Additive Manufacturing 48 (2021) 102387

Contents lists available at ScienceDirect

Additive Manufacturing
journal homepage: www.elsevier.com/locate/addma

Research Paper

Prediction of the interpass temperature of a wire arc additive manufactured

wall: FEM simulations and artificial neural network
Francisco Werley Cipriano Farias *, João da Cruz Payão Filho , Victor Hugo Pereira Moraes e
Oliveira
Program of Metallurgical and Materials Engineering, Federal University of Rio de Janeiro (UFRJ), CEP 21941-972 Rio de Janeiro RJ, Brazil

A R T I C L E I N F O A B S T R A C T

Keywords: Finite element method (FEM) simulations are a powerful tool for understanding the thermal­
Wire arc additive manufacturing (WAAM) –metallurgical–mechanical effects of wire arc additive manufacturing (WAAM). Nonetheless, owing to the
Idle time multiphysical nonlinear nature of welding coupled with the longer deposition time of WAAM, FEM simulations
Finite element method
can be laborious and time-consuming, which makes it difficult to simulate the numerous procedural parameters
Artificial neural network
Thermal management
of WAAM. Therefore, the present work aimed to employ an FEM mode to analyze the influence of idle time
(30–240 s) on the interpass temperature (IT) of 20-layer single-bead walls produced via WAAM and use the FEM
results to develop a predictive model for the IT based on an artificial neural network (ANN). The FEM simulations
were performed using a heat source and a 20-layer single-bead wall model that was experimentally calibrated
and validated. The first layers exhibited similar energy accumulation; however, as the wall height increased, the
IT rapidly increased under to low idle times (≤120 s). The ANN was trained using the FEM simulations results,
validated with FEM results (not included in the training database), and used to establish a process map (including
the idle time, number of layers, and IT). This can help the manufacturers to obtain a suitable balance between
productivity (lower idle times) and part behavior (e.g., microstructure and mechanical properties).

1. Introduction in the part cross-section as it is printed. In the first layers, conduction is

the predominant heat transfer mechanism. However, as the part is built,
The advent of additive manufacturing (AM) has increased conduction becomes progressively less efficient (owing to the reduced
manufacturing flexibility (through the absence of molds and dies), cross-section, distance from the substrate, and increasing deposit tem­
reduced material waste (near-net-shape parts) [1], and promoted the perature), and heat transfer through radiation and convection (naturally
development of new component designs (e.g., functionally graded ma­ limited compared to conduction in metals) becomes relevant, resulting
terials and topology optimization) [2]. Among the AM processes, in heat accumulation in the part [9,10].
directed energy deposition, particularly the wire arc additive Energy accumulation has negative consequences, e.g., coarse and
manufacturing (WAAM), is notable because of its high productivity and oriented microstructure, anisotropic properties, low fracture toughness,
low initial capital investment, allowing the printing of large parts with high levels of distortion, loss of dimensional control, and oxidation
highly/moderately complex geometries and a high relative bulk density [11–15]. To solve these problems, forced cooling techniques have been
(low porosity) [3]. WAAM uses an electric arc (commonly gas metal arc developed to control the energy accumulation and IT, e.g., the use of
welding – GMAW – with controlled short-circuit and/or pulse metal refrigerated CO2 jets [16], immersion tanks [17], and cooled substrates
transfer modes, plasma arc welding, or gas tungsten arc welding) as the [18]. However, these techniques increase the number of process pa­
heat source to melt the wire and build parts [4–7]. In this WAAM, a large rameters and are a challenging for large/complex geometries; therefore,
amount of heat is introduced into the part, particularly under high the IT is commonly controlled by increasing idle time [19] (natural heat
productivity/deposition rate conditions (high wire feed speed and cur­ transfer between the arc off – arc on cycle), despite this having a dele­
rent) [8]. Thus, during deposition, energy accumulates in the part, terious effect on the effective deposition rate (productivity). In contrast,
increasing the interpass temperature (IT). This is intensified by changes when multiple parts are deposited simultaneously, controlling the IT via

* Corresponding author.
E-mail address: [email protected] (F.W.C. Farias).

https://doi.org/10.1016/j.addma.2021.102387
Received 14 May 2021; Received in revised form 7 September 2021; Accepted 2 October 2021
Available online 6 October 2021
2214-8604/© 2021 Elsevier B.V. All rights reserved.
F.W.C. Farias et al. Additive Manufacturing 48 (2021) 102387

Fig. 1. (a) Experimental setup, (b) side view of the 20-layer single-bead wall (b), and (c) cross-section macrograph.

the idle time becomes economically favorable because the total idle time use of forced air cooling for various idle times and reported that the
becomes the sum of the layer deposition times of the other parts. Thus, forced air cooling effectively mitigated the heat accumulation. Mon­
the 3D printer does not stop deposition and high productivity is main­ tevecchi et al. [22] developed an algorithm to calculate the idle time
tained. In addition, despite the abovementioned deleterious effects of based on an FEM model; the results indicated that the idle time required
heat accumulation and an increase in the IT, Lu et al. [20] and Mis­ for IT stabilization increased continuously and had an asymptotic pro­
hchenko and Scotti [21] reported that controlled heat management can file. Xiong et al. [32] modeled the influence of the preheating temper­
promote stress relaxation during deposition and reduce the overall re­ ature on the IT and reported that the IT became almost constant under
sidual stress. preheating temperatures exceeding 400 ◦ C. Lei et al. [19] showed that
IT monitoring is commonly carried out using sensors, particularly the maximum temperature gradient increased with the idle time and
thermocouples welded to the substrate [22]. However, as the part height became almost constant after the 40th layer. Hejripour et al. [33]
increases during deposition, the substrate rapidly reaches a temperature controlled the phase proportion (δ-ferrite content) for a 2209 duplex
plateau, which can lead to erroneous interpretations of the temperature stainless steel wall and tube deposited by GMAW-WAAM using an FEM
evolution of the part. In addition, the use of multiple thermocouples at model and a thermal management methodology.
various part heights can have a similar effect and increase the opera­ The abovementioned studies described the temperature distribution,
tional difficulty (welding the thermocouples during the deposition) heat accumulation, and IT increase using FEM models. However, even
[23]. Non-contact sensors (e.g., infrared cameras) have also been the simplified models require long computational times, as noted by
applied for IT monitoring; however, the difference in emissivity between Ding et al. [34], because the instantaneous thermal profile and IT in each
the liquid and solid metals, oxidation of the previously deposited layers, layer require the use of transient thermal models (incremental temper­
the necessity of calibration, and surface roughness make it difficult to ature histories) with a small timestep (commonly 0.2–1 s for the “arc on”
measure the IT accurately [24–26]. Therefore, the use of the nonlinear load case) and relatively fine meshes (the maximum size around the
multiphysics finite element method (FEM) can more clearly elucidate molten pool/moving heat source width/penetration is restricted to
the experimentally complex or difficult to quantify effects that occur in ensure a minimum number of ~ 2–4 elements in areas of high thermal
WAAM, such as the increase in the IT and cooling rate during the gradient and heat flux). There have been efforts to reduce the FEM
deposition [27–29]. Oyama et al. [30] showed that an increase in the computational time for welding/WAAM [35,36]; however, this remains
idle time homogenized the temperature distribution, and heat source an open subject. The computational time can increase the FEM simula­
management (changing the deposition parameters) during deposition tion costs, which have led some authors to use FEM simulations coupled
was suitable for controlling the IT. Hackenhaar et al. [31] simulated the with data analysis techniques to describe the complex relationship

