Additive Manufacturing 54 (2022) 102708

Contents lists available at ScienceDirect

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

Research paper

A thermomechanical finite element model and its comparison to inherent

strain method for powder-bed fusion process
Can Bayraktar a,b , Eralp Demir c ,∗
a Faculty of Engineering and Natural Sciences, Sabanci University, Tuzla, Istanbul, Turkey
b
Integrated Manufacturing Tech. Research and App. Center, Sabanci University, Tuzla, Istanbul, Turkey
c
Department of Mechanical Engineering, University of Bristol, Bristol, BS8 1TR, UK

ARTICLE INFO ABSTRACT

Keywords: In this study, a thermomechanical model is developed to predict the melt-pool dimensions and residual stresses
Powder bed fusion for laser powder bed fusion process. Inherent strain method is also used to predict residual stresses by using
Thermomechanical process model only the thermal solution that is free the complexities involved with the mechanical solution. A unique
FEM
approach is developed to define the surface heat losses as volumetric heat losses in order to avoid the definition
Inherent strain method
of traction-free surfaces and their re-definition after layer deposition. The thermal process simulations predict
melt-pool dimensions of experimental cross-sections of single tracks within approximately 10% agreement.
The thermomechanical process model is used to forecast the effect of process parameters on the melt-pool
dimensions and residual stresses. The inherent strain method reproduces the residual stresses within 15%
accuracy, approximately six times faster in comparison to the thermomechanical model, and free of any
convergence issues related with the displacement field solution.

1. Introduction that cause deviations from the desired geometry [3]. High thermal gra-
dients during the LPBF process lead to the formation of tensile stresses
Additive manufacturing (AM) has been referred to as the manu- that adversely affect the mechanical performance by promoting crack
facturing method of the future. AM has become a great alternative initiation or formation. Residual stresses also cause loss of dimensional
to conventional manufacturing processes in various industries because accuracy owing to the distortions. Accordingly, the process parameters
of its numerous advantages, such as process automation, reduction of need to be optimized to ascertain the penetration of melt zones with
multi-stage manufacturing processes, elimination of tooling, consum- each other and integration of each layers for different materials [4].
ables, and waste products [1]. In addition, AM makes it possible to Experimental trial–error methods to correlate the process param-
produce highly complex geometries and lightweight structures to signif- eters with the quality of product for LPBF process is a tedious and
icantly improve the design performance. Currently, AM has found wide costly task because of the high machine costs involved with the AM
use in high-end applications in aerospace and biomedical industries.
process [5]. Therefore, researchers have turned their attention on
However, still the applications of AM in critical industries require a
developing efficient numerical methods to simulate the PBF process
better understanding of the effect of process parameters on the product
instead [6], which is consistent with the current research trends related
quality.
to the digitization and digital twinning of AM processes. Process simu-
The laser powder bed fusion (LPBF) process is an AM process for
lations have been used to unravel: the link between process parameters
manufacturing parts with fine precision. In the LPBF process, a laser is
and the structural properties [7], anisotropic microstructure evolu-
used as the heat source to melt the metal powder particles selectively
and to solidify them by cooling in a layer-by-layer manufacturing [2]. tion [8], melt-pool kinetics [9], defect formation mechanisms [10],
After selective melting of each layer over a powder substrate, a new phase transformations [11], and solidification microstructure [12].
layer of powder material is added on top of the former layer, and Thermomechanical finite element (FE) process simulations have
the same procedure continues until the part geometry is complete. been extensively used to determine the temperatures and displacements
Although this process has great advantages, AM suffers from the for- and to compute the melt-pool size and residual stresses after the AM
mation of defects, residual stresses during deposition, and distortions process [13,14]. Beuth et al. investigated the prediction of distortion

∗ Corresponding author.
E-mail address: eralp.demir@bristol.ac.uk (E. Demir).

https://doi.org/10.1016/j.addma.2022.102708
Received 16 May 2021; Received in revised form 8 February 2022; Accepted 19 February 2022
Available online 4 March 2022
2214-8604/© 2022 Elsevier B.V. All rights reserved.
C. Bayraktar and E. Demir Additive Manufacturing 54 (2022) 102708

