Finite Elements in Analysis and Design 221 (2023) 103949

Contents lists available at ScienceDirect

Finite Elements in Analysis & Design

journal homepage: [Link]/locate/finel

Towards an automated framework for the finite element computational

modelling of directed energy deposition
Dejan Kovšca, Bojan Starman, Damjan Klobčar, Miroslav Halilovič, Nikolaj Mole *
Faculty of Mechanical Engineering, University of Ljubljana, Aškerčeva 6, SI-1000, Ljubljana, Slovenia

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

Keywords: In the present work, a comprehensive framework for finite element-based computational modelling of Directed
Directed energy deposition Energy Deposition (DED) process is presented. The proposed approach can be fully automated and implemented
Multi-pass welding on a complex real-life part geometry to accurately predict a thermo-mechanical response during the full-scale
Computational modelling
deposition process. The discrete material deposition modelling in Finite Element Analysis (FEA) leads to artifi­
Finite element method
Thermo-mechanical analysis
cial increases in temperature gradients in the melt pool domain. A new method is therefore proposed that aims to
mitigate these gradients. Additionally, an easy-to-implement free-surface detection algorithm to accurately
prescribe the evolving heat transfer boundary conditions is presented. A three-dimensional sequentially coupled
thermo-mechanical model of the process is then validated against experimental data obtained in a deposition
case study. The simulation results show good agreement with the in-situ temperature measurements taken during
the actual deposition. In addition, result analysis showed that the largest tensile residual stresses form in the
hoop and axial direction on the outer domain of the thin-wall cylindrical part near the base plate while the
inward material is compressed.

1. Introduction and the actual distortion of the part during AM processes [2]. Such
simulations can also be utilised to optimise the process parameters [3]
Additive Manufacturing (AM) is a manufacturing technology and predict the microstructural evolution through phase-field modelling
increasingly used for rapid prototyping and producing near-net-shape [4]. In addition, several researchers have studied the effects of path
parts with intricate geometry in a wide variety of industries and appli­ strategy on residual stress and final part distortion [5,6]. For example,
cations. The desired shape is often built layer-by-layer through the Afazov et al. [7] used Finite Element Analysis (FEA) to compensate for
deposition of various materials. Wire-Arc Additive Manufacturing the shrinkage of the final part by modifying its initial geometry.
(WAAM) is a type of Directed Energy Deposition (DED) process where a Similarly, Nguyen et al. [8] developed a method to correct the cen­
consumable metal wire is deposited using an electric arc as a heat source tral axis distortion of a thin-wall hollow section by applying a virtual
[1]. In contrast to the conventional subtractive manufacturing processes bending moment. Furthermore, state-of-the-art numerical simulations
and other AM processes such as Laser Powder Bed Fusion (LPBF), it can be employed to compensate and mitigate the distortions of the
enables high deposition rates and thus reduced fabrication time, higher produced part in the first place. For example, Ramani et al. [9] presented
material utilisation and low operational costs. The heterogeneous heat a new approach for specifying LPBF scan sequences to attain a more
distribution in metal deposition does, however, lead to the well-known homogeneous temperature field distribution and reduce residual
issue of thermally induced residual stresses. These stresses result in se­ stresses and strains. Similarly, Sun et al. [10] proposed a bead
vere workpiece distortion, as well as other defects, such as thermal sequence-driven method to evaluate and optimise the deposition
cracking or delamination of the deposited layers. pattern. Meanwhile, Montevecchi et al. [11] developed an FEA-based
Due to the complex physical phenomena, computational modelling algorithm to compute specific dwell times for the deposition of each
of DED has proven to be an essential tool for quality improvement and a layer to ensure a constant interlayer temperature.
better understanding of the deposition process. Numerical simulations Currently, many commercial FEA software packages offer thermo-
have found wide use in the prediction of thermal cycles, stress evolution mechanical computational modelling of AM processes. However, their

* Corresponding author.
E-mail address: [Link]@[Link] (N. Mole).

[Link]
Received 6 January 2023; Received in revised form 2 March 2023; Accepted 30 March 2023
Available online 3 May 2023
0168-874X/© 2023 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license ([Link]
nc-nd/4.0/).
D. Kovšca et al. Finite Elements in Analysis & Design 221 (2023) 103949

capabilities are mostly limited to depositing a whole layer of FE simul­ shown in Fig. 1. The worktable moves in three directions under the fixed
taneously, obtaining an average temperature field evolution [12]. To deposition torch, where overall build volume dimensions are 165 mm ×
capture the local thermal history, researchers still have to develop 180 mm x 220 mm. The interlayer temperature control and forced
high-fidelity deposition models on their own. Such a task, however, water-cooling of the base plate improve the system productivity. In
requires the user to have extensive Finite Element Method (FEM) addition, the system enables higher geometric accuracy through online
knowledge and coding expertise. layer height control and in-process toolpath replanning [16]. All system
To date, many researchers have successfully implemented complex operations are controlled and synchronised by an open-source
models to simulate the deposition process. In many cases, however, LinuxCNC motion controller.
these models are not general purpose, requiring considerable additional
coding to adapt them for a specific study case. This becomes distinctly
2.2. Description of the thin-wall cylindrical part geometry
harder and less practical when simulating manufacturing of real-life
products with complex geometry, whether due to meshing ability, FE
A 10-layer thin-wall cylindrical part with an outside diameter of 52
activation criteria, or just accurately prescribing thermal boundary
mm was built on a 5 mm thick square base plate with dimensions 120
conditions to the model. Specifically, many researchers face problems
mm × 120 mm (Fig. 2). By employing the WAAM process parameters
accurately defining heat loss conditions on the constantly evolving free
presented in Table 1, the final dimensions of a welded layer were 4.2 mm
surfaces of the model. Convection on the currently deposited layer is
in width and 1.5 mm in height. Each layer deposition started at the same
therefore frequently neglected [13] or applied by stopping the simula­
location on the circle after a 30 s interlayer cooling time.
tion and recalculating the free-surfaces at each time step [12].
We clamped the S235 structural steel square base plate to the
Furthermore, thermal boundary conditions are often implemented using
worktable on opposite sides. Once clamped, the material deposition was
tie contact between the beads, activating them at once for the whole
done in a counterclockwise direction with a 1.2 mm diameter G3Si1
bead surface [14], which can lead to many numerical problems.
welding wire using a low spatter welding protocol and a shielding gas
This paper presents a fully automated approach to advanced
mixture of Ar + 18% CO2. The chemical compositions of the two ma­
computational modelling of DED process that is compatible with most
terials used in the experiment are given in Table 2, while Fig. 3 shows
commercially available FEA-solving software packages. Novel in our
the completed part and the clamping arrangement of the base plate.
model is a new, easy-to-implement free-surface detection algorithm that
can accurately prescribe heat-loss boundary conditions during deposi­
tion, with no extra computational effort needed during the analysis. The 2.3. In situ temperature measurement
proposed model can therefore accurately predict the thermo-mechanical
response of a complex part during and after a full-scale deposition During the deposition, we measured the surface temperature of each
process. First, we present the experimental setup of the WAAM process layer using a K-type thermocouple, a National Instruments (NI) CDAQ
for a 10-layer thin-wall cylindrical component build. Next, we describe a 9174 chassis, an NI 9213 temperature measuring card and Labview
FE-based thermo-mechanical model validated against in-situ temporal software on a second control computer. Throughout the welding pro­
temperature variation measurements. Finally, we discuss the model’s cess, the thermocouple was manually placed on top of the deposited
ability to predict thermomechanical behaviour, finishing with an anal­ material and removed while the deposition torch passed the measuring
ysis of part distortion and residual stress field distribution. location, which was set approximately 20 mm from the welding start
point (Fig. 3). This procedure was repeated with each added layer using
2. Experimental study a thermally conductive paste on the contact between the thermocouple
and the material surface. Later, we used the measured layer-by-layer
2.1. Experimental setup temporal temperature variation to validate the proposed numerical
model.
We performed the experimental case study of the welding process
using an in-house developed WAAM prototype system [15]. The welding 3. DED process modelling
source is a Daihen Welbee P500L with a CNC positioning mechanism, as
In this section, we present the macroscopic thermo-mechanical
modelling of the DED process. We developed the FE-based numerical
model in SIMULIA Abaqus/Standard software using an implicit inte­
gration scheme. The thermal and mechanical problems are solved
sequentially as weakly coupled (i.e., the mechanical response does not
affect the thermal history) to reduce computational effort [17].
The flowchart presents the proposed computational model of the
DED process in Fig. 4. The modelling process starts with the preparation
of part geometry, represented by the base plate and the welded struc­
ture. Then we mesh the part in SIMULIA Abaqus/CAE software. At this
point, we also export the geometry to an open-source slicing and path