2
F.W.C. Farias et al. Additive Manufacturing 48 (2021) 102387

Table 1 with an idle time of 180 s. Second, several simulations with varying idle
Chemical composition of the AWS ER90S-B3 wire. times (30–240 s) are performed, and the layer number (input data), the
Element Chemical composition [wt%] idle time (input data), and IT at the midpoint of each layer (output data)
are used to train the ANN. Third, the ANN prediction is validated using
C 0.08
Mn 0.90 additional FEM simulations (not included in the training database).
Si 0.60 Finally, the ANN is used to build a process map (layer number × idle
Cr 2.50 time × interpass temperature).
Mo 1.00
Fe 94.92
2. Materials and methods

2.1. Materials and deposition

Table 2
Bead-on-plate and 20-layer single-bead wall deposition parameters.
The bead-on-plate and 20-layers single-bead wall were deposited
Parameter Value using a Fronius cold metal transfer (CMT) RCU 5000i machine with the
Deposition speed, v 10 mm∙s− 1 synergic curve CMT-1220 (available in the Fronius database). This curve
Wire feed speed 8 m∙min − 1 correlates the wire feed speed with the instantaneous arc power and
Average arc power, P 5400 W electric parameters (current and arc voltage) waveform [46]. The
Contact tip to work distance 15 mm welding torch was controlled using a KUKA KR90 robot. Fig. 1 shows the
1
Shielding gas / flow rate 98% Ar + 2% CO2 / 20 L∙min− experimental setup and the deposited wall. An AWS ER90S-B3 wire with
a diameter of 1.2 mm was used. Table 1 lists the chemical composition of
the wire, and Table 2 lists the deposition parameters. The average arc
power (P; Table 2) was calculated using Eq. 1. The current and voltage
oscillograms (Fig. 2) were obtained using an IMC SAP acquisition system
with sampling rate of 5 kHz.
∫ ttotal
1
P= Ui (t)Ii (t)dt (1)
ttotal 0

Where Ui , Ii , t, and ttotal are the instantaneous arc voltage (Fig. 2),
instantaneous current (Fig. 2), time, and total deposition time,
respectively.

2.2. WAAM modeling

A thermometallurgical model of 20-layers single-bead walls depos­

ited by WAAM was developed using the Simufact Welding® FEM com­
mercial software [47–49]. The model is based on the first law of
thermodynamics (energy conservation) and the solution of the solid
multiphase transient heat conduction equation for a moving heat source
Fig. 2. Current and arc voltage oscillograms. (Eq. 2) [45,50], where T is the temperature, v is the heat source velocity,
k is the thermal conductivity, ρ is the density, cp is the specific heat,
between the WAAM parameters and outputs results. For example, Fang and Q̇ is the volumetric heat source. The term − vρcp ∂∂Tx in Eq. 2 arises
et al. [37] developed a Box–Behnken response surface experimental from the transformation of the coordination system as the heat source
design with the FEM simulation results to correlate the substrate ge­ moves through the deposition path [50–53]. The phase fraction (P(T)i ) is
ometry (e.g., thickness and length) and IT, which allowed a reduction in solved at each time step using the FEM software phase transformation
the number of FEM simulations. The results indicated that the substrate model (Johnson-Mehl-Avrami-Kolmogorov – JMAK [54–56]), which
thickness had the most significant influence on the IT. Wu et al. [38] uses the time temperature transformation (TTT) and continuous cooling
used an FEM model to predict the residual stress in parts deposited by transformation (CCT) diagrams of the material. In addition, the isotropic
WAAM and an artificial neural network (ANN) to evaluate the hierar­ physical properties (k, ρ, and cp ; Eqs. 3 – 5) depend on the phase
chical influences of process variables (e.g., material properties, arc fraction and temperature, which are also updated at every time step. The
power, preheating temperature, scanning speed, and substrate thick­ physical properties (Fig. 3) and the CCT diagram (Fig. 4) were obtained
ness) on the residual stress. Shojaeefard et al. [39] optimized the friction using the JMatPro® thermodynamic simulation commercial software
stir welding parameters using an FEM model coupled with an ANN. This considering the wire chemical composition (Table 1) and the prior
procedure was also adopted to analyze and optimize laser-assisted austenite grain size of 15 µm. JMatPro® enables direct exportation of
bending [40], drawing [41], and injection [42] processes, as well as the material model (phase transformation and physical data) to the
the corrosion resistance [43] and other material properties [44]. Simufact Welding®. In addition, the thermal conductivity was artifi­
Given the detrimental effects of IT on WAAM (e.g., anisotropic cially increased (effective thermal conductivity) above the melting
properties and loss of dimensional control), the difficulty of IT mea­ temperature (~ 1511 ◦ C) to simulate the effects of convective heat
surement, and the high FEM computational cost to simulate large parts transfer in the weld pool [57–59], and the latent heat for melting was set
deposited by WAAM, the present study proposes the following: (1) to at a constant value of 221020 J/kg (extracted from JMatPro®).
analyze, through an FEM model, the influence of idle time on the IT of ( ) ( ) ( ) ( )
20-layer single-bead walls produced via WAAM and (2) to use the results ∂ ∂T ∂ ∂T ∂ ∂T ∂T ∂T
k + k + k + Q̇ = ρcp − v (2)
of this FEM model to train an ANN and predict the IT. First, the heat- ∂x ∂x ∂y ∂y ∂z ∂z ∂t ∂x
source (Goldak double-ellipsoid [45]) is initially calibrated (bead-on-­ ∑
plate); then, the FEM model is experimentally (thermocouples and an k= P(T)i k(T)i (3)
infrared camera) validated for a 20-layer single-bead wall deposited
i

3
F.W.C. Farias et al. Additive Manufacturing 48 (2021) 102387

Fig. 3. Density, specific heat capacity, and thermal conductivity (JMatPro®) of the AWS ER90S-B3 wire.

∑
ρ= P(T)i ρ(T)i (4)
i

∑
cp = P(T)i cp (T)i (5)
i

The initial condition assumes the substrate and external environment

temperatures (Text ) are 25 ◦ C (Eq. 6). The boundary condition considers
the radiation and convection heat fluxes on the surface (Eq. 7):
T(x, y, z, t = 0) = Text = 25 ◦
C (6)
〈 〉
∂T ∂T ∂T → ( )
− , , ⋅ n = εσ T 4 − T 4ext + h(T − Text ) (7)
∂x ∂y ∂z

where → n , σ, ε, and h are the unitary normal to surface vector, Ste­

fan–Boltzmann constant (5.6704 × 10− 8 W∙m− 2∙K− 4), emissivity of a
“gray” body (0.73), and the local heat exchange coefficient
(20 W∙m− 2∙K− 1), respectively. In addition, h is defined as 400 W m− 2
Fig. 4. Simulated (JMatPro®) continuous cooling transformation diagram of
the AWS ER90S-B3 wire. K− 1 on the surface in contact with the table (bearing; Fig. 6).
The CMT welding heat source is modeled using the Goldak et al. [60]
double-ellipsoid (Eqs. 8, 9, and 10):
( )
√̅̅̅ 2 y2 (z− vt)2
− 3 x2 − 3 − 3
6 3ff ηP a b2 c2
(8)
f
qf (x, y, z, t) = √̅̅̅e
abcf π π

4
F.W.C. Farias et al. Additive Manufacturing 48 (2021) 102387

Fig. 5. Schematic representation of the finite element method (FEM) model. The thermocouples and interpass temperature (IT) measurements are represented by the
white and orange points, respectively.

Fig. 7. Schematic representation of the multilayer artificial neural network

adopted in this study.