Nomenclature 𝐴, 𝐴𝑠 , 𝐶 Constants used in inherent strain & stress

calculation
𝛼 Thermal expansion coefficient
𝑎, 𝑓0 Scaling factors to find powder conductivity
𝛽 Rigidity constraint in ISM
𝐴(𝐻) Areal fraction of voids in powder state
𝝐, 𝛥𝝐 Total strain and increment
𝐶𝑝 Specific heat
𝝐𝑒 Elastic strain
𝑑𝑥 , 𝑑𝑦 , 𝑑𝑧 Distances from the heat source along x, y,
𝝐 𝑝 , 𝛥𝝐 𝑝 Plastic strain and increment
and z directions
𝝐 𝑇 , 𝛥𝝐 𝑇 Thermal strain and increment
𝐸 Young’s modulus
𝝈 Stress
′ ℎ Convection coefficient
𝝈 𝑡𝑟 Trial deviatoric stress
𝑘(𝑠) , 𝑘(𝐴𝑟) , 𝑘(𝜑) Conductivity of solid, Argon gas, and pow-
𝝈 𝑡𝑟 Trial stress der, respectively.
𝛥𝑑 Hatch spacing 𝐾𝐵 Boltzmann constant
𝛥ℎ Layer thickness 𝑘𝑥 , 𝑘𝑦 , 𝑘𝑧 Thermal conductivity constants x, y, and z
𝛥𝐻𝜈∗ Effective enthalpy of metal vapor directions
𝛥𝑝 Plastic strain increment 𝐿 Characteristic element size along scan direc-
𝛥𝑡𝑟𝑒𝑐𝑜𝑎𝑡 Time step at the recoating time interval tion
𝛥𝑡𝑠𝑐𝑎𝑛 Time step at the scan time interval 𝑀 Molar mass of metal vapor
𝑄̇ Heat input of the moving heat source 𝑃 Laser power
𝑞̇ 𝑐𝑜𝑛𝑣 Convective heat loss from surfaces 𝑃0 Ambient pressure
𝑞̇ 𝑒𝑓 𝑓 Effective surface heat loss 𝑅 Universal gas constant
𝑞̇ 𝑒𝑣𝑎 Evaporative heat loss from surfaces 𝑟 Effective radius of the heat source
𝑞̇ 𝑟𝑎𝑑 Radiative heat loss from surfaces 𝑟𝑥 , 𝑟𝑦 , 𝑟𝑧 Heat source radii
𝜖𝐼𝑆 Scalar inherent strain 𝑆 Surface of the domain
𝜂 Efficiency factor/absorptivity 𝑇 Temperature
̂𝐭
𝐓 Nodal temperatures at former time incre- 𝑡 Time
ment 𝑇0 Ambient temperature
̂
𝐓 Nodal temperatures 𝑇2 , 𝑇1 Upper and lower limit temperature for ISM
̂
𝐮 Node displacements 𝑇𝜈 Boiling temperature
𝐁 Strain–displacement mapping 𝑇𝑠 Surface temperature
𝐂 Elasticity matrix 𝑉 Volume of the domain
𝐄 Inherent strain 𝑣 Scan speed
𝐤 Conductivity matrix
𝐧 Area normal of a surface
𝐍, 𝐍𝐬 Volume and surface interpolation functions
𝐧𝑝 Plastic flow direction Melt-pool characteristics such as dimensions and geometry have a
𝐪, 𝐪𝐬 Volumetric and surface heat flux considerable effect on the overall part quality, hence it has been the
𝐑 Rotation matrix subject of many studies [22]. Andreotta et al. developed a thermal
model to predict melt-pool dimensions for various process parameter
𝐒 Inherent stress
combinations and they validated their model with an experimental
𝐮 Displacements
study. According to their results, melt-pool width and depth increases
𝐱𝐬 Surface position vector with laser power and decreases with scan speed [20]. A similar finding
𝐱 Position vector was recorded in the experimental studies that were performed by Scime
∇𝐍 Interpolation function derivatives et al. in which the effect of process parameters on the formation of
𝜔 Material state indicator, solid volume frac- defects such as keyhole and balling were emphasized [23].
tion at mushy state The process parameters significantly influence the formation of
𝜓 Yield function residual stresses and distortion of the deposited part [24,25]. Accord-
𝜌 Mass density ingly, Mukherjee et al. concluded that thermal strains being directly
𝜎𝑦 Yield stress proportional to the laser power and inversely proportional to the scan
𝜎𝑉 𝑀 Equivalent Von Mises stress speed. Similarly, Simsons et al. studied the influence of energy density
on the residual stresses which depended on the process parameters of
𝜃 Scan direction with respect to sample x-axis
laser power, scan speed, layer thickness, and hatch spacing [26]. They
𝜀 Emissivity
reported an increase in the residual stresses with increasing energy
𝜑 Solid volume fraction of powder state density. However, according to Bartlett et al. generic energy density
𝝈
̂ Node values of stress expression was not a precise formulation for the actual energy density
applied, and each parameter should be considered individually, because
each parameter had a different effect on the residual stress [27]. Shiomi
et al. reported a slight decrease in residual stresses with increasing
after deposition as a pioneering study in this field [15]. The model- scan speed [28]. The layer thickness is another important process
ing of AM processes includes specific FE procedures such as element parameter for the residual stresses of the deposited part. Mukherjee
birth/death methods [16], ghost elements [17], and hybrid element et al. concluded that using thin layers reduces the build-up of residual
stresses while increasing distortions [24]. Anderson et al. reported
activation methods [18,19]. All of these process models were utilized in that to reduce the residual stress and its gradients, the layer thickness
conjunction with temperature-dependent material properties [20,21]. shall be increased based on their experimental measurements, which