Fig. 1. The experimental setup for the WAAM prototype system consists of the
Daihen Welbee P500L welding source and a three-axis CNC positioning Fig. 2. Thin-wall cylindrical part geometry is composed of a square base plate
table [15]. and 10 deposited layers.

2
D. Kovšca et al. Finite Elements in Analysis & Design 221 (2023) 103949

Table 1
Welding process parameters used in the experimental case
study.
Parameter Value

Welding velocity (mm/min) 240.0

Interlayer dwell time (s) 30.0
Welding current (A) 65.0
Welding voltage (V) 14.8
Wire diameter (mm) 1.2
Gas flow (l/min) 12.0

Table 2
Chemical composition of G3Si1 (EN 440) filler wire and S235 (EN 10025) base
plate.
Material C Si Mn P S Fe

G3Si1 0.08% 0.90% 1.50% <0.025% <0.025% Balance

S235 <0.22% <1.60% <1.60% <0.050% <0.050% Balance

Fig. 4. Proposed flowchart for the complete DED process modelling approach
consists of nine main consecutive steps.

material properties to quiet FEs that do not affect the analysis results.
However, we should note that all FEs representing the deposited regions
are present in the analysis from the start. As a result, the number of
unknowns in the numerical model remains constant throughout the
computation, and the procedure is known to be time-consuming.
In contrast to the Quiet FE method, only the active degrees of
freedom are calculated in the Inactive FE method. The inactive FEs are
not included in the analysis until they are activated. Although this
approach reduces the computational effort, each activation step requires
Fig. 3. The completed thin-wall part with the base plate clamped to the additional equation re-numbering and frequent solver initialisations.
worktable. The temperature measuring location was set approximately 20 mm Our research adopts a hybrid Inactive/Quiet activation strategy
from the welding start point in a counterclockwise (welding) direction. proposed by Michaleris [18]. In his approach, all FEs representing the
deposited material are initialised in an inactive state by excluding them
generation software, e.g., Ultimaker Cura engine, where a G-code data from the numerical model. Then they are sequentially introduced into
file is generated, taking into consideration the actual welding process the analysis as quiet FEs using a layer-by-layer technique. Finally, in­
parameters. This file is used as an input for the CNC machine controlling dividual FEs are activated when the activation criteria are satisfied. This
of the actual welding deposition, as well as for the in-house developed strategy results in improved computation efficiency and smoother FE
Python/Fortran software for automatic FE activation and free-surface activation. In Abaqus/Standard, the Model change function is used to
detection during the simulation of the DED process. The latter is implement the Inactive FE method, supplemented with the USDFLD
accomplished through pre-processing the FE activation criteria (i.e., Abaqus/Standard subroutine for quiet FE activation. Quiet FE material
activation time) and adjacent FE identification from the generated FE properties and activation criteria are discussed in the following sections.
mesh and the G-code data file. In the next step, we use the developed
Python code to automatically prepare the SIMULIA Abaqus/Standard 3.2. Material properties
input files for weakly coupled transient heat transfer and quasi-static
mechanical analysis. After the heat transfer analysis, we mitigate the Temperature-dependent material properties of the welding wire
artificial temperature gradients with our newly proposed technique. (G3Si1) utilised in the study are taken from literature [19]. The data in
Finally, we import a smoothed temporal temperature evolution map as a Table 3 and Table 4 correspond to the dendritic microstructure in the
thermal load in the mechanical analysis. welded structure. Since the chemical compositions of the filler and the
base plate material are similar (c.f. Table 2), we assumed identical
3.1. Material deposition modelling properties for both materials. We also assumed the properties to be
homogeneous and isotropic. During the analysis, the values between the
In FEA, continuous material deposition modelling is achieved by tabulated values were linearly interpolated.
discrete and sequential activation of individual FE segments (i.e., the The material used has a characteristic microstructure phase trans­
birth and death technique). This can be accomplished using one of the formation, observed as a rise of the specific heat in the temperature
two available strategies: the Quiet FE Method and the Inactive FE Method interval between 700 ◦ C and 800 ◦ C. The melting and solidification
[18]. The former is based on assigning unique thermo-mechanical process is modelled by directly defining the latent heat of fusion as 100

3
D. Kovšca et al. Finite Elements in Analysis & Design 221 (2023) 103949

Table 3
Temperature-dependent material properties of G3Si1 (EN 440) welding wire.
Temperature Density (kg/ Thermal Conductivity (W/(m Specific Heat Capacity (J/(kg Thermal Expansion Coefficient (μm/(m Young Modulus
(◦ C) m3) K)) K)) K)) (MPa)