Fig. 6. Finite element model mesh; the substrate, 20-layers single-bead wall, ( )
clamps, and bearing are represented by the gray, blue, brown, and yellow √̅̅̅ −
2
3 x2 −
y2
3 2 −
(z− vt)2
3 2
colors, respectively. 6 3fr ηP a b cr
qr (x, y, z, t) = √̅̅̅e (9)
abcr π π

Table 3 Q̇ = qf + qr (10)
Goldak’s double-ellipsoid heat-source
parameters. where qf and qr represent the power density distribution in the front
Parameter Value and rear quadrants of the double-ellipsoid, respectively; v and t are the
a 1.0 mm deposition speed (Table 2) and time, respectively; parameters a (molten
b 3.8 mm pool depth), b (molten pool width), cf , and cr (Table 3) are correlated
cf 2.5 mm with the dimensions of the double-ellipsoid semi-axes. ff and fr (fr =
cr 6.5 mm 2− ff ) represent the energy fraction at the front and rear of the double-
ff 0.56 ellipsoid (Table 3), respectively; and η and P are the thermal efficiency
fr 1.44 (0.85) [61] and average arc power (Table 1), respectively.
A single-bead model representing the bead-on-plate experiment (see
Section 2.1) was used to calibrate and validate the heat-source param­
eters (see Section 3.1). This experiment included the use of thermo­
couples positioned at 2, 6, and 10 mm from the edge of the bead (white
points; Fig. 5), whose thermal cycles and macrograph (Fig. 7, see Section
3.1) were used to validate the simulated thermal results. The model for
the 20-layer single-bead wall (Figs. 5 and 6) was validated using ther­
mocouples (welded on the substrate and deposited wall) and an infrared
camera (located 25 cm from the wall in a sideview position). The ther­
mocouples on the substrate were welded at 2 and 6 mm from the edge of

5
F.W.C. Farias et al. Additive Manufacturing 48 (2021) 102387

Fig. 8. Simulated thermal profile (left) and macrograph of the bead-on-plate (right). The temperature scale is in ◦ C.

Fig. 9. Thermal cycles simulated by the FEM (○) and measured (—) by the
thermocouples located 2, 6, and 10 mm from the fusion line.

the first bead (assumed from the planned trajectory and previously
determined bead-on-plate width) and at the middle bead length; the wall
thermocouple, in the middle length of the 5th layer, immediately after
its deposition (during the idle time – 180 s – before the deposition of the
6th layer). An Optris PI 08M infrared camera with a resolution of
382 × 288 pixels, acquisition rate of 80 Hz, thermal resolution of 0.1 K,
and temperature range of 625–1900 ◦ C was used to acquire images of
half of the wall length during the deposition. The emissivity (0.73; the
same as in the model simulation) was selected based on previous Fig. 10. FEM simulation of the thermal profile for the bead-on-plate (top) and
high-speed image obtained during deposition of the bead-on-plate (bottom).
experience and is in good agreement with the resultant temperature
The red dashed line represents the fusion line (FL).
(~1502.5 ◦ C) near the visually detectable molten pool end. The points
for the thermal cycle analyses had dimensions of 2 × 2 pixels, and data
post-processing was performed using Pix Connect software. [65], a rectangular bead shape (6.1 × 2.7 mm2) was used to model each
Fig. 5 schematically depicts the developed model. The model consists deposited bead in the 20-layers single-bead wall. This method offers
of a substrate (15 × 150 × 300 mm3, with constant dimensions for all lower error, simpler meshing, and faster computation compared to the
simulations) and a 20-layers single-bead wall (54 × 6.2 × 200 mm3). arc-shaped bead method. In addition, the wall dimensions (width and
The orientation of the unidirectional deposition sequence was reversed height are 6.1 and 2.7 mm, respectively) were measured through the
in each layer (Fig. 5), i.e., the nth layer was deposited from left to right macrograph (Fig. 1c) and maintained constant for all layers and idle
(→) and the (n +1)th layer from right to left (←). This deposition times simulations. Despite the thermal condition (e.g., interpass tem­
sequence was adopted because of the imbalance in the heat accumula­ perature and idle time) effectively alter the bead geometry [66], Lu et al.
tion (i.e., the arc ignition point becomes hotter than the opposite side) [67], Montevecchi et al. [22], Geng et al. [68], Lee [69], and Xiong et al.
and loss of dimensional control in non-reversed deposition [62]. The [32] developed accurate thermal FEM models considering constant the
thermal cycle and IT were retrieved at midpoint of each layer (orange model geometry for all thermal condition simulated.
dots; Fig. 5). Owing to the symmetry of the 20-layers single-bead wall The FEM mesh consisted of 36,358 3D hexahedral elements and
model, the symmetry plan (Neuman boundary condition) along the 36,368 nodes (Fig. 6). The areas adjacent to the wall and substrate were
torch path was used to reduce the computational time. Similarly, finely meshed (element size = 0.9 × 0.9 mm2) to allow accurate tem­
following Prajadhiama et al. [63], Bock et al. [64], and Ahmad et al. perature prediction during the deposition and cooling (abrupt thermal

6
F.W.C. Farias et al. Additive Manufacturing 48 (2021) 102387

Fig. 11. Simulated and measured (thermocouples) thermal cycles in the 20-layer single-bead wall wire arc additive manufactured.

gradients) [70,71]. The wall elements were simulated using the data), as shown in Fig. 7. The ANN was developed using the Lev­
birth-and-death technique. This model was used to simulate the influ­ enberg–Marquardt backpropagation algorithm (TRAINLM network
ence of idle time (30, 45, 60, 90, 120, 150, 180, 210, and 240 s) on the training function) in the MATLAB® Neural Network toolbox. In addi­
IT. Moreover, the execution, documentation, and validation of the tion, the tangent sigmoid and mean square error were used as the
WAAM simulations followed the ISO 18166 standard. transfer and performance functions, respectively. As indicated by Xiong
et al. [77] and adopted in the present study, an ANN with one hidden
layer is suitable for describing the nonlinear correlation in welding and
2.3. Artificial neural network
WAAM because of its small database and restricted process window. The
number of neurons tested in the hidden layer varied from 8 to 14.
Many methods have been used to correlate the input and output
process parameters (e.g., linear regression, factorial design, response
3. Results and discussion
surface, and ANNs). Nevertheless, each method has an accuracy limi­
tation, and thus some linear regression and response surface methods
3.1. Heat-source calibration
are not suitable for highly nonlinear problems [72]. For optimization
problems, some techniques are also not applicable, as observed by
The Goldak’s double-ellipsoid parameters (Table 3) were a model
Dinovitzer et al. [73]. In addition, factorial designs with high factor and
input variable and needed adjustments for the adopted process param­
level numbers requires a large number of experiments and, considering
eters (Table 1). Initially, the bead-on-plate cross-section dimensions and
the simulation problem of the present study, long computational times.
morphology were used to describe the geometry of the model (Fig. 8)
In addition, ANNs have been shown to be suitable for welding and AM
and adjust the double-ellipsoid parameters (Table 3). In addition, the
processes prediction owing to the low number of experimental data
experimental thermal cycles were measured using thermocouples and
required and high accuracy of the trained models [74–76]. Thus, owing
compared with the data obtained from the simulation. As can be seen in
to the complex correlation between the idle time, layer number, and IT,
Fig. 9, the heat source was accurately calibrated and validated, with root
as well as the high computational cost of the FEM model, the use of
mean square errors (RMSEs) of 8.78, 10.36, and 5.41 between the
ANNs has been shown to be an efficient tool for decision-making and
simulated values and those measured by the thermocouple thermal cy­
process parameter selection. In the present study, an ANN was adopted
cles (Fig. 9). A similar methodology was also used by Bate et al. [78], as
to correlate the idle time and layer number (input data) with IT (output