2
C. Bayraktar and E. Demir Additive Manufacturing 54 (2022) 102708

were performed on LPBF manufactured specimens by using neutron position 𝐱 and at time 𝑡. 𝑞̇ 𝑐𝑜𝑛𝑣 , 𝑞̇ 𝑟𝑎𝑑 , and 𝑞̇ 𝑒𝑣𝑎 represent the surface heat
diffraction [29]. The influence of process parameters on the residual losses due to convection, radiation, and evaporation, respectively.
stresses still not well understood that requires further research. 𝜕𝑇
𝜌 𝐶𝑝 = −𝛁 · 𝐪(𝐱, 𝑡) + 𝑄(𝐱,
̇ 𝑡) − 𝑞̇ 𝑐𝑜𝑛𝑣 − 𝑞̇ 𝑟𝑎𝑑 − 𝑞̇ 𝑒𝑣𝑎 (1)
The scan strategy has an impact on the residual stresses after 𝜕𝑡
AM [30]. In practice, the scan direction rotates after completion of Goldak’s double ellipsoid model is used as the heat input, as shown
each layer with a default angular rotation of 90◦ or 67◦ about the in Eq. (2). 𝑄̇ is the heat input per unit volume, 𝑃 is the laser beam
build direction [31]. Parry et al. performed process simulations using power, and 𝜂 is the lumped efficiency factor. 𝑟𝑥 , 𝑟𝑦 , and 𝑟𝑧 represent
a commercial software and its user subroutines to conclude that the the laser spot radii in the 𝑑𝑥 , 𝑑𝑦 , and 𝑑𝑧 directions, respectively.
scan pattern affects the residual stress distribution but not its mag- √
6𝜂 3𝑃 3𝑑 2 3𝑑𝑦2 3𝑑 2
nitude [32]. Similarly, Bagg et al. conducted experiments to relate ̇
𝑄(𝐱, 𝑡) = √ exp(− 𝑥 ) exp(− ) exp(− 𝑧 ) (2)
different scan strategies, such as continuous, island, and chess patterns, 𝜋 𝜋 𝑟𝑥 𝑟𝑦 𝑟𝑧 𝑟2𝑥 𝑟2𝑦 𝑟2𝑧
to compare the part distortion and residual stresses [33]. In that study, The heat flux vector, 𝐪, is computed using the temperature gradients
the largest distortions were found with the chess scan strategy, and as given in Eq. (3).
the largest tensile stresses appeared with the continuous scan strategy.
There is still a demand for a better understanding of the effect of the 𝐪(𝐱, 𝑡) = −𝐤 𝛁𝑇 (3)
scan strategy on distortions and residual stresses. Conductivity is assumed for homogeneous media with only the
Part-scale distortion and residual predictions require numerically ef- diagonal terms along the longitudinal, lateral, and depth directions that
ficient process models. For this reason, layer-by-layer simulation meth- are represented by 𝑘𝑥 , 𝑘𝑦 , and 𝑘𝑧 in Eq. (4).
ods have been used for distortion predictions [34]. In addition, semi-
quiet element method activation methods have been used [35]. Patil ⎡𝑘𝑥 0 0⎤
et al. developed an FE method that used two separate mesh refinement 𝐤 = ⎢0 𝑘𝑦 0⎥ (4)
⎢ ⎥
levels in one model [36]. Some models divide the part into sub-domains ⎣0 0 𝑘𝑧 ⎦
and apply a heat source simultaneously over the domain to simulate Convection is considered in the model. The corresponding heat loss
the process efficiently [37,38]. Similar methods, such as simultaneous from the surfaces, 𝑞̇ 𝑐𝑜𝑛𝑣 (𝐱𝐬 , 𝑡), is computed using an effective convection
activation of multiple layers [18], adaptive meshing [39] and domain coefficient, ℎ, and ambient temperature, 𝑇0 , as given in Eq. (5). 𝐱𝐬
ecomposition [40], have been developed to perform simulations in a denotes the surface-position vector.