20 7700 42.0 480 12.0 212,000

100 – 41.7 482 12.9 206,000
200 – 39.7 515 14.0 198,000
400 7680 38.2 610 14.5 180,000
500 – 35.2 680 14.6 168,000
600 – 32.0 775 15.0 154,600
700 – 30.0 985 14.5 142,000
730 – – 1050 14.3 –
760 7670 – 1580 14.2 –
770 – – 1580 14.2 –
800 – 29.5 940 13.9 125,000
870 – – 985 13.6 –
900 – 27.0 615 13.6 112,100
1000 7580 27.7 632 13.6 103,000
1200 – 29.9 675 15.7 84,300
1300 – 30.8 690 16.8 –
1400 7380 32.5 700 17.7 69,700
1450 – 32.7 705 18.3 200
1500 7250 33.0 710 19.0 50
Quiet 1e− 3 1e− 6 1e8 1e− 6 5

mesh consists of 32,000 FEs, with mesh refined in the vicinity of the
Table 4 deposition domain, as illustrated in Fig. 5. The mesh of each welding
Temperature-dependent yield stress of G3Si1 (EN 440) welding wire at specific
layer consists of 15,000 FEs with a characteristic length of 1.4 mm (108
equivalent plastic strain values.
FEs activation steps per layer). Moreover, at least two FEs per heat
Temperature Yield Stress (MPa) source radius and three FEs through thickness are required, according to
(◦ C)
Equivalent Plastic Strain (/) Zhang et al. [17]. In the layer’s cross-section, the mesh is refined where
0.0 0.02 0.05 0.1 0.2 0.5 1.0 it contacts the underlying layer, due to the large temperature gradients
expected upon activation. Additionally, this minimises the artificial
20 362.0 375.0 390.0 410.0 425.0 437.0 457.0
450 212.1 226.0 240.4 257.1 269.8 275.0 269.1
energy gains or losses at nodes shared by active and quiet FEs [18]. The
900 62.2 77.0 90.7 104.1 114.6 102.2 93.0 contact conditions between the base plate and all the sequentially
1050 48.0 58.0 64.6 72.1 74.0 71.2 71.2 deposited layers are modelled as a uniform material state with a con­
1150 46.0 50.0 53.1 55.1 51.7 50.9 50.9 forming mesh topology.
1300 25.0 25.2 25.5 26.0 27.0 30.0 35.0
While a denser FE mesh results in a more continuous material
1400 10.0 10.1 10.3 10.5 11.0 12.5 15.0
1500 5.0 5.1 5.3 5.5 6.0 7.5 10.0 deposition and thus more realistic simulation results, the computational
Quiet 1e6 1e6 1e6 1e6 1e6 1e6 1e6 burden is not to be neglected. Therefore, a dynamic mesh refining
strategy is often implemented [27].

kJ/kg for the solidus and liquidus temperatures at 1450 ◦ C and 1500 ◦ C,
respectively. 3.4. Pre-processing of activation criterion
Quiet FE properties, also included in Tables 3 and 4, are set as low as
possible to ensure that they do not affect the analysis. However, they The literature proposes a variety of activation criteria. Most
should be high enough to avoid ill-conditioning the system of equations. commonly, the activation temperature [8], activation time [28] or
The scaling factor of 1e− 6 is commonly applied to all properties [20,21]. Gauss point evaluated heat source [29] model-based criteria are
In our research, a significant specific heat capacity value of 1e8 J/(kg K) implemented. While the activation time criterion is convenient for DED
has been chosen because it immensely improves convergence at the cost modelling, the activation temperature criterion offers a better repre­
of negligible computational error. Also, the value of yield stress is set sentation of the LPBF process. We choose the activation time criterion in
high enough that it cannot be reached unintentionally. Mechanical
properties vary gradually within a temperature range for smoother FE
activation in order to avoid convergence issues [22].

3.3. Finite element mesh

The model geometry should represent the actual state as closely as

possible (Fig. 2). Realistic weld cross-section geometry is generally more
challenging to mesh, so the bead shape is often simplified [23]. In order
to obtain valid thermo-mechanical state results, however, it is crucial to
impose accurate boundary conditions. Therefore, in this study, we as­
sume an elliptic profile approximation as accounted for by Lambiase
et al. [24]. Furthermore, appropriate modelling of the base plate is
essential, since it mechanically restricts the part and influences heat Fig. 5. The finite element (FE) mesh considered in the numerical case study is a
accumulation [25,26]. realistic representation of the manufactured experimental part geometry. Each
We built the DED numerical model from 182,000 three-dimensional welding layer mesh cross-section is refined at the interface with the underlying
8-node hexahedral FEs with 194,000 nodes connected. The base plate base material.