7
F.W.C. Farias et al. Additive Manufacturing 48 (2021) 102387

thermal cycle. Finally, there may be a small error related to the differ­
ence in the exact location of the thermocouples and the mesh nodes in
the model (the points where the simulated thermal cycles were ob­
tained). Increasing the model fidelity would requires calibration of the
heat source and an evaluation of the bead shape for each layer, as well as
a possible mesh refinement; however, these were not practical owing to
the corresponding increase in the experimental matrix, modeling efforts,
and computational time.
To obtain the thermal cycles far from the substrate during the
deposition (close to the molten pool), an infrared camera was used
owing to the limitations of the thermocouples. Figs. 12 and 13 show the
thermographs during deposition and thermal cycles at the middle point
of the 1st, 11th, 15th, and 19th layers (orange dots in Fig. 5), respec­
tively. The simulated and measured thermal cycles also showed good
agreement, which complements the thermocouples results (Fig. 11) and
validated the heat source parameters and the 20-layers single-bead wall
FEM model. In addition, as expected, during the deposition, the highest
layer (layer 19; Fig. 12) became hotter owing to the heat accumulation
phenomena in WAAM [9,10]. This result was also observed by Froend
et al. [24] using an infrared camera. It is worth noting that owing to the
difference in emissivity between the liquid and solid metals [25], some
discrepancies relative to the simulation were observed for temperatures
higher than the melting temperature (1511 ◦ C). However, in the
solid-state (below the melting temperature), the temperatures measured
by the infrared camera and simulated were closer (RMSEs of 124.16,
38.39, 101.69, and 141.33 for the 1st, 11th, 15th, and 19th layers,
respectively) and had similar profiles, despite the influence of oxides
formed during deposition and surface roughness on emissivity [83].
Fig. 12. Infrared camera temperature profile during the deposition for the 1st, After the heat source and FEM model accurately validated, simulations
11th, 15th, and 19th layers. with idle times of 30, 45, 60, 90, 120, 150, 180, 210, and 240 s were
performed.
well as indicated by the ISO 18166 standard, to validate welding sim­
ulations. In addition, Fig. 10 compares the predicted molten pool length 3.2.2. Extended simulation
(20.02 mm) with that measured by the high-speed camera (18.82 mm); Fig. 14 outlines the influence of idle time on the energy accumulation
the resulting low error (9.1%) corroborates the above results. This error in a 20-layer single-bead wall produced by WAAM. As can be seen,
was attributed to the difficulty in obtaining a precise measurement from during deposition of the first few layers, there were no significant dif­
the experimental image, including an accurate fusion line (FL) defini­ ferences in heat accumulation between the conditions. However, with
tion, dimensional calibration using the nominal wire diameter the deposition of the part and the consequent reduction of the heat
(1.2 mm), and camera view distortions (parallax error) [79]. Moreover, transfer mechanism via conduction to the substrate, a lower idle time
the applied FEM model did not consider mass and moment transport in caused a higher energy accumulation and, consequently, increased the
the molten pool [80]. IT. An increase in the idle time thus enhanced the energy loss between
the deposition cycles and decreased the overall temperature. Fig. 15
shows the effects of idle time on the IT of each layer. As expected, a
3.2. WAAM simulation lower idle time significantly increased the IT. In addition, for idle time
≤120 s, the IT tended to reach a plateau, which decreased as the idle
3.2.1. Validation time increased. Conversely, idle times ≥150 s tended to quickly reach a
After validation of the heat source (Section 3.1), an FEM model was first IT threshold; then, as the number of layers increased, the IT tended
built (Section 2.2) consisting of a 20-layers single-bead wall deposited to rise again. This is attributed to a decrease in the efficiency of the
with the same parameters as the bead-on-plate (Table 1). In this model, conductive heat transfer mechanism [9,10]. Furthermore, as shown in
an idle time of 180 s was used, the same idle time was used for experi­ Fig. 14 and infrared images (Fig. 12), only the final layers remained
mental validation. Fig. 11 compares the thermal cycles obtained in the expressively heated; thus, the efficiency of the radiation and convection
simulation and measured (thermocouples); the results show good transfer mechanisms were also reduced even with an increase in the idle
agreement in terms of the overall heating and cooling rate (cycle profile) time. This resulted in the accumulation of energy in the part and a
between them, although some irregularities should be highlighted in consequent increase in IT. Similar to the IT, the cooling rate (Fig. 16)
terms of the peak temperature. First, it is worth noting that the ther­ was also influenced by the idle time and was similar in all conditions for
mocouples (at 2 and 6 mm from the wall) presented intense noise after the first five layers. However, as the wall was deposited, the difference in
the 6th layer, and only the clear part of the data is shown in Fig. 11. the cooling rate increase. The value was higher for longer idle times
Second, a progressive error (especially clear for the thermocouple in the owing to the higher temperature gradient in the part (Fig. 14). The
5th layer) observed in the peak temperature was attributed to the above results are corroborated by the experimental results reported by
variation in layer height (considered constant in the present model) that of Froend et al. [24], Yang et al. [25], and Silva et al. [26], which also
directly affected the distance of the peak isotherms [81]. This error used infrared cameras to evaluate the IT during the deposition.
could also be attributed to possible changes in the contact tip to work From a practical perspective, the AWS A5.28/A 5.28M:2005 stan­
distance that influenced the actual arc power and penetration of each dard (Specification for Low-Alloy Steel Wire for GMAW) indicates a
layer (both considered constant in the model) [82], which would also maximum IT of 215 ◦ C for ER90S-B3 wire (used in the present study).
affect the peak temperatures. Nonetheless, the thermocouple in the 5th Thus, considering the results of Fig. 15, idle times ≤120 s will result in
layer exhibited the same profile and good agreement with the simulated an IT higher than that specified by AWS A5.28/A 5.28M:2005, while idle

8
F.W.C. Farias et al. Additive Manufacturing 48 (2021) 102387

Fig. 13. Simulated and measured (infrared camera) thermal cycles during deposition of the 1st, 11th, 15th, and 19th layers.

Fig. 14. Temperature evolution of the single bead wall using the FEM model for idle times of 60, 120, and 180 s. The frames were taken from deposition of the 1st,
5th, 10th, 15th, and 20th layers from top to bottom, respectively. The temperature scale is in Celsius degree [◦ C].

times >120 s are a good choice that can balance productivity and microstructure along the build direction. However, it is worth noting
properties. Based on A5.28/A 5.28M:2005, Rafieazad et al. [84] reduced that each industrial application (e.g., oil and gas, aerospace, and auto­
the effects of consecutive thermal WAAM cycles on the microstructure of mobilist) have specific requirements, which also need to be considered
previously deposited layers (i.e., reduced the softened regions) for an during the selection of process parameters.
AWS ER70S-6 wire. The authors claimed that maintaining the IT in the
range specified by AWS A5.28/A 5.28 M:2005 promoted a homogenous

9
F.W.C. Farias et al. Additive Manufacturing 48 (2021) 102387

number of ANN neurons, the ANN results were compared (Fig. 19) with
those of the FEM model for idle times of 75, 135, and 195 s (not included
in the training database in Fig. 15). The ANN predictions of the IT were
similar to those of the FEM model (RMSE of 10.89, 6.58, and 3.91,
respectively), validating the ANN accuracy. Similar to the results ob­
tained by Xiong et al. [77] and Wu et al. [38], the ANN proved to be a
suitable tool for analyzing the complex relationship between the depo­
sition parameters and energy accumulation in WAAM parts. Given the
good agreement between the ANN and FEM results and the ability of the
ANN to enable rapid prediction of the results over a wide range of idle
times, a process map describing the influence of the idle time on the IT
was obtained (Fig. 20). This map can help manufacturers to regulate the
process and obtain a better compromise between productivity (low idle
times) and IT (quality [11–13, 87]).