computationally efficient manner.
𝑞̇ 𝑐𝑜𝑛𝑣 (𝐱𝐬 , 𝑡) = ℎ (𝑇 − 𝑇0 ) (5)
The inherent strain method has been used to predict the residual
deformations after the welding process efficiently [41,42]. Similarly, The radiative heat loss from the exposed surfaces, 𝑞̇ 𝑟𝑎𝑑 (𝐱𝐬 , 𝑡), is com-
eigenstrain construction methods have been developed to find stresses puted using Eq. (6), where 𝜖 and 𝐾𝐵 are the emissivity and Boltzmann
for processes involving thermal gradients [43,44]. Ding et al. used the constant, respectively.
inherent strain method to determine the residual stresses after the wire
𝑞̇ 𝑟𝑎𝑑 (𝐱𝐬 , 𝑡) = 𝜀 𝐾𝐵 (𝑇 4 − 𝑇04 ) (6)
and arc AM process approximately a hundred times faster than with
thermomechanical process simulations [45]. The method was used with The evaporative heat loss, 𝑞̇ 𝑒𝑣𝑎 (𝐱𝐬 , 𝑡), is calculated using Eq. (7) in
multi-scale models to compute the distortions and residual stresses at accordance with [50]. 𝛥𝐻𝜈∗ , 𝑀, 𝑅, 𝑇𝑠 , 𝑇𝜈 , and 𝑃0 are the effective
the part scale using inherent strains that were computed from meso enthalpy of the metal vapor, molar mass of the metal vapor, universal
and micro-scale simulations [46,47]. The limitations of the inherent gas constant, temperature of the surface, boiling temperature, and
strain method have also been investigated [48], and different modified atmospheric pressure, respectively.
versions of this method have been presented [49]. 0.82 𝛥𝐻𝜈∗ 𝛥𝐻𝜈∗ (𝑇𝑠 − 𝑇𝜈 )
𝑞̇ 𝑒𝑣𝑎 (𝐱𝐬 , 𝑡) = √ 𝑃0 exp( ) (7)
In this study, a thermomechanical simulation model is developed, 2 𝜋𝑀 𝑅 𝑇𝑠 𝑅 𝑇𝑠 𝑇𝜈
and it is implemented in a FE environment considering the temperature
dependent properties of powder, solid, and liquid states. In addition, 2.2. Mechanical model
the inherent strain method (ISM) is implemented to predict residual
stresses from the thermal LPBF process simulations. The ISM and The local form of equilibrium is expressed with Eq. (8) for a static
thermomechanical model results are compared to show the deliverables case without any body force.
and drawbacks of the two approaches. Furthermore, a new method
∇·𝝈 = 𝟎 (8)
is used to represent surface heat losses as volumetric heat fluxes to
eliminate the definition of surface fluxes after deposition of a layer. The The total strain, 𝝐, can be decomposed into its elastic, 𝝐 𝑒 , plastic,
thermomechanical model was used to predict the melt-pool dimensions, 𝝐 𝑝 , and thermal 𝝐 𝑇 strain components, as expressed by Eq. (9).
states, and residual stresses for different process parameters.
𝝐 = 𝝐𝑒 + 𝝐𝑝 + 𝝐𝑇 (9)
The paper is structured as follows. Sections 2 and 3 explain the
thermomechanical and inherent strain modeling approaches in detail, The numerical solution is based on incremental form of strain,
respectively. Section 4 shows the experimental methods that were used hence incremental form of Eq. (9) becomes as in Eq. (10) in which 𝛥𝝐
provide comparison to the simulation findings. Sections 5 and 6 include represents the given increment of total strain.
the discussion of the results and conclusions, respectively.
𝛥𝝐 = 𝛥𝝐 𝑒 + 𝛥𝝐 𝑝 + 𝛥𝝐 𝑇 (10)

