very low variability response due to all the inputs being
known constants. The time taken was calculated by importing
the video from the camera placed in the weld cell into a video
editing software. By finding the frame where the weld started
and ended, the time difference was able to be found for each
weld. The time taken was then multiplied by the weight per
Fig. 11. (b) Weld penetration Left and Right for Expt. 2 in Table in 13a
Iqbal Shareef et al. / Procedia Manufacturing 48 (2020) 358–371 369
12 Iqbal Shareef and Christopher Martin/ Procedia Manufacturing 00 (2019) 000–000

second to find the theoretical weight of the wire that was put pieces had severe undercut along much of their lengths. Any
into the weld. sets of parameters that produced undercut were not considered
Percent loss of weld wire during welding process: The for the team’s recommendations as it was considered a huge
percent loss was based on the difference between the quality issue.
calculated weight and the actual weight of the weld. The Weld Spatter Baseline: The weld spatter baseline provided
percent loss correlated pretty well to most of the other a good baseline for understanding weld spatter loss and
responses, but did not correlate directly with weld quality. A resultant costs on weld robots. This cost is different from
lower percent loss did not necessarily lead to better weld manual weld operations due to the extensive use of anti-
quality. If the voltage was a little too high for the wire feed spatter spray in manual vs robotic weld operations which
speed to where it melts too much of the metal but not to the costs an additional $11,781 annually. This cost was not
point where it is vaporizing the metal, the result led to factored in this study. Costs factored include time spent
undercut on the top plate. scraping and cleaning the robot cells. These times are based
Spatter collection: The spatter collected from the tray on operator surveys and are deemed as necessary to maintain
placed in the welding robot varied between certain trials. the standards of the facility. Spatter reduction will have
While the maximum weight of spatter collected from the tray positive impact on the overall cost.
after a trial was 0.0184 lb., the minimum weight was only Optimization testing: Experiments 1 and 2 were repeat
0.0002 lb. and the average was 0.00438 lbs. The spatter experiments meant to verify the results from the screening
produced in the designed experiment trials was not as experiments as being accurate and repeatable. Those two
significant as the 0.01387 lb. per piece average of the experiments were run at the same settings as experiments 5
baseline. The primary reason lies in the nature of the T-joint and 6 from the L16 Designed Experiment. After analysing
used for the experiments which was an ideal reach for the and comparing the responses for the identical experiments,
robot arm and the distance from the nozzle to the work piece they were found to be very similar, verifying the responses for
remained constant. Despite this, the experiment loss variation their repeatability. The other six experiments, 3 to 6 in Table
and multiplication effect validate that a comparison between 12a, were conducted to optimize parameter settings. It was
baseline and experiment is valid. It can also be extrapolated found that increasing the voltage by small increments and
that more complex production welds will present more spatter utilizing pulsed MIG spray transfer welding, spatter
and more opportunity for spatter reduction in a real generation was reduced, and weld quality was maintained.
production environment. Switching to pulse did impact the voltage, and the voltage did
Optical evaluation: There is a great deal of variability with not remain constant. A trim was set at a certain value that
the nature of spatter collection itself, but by taking multiple corresponds to arc length. The trim relationship to arc length
images of the same weld and averaging the particle count, and voltage differed from machine to machine. A trim value
variability was decreased. It was decreased again by taking of 1 was the recommended voltage for a given wire feed
the average of all the trials that contain parts ran at the same speed by the welding machine’s original equipment
parameters. The results showed some correlation between manufacturer. The goal was to get the voltage and trim as
other responses that were measured. Also, spatter particle high as possible without creating undercut on the welded
count correlated with what was expected. For example, piece. The results from the optimization tests conducted were
running parts at low wire feed speeds with high voltages analysed to determine recommended settings for our industry
resulted in low spatter counts because the wire itself was partner.
vaporizing and not being ejected as spatter. This was the case More experiments were conducted to verify the three
in other experiments in this investigation, although the lower recommended settings at different wire feed speeds. Although
spatter count experiments did not always result in high quality the optimization experiments ran pulsed MIG with a trim of
welds. As the best parameter sets were selected, the goal was 1.1 had no quality problems, it was noticed that when trying
kept on reducing spatter count while also maintaining weld to set three recommended parameter settings on a different
quality. day, the voltages generated were a little higher than desired. It
Audio evaluation: The audio data had variability across was determined that a trim value of 1.05 would be the best.
different experiments. The variability proved there was Although spatter is not minimized as much as possible with
something to be learned from the audio data. The average this, quality cannot be sacrificed as a result. Essentially a trim
value of the maximums showed the relative loudness of the 2 value of 1.05 could be transferred to all machines without
second weld clip. The louder welds produced more spatter. having to fret about compromised weld quality. Using a trim
The loudest experiment was 12. Trial 12 produced the most greater than 1.0 still leads to spatter reduction. Leveraging
spatter according to optical analysis. There was a high level of pulsed MIG welding versus the traditional constant voltage
confidence that the average of the maximums correlated to the welding is the cause of the improvement. The verification
spatter generation. trials included tests in both constant voltage and pulse. The
Weld Quality: The quality in welds varied considerably reasoning behind the duplication of similar wire feed speeds
between test pieces. The toe-to-toe lengths in comparison to and voltages was to verify that pulsed MIG did indeed
the profile of the welds were mostly acceptable and porosity drastically impact the spatter generation. The wire feed speeds
was not found on many pieces. The pieces that had porosity of 425 and 450 inches per minute did not appear to be the best
were ones having an oily surface condition. Also, no piece settings to be disseminated across all possible welds.
had any cracks along the welds. However, several of the test However, the wire feed speed of 475 inches per minute was
370 Iqbal Shareef et al. / Procedia Manufacturing 48 (2020) 358–371
Iqbal Shareef and Christopher Martin/ Procedia Manufacturing 00 (2019) 000–000 13