4. Conclusions
Fig. 15. Influence of the idle time on the interpass temperature.
A finite element method (FEM) model was developed to estimate the
3.3. Interpass temperature prediction effect of the idle time (30, 45, 60, 90, 120, 150, 180, 210, and 240 s) on
the interpass temperature (IT) for a 20-layer single-bead wall deposited
First, a neuron number test was carried out (Fig. 17) to avoid over­ by wire arc additive manufacturing (WAAM). The FEM model was
fitting and/or underfitting [85,86]; the use of 12 neurons provided a accurately calibrated and experimentally validated for a single bead-on-
lower RMSE (Fig. 18) and the best performance. The database adopted plate and a 20-layer single-bead wall with and idle time of 180 s. The
for ANN training is shown in Fig. 15. After training and selecting the first layers exhibited similar energy accumulation; however, as the part

Fig. 16. Influence of the idle time on the cooling curve for the 5th (a), 10th (b), and 15th (c) layers. Moreover, the average cooling rate was calculated between 1300
and 800 ◦ C.

10
F.W.C. Farias et al. Additive Manufacturing 48 (2021) 102387

Fig. 17. Scatter plots of the finite element method (FEM) data versus the artificial neural network training test for various neuron numbers.

Fig. 18. Artificial neural network neuron number tests. Fig. 19. Comparison of the ITs predicted by the FEM model and the artificial
neural network (12 neurons) for idle times of 75, 135, and 195 s. The root mean
squared error (RMSE) were 10.89, 6.58, and 3.91, respectively.

11
F.W.C. Farias et al. Additive Manufacturing 48 (2021) 102387

Acknowledgements

The authors thank the LEAD/COPPE/UFRJ team for its contributions

to the ANN topics and Labsolda/UFSC for its partnership and high-speed
filming of the bead-on-plate deposition.

References

[1] B. Wu, Z. Pan, D. Ding, D. Cuiuri, H. Li, J. Xu, J. Norrish, A review of the wire arc
additive manufacturing of metals: properties, defects and quality improvement,
J. Manuf. Process. 35 (2018) 127–139, https://doi.org/10.1016/j.
jmapro.2018.08.001.
[2] Y. Li, Z. Feng, L. Hao, L. Huang, C. Xin, Y. Wang, E. Bilotti, K. Essa, H. Zhang, Z. Li,
F. Yan, T. Peijs, A review on functionally graded materials and structures via
additive manufacturing: from multi-scale design to versatile functional properties,
Adv. Mater. Technol. 5 (2020), 1900981, https://doi.org/10.1002/
admt.201900981.
[3] S.W. Williams, F. Martina, A.C. Addison, J. Ding, G. Pardal, P. Colegrove, Wire +
Arc additive manufacturing, Mater. Sci. Technol. 32 (2016) 641–647, https://doi.
org/10.1179/1743284715Y.0000000073.
[4] N. Rodriguez, L. Vázquez, I. Huarte, E. Arruti, I. Tabernero, P. Alvarez, Wire and
arc additive manufacturing: a comparison between CMT and TopTIG processes
applied to stainless steel, Weld. World 62 (2018) 1083–1096, https://doi.org/
10.1007/s40194-018-0606-6.
[5] T.A. Rodrigues, V.R. Duarte, R.M. Miranda, T.G. Santos, J.P. Oliveira, Ultracold-
Fig. 20. Influence of the idle time and layer number on the interpass temper­ wire and arc additive manufacturing (UC-WAAM), J. Mater. Process. Technol. 296
ature obtained using the ANN. (2021), 117196, https://doi.org/10.1016/j.jmatprotec.2021.117196.
[6] Z. Zeng, B.Q. Cong, J.P. Oliveira, W.C. Ke, N. Schell, B. Peng, Z.W. Qi, F.G. Ge,
W. Zhang, S.S. Ao, Wire and arc additive manufacturing of a Ni-rich NiTi shape
was built, the efficiency of the heat transfer mechanism via conduction memory alloy: microstructure and mechanical properties, Addit. Manuf. 32 (2020),
to the substrate was reduced, causing a rapid increase in IT for low idle 101051, https://doi.org/10.1016/j.addma.2020.101051.
times (30, 45, 60, 90, and 120 s). An increase in the idle time could [7] K.S. Derekar, A. Addison, S.S. Joshi, X. Zhang, J. Lawrence, L. Xu, G. Melton,
D. Griffiths, Effect of pulsed metal inert gas (pulsed-MIG) and cold metal transfer
ensure high energy losses during the deposition cycles and a decrease in (CMT) techniques on hydrogen dissolution in wire arc additive manufacturing
the overall temperature of the built part, which reduced the IT. In (WAAM) of aluminium, Int. J. Adv. Manuf. Technol. 107 (2020) 311–331, https://
addition, an artificial neural network (ANN) was trained using the FEM doi.org/10.1007/s00170-020-04946-2.
[8] I. Tabernero, A. Paskual, P. Álvarez, A. Suárez, Study on Arc welding processes for
results and cloud accurately predict the IT evolution during deposition. high deposition rate additive manufacturing, Procedia CIRP 68 (2018) 358–362,
The ANN was used to obtain a process map considering the idle time, https://doi.org/10.1016/j.procir.2017.12.095.
layer number (inputs), and IT (output). This process map will allow [9] F.M. Scotti, F.R. Teixeira, L.J. da Silva, D.B. de Araújo, R.P. Reis, A. Scotti, Thermal
management in WAAM through the CMT advanced process and an active cooling
manufacturers to regulate the process to obtain a suitable compromise
technique, J. Manuf. Process. 57 (2020) 23–35, https://doi.org/10.1016/j.
between the productivity (low idle times) and IT-related quality char­ jmapro.2020.06.007.
acteristics (e.g., cooling rate, mechanical behavior, microstructure, bead [10] J. Xiong, R. Li, Y. Lei, H. Chen, Heat propagation of circular thin-walled parts
fabricated in additive manufacturing using gas metal arc welding, J. Mater.
morphology, and roughness) of a single-bead-thin parts fabricated by
Process. Technol. 251 (2018) 12–19, https://doi.org/10.1016/j.
WAA. This study highlighted the synergic use of FEM simulation coupled jmatprotec.2017.08.007.
with data analysis techniques for process optimization and control, [11] X. Xu, S. Ganguly, J. Ding, C. Er, S. Williams, Enhancing mechanical properties of
which combine the precision and reduced cost of virtual experiments wire + arc additively manufactured INCONEL 718 superalloy through in-process
thermomechanical processing, Mater. Des. 160 (2018) 1042–1051, https://doi.
(FEM simulations) with fast and adaptive calculations (ANNs). org/10.1016/j.matdes.2018.10.038.
[12] J. Wang, X. Lin, J. Wang, H. Yang, Y. Zhou, C. Wang, Q. Li, W. Huang, Grain
Funding morphology evolution and texture characterization of wire and arc additive
manufactured Ti-6Al-4V, J. Alloy. Compd. 768 (2018) 97–113, https://doi.org/
10.1016/j.jallcom.2018.07.235.
Shell Brasil Petróleo Ltda (Shell Brasil; grant DPW019-W01) and [13] P. Dirisu, S. Ganguly, A. Mehmanparast, F. Martina, S. Williams, Analysis of
Empresa Brasileira de Pesquisa e Inovação Industrial (Embrapii; PCOP- fracture toughness properties of wire + arc additive manufactured high strength
low alloy structural steel components, Mater. Sci. Eng. A. 765 (2019), 138285,
1904.0016,) with support from Agência Nacional de Petróleo, Gás https://doi.org/10.1016/j.msea.2019.138285.
Natural e Biocombustíveis (ANP). [14] B. Wu, D. Ding, Z. Pan, D. Cuiuri, H. Li, J. Han, Z. Fei, Effects of heat accumulation
on the arc characteristics and metal transfer behavior in Wire Arc additive
manufacturing of Ti6Al4V, J. Mater. Process. Tech. 250 (2017) 304–312, https://
CRediT authorship contribution statement
doi.org/10.1016/j.jmatprotec.2017.07.037.
[15] C. Shen, M. Reid, K. Liss, Z. Pan, Y. Ma, D. Cuiuri, S. Van Duin, H. Li, Neutron
Francisco Werley Cipriano Farias: Conceptualization, methodol­ diffraction residual stress determinations in Fe3Al based iron aluminide
components fabricated using wire-arc additive manufacturing, 29 (2019) 100774.
ogy, validation, software, investigation, and Writing – original draft.
https://doi.org/10.1016/j.addma.2019.06.025.
João da Cruz Payão Filho: methodology, investigation, Writing – re­ [16] B. Wu, Z. Pan, D. Ding, D. Cuiuri, H. Li, Z. Fei, The effects of forced interpass
view & editing, and supervision. Victor Hugo Pereira Moraes and cooling on the material properties of wire arc additively manufactured Ti6Al4V
Oliveira: Conceptualization, software, methodology, validation, inves­ alloy 258 (2018) 97–105, https://doi.org/10.1016/j.jmatprotec.2018.03.024.
[17] L.J. da Silva, D.M. Souza, D.B. de Araújo, R.P. Reis, A. Scotti, Concept and
tigation, and Writing – review & editing. validation of an active cooling technique to mitigate heat accumulation in WAAM,
Int. J. Adv. Manuf. Technol. 107 (2020) 2513–2523, https://doi.org/10.1007/
Declaration of Competing Interest s00170-020-05201-4.
[18] L. Vázquez, N. Rodríguez, I. Rodríguez, E. Alberdi, P. Álvarez, Influence of
interpass cooling conditions on microstructure and tensile properties of Ti-6Al-4V
The authors declare that they have no known competing financial parts manufactured by WAAM, Weld. World 64 (2020) 1377–1388, https://doi.
interests or personal relationships that could have appeared to influence org/10.1007/s40194-020-00921-3.
[19] Y. Lei, J. Xiong, R. Li, Effect of inter layer idle time on thermal behavior for multi-
the work reported in this paper. layer single-pass thin-walled parts in GMAW-based additive manufacturing, Int. J.
Adv. Manuf. Technol. 96 (2018) 1355–1365, https://doi.org/10.1007/s00170-
018-1699-1.
[20] X. Lu, X. Lin, M. Chiumenti, M. Cervera, Y. Hu, X. Ji, L. Ma, H. Yang, W. Huang,
Residual stress and distortion of rectangular and S-shaped Ti-6Al-4V parts by