2. Thermomechanical model Thermal strain increment is computed by multiplying the thermal

expansion coefficient, 𝛼(𝑇 ), which is a function of temperature, with
the temperature increment, 𝛥𝑇 , as in Eq. (11).
2.1. Thermal model
𝛥𝝐 𝑇 = 𝛼(𝑇 ) 𝛥𝑇 (11)
The thermal model includes a solution of the transient energy
Trial stress, 𝝈 𝑡𝑟 ,
is calculated assuming all of the strain increment is
balance relation. In Eq. (1), 𝜌, 𝐶𝑝 , 𝑇 , 𝐪, and 𝑄̇ represent the mass
elastic using elasticity, as in Eq. (12).
density, specific heat, temperature, heat flux vector, and heat input,
respectively. The heat flux vector and heat input are computed at 𝝈 𝑡𝑟 = 𝐂 ∶ 𝛥𝝐 (12)

3
C. Bayraktar and E. Demir Additive Manufacturing 54 (2022) 102708

′
Deviatoric part of trial stress, 𝝈 𝑡𝑟 , is calculated using Eq. (13).
′ 𝐈 𝑡𝑟
𝝈 𝑡𝑟 = 𝝈 𝑡𝑟 − (𝝈 ∶ 𝐈) (13)
3
Von Mises trial stress, 𝜎𝑉𝑡𝑟𝑀 , is the scalar measure for the driving
stress for plasticity that is computed from the deviatoric part of the
trial stress, 𝝈 𝑡𝑟
𝑑𝑒𝑣
, as in Eq. (14).
√
3 𝑡𝑟′
𝜎𝑉𝑡𝑟𝑀 = 𝝈 ∶ 𝝈 𝑡𝑟′ (14)
2
The plastic flow direction, 𝐧𝐩 , is computed from the trial deviatoric
stress using Eq. (15) based on normality principle.
′
3 𝝈 𝑡𝑟
𝐧𝐩 = (15)
2 𝜎𝑉𝑡𝑟𝑀 Fig. 1. Conversion of surface fluxes to volumetric heat loss term (a) general surface
losses from a body, (b) a finite element with surface flux from its top surface.
The plastic strain increment is computed using the flow direction,
𝐧𝐩 , based on normality rule with plastic strain increment, 𝛥𝑝, as shown
in Eq. (16).
surface flux which is applied only on the top surface. Therefore, the
𝛥𝝐 𝑝 = 𝐧𝐩 𝛥𝑝 (16) surface integral is converted into volume integrals by dividing the
flux by the thickness of the element, as in Eq. (22). In this study,
The plastic strain increment is coupled with the stresses through
one element per thickness direction is used for each layer; hence, the
a flow rule or yield function, 𝜓, that is a function of temperature.
Consistency requires Eq. (17) to hold during plasticity. surface heat flux, 𝐪𝐬 , is obtained by division of the effective surface flux
to the layer thickness, 𝛥ℎ. This term is applied to the model together
𝜓(𝝈, 𝛥𝑝) = 𝜎𝑉 𝑀 − 𝜎𝑦 (𝛥𝑝, 𝑇 ) = 0 (17) with the volumetric heat input.
Once the thermal and plastic strain increments are calculated, the 𝑞𝑒𝑓 𝑓
𝐪𝑠 · 𝐧 d𝑆 = ∇ · 𝐪𝑠 d𝑉 = d𝑉 (22)
elastic strain increment and the corresponding stress are computed ∫𝑆 ∫𝑉 ∫𝑉 𝛥ℎ
using Eqs. (10) and (12), respectively. The solution for the plastic strain The energy balance is solved at each time step until the temperature
increment (or stress) is obtained by iteratively until the increment is update for an iteration cycle becomes smaller than a tolerance; in this
less than a convergence tolerance. In this study, an isotropic elastic- study, 100 K was used as the tolerance value.
perfectly plastic model is employed. The same model is applied to the The mechanical equilibrium invokes a solution for the displace-