4
D. Kovšca et al. Finite Elements in Analysis & Design 221 (2023) 103949

our research because it allows pre-processing. Consequently, no further a higher global average film coefficient value of 20 W/(m2 K) was used
computational steps are needed during analysis. Another advantage of instead, as further described in Ref. [33].
this activation criterion is that it permits the temperature of the quiet The thermal radiation is described by the Stefan-Boltzmann law
FE’s to be initialised at the melting value. defined by the relation (4):
In our approach, individual FE activation time is determined ahead [ ]
qrad (x, t) = εσ Ts4 (x, t) − T∞
4
(4)
of time by the heat-affected-volume method [30] via the developed
Python/Fortran code. The user-defined geometry region of the deposited
where ε is the surface emissivity, and σ the Stefan-Boltzmann constant.
material—an elliptical shape in our case—is run across the FE mesh
The radiation model is often linearised and coupled with film convection
based on the generated G-code data file, considering the actual process
into effective heat transfer through (5) and (6) to avoid convergence
parameters. For each time step, our search algorithm identifies the FEs
errors:
that are within this region and assigns them the current time as the
[ ]
activation criterion. During a simulation, individual FEs get activated heff (x, t) = hconv + εσ [Ts (x, t) + T∞ ] Ts2 (x, t) + T∞
2
(5)
based on these element-specific activation times, which are compared to
the current analysis time at each time step, using the USDFLD Aba­ qeff (x, t) = heff (x, t) [Ts (x, t) − T∞ ] (6)
qus/Standard subroutine. For instance, a FE is activated if the analysis
The base plate accumulates significant heat during the deposition,
time exceeds its activation time. Since the base plate is active from the
and we set its temperature to the ambient temperature of 25 ◦ C. We
start of the analysis, we assign a negative activation time (i.e., − 1) to the
define the conduction heat flux through the base plate to the unmodelled
associated FEs for a more straightforward code implementation.
support table as an equivalent contact film transfer described by the
relation (6), where a heat transfer coefficient of 50 W/(m2 K) is assumed.
3.5. Transient thermal analysis
We should emphasise that at the beginning of the deposition and on
smaller build parts, the heat loss through the support table
The transient thermal analysis comprises three sub-steps per layer:
predominates.
inactive layer inclusion, material deposition modelled through quiet FE
All the parameters for specifying the boundary conditions (c.f.
activation, and an interlayer cooling step. The first step in the current
Table 5) are averaged and assumed to remain constant throughout the
case study—inactive layer inclusion—is instantaneous. After that, the
analysis. Calibration of the input parameters through inverse analysis
duration of the welding of each layer is 37.5 s, followed by a 30.0 s
was performed by Bai et al. [34], while the film coefficient sensitivity
cooling break. Finally, after the 10th layer has been deposited, the part is
was studied by Dunbar et al. [35].
cooled down to a steady state. During the analysis, the maximum time
Abaqus/Standard uses the FILM subroutine to prescribe the user-
increment is set to 0.04 s, while during the cooling step, the maximum
defined surface heat flux to the model. However, the film boundary
time increment is increased to 5.0 s, owing to the smaller thermal gra­
condition should only be applied to the free surfaces of the model, as
dients expected.
discussed in the following section.
3.5.1. Heat transfer governing equation
3.5.3. Free-surface (boundary condition) detection
Transient heat transfer follows the energy balance governing equa­
As material deposition progresses, the free surface of the model is
tion defined by the following relation (1):
constantly evolving. In light of this fact, convection heat transfer from
dT(x, t) the internal free-surfaces at the interface between the active and inactive
ρc p = − ∇⋅q(x, t) + Q(x, t) (1)
dt FEs is often neglected in the literature, as it is hard to implement in
general-purpose FEA software. As shown by Michaleris [18], neglecting
where ρ is the material density, cp specific heat capacity, T temperature, the free-surface heat convection and radiation can be a significant
t time, q heat flux, Q volumetric heat generation, and x denotes the source of error.
spatial coordinate. The temperature dependence have been omitted A new, easy-to-implement free-surface detection algorithm that uses
from the notation for clarity. Moreover, the heat flux q is described by the Abaqus/Standard FILM subroutine and requires no extra computa­
Fourier’s law (2): tional effort during analysis, is presented. Before starting thermal
q(x, t) = − k⋅∇T(x, t) (2) analysis, we identify the adjacent FEs for each FE by comparing mutual
nodes using the Python/Fortran code we had developed. Consequently,
where k is the thermal conductivity. the same face of the two neighbouring FEs is also known.
Following the FEM, a system of equations is derived from a weak During thermal analysis in Abaqus/Standard software, the FILM
formulation of the governing equation using the Galerkin method. The subroutine is called individually for each face of each active FE.
applied boundary conditions addressing the surface heat losses are dis­ Knowing the activation state of the corresponding adjacent FE, the exact
cussed in the following subsections. boundary condition can be established for each time step. For example,
we prescribe surface heat convection and radiation if the neighbouring
3.5.2. Thermal boundary conditions/thermal losses FE is inactive (quiet), while omitting this if the latter is active. Faces not
Boundary conditions in heat transfer analysis comprise thermal ra­ shared with any other FE belong to the external surface of the model.
diation, convection at the free surface, and heat conduction through the Moreover, we introduce a new fictitious FE with out-of-reach activation
mounting and the worktable. We model the natural convection using the time as an outlying face of an adjacent FE, for a more straightforward
following relation (3):
qconv (x, t) = hconv [Ts (x, t) − T∞ ] (3) Table 5
Heat transfer boundary condition parameters as used in the numerical case
where hconv is the surface film coefficient, Ts surface temperature, and study.
T∞ the average ambient temperature. Natural convection rates typically Parameter Symbol Value
vary between 5 and 15 W/(m2 K). It is important to note that the Ambient temperature (◦ C) T∞ 25.0
shielding gas flow results in forced free-surface convection near the Convective film coefficient (W/(m2 K)) hconv 20.0
welding torch during the deposition. This effect can be accounted for as Surface emissivity (/) ε 0.5
proposed by Abusalma et al. [31] or Gouge et al. [32]. However, Stefan-Boltzmann constant (W/(m2 K4)) σ 5.67e− 8
Contact film coefficient (W/(m2 K)) hbase 50.0
considering the relatively small-scale dimensions of the case study part,

5
D. Kovšca et al. Finite Elements in Analysis & Design 221 (2023) 103949

implementation. This FE cannot be activated; the outer surface bound­ Table 6

ary condition is therefore always present. Parameters as used in Goldak’s heat source model.
The described boundary condition detection algorithm is closely Parameter Symbol Value
related to the immersed boundary approach found in Ref. [36] but
Weld pool length (mm) cf , cr 2.1
implemented for the conforming FE mesh. However, our method uses a Weld pool width (mm) a 2.1
slightly different approach by working with FE surfaces instead of Finite Weld pool depth (mm) b 1.5
Cell integration points. Fig. 6 depicts the FEs activation using activation Layer height (mm) hlayer 1.5
time as the activation criterion and free-surface boundary condition Forward heat factor (/) ff 1.0
Rearward heat factor (/) fr 1.0
evolution during the deposition as the effective nodal heat sink. The Welding current (A) I 65.0
moving heat source is represented as described in the following section. Welding voltage (V) U 15.0
Arc efficiency (/) η 0.5
3.5.4. Heat source model Welding source power (W) P 975.0
Heat source velocity (mm/min) vt 240.0
In the literature, the welding heat source power density Qα is often
Filler melting temperature (◦ C) Tmelt 1500.0
implemented using the double ellipsoid volumetric model proposed by
Goldak et al. [37]:
√̅̅̅ [ ( 2 )] remainder is applied to the filler material with a constant distribution, as
6 3Pfα x y2 z2
Qα (x, y, z) = √̅̅̅ ⋅exp − 3 + + , α ∈ {f , r} (7) shown in Fig. 7.
π πabcα cα 2 a2 b2 To simplify the computational model further, we assign the quiet FEs
the melting temperature Tmelt of the filler material at the initial step of
where a, b, and c represent the weld pool geometry and P the welding
the simulation [41]. Upon activation, they immediately start to cool
source power calculated from the arc efficiency, welding current and
down while the surrounding volume gets heated by the Goldak heat
welding voltage as P = ηIU. Because of the relatively small welding
source, as depicted in Figs. 6 and 7. Therefore, the realistic bead
velocity (vt), the shape of the heat source is assumed to be circular. The
cross-section should be considered to apply the proper heat input to the
same heat factor values therefore apply to both the front (x > 0; α = f)
model [19]. Although not frequently used in the literature, this
and the rear half (x < 0; α = r) of the ellipsoid. The heat source
approach gives us better numerical convergence than the instantaneous
parameter values used, presented in Table 6, were experimentally ob­
heating of the newly added FEs to the melting point. Furthermore,
tained and are considered constant throughout the welding process. In
adding molten material more accurately represents DED processes like
general, the heat source geometry parameters can be determined by
WAAM. By contrast, it may not be the most appropriate approach to
cutting the cross-section of the weld bead, as shown by Giarollo et al.
model the LPBF process, as it cannot be used, for example, to model
[38] and Liu et al. [39]. Moreover, the heat source parameter sensitivity
porosity due to improper powder melting [42].
study was performed by Irwin et al. [40].
Abaqus/Standard uses the DFLUX subroutine to model user-defined
3.5.5. Mitigation of artificially induced thermal gradients
volumetric heat input. The heat source follows the activation front of the
Because of FEA’s discrete material deposition modelling, the
FEs according to the G-code used in the experiment. However, to
computed temperature gradients are significantly higher than those
simulate the continuous heat source motion along the welding path, the
occurring in reality, resulting from a continuous welding process.
time increment should be restricted by the size of the ellipsoid [39]. The
Bearing this in mind, we want to mitigate the artificially induced ther­
distance travelled in one increment should therefore not exceed the
mal gradients that occur just after the FE’s activation in the melting pool
length of the melting pool itself.
domain. To this end, we developed a method where we smooth the
In this paper, we consider a modified Goldak model proposed by
computed temporal variation of the temperature at the FE’s activation
Montevecchi et al. [22] to get a more realistic heat source power density
time (tact ) by computing a weighted average of temperature values a step
distribution. Approximately 50% of the total arc power is transferred
before (tact − Δt) and step after the activation (tact + Δt), using relation
into the underlying base material, as described by expression (7). The
(8):

Fig. 6. Schematic representation of the proposed numerical model showing the Fig. 7. Modified heat source as used in the numerical model employs the
FE activation mechanics, moving heat source distribution and free-surface double ellipsoid heat distribution for the base material and constant tempera­
boundary condition evolution in consecutive time steps. ture distribution for the filler material.