recommended for implementation. It was further determined 7. Conclusions


that whenever a weld design is not completed enough to
create a situation requiring the welding robot to reach difficult In conclusion, the L16 Taguchi Orthogonal Array was
to reach places, pulse welding is a better fit than constant useful in identifying parameters influential to spatter
voltage. When welds require extreme angles or stick out, production in the welding process. The test also gave the
constant voltage is the better selection. The reason for optimal set of parameters where virtually no spatter was being
constant voltage being better in these challenging welding produced. The issue with utilizing the mean value table to
locations is because the trim value only regulates the arc determine the optimal setting was that quality would have
length. When arc length has consistent variation, the pulsed been violated. A weld with a wire feed speed of 325 inches
current jumps to extremely high voltages leading to undercut. per minute and a voltage of 30 volts had massive undercut in
Upon inspecting the etched parts from the optimization the designed experiment, which were the parameters the mean
tests, it was determined that all of the weld penetrations for value table indicated to leverage.
the test pieces were sufficient; the plates were melded The ImageJ and Audacity software provided insight into
together properly for a strong weld. This was clearly evident the spatter generation for the designed experiments. The
by the extent to which the weld penetration extended in the ImageJ software was used for the optical analysis. The optical
weld joint area on both sides of each test piece shown in analysis served as the best response for spatter quantification.
Figures 11.1 to 11.6. While the penetrations varied slightly It was the most effective because the response was not
between pieces and sides, all of them displayed high quality. affected by any outside noise. The average of 10 images was a
sufficient sample size to discern the difference between
6.3 Cost savings justification averages and experiments. Audacity was used for audio
The cost savings for the project were based off the 3 best analysis. Leveraging audio data was a challenge due to the
responses from the experiments. The three responses are size of the samples. However, after analysis, Audacity could
percent loss of weld wire, audio out of the average of the differentiate different L16 experiments. If the team had access
relative maximums, and spatter count from optical analysis. to computers with greater processing power, more
The responses were given weights to find the weighted information could have been derived from the audio data.
average of the percent improvement from the current most The primary conclusion is that maximizing voltage,
used schedule. The weight for percent loss of weld wire was without causing undercut, minimizes spatter. With the given
30%. The weight for audio output was 10%. The weight for constraints of certain parameters (weld wire, gas mixture), the
particle count was 60%. Weights were assigned based on the project revolved around finding a sweet spot where quality
accuracy and consistency of the response throughout all was not compromised but rather improved along with the
experiments. Once weights were assigned to the response, a primary goal of reduction in spatter. It is recommended to
weighting factor was also assigned to each recommendation. increase voltage to match drastically increased wire feed
The weighting factors on the recommendations were assigned speeds that stemmed from previous designed experiments.
based on DOE results at a given wire feed speed. The 425, Pulsed MIG welding was found to produce less spatter by
450 and 475 inches per minute of wire feed speed were given 9.02% or approximately 9%. Pulse welding is recommended
45%, 30%, and 25% weights respectively. The 475 inches per to be used when applicable. The determination of 9% is
minute schedule corresponds to the test industry sites current explained in the justification of cost savings in Table 14. The
schedule 12. Table 14 shows the computation of the percent 9% translates to an annual cost savings of $6,490 for the
reduction from each process improvement as well as overall industrial facility, not including the significant impact it has
percentage savings and cost savings form each on the overall quality of the welding operations resulting in a
recommendation. Columns 1 and 2 of Table 14 lists the three robust product. In addition, an uncalculated resultant cost
recommendations for implementation and the corresponding savings was the energy consumption difference between
weighting factor for each recommendation. Columns 3 to 5 pulsed MIG and traditional constant voltage. Pulse used a low
lists the percentage savings from loss of weld wire, reduction background voltage and high peak current so that average
in noise and number of spatter particles A full display of these current was less than using constant voltage. The energy
percent improvements and results of the trials conducted are consumption is a subject of ongoing future work that is not
shown in Table 13b. reported in this manuscript. An interesting conclusion related
to the cleanliness of the parts was that parts contaminated
Table 14. Cost savings with an oily substance do not suffer from additional spatter
Savings from process generation.
Recommendation