baseplate and build material considering different material properties. ments; hence, the displacements are discretized, as in Eq. (23).
In case the material state of active element is either powder or liquid, a
low value of elastic modulus (0.1 GPa) is used to compute the material 𝐮(𝐱) = 𝐍(𝐱) ̂
𝐮 (23)
tangent and the stresses are assumed to vanish.
Strains are obtained through strain–displacement coupling, 𝐁, as in
2.3. Finite element model Eq. (24).

𝝐 = 𝐁̂
𝐮 (24)
The local problem is converted into an FE problem by discretization
of temperatures, as in Eq. (18) [51]. The weak form of equilibrium is obtained by multiplying the local
form in Eq. (8) with the weight function and integrating the result
̂
𝑇 (𝐱) = 𝐍(𝐱) 𝐓 (18)
over the volume of the domain. Eq. (25) shows the discrete form of
The weak form of the energy balance is obtained by discretizing mechanical equilibrium.
the temperatures and multiplying the energy balance with a weight
function followed by integration. The solution of Eq. (19) gives nodal 𝐁𝑇 𝝈 d𝑉 = 𝟎 (25)
∫𝑉
̂ for boundary conditions [52].
temperatures 𝐓,
MATLAB® is used to generate the inputs for the FE model, mesh file,
̂−𝐓
𝐓 ̂𝑡
𝑇
[ 𝐍 𝜌 𝐶𝑝 𝐍 d𝑉 ] ̂
+ [ ∇𝐍𝑇 𝐤 ∇𝐍 d𝑉 ] 𝐓 input file for analysis (i.e., process parameters, material constants), and
∫𝑉 𝛥𝑡 ∫𝑉 parameters for path generation at the initialization stage for efficient
= 𝐍𝑇 𝑄̇ d𝑉 − 𝐍𝑇𝑠 𝐪𝑠 · 𝐧 d𝑆 (19) and consistent generation of the model. MSC MARC® is used as the FE
∫𝑉 ∫𝑆 environment with its available user subroutines that are in Fortran®
𝑞𝑒𝑓 𝑓 represents effective surface heat losses due to convection, radi- language, as shown in Fig. 2(a).
ation, and evaporation through the top surface of the topmost elements Fig. 2(b) shows the FE model and boundary conditions. Fixed dis-
(Eq. (20)). placements and a fixed temperature of 323 K (50 ◦ C) are applied to the
bottom surface of the baseplate as the boundary conditions.
𝑞̇ 𝑒𝑓 𝑓 = 𝑞̇ 𝑐𝑜𝑛𝑣 + 𝑞̇ 𝑟𝑎𝑑 + 𝑞̇ 𝑒𝑣𝑎 (20)
Several subroutines are used together in the simulations. Finite
In the FE model, after the addition of a new layer of material, the element solver visits the subroutines UACTIVE, USPCHT, ANKOND,
traction-free surfaces must be defined to identify heat losses. FLUX, and HYPELA2 for every element, integration point, during every
Fig. 1(a) shows the surface losses of 𝐪∗ and 𝐪∗∗ from two surfaces increment, and iteration cycle. Detailed description of the user-defined
𝑑𝑆 and 𝑑𝑆 ∗∗ with the corresponding surface normals of 𝐧∗ and 𝐧∗∗ ,
∗ subroutines is as follows:
respectively. The total loss can be expressed as in Eq. (21) [53].
• UBINC sets up the initial values of global variables and updates
(𝐪∗ · 𝐧∗ + 𝐪∗∗ · 𝐧∗∗ ) d𝑆 = ∇ · 𝐪 d𝑉 (21) the state variables used in the thermal and mechanical field
∫𝑆 ∫𝑉 solvers at the beginning of each increment. The first call and
Fig. 1(b) shows the surface losses on a single finite element during corresponding initialization of the software is performed within
the AM process in which the effective heat loss is in the form of a this subroutine.