6
D. Kovšca et al. Finite Elements in Analysis & Design 221 (2023) 103949

w1 ⋅T(x, tact − Δt) + w2 ⋅T(x, tact + Δt) alent plastic strain rate ε̇p . In the case of the associated flow rule, the
T(x, tact ) = (8)
w1 + w2 plastic flow occurs in a direction normal to the yield surface, i.e.:
( )
where w1 and w2 are the weight factors. The smoothed temporal tem­ ∂Φ σ, σy
N≡ (15)
perature field distribution is then used as a thermal load on the base ∂σ
plate and the deposited layers in the mechanical analysis. Plastic behaviour is the principal source of nonlinearity in a FEM
implementation of the above equations. The Newton-Raphson algorithm
is therefore used to solve the equilibrium equations iteratively.
3.6. Quasi-static mechanical analysis

3.6.2. Mechanical boundary conditions

Even though AM is a dynamic process, quasi-static analysis is often
The smoothed temporal temperature field variation presented in
adopted in a mechanical simulation. The mechanical equilibrium is
Section 3.5.5 is considered a step-by-step thermal load in a quasi-static
found in each time step for the transient thermal loadings applied. Yang
mechanical analysis. The base plate is clamped on the opposite sides in
et al. [21] showed that quasi-static modelling produces similar results
the same way as in the experimental case study in Fig. 3. After cooling to
and saves computational cost compared to a dynamical simulation of the
room temperature, the constraints are released, and the base plate is
material deposition process.
removed from the part. The cutting process is not modelled directly.
Unbalanced thermal loading results in nonhomogeneous material
However, the separation is achieved by excluding the corresponding FEs
expansion; therefore, we implemented thermo-elastic-plastic material
from the analysis (the Inactive FE method). Even though the given method
behaviour with isotropic hardening under a small strain assumption to
is not an accurate representation of the cutting process, it still provides a
model the evolution of residual stresses. Moreover, due to the relatively
good insight into the residual stress and displacement fields after the
dense FE mesh used in our case study, reduced Gauss integration of the
removal.
stress state is considered to improve computational efficiency without a
significant loss of accuracy.
4. Results and discussion
In the mechanical analysis, we used the same computational step
sequence as in the thermal analysis (the layer deposition process fol­
4.1. Thermal analysis and validation
lowed by cooling to room temperature), additionally followed by the
base plate removal step. However, the mechanical time step upper limit
The computed temperature field history during the 10th layer
is set approximately ten times lower (0.005 s) due to the nonlinearity of
deposition is presented in Fig. 8. At the beginning in Fig. 8a (at 0.35 s of
the plasticity model.
welding time), only the arc preheating of the underlying layer can be
observed, followed by the molten material deposition at the same
3.6.1. Governing equations in constitutive modelling
location in Fig. 8b (at 0.5 s of welding time). This is a valid represen­
The mechanical state variables (stresses, strains, and displacements)
tation of the DED processes like WAAM. The welding progression in a
are computed from the linear momentum equilibrium equation:
counterclockwise direction and the heat conduction to the base plate are
∇⋅σ(x, t) + b(x, t) = 0 (9) visible in successive time steps in Fig. 8c–f.
Fig. 9 shows four temperature profiles, representing the surface of
where σ is the Cauchy stress tensor and b denotes the volumetric body the 1st (grey line), 4th (blue line), 7th (orange line), and 10th layer (red
force. Hooke’s law defines the relationship between the stress and elastic line) at a point 20 mm from the welding start point (validation point in
strain: Fig. 8a), where solid and dotted lines represent computed and measured
σ = D : εe (10) temperature profiles respectively. Fig. 9a shows large temperature gra­
dients due to the significant variations between the surface temperature
where D is the elasticity tensor. Under the assumption of small strains, and melting point at the FEs activation. It can also be noticed that the
the total strain ε can be additively decomposed into elastic, plastic, and computed average part temperature is asymptotically increasing
thermal strain tensors as: throughout the welding process. This is because each deposited layer
results in an increase in the free-surface area, leading to higher heat
ε̇ = ε̇e + ε̇p + ε̇t (11) dissipation through film convection and radiation.
The thermal part can be computed as: Meanwhile, the excessive heat input remains constant, so the rise in
average temperature decreases with each layer added, approaching
ε̇t = α(T)⋅Ṫ⋅δ (12) thermal equilibrium. Eventually, however, a decline in average layer
temperature is expected, once energy dissipation starts exceeding the
where α is the thermal expansion coefficient, Ṫ the temperature rate and input source. Furthermore, a slight temperature rise is evident in the
δ the Dirac delta tensor. cooling break intervals between layers due to the lag in heat transfer
The presence of plastic strains results in permanent part deformation. from the welding endpoint to the measuring location.
Therefore, a von Mises yield criterion is implemented for ductile mate­ The temperature measurements are compared to the computed
rials, e.g., metals: temperature profiles to validate the presented numerical model of the
( ) ( ) DED process. Fig. 9b shows four detailed sections of the temperature
Φ σ, σy = q(σ) − σy εp , T ≤ 0 (13)
profiles (also marked in Fig. 9a) in time relative to the beginning of the
where q(σ) represents the von Mises equivalent stress, and σy is the yield corresponding layer deposition. It is evident that the temperature gra­
stress as a function of the accumulated equivalent plastic strain εp and dients are higher throughout the deposition of an individual layer
(0–37.5 s) than during the cooling break intervals (37.5–67.5 s). During
temperature T. Plasticity occurs when the yield condition is fulfilled:
the first few seconds of the layer welding process, measuring the surface
Φ(σ, σy ) = 0. The associated flow rule governs the evolution of plastic
temperature is impossible, as it would damage the thermocouple.
strains, generally expressed as:
Arc extinguishing at the end of welding intervals (Fig. 9b at 37.5 s
ε̇p = γ̇N (14) relative to the welding start) is accompanied by a sudden drop in
experimental measurements for which there is no thermal explanation.
where ε̇p is the plastic strain rate tensor. Since q(σ) is a first order ho­ The high welding current is known to produce a strong magnetic field
mogeneous function, the plastic multiplier γ̇ corresponds to the equiv­