12
F.W.C. Farias et al. Additive Manufacturing 48 (2021) 102387

directed energy deposition: modelling and experimental calibration, Addit. Manuf. [46] C.G. Pickin, S.W. Williams, M. Lunt, Characterisation of the cold metal transfer
26 (2019) 166–179, https://doi.org/10.1016/j.addma.2019.02.001. (CMT) process and its application for low dilution cladding, J. Mater. Process.
[21] A. Mishchenko, A. Scotti, Residual stresses in arc welding: a holistic vision, Soldag. Technol. 211 (2011) 496–502, https://doi.org/10.1016/j.jmatprotec.2010.11.005.
e Insp. 23 (2018) 93–112, https://doi.org/10.1590/0104-9224/SI2301.10. [47] M. Bielik, Thermo-mechanical analysis of plasma-based additive manufacturing of
[22] F. Montevecchi, G. Venturini, N. Grossi, A. Scippa, G. Campatelli, Idle time Ti-6Al-4V components using Simufact Welding 8.0, Diss. Wien. (2020), https://doi.
selection for wire-arc additive manufacturing: a finite element-based technique, org/10.34726/hss.2020.65940.
Addit. Manuf. 21 (2018) 479–486, https://doi.org/10.1016/j.addma.2018.01.007. [48] M. Biegler, B.A.M. Elsner, B. Graf, M. Rethmeier, Geometric distortion-
[23] J.F. Wang, Q.J. Sun, H. Wang, J.P. Liu, J.C. Feng, Effect of location on compensation via transient numerical simulation for directed energy deposition
microstructure and mechanical properties of additive layer manufactured Inconel additive manufacturing, Sci. Technol. Weld. Join. 25 (2020) 468–475, https://doi.
625 using gas tungsten arc welding, Mater. Sci. Eng. A. 676 (2016) 395–405, org/10.1080/13621718.2020.1743927.
https://doi.org/10.1016/j.msea.2016.09.015. [49] K. Oyama, S. Diplas, M. M’hamdi, A.E. Gunnæs, A.S. Azar, Heat source
[24] M. Froend, V. Ventzke, N. Kashaev, B. Klusemann, J. Enz, Thermal analysis of wire- management in wire-arc additive manufacturing process for Al-Mg and Al-Si alloys,
based direct energy deposition of Al-Mg using different laser irradiances, Addit. Addit. Manuf. 26 (2019) 180–192, https://doi.org/10.1016/j.addma.2019.01.007.
Manuf. 29 (2019), 100800, https://doi.org/10.1016/j.addma.2019.100800. [50] J.-M. Bergheau, R. Fortunier. Finite Element Simulation of Heat Transfer, first ed..,,
[25] D. Yang, G. Wang, G. Zhang, Thermal analysis for single-pass multi-layer GMAW John Wiley & Sons,, Nova Jersey, 2013.
based additive manufacturing using infrared thermography, J. Mater. Process. [51] M. Arunkumar, V. Dhinakaran, N. Siva Shanmugam, Numerical prediction of
Technol. 244 (2017) 215–224, https://doi.org/10.1016/j.jmatprotec.2017.01.024. temperature distribution and residual stresses on plasma arc welded thin titanium
[26] R.H.G. E Silva, P.C.J. Rocha, M.B. Rodrigues, M. Pereira, D. Galeazzi, Analysis of sheets, Int. J. Model. Simul. 41 (2021) 146–162, https://doi.org/10.1080/
interlayer idle time as a temperature control technique in additive manufacturing 02286203.2019.1700089.
of thick walls by means of cmt and cmt pulse welding processes, Soldag. e Insp. 25 [52] M.N. Ozisik, Heat conduction, Wiley, 1993, 2., in: R.M. Cotta, M.D. Mikhailov
(2020) 1–9, https://doi.org/10.1590/0104-9224/SI25.01. (Eds.), Heat Conduction: - Lumped Analysis, Integral Transforms, Symbolic
[27] L.E. Lindgren, Finite element modeling and simulation of welding part 1: increased Computation, 2nd ed..,, Wiley, 1997.
complexity, J. Therm. Stress. 24 (2001) 141–192, https://doi.org/10.1080/ [53] D. Das, S. Bag, S. Pal, A finite element model for surface and volumetric defects in
01495730150500442. the FSW process using a coupled Eulerian–Lagrangian approach, Sci. Technol.
[28] P. Michaleris, Modeling metal deposition in heat transfer analyses of additive Weld. Join. 26 (2021) 412–419, https://doi.org/10.1080/
manufacturing processes, Finite Elem. Anal. Des. 86 (2014) 51–60, https://doi. 13621718.2021.1931760.
org/10.1016/j.finel.2014.04.003. [54] M. Fanfoni, M. Tomellini, The Johnson-Mehl-Avrami-Kolmogorov model: a brief
[29] T.R. Walker, C.J. Bennett, T.L. Lee, A.T. Clare, A novel numerical method to predict review, Nuovo Cim. Della Soc. Ital. Di Fis. D. - Condens. Matter, . Mol. Chem. Phys.,
the transient track geometry and thermomechanical effects through in-situ Biophys. 20 (1998) 1171–1182, https://doi.org/10.1007/BF03185527.
modification of the process parameters in Direct Energy Deposition, Finite Elem. [55] F. Liu, G.C. Yang, J.N. Liu, Comparison between an analytical model and JMA
Anal. Des. 169 (2020), 103347, https://doi.org/10.1016/j.finel.2019.103347. kinetics for isothermally and isochronally conducted transformations,
[30] K. Oyama, S. Diplas, M. M’hamdi, A.E. Gunnæs, A.S. Azar, Heat source Thermochim. Acta 438 (2005) 83–89, https://doi.org/10.1016/j.tca.2005.08.019.
management in wire-arc additive manufacturing process for Al-Mg and Al-Si alloys, [56] X. Lei, J. Huang, X. Jin, S. Chen, X. Zhao, Application of Johnson-Mehl-Avrami-
Addit. Manuf. 26 (2019) 180–192, https://doi.org/10.1016/j.addma.2019.01.007. Kolmogorov type equation in non-isothermal phase process: re-discussion, Mater.
[31] W. Hackenhaar, J.A.E. Mazzaferro, F. Montevecchi, G. Campatelli, An Lett. 181 (2016) 240–243, https://doi.org/10.1016/j.matlet.2016.06.039.
experimental-numerical study of active cooling in wire arc additive manufacturing, [57] A.S. Shrestha, B. Cheng, K. Chou, An investigation into melt pool effective thermal
J. Manuf. Process. 52 (2020) 58–65, https://doi.org/10.1016/j. conductivity for thermal modeling of powder-bed electron beam additive
jmapro.2020.01.051. manufacturing. International Solid Freeform Fabrication Symposium-An Additive
[32] J. Xiong, Y. Lei, R. Li, Finite element analysis and experimental validation of Manufacturing Conference. (2016) 207–218.
thermal behavior for thin-walled parts in GMAW-based additive manufacturing [58] A. De, T. DebRoy, A smart model to estimate effective thermal conductivity and
with various substrate preheating temperatures, Appl. Therm. Eng. 126 (2017) viscosity in the weld pool, J. Appl. Phys. 95 (2014) 5230–5240, https://doi.org/