4
C. Bayraktar and E. Demir Additive Manufacturing 54 (2022) 102708

Fig. 2. (a) Finite element model generation including the inputs, constants, and a list of the user-subroutines that are utilized in the thermomechanical finite element analysis.
(b) Mesh and boundary conditions of finite element model.

• UTIMESTEP is used to set the minimum and maximum limits for refinement level significantly effects the magnitude of the minimum
the time increment based on the position of the heat source which and maximum temperatures. The minimum temperature becomes lower
is determined by the absolute time. The time stepping is deter- than the baseplate temperature (323 K) for mesh refinement levels less
mined based on the convergence criteria. During the recoating or than 60%. Accordingly, minimum temperature reached almost absolute
cooling stages larger initial time steps are enforced. zero for the mesh refinements coarser than 30%, that is highly inac-
• UACTIVE activates the elements at the powder state after recoat- curate. This observation is attributed to the intense heat source input,
ing cycle before the deposition of a layer. The build elements are causing a very sharp temperature gradient near the melt-pool. Both the
deactivated initially only and baseplate elements are active all minimum and maximum temperatures have convergent behavior for a
the time during the process simulations. The activation criteria mesh refinement level finer than 60%.
is based on the absolute time of the simulation depending on the The mesh refinement does not impact the equivalent Von Mises
position of the heat source along build direction. stresses significantly because the stresses were approximately equal to
• INTCRDS gives the integration point coordinates of each element the yield strength as shown in Fig. 3(b). In contrast, the longitudinal
which is used only once at the beginning of the calculations. stress component reached a steady value at 50% mesh refinement,
• USPCHT is used to enter the specific heat that depends on the ma- as shown in Fig. 3(b). Considering the computational time, a mesh
terial type (baseplate or build layer), state (powder/liquid/solid), refinement of 60% is selected as the minimum refinement necessary to
and temperature. obtain temperatures and the stresses accurately. The converged solution
• ANKOND is used to enter the anisotropic conductivity constants consists of a mesh with 113 409 number of nodes and 99 200 elements
along x, y, and z directions that depends on the material type in total.
(baseplate or build layer), state (powder/liquid/solid), and tem-
perature. This subroutine is used to incorporate Maragoni effect 2.5. Material properties
by increasing conductivity along longitudinal direction with a
factor at liquid state, 𝑘𝑥 . Two types of material models are used in the model: baseplate and
• FLUX is used enter the coordinates and magnitude of the volu- build material. Table A.1 show the constant properties of baseplate
metric moving heat flux. Surface heat losses are also entered as (SS316). Figs. A.1 and A.2 show the temperature-dependent build
negative heat sources for the topmost elements in the layer only. material (IN718) properties. The following definitions for material
This subroutine is defined for only the build elements for time properties are implemented in the user-defined subroutines of the finite
efficiency. Thermal solver visits this subroutine only for the active element solver.
elements in the mesh. The build material consists of three different states: powder, liquid,