7
D. Kovšca et al. Finite Elements in Analysis & Design 221 (2023) 103949

Fig. 8. Computed thermal history in successive time steps relative to the beginning of the 10th layer deposition. At the beginning only the arc preheating is observed,
followed by the molten material deposition in a counterclockwise direction. Validation point marks the spot where the computed and experimentally measured
values are compared in Fig. 9.

that can induce an additional electrical voltage in the thermocouple. material gets partially remelted. Since the base plate is clamped on the
Therefore, a slightly higher temperature than expected is measured upper left and bottom right sides, tensile stresses occur in this direction
during the deposition. Other minor causes of thermal deviations include after cooling to room temperature (Fig. 11e). When the base plate is
thermal inertia and manual positioning of the thermocouple. Consid­ finally removed, a reduction of the residual stresses is observed in
ering these limitations, the experimental and computed values generally Fig. 11f, owing to the release of the constrained displacements.
show good agreement. Table 7 provides mutual temperature differences Residual stress monitoring is vital due to the possibility of crack
and relative errors at the end of the cooling break intervals (at 67.5 s propagation in the tensile-loaded areas. Tensile hoop stress is primarily
relative to the welding start). The maximal relative error between the responsible for vertical crack propagation, while the tensile axial
numerical results and thermocouple measurements is comparable to stresses result in inter-layer cracking and delamination. Fig. 12 shows
that found in the literature (e.g., Giarollo et al. [38]). the computed radial, hoop, and axial stress components in three
different time steps. The first row of images shows residual stress com­
4.2. Mechanical analysis ponents just after completing the 10th layer deposition while still in a
hot state. In general, the most significant tensile stresses are observed in
When the welding beads cool down, they contract most noticeably in the hoop and axial directions, always on the outer region of the layers
a longitudinal direction. In the case of cylindrically shaped parts lon­ close to the base plate, where the cold plate limits thermal contractions.
gitudinal contraction corresponds to hoop strains, which result in radial By contrast, the compressive stresses are often found in the inner
contraction owing to the circumference reduction. Fig. 10a and b shows domain. Meanwhile, residual shear stress components are negligible
the radial displacements in the part cross-section immediately after compared to the normal stresses.
completing the 10th layer deposition, while still in a hot state, and then Additionally, a noticeable difference is evident between the left and
after cooling to room temperature. Each bead tends to volumetrically right side’s cross-section stress distribution, where the right side corre­
contract towards its mass centre. In addition, we see an inward defor­ sponds to the welding start point. The residual stresses increase after
mation of the structure. Consequently, a more significant radial cooling the part to room temperature, as seen in Fig. 12 in the second
displacement occurs on the outer side of the bead, where these two ef­ row of images. However, they decrease significantly after the base plate
fects add up. is removed, as shown in the third row of images.
Additionally, the colder base plate restrains the contraction of the With each pass of the deposition torch, the underlying material is
lower layers. The part’s final diameter is therefore smaller after compressed in the radial direction, because it cannot thermally expand
unclamping and removing the base plate in Fig. 10c. The maximal properly. Then the cooling begins, and tensile stresses build up due to
computed reduction of the outside diameter in the 5th layer is 0.34 mm. the contraction. Since the base is colder and cannot contract as much as
Also noticeable is a considerable upward bending of the base plate after the newly deposited material, tensile stresses form in the lower part of
unclamping, where the total vertical deflection between the plate’s the molten bead. The remelted top part of the layer below is also under
centre and its corners is 0.72 mm. tensile stress, resulting in compression of the lower part of the same
The computed spatial and temporal variation in von Mises stress is layer. Fig. 12a shows the part’s cross-sectional radial stresses field dis­
presented in Fig. 11, where the white circle indicates the current loca­ tribution, where tensile and compressive loaded areas alternate in a fish
tion of the 10th layer deposition. Residual stresses form on a layer-by- scale-like pattern. After the cool-down phase and the removal of the base
layer basis due to thermal contraction. In Fig. 11a–d, a temporary plate, no further changes in the part’s radial stresses are observed, with
stress release is observed in the domain below the welding area as the the maximal tensile values reaching up to 50 MPa.

8
D. Kovšca et al. Finite Elements in Analysis & Design 221 (2023) 103949

Fig. 10. Computed radial displacement field: (a) in a hot state after completing
the 10th layer, (b) after cooling to room temperature, and (c) after the base
plate removal. Times in parentheses are relative to the beginning of the 10th
layer deposition.
Fig. 9. Comparison of the computed (FEM, solid line) and measured (exp,
dotted line) surface temperature profiles at the validation point, 20 mm from
the welding start point. (a) Temperature profiles during the entire contraction. As a result, the outer region is in tension, reaching up to
manufacturing process of a whole part. Different coloured lines represent the 500 MPa near the base plate. Once the plate is removed, tensile hoop
temperature profile for a particular layer (grey – 1st layer, blue – 4th layer, stresses are still present in the undistorted top and bottom layer but peak
orange – 7th layer, red – 10th layer). (b) For better comparison with the at less than 300 MPa, while the rest is compressed due to some addi­
experimental results, the temperature profiles of the 1st, 4th, 7th and 10th tional radial displacements.
layers are collected and plotted on the scale of the manufacturing time of each The part’s cross-sectional axial stress field distribution in Fig. 12c
layer. During the first few seconds of the layer welding process, the measure­
resembles one found in a cantilever beam, with each layer contributing
ment of the surface temperature is not possible, as the thermocouple would
to an inward bending load due to radial contraction. The maximum
be damaged.
tensile stresses are thus observed in the outer region of the structure,
close to the base plate, while the compressive stresses are present in the
Table 7 inner domain. Before the removal of the base plate, axial stress peaks at
Computed versus measured temperature difference Δ and relative error δ about 550 MPa but drastically decreases afterwards, locally not
for the 1st, 4th, 7th and 10th layer at the end of the cooling break exceeding 200 MPa.
intervals.
Layer Δ (◦ C) δ(%) 5. Conclusions
1st 3.6 6.14
4th 17.0 21.41 This paper presents, in detail, a comprehensive framework for FE-
7th 11.5 11.81 based computational modelling of the DED process. A new method for
10th 2.1 1.76 mitigating the artificial thermal gradients has been proposed, as well as
an easy-to-implement free-surface detection algorithm for accurately
The most significant radial tensile stresses, at about 200 MPa, are prescribing the heat transfer boundary conditions. A 10-layer thin-wall
observed on the bottom side of the base plate. They occur under the cylindrical part deposition was simulated and then experimentally
part’s geometry area due to the radial contraction and downward reproduced in order to validate the proposed numerical model. The
bending of the plate. The same is true for the hoop stresses as well. main conclusions are as follows:
Meanwhile, radial stresses are inverted in the base plate’s outer region
due to the constrained upward bending. • With the noted thermocouple limitations taken into account, the in-
As stated before, a significant thermal contraction of the weld bead is situ temperature measurements taken during the deposition showed
present in the longitudinal direction. Therefore, extensive tensile hoop good agreement with the simulated results.
stresses form along the bead length as the cold base opposes these dis­ • The simulation result shows that the residual stresses are the largest
placements. Simultaneously, the diameter of the middle part shrinks due in the circumferential and axial directions due to thermal contrac­
to compression by subsequent layers. Consequently, the tensile stresses tion. The tensile stresses form on the outer region of the deposited
in areas where the radial displacement is present are released. Fig. 12b layers close to the base plate, while the inward material is com­
shows the cross-sectional inner material in compression due to radial pressed. Consequently, thermal cracks can be expected to form on
the outer side of the thin-wall cylindrical part. The removal of the

9
D. Kovšca et al. Finite Elements in Analysis & Design 221 (2023) 103949

Fig. 11. Computed von Mises stress field history: (a) before the 10th layer welding start, (b) and (c) during the 10th layer deposition, (d) after completion of the part,
(e) after cooling to room temperature, and (f) after the base plate removal, where the white circle indicates the current location of the deposition. Times are given
relative to the beginning of the 10th layer deposition.