43–52, https://doi.org/10.1016/j.applthermaleng.2017.07.168. 10.1063/1.1695593.
[33] F. Hejripour, F. Binesh, M. Hebel, D.K. Aidun, Thermal modeling and [59] Q. Liu, L. Wang, L. Zhang, L. Cao, X. Ding, M. Liang, Y. Qi, Mathematical model of
characterization of wire arc additive manufactured duplex stainless steel, J. Mater. heat transfer for bloom continuous casting, J. Univ. Sci. Technol. Beijing Miner.
Process. Technol. 272 (2019) 58–71, https://doi.org/10.1016/j. Metall. Mater. 15 (2008) 17–23, https://doi.org/10.1016/S1005-8850(08)60004-
jmatprotec.2019.05.003. 3.
[34] J. Ding, P. Colegrove, J. Mehnen, S. Ganguly, P.M.S. Almeida, F. Wang, [60] J.A. Goldak, A. Chakravarti, M. Bibby, A new finite element model for welding heat
S. Williams, Thermo-mechanical analysis of Wire and arc additive layer sources, Met. Mater. Trans. B 15 (1984) 299–305, https://doi.org/10.1007/
manufacturing process on large multi-layer parts, Comput. Mater. Sci. 50 (2011) BF02667333.
3315–3322, https://doi.org/10.1016/j.commatsci.2011.06.023. [61] N. Pépe, S. Egerland, P.A. Colegrove, D. Yapp, A. Leonhartsberger, A. Scotti,
[35] H. Huang, N. Ma, J. Chen, Z. Feng, H. Murakawa, Toward large-scale simulation of Measuring the process efficiency of controlled gas metal arc welding processes, Sci.
residual stress and distortion in wire and arc additive manufacturing, Addit. Technol. Weld. Join. 16 (2011) 412–417, https://doi.org/10.1179/
Manuf. 34 (2020), https://doi.org/10.1016/j.addma.2020.101248. 1362171810Y.0000000029.
[36] L.-E. Lindgren, A. Malmelöv, A. Lundbäck, History reduction by lumping for time- [62] V.R. Duarte, T.A. Rodrigues, N. Schell, R.M. Miranda, J.P. Oliveira, T.G. Santos,
efficient simulation of additive manufacturing, Met. (Basel). 10 (2020) 1–17, Hot Forg. wire arc Addit. Manuf. (HF-WAAM), Addit. Manuf. 35 (2020), 101193,
https://doi.org/10.3390/met10010058. https://doi.org/10.1016/j.addma.2020.101193.
[37] X. Fang, L. Zhang, J. Yang, H. Bai, L. Zhao, K. Huang, B. Lu, Effect of characteristic [63] K.P. Prajadhiama, Y.H. Manurung, Z. Minggu, F.H. Pengadau, M. Graf, A. Haelsig,
substrate parameters on the deposition geometry of CMT additive manufactured T.-E. Adams, H.L. Choo, Development of bead modelling for distortion analysis
Al-6.3%Cu alloy, Appl. Therm. Eng. 162 (2019), 114302, https://doi.org/ induced by wire arc additive manufacturing using FEM and experiment, MATEC
10.1016/j.applthermaleng.2019.114302. Web Conf. 269 (2019) 05003, https://doi.org/10.1051/matecconf/
[38] Q. Wu, T. Mukherjee, A. De, T. DebRoy, Residual stresses in wire-arc additive 201926905003.
manufacturing – hierarchy of influential variables, Addit. Manuf. 35 (2020), [64] F.E. Bock, J. Herrnring, M. Froend, J. Enz, N. Kashaev, B. Klusemann, Experimental
https://doi.org/10.1016/j.addma.2020.101355. and numerical thermo-mechanical analysis of wire-based laser metal deposition of
[39] M.H. Shojaeefard, M. Akbari, P. Asadi, Multi objective optimization of friction stir Al-Mg alloys, J. Manuf. Process. 64 (2021) 982–995, https://doi.org/10.1016/j.
welding parameters using FEM and neural network, Int. J. Precis. Eng. Manuf. 15 jmapro.2021.02.016.
(2014) 2351–2356, https://doi.org/10.1007/s12541-014-0600-x. [65] S.N. Ahmad, Y.H.P. Manurung, M.F. Mat, FEM Simulation Procedure for Distortion
[40] B.N. Fetene, R. Shufen, U.S. Dixit, FEM-based neural network modeling of laser- and Residual Stress Analysis of Wire Arc Additive Manufacturing FEM Simulation
assisted bending, Neural Comput. Appl. 29 (2018) 69–82, https://doi.org/ Procedure for Distortion and Residual Stress Analysis of Wire Arc Additive
10.1007/s00521-016-2544-9. Manufacturing, (2020). 〈https://doi.org/10.1088/1757–899X/834/1/012083〉.
[41] J.C. Lin, Using FEM and neural network prediction on hydrodynamic deep drawing [66] S. Manokruang, F. Vignat, M. Museau, M. Limousin, Process parameters effect on
of T-piece maximum length, Finite Elem. Anal. Des. 39 (2003) 445–456, https:// weld beads geometry deposited by wire and arc additive manufacturing (WAAM),
doi.org/10.1016/S0168-874X(02)00092-6. in: L. Roucoules, M. Paredes, B. Eynard, P. Morer Camo, C. Rizzi (Eds.), Adv. Mech.
[42] K.S. Lee, J.C. Lin, Design of the runner and gating system parameters for a multi- Des. Eng. Manuf. III, Springer International Publishing, Cham, 2021, pp. 9–14.
cavity injection mould using FEM and neural network, Int. J. Adv. Manuf. Technol. [67] X. Lu, M.V. Li, H. Yang, Simulation of precipitates evolution driven by non-
27 (2006) 1089–1096, https://doi.org/10.1007/s00170-004-2287-0. isothermal cyclic thermal history during wire and arc additive manufacturing of
[43] R.C.C. Silva, J.N.C. Guerreiro, A.F.D. Loula, A study of pipe interacting corrosion IN718 superalloy, J. Manuf. Process. 65 (2021) 258–270, https://doi.org/10.1016/
defects using the FEM and neural networks, Adv. Eng. Softw. 38 (2007) 868–875, j.jmapro.2021.03.032.
https://doi.org/10.1016/j.advengsoft.2006.08.047. [68] R. Geng, J. Du, Z. Wei, S. Xu, N. Ma, Modelling and experimental observation of the
[44] A. Saoudi, M. Fellah, N. Hezil, D. Lerari, F. Khamouli, L. Atoui, K. Bachari, deposition geometry and microstructure evolution of aluminum alloy fabricated by
J. Morozova, A. Obrosov, M. Abdul Samad, Prediction of mechanical properties of wire-arc additive manufacturing, J. Manuf. Process. 64 (2021) 369–378, https://
welded steel X70 pipeline using neural network modelling, Int. J. Press. Vessel. Pip. doi.org/10.1016/j.jmapro.2021.01.037.
186 (2020), 104153, https://doi.org/10.1016/j.ijpvp.2020.104153. [69] S.H. Lee, CMT-based wire arc additive manufacturing using 316L stainless steel:
[45] J.A. Goldak, M. Akhlaghi. Computational Welding Mechanics, first ed..,, Springer,, effect of heat accumulation on the multi-layer deposits, Met. (Basel). 10 (2020)
US, 2005. 278, https://doi.org/10.3390/met10020278.
[70] P.H.G. Dornelas, F.W.C. Farias, V.H.P.M. e Oliveira, D. de Oliveira Moraes, P.
Z. Júnior, J. da Cruz Payão Filho, Influence of welding interpass temperature on