Weighting Factor

Overall % Overall $ Furthermore, it was found that practically it was


Improvements
Savings Savings impossible to eliminate weld spatter entirely while still
Audio Particle
% Loss from Each from Each maintaining weld quality with the constraints of high wire
Output Count
Rec Rec feed speeds, constant gas mixture, and constant wire type.
30% 10% 60%
Nevertheless, optimization of the salient weld parameters has
1 45% 9.40% 7.90% 27.40% 9.02% $6,490 been completed, and the recommended schedules that were
implemented in the industrial production are shown in Table
2 30% 0.50% 4.20% 25.20% 4.71% $3,386
13b, along with the current highest used schedule 12 listed in
3 25% 1.60% 3.90% 24.10% 3.83% $2,757
the first column of table 13b.
Iqbal Shareef et al. / Procedia Manufacturing 48 (2020) 358–371 371
14 Iqbal Shareef and Christopher Martin/ Procedia Manufacturing 00 (2019) 000–000

Acknowledgements Advances in Mechanical Engineering, 9(9), 1-9.

[11]Sumesh, A., Rameshkumar, K., Mohandas, K., & Babu, R. S. (2015). Use
Special thanks to our great research team of Michael of machine learning algorithms for weld quality monitoring using
Richey, Kyle Roth, Eric Statz, Logan Welling, and Andy acoustic signature. Procedia Computer Science, 50, 316-322.
[12]Lipei, J., Jingchang, L., & Zhihuan, W. (1988). The effect of the dynamic
Routley. This study would have not been possible without
behaviour of welding rectifiers on spatter. Welding International, 2(3),
their invaluable input. Also, thanks to our Industry partner and 263-268.
their many representatives, in particular Ed Miller, Brad [13]Mita, T. (1989). Reducing spatter in CO2 gas-shielded arc welding—
Monk, and Brian Hansche, who allowed us to conduct this Waveform control. Welding International, 3(3), 227-232.
study and have subsequently implemented recommendations. [14]Mita, T. (1989). Reducing spatter in CO2 gas-shielded arc welding (2)—
The authors are also thankful to welding instructors Bob relationship between spatter and welding conditions. Welding
Branon, from Illinois Central College, and Chris Winn from International, 3(6), 503-508.
Illinois Valley Community College. Lastly, by no means the [15]Yamamoto, H., Harada, S., & Yasuda, T. (1990). The development of
least, the authors are thankful to Dr. Rick Polanin, Vice welding current control systems for spatter reduction. Welding
International, 4(5), 398-407.
President of the American Welding Society and CoPI of The
[16]Mita, T. (1991). Spatter reduction-power source considerations. Welding
National Center for Welding Education and Training (Weld- international, 5(11), 847-850.
Ed) for not only providing his valuable suggestions during the
[17]Takeuchi, Y., & Shinoda, T. (1991). Spatter and blowhole formation
course of this study but also providing his editorial comments phenomena in pulsed gas shielded metal arc welding. Materials science
to this manuscript. and technology, 7(9), 869-876.
[18]Fan, D., Shi, Y., & Ushio, M. (2001). Investigation of CO₂ Welding Arc
Sound: Correlation of Welding Arc Sound Signal with Welding Spatter
References (Physics, Processes, Instruments & Measurements). Transactions of
JWRI, 30(1), 29-33.
[1] Ersoy, U., Hu, S. J., & Kannatey-Asibu, E. (2008). Observation of arc [19]Baune, E., Bonnet, C., & Liu, S. (2001). Assessing metal transfer stability