Fig. 12. Computed (a) radial, (b) hoop, and (c) axial stress fields consecutively after completing the 10th layer still in a hot state, after cooling to room temperature,
and after the base plate removal. Times in parentheses are relative to the beginning of the 10th layer deposition.

base plate significantly reduces the residual stress in the part. Credit author statement
Noticeable differences in the cross-section stress distribution can be
observed near the welding start/endpoint compared to the rest of the Dejan Kovšca: Conceptualization, Methodology, Software, Writing –
deposited layer. Original Draft. Bojan Starman: Conceptualization, Data Curation,
• The use of thermo-mechanical FEA in the DED process provides an Writing – Review & Editing, Visualization. Damjan Klobčar: Valida­
important insight into the temporal and spatial variations in the tion, Investigation, Resources. Miroslav Halilovič: Resources, Super­
thermal and stress-strain state of the part. vision. Project administration, Funding acquisition. Nikolaj Mole:
Conceptualization, Methodology, Formal analysis, Data Curation,
Writing – Review & Editing. All authors have read and agreed to the

10
D. Kovšca et al. Finite Elements in Analysis & Design 221 (2023) 103949

published version of the manuscript. Int. J. Adv. Manuf. Technol. 110 (2020) 1955–1968, [Link]
s00170-020-05958-8.
[16] A. Ščetinec, D. Klobčar, D. Bračun, In-process path replanning and online layer
height control through deposition arc current for gas metal arc based additive
Declaration of competing interest
manufacturing, J. Manuf. Process. 64 (2021) 1169–1179, [Link]
10.1016/[Link].2021.02.038.
The authors declare that they have no known competing financial [17] L. Zhang, E.W. Reutzel, P. Michaleris, Finite element modeling discretization
requirements for the laser forming process, Int. J. Mech. Sci. 46 (2004) 623–637,
interests or personal relationships that could have appeared to influence
[Link]
the work reported in this paper. [18] P. Michaleris, Modeling metal deposition in heat transfer analyses of additive
manufacturing processes, Finite Elem. Anal. Des. 86 (2014) 51–60, [Link]
Data availability org/10.1016/[Link].2014.04.003.
[19] M. Graf, A. Hälsig, K. Höfer, B. Awiszus, P. Mayr, Thermo-mechanical modelling of
wire-Arc Additive manufacturing (WAAM) of semi-finished products, Metals 8
Data will be made available on request. (2018), [Link] Article 1009.
[20] H. Fawad, Heat transfer modeling of metal deposition employing welding heat
source, Appl. Mech. Mater. 315 (2013) 463–467, [Link]
Acknowledgements [Link]/AMM.315.463.
[21] Q. Yang, P. Zhang, L. Cheng, Z. Min, M. Chyu, A.C. To, Finite element modeling
The authors acknowledge the financial support from the Slovenian and validation of thermomechanical behavior of Ti-6Al-4V in directed energy
deposition additive manufacturing, Addit. Manuf. 12 (2016) 169–177, [Link]
Research Agency (research core funding P2-0263 and L2-3172). org/10.1016/[Link].2016.06.012.
[22] F. Montevecchi, G. Venturini, N. Grossi, A. Scippa, G. Campatelli, Finite element
mesh coarsening for effective distortion prediction in wire Arc Additive
Appendix A. Supplementary data
manufacturing, Addit. Manuf. 18 (2017) 145–155, [Link]
addma.2017.10.010.
Supplementary data to this article can be found online at [Link] [23] K.P. Prajadhiama, Y.H. Manurung, Z. Minggu, F.H. Pengadau, M. Graf, A. Haelsig,
et al., Development of bead modelling for distortion analysis induced by wire arc
org/10.1016/[Link].2023.103949.
additive manufacturing using FEM and experiment, MATEC Web Conf 269 (2019),
05003, [Link]
References [24] F. Lambiase, S.I. Scipioni, A. Paoletti, Accurate prediction of the bead geometry in
wire arc additive manufacturing process, Int. J. Adv. Manuf. Technol. 119 (2022)
7629–7639, [Link]
[1] B. Wu, Z. Pan, D. Ding, D. Cuiuri, H. Li, J. Xu, J. Norrish, A review of the wire arc
[25] P. Prabhakar, W.J. Sames, R. Dehoff, S.S. Babu, Computational modeling of
additive manufacturing of metals: properties, defects and quality improvement,
residual stress formation during the electron beam melting process for Inconel 718,
J. Manuf. Process. 35 (2018) 127–139, [Link]
Addit. Manuf. 7 (2015) 83–91, [Link]
jmapro.2018.08.001.
[26] X. Lu, M. Chiumenti, M. Cervera, G. Zhang, X. Lin, Mitigation of residual stresses
[2] R. Li, G. Wang, X. Zhao, F. Dai, C. Huang, M. Zhang, X. Chen, H. Song, H. Zhang,
and microstructure homogenization in directed energy deposition processes, Eng.
Effect of path strategy on residual stress and distortion in laser and cold metal
Comput. (2022), [Link]
transfer hybrid additive manufacturing, Addit. Manuf. 46 (2021), 102203, https://
[27] S. Jayanath, A. Achuthan, A computationally efficient finite element framework to
[Link]/10.1016/[Link].2021.102203.
simulate additive manufacturing processes, J. Manuf. Sci. Eng. 140 (2018),
[3] X. Lu, X. Lin, M. Chiumenti, M. Cervera, Y. Hu, X. Ji, L. Ma, H. Yang, W. Huang,
041009, [Link]
Residual stress and distortion of rectangular and S-shaped Ti-6Al-4V parts by
[28] A. Anca, V.D. Fachinotti, G. Escobar-Palafox, A. Cardona, Computational
Directed Energy Deposition: modelling and experimental calibration, Addit.
modelling of shaped metal deposition, Int. J. Numer. Methods Eng. 85 (2011)
Manuf. 26 (2019) 166–179, [Link]
84–106, [Link]
[4] R. Martukanitz, P. Michaleris, T. Palmer, T. DebRoy, Z.-K. Liu, R. Otis, T.W. Heo,
[29] E.R. Denlinger, M. Gouge, J. Irwin, P. Michaleris, Thermomechanical model
L.-Q. Chen, Toward an integrated computational system for describing the additive