13
F.W.C. Farias et al. Additive Manufacturing 48 (2021) 102387

Charpy V-notch impact energy of coarse-grain heat-affected zone of AISI 4130 steel [79] J. Xiong, Y. Liu, Z. Yin, Passive vision measurement for robust reconstruction of
pipe, Int. J. Adv. Manuf. Technol. 108 (2020) 2197–2211, https://doi.org/ molten pool in wire and arc additive manufacturing, Measurement 153 (2020),
10.1007/s00170-020-05542-0. 107407, https://doi.org/10.1016/j.measurement.2019.107407.
[71] C. Heinze, C. Schwenk, M. Rethmeier, Influences of mesh density and [80] D. Wu, S. Tashiro, Z. Wu, K. Nomura, X. Hua, M. Tanaka, Analysis of heat transfer
transformation behavior on the result quality of numerical calculation of welding and material flow in hybrid KPAW-GMAW process based on the novel three
induced distortion, Simul. Model. Pract. Theory 19 (2011) 1847–1859, https://doi. dimensional CFD simulation, Int. J. Heat. Mass Transf. 147 (2020), 118921,
org/10.1016/j.simpat.2011.05.001. https://doi.org/10.1016/j.ijheatmasstransfer.2019.118921.
[72] L. Pronzato, A. Pázman, Design of experiments in nonlinear models (https://link. [81] Y. Yehorov, L.J. da Silva, A. Scotti, Balancing WAAM production costs and wall
springer.com/book/), Lect. Notes Stat. 212 (2013) 1–404, https://doi.org/ surface quality through parameter selection: a case study of an Al-Mg5 alloy
10.1007/978-1-4614-6363-4. multilayer-non-oscillated single pass wall, J. Manuf. Mater. Process 3 (2019),
[73] M. Dinovitzer, X. Chen, J. Laliberte, X. Huang, H. Frei, Effect of wire and arc https://doi.org/10.3390/jmmp3020032.
additive manufacturing (WAAM) process parameters on bead geometry and [82] A. Scotti, V. Ponomarev, W. Lucas, Interchangeable metal transfer phenomenon in
microstructure, Addit. Manuf. 26 (2019) 138–146, https://doi.org/10.1016/j. GMA welding: features, mechanisms, classification, J. Mater. Process. Technol. 214
addma.2018.12.013. (2014) 2488–2496, https://doi.org/10.1016/j.jmatprotec.2014.05.022.
[74] C. Rubio-Ramirez, D.F. Giarollo, J.E. Mazzaferro, C.P. Mazzaferro, Prediction of [83] H.-J. Jo, J.L. King, K. Blomstrand, K. Sridharan, Spectral emissivity of oxidized and
angular distortion due GMAW process of thin-sheets Hardox 450® steel by roughened metal surfaces, Int. J. Heat. Mass Transf. 115 (2017) 1065–1071,
numerical model and artificial neural network, J. Manuf. Process. 68 (2021) https://doi.org/10.1016/j.jmatprotec.2017.01.024.
1202–1213, https://doi.org/10.1016/j.jmapro.2021.06.045. [84] M. Rafieazad, M. Ghaffari, A.V. Nemani1, A. Nasiri, Microstructural evolution and
[75] D. Zhao, D. Ren, K. Zhao, S. Pan, X. Guo, Effect of welding parameters on tensile mechanical properties of a low-carbon low-alloy steel produced by wire arc
strength of ultrasonic spot welded joints of aluminum to steel – by experimentation additive manufacturing, Int. J. Adv. Manuf. Technol. 105 (2019) 2121–2134,
and artificial neural network, J. Manuf. Process. 30 (2017) 63–74, https://doi.org/ https://doi.org/10.1007/s00170-019-04393-8.
10.1016/j.jmapro.2017.08.009. [85] H. Demuth, M. Beale, Neural network toolbox-for use with MATLAB, Math. Works
[76] R. Hong, L. Zhang, J. Lifton, S. Daynes, J. Wei, S. Feih, W.F. Lu, Artificial neural Inc. (1998).
network-based geometry compensation to improve the printing accuracy of [86] Q. Wang, W. Jiao, P. Wang, Y.M. Zhang, A tutorial on deep learning-based data
selective laser melting fabricated sub-millimetre overhang trusses, Addit. Manuf. analytics in manufacturing through a welding case study, J. Manuf. Process (2020)
37 (2021), 101594, https://doi.org/10.1016/j.addma.2020.101594. 1–12, https://doi.org/10.1016/j.jmapro.2020.04.044.
[77] J. Xiong, G. Zhang, J. Hu, L. Wu, Bead geometry prediction for robotic GMAW- [87] J. Xiong, Y. Li, R. Li, Z. Yin, Influences of process parameters on surface roughness
based rapid manufacturing through a neural network and a second-order of multi-layer single-pass thin-walled parts in GMAW-based additive
regression analysis, J. Intell. Manuf. 25 (2014) 157–163, https://doi.org/10.1007/ manufacturing, J. Mater. Process. Technol. 252 (2018) 128–136, https://doi.org/
s10845-012-0682-1. 10.1016/j.jmatprotec.2017.09.020.
[78] S.K. Bate, R. Charles, A. Warren, Finite element analysis of a single bead-on-plate
specimen using SYSWELD, Int. J. Press. Vessel. Pip. 86 (2009) 73–78, https://doi.
org/10.1016/j.ijpvp.2008.11.006.

14