start instability and spatter generation in GMAW. Welding Journal, and spatter severity in flux cored arc welding. Science and Technology of
87(2), 51-56. Welding and Joining, 6(3), 139-148.
[2] Schwab, G., Steele, J. P. H., & Vincent, T. L. (2009). Vision-based [20]Lee, S. H., Kim, J. S., Lee, B. Y., & Lee, S. Y. (2005). The effect of
spatter classification for contaminant detection. Welding Journal, 88(6), external electromagnetic force in gas metal arc welding on the transfer
121-30. mode. In Key Engineering Materials(Vol. 297, pp. 2825-2830). Trans
Tech Publications Ltd.
[3] Kang, M. J., Kim, Y., Ahn, S., & Rhee, S. (2003). Spatter rate estimation
in the short circuit transfer region of GMAW. Welding Journal, 82(9), [21]Era, T., & Ueyama, T. (2007). Spatter reduction in GMAW by current
238-247. waveform control. Welding International, 21(7), 496-501.
[4] Chen, J. H., Sun, Z. C., & Fan, D. (1996). Study on the mechanism of [22]Kataoka, T., Ikeda, R., Yasuda, K., & Hirata, Y. (2009). Development of
spatter produced by basic welding electrodes. Welding Journal-Including a low-spatter CO2 arc welding process with a high-frequency pulse
Welding Research Supplement, 75(10), 311-316. current. Welding International, 23(5), 353-359.
[5] Chavda, S. P., Desai, J. V., & Patel, T. M. (2014). A review on [23]de Meneses, V. A., Gomes, J. F. P., & Scotti, A. (2014). The effect of
optimization of MIG Welding parameters using Taguchi’s DOE method. metal transfer stability (spattering) on fume generation, morphology and
International Journal of Engineering and Management Research, composition in short-circuit MAG welding. Journal of Materials
4(1),16-21. Processing Technology, 214(7), 1388-1397.
[6] Tay, K. M., & Butler, C. (1997). Modelling and optimizing of a MIG [24]Guo, N., Xu, C., Guo, W., Du, Y., & Feng, J. (2015). Characterization of
welding process—a case study using experimental designs and neural spatter in underwater wet welding by X-ray transmission
networks. Quality and Reliability Engineering International, 13(2), 61- method. Materials & Design, 85, 156-161.
70.
[25]Xue, L., Wu, J., Huang, J., Huang, J., Zou, Y., & Liu, J. (2016). Welding
[7] Woods, S. (2017). Spatter Control, Along with Other Welding Issues, polarity effects on weld spatters and bead geometry of hyperbaric dry
Can be Solved by Inverter and Waveform Technology. Welding GMAW. Chinese Journal of Mechanical Engineering, 29(2), 351-356.
Productivity, May 2017, 12 to 16. [26]Cheng, F., Zhang, S., Di, X., Wang, D., & Cao, J. (2017). Arc
[8] Sapakal, S. V., & Telsang, M. T. (2012). Parametric optimization of MIG Characteristic and Metal Transfer of Pulse Current Horizontal Flux-
welding using Taguchi design method. Int J Adv Eng Res Stud, 1(4), 28- Cored Arc Welding. Transactions of Tianjin University, 23(2), 101-109.
30. [27]Fu, Y., Guo, N., Du, Y., Chen, H., Xu, C., & Feng, J. (2018). Effect of
metal transfer mode on spatter and arc stability in underwater flux-cored
[9] Sivasakthivel, K., Rajkumar, R., & Yathavan, S. (2015). Optimization of wire wet welding. Journal of Manufacturing Processes, 35, 161-168.
welding parameter in MIG welding by Taguchi method. In Proceedings
of international conference on advances in materials, manufacturing and
applications (Vol. 761, pp. 761-765).

[10]Jeong, Y. C., Jung, Y. G., & Cho, Y. T. (2017). Simple monitoring of


welding spatter for quantification and observation using a mobile phone.

You might also like