development and in situ experimental validation of the Laser Powder-Bed Fusion
manufacturing process for metallic materials, Addit. Manuf. 1–4 (2014) 52–63,
process, Addit. Manuf. 16 (2017) 73–80, [Link]
[Link]
addma.2017.05.001.
[5] L. Parry, I.A. Ashcroft, R.D. Wildman, Understanding the effect of laser scan
[30] M. Chiumenti, M. Cervera, A. Salmi, C. Agelet de Saracibar, N. Dialami, K. Matsui,
strategy on residual stress in selective laser melting through thermo-mechanical
Finite element modeling of multi-pass welding and shaped metal deposition
simulation, Addit. Manuf. 12 (2016) 1–15, [Link]
processes, Comput. Methods Appl. Mech. Eng. 199 (2010) 2343–2359, [Link]
addma.2016.05.014.
org/10.1016/[Link].2010.02.018.
[6] K. Ren, Y. Chew, J.Y.H. Fuh, Y.F. Zhang, G.J. Bi, Thermo-mechanical analyses for
[31] H. Abusalma, H. Eisazadeh, F. Hejripour, J. Bunn, D.K. Aidun, Parametric study of
optimized path planning in laser aided additive manufacturing processes, Mater.
residual stress formation in wire and Arc Additive manufacturing, J. Manuf.
Des. 162 (2019) 80–93, [Link]
Process. 75 (2022) 863–876, [Link]
[7] S. Afazov, W.A.D. Denmark, B.L. Toralles, A. Holloway, A. Yaghi, Distortion
[32] M.F. Gouge, J.C. Heigel, P. Michaleris, T.A. Palmer, Modeling forced convection in
prediction and compensation in selective laser melting, Addit. Manuf. 17 (2017)
the thermal simulation of laser cladding processes, Int. J. Adv. Manuf. Technol. 79
15–22, [Link]
(2015) 307–320, [Link]
[8] L. Nguyen, J. Buhl, R. Israr, M. Bambach, Analysis and compensation of shrinkage
[33] M. Gouge, P. Michaleris, E. Denlinger, J. Irwin, Chapter 2 — the finite element
and distortion in wire-arc additive manufacturing of thin-walled curved hollow
method for the thermo-mechanical modeling of additive manufacturing processes,
sections, Addit. Manuf. 47 (2021), [Link]
in: M. Gouge, P. Michaleris (Eds.), Thermo-Mechanical Modeling of Additive
addma.2021.102365. Article 102365.
Manufacturing, Butterworth-Heinemann, 2018, pp. 19–38, [Link]
[9] K.S. Ramani, C. He, Y.-L. Tsai, C.E. Okwudire, SmartScan: an intelligent scanning
10.1016/B978-0-12-811820-7.00003-3.
approach for uniform thermal distribution, reduced residual stresses and
[34] X. Bai, H. Zhang, G. Wang, Improving prediction accuracy of thermal analysis for
deformations in PBF additive manufacturing, Addit. Manuf. 52 (2022), [Link]
weld-based additive manufacturing by calibrating input parameters using IR
org/10.1016/[Link].2022.102643. Article 102643.
imaging, Int. J. Adv. Manuf. Technol. 69 (2013) 1087–1095, [Link]
[10] L. Sun, X. Ren, J. He, Z. Zhang, A bead sequence-driven deposition pattern
10.1007/s00170-013-5102-y.
evaluation criterion for lowering residual stresses in additive manufacturing,
[35] A.J. Dunbar, E.R. Denlinger, M.F. Gouge, P. Michaleris, Experimental validation of
Addit. Manuf. 48 (2021), [Link] Article
finite element modeling for laser powder bed fusion deformation, Addit. Manuf. 12
102424.
(2016) 108–120, [Link]
[11] F. Montevecchi, G. Venturini, N. Grossi, A. Scippa, G. Campatelli, Idle time
[36] M. Carraturo, S. Kollmannsberger, A. Reali, F. Auricchio, E. Rank, An immersed
selection for wire-arc additive manufacturing: a finite element-based technique,
boundary approach for residual stress evaluation in selective laser melting
Addit. Manuf. 21 (2018) 479–486, [Link]
processes, Addit. Manuf. 46 (2021), 102077, [Link]
[12] M. Chiumenti, E. Neiva, E. Salsi, M. Cervera, S. Badia, J. Moya, Z. Chen, C. Lee,
addma.2021.102077.
C. Davies, Numerical modelling and experimental validation in selective laser
[37] J. Goldak, A. Chakravarti, M. Bibby, A new finite element model for welding heat
melting, Addit. Manuf. 18 (2017) 171–185, [Link]
sources, Metall. Trans. B 15 (1984) 299–305, [Link]
addma.2017.09.002.
BF02667333.
[13] S. Bontha, N.W. Klingbeil, P.A. Kobryn, H.L. Fraser, Thermal process maps for
[38] D.F. Giarollo, C.C.P. Mazzaferro, J.A.E. Mazzaferro, Comparison between two heat
predicting solidification microstructure in laser fabrication of thin-wall structures,
source models for wire-arc additive manufacturing using GMAW process, J. Braz.
J. Mater. Process. Technol. 178 (2006) 135–142, [Link]
Soc. Mech. Sci. Eng. 44 (2022), [Link]
jmatprotec.2006.03.155.
Article 7.
[14] R. Israr, J. Buhl, L. Elze, M. Bambach, Simulation of Different Path Strategies for
[39] M. Liu, L.N.S. Chiu, C. Vundru, Y. Liu, A. Huang, C. Davies, X. Wu, W. Yan,
Wire-Arc Additive Manufacturing with Lagrangian Finite Element Methods, LS-
A characteristic time-based heat input model for simulating selective laser melting,
DYNA Forum 2018, Bamberg, 2018.
Addit. Manuf. 44 (2021), 102026, [Link]
[15] N. Kozamernik, D. Bračun, D. Klobčar, WAAM system with interpass temperature
control and forced cooling for near-net-shape printing of small metal components,

11
D. Kovšca et al. Finite Elements in Analysis & Design 221 (2023) 103949

[40] J.E. Irwin, Q. Wang, P. Michaleris, A.R. Nassar, Y. Ren, C.B. Stutzman, Iterative [42] A. Olleak, Z. Xi, Part-scale finite element modeling of the selective laser melting
simulation-based techniques for control of laser powder bed fusion additive process with layer-wise adaptive remeshing for thermal history and porosity
manufacturing, Addit. Manuf. 46 (2021), 102078, [Link] prediction, J. Manuf. Sci. Eng. 142 (2020), 121006, [Link]
addma.2021.102078. 1.4047733.
[41] R.J. Williams, C.M. Davies, P.A. Hooper, A pragmatic part scale model for residual
stress and distortion prediction in powder bed fusion, Addit. Manuf. 22 (2018)
416–425, [Link]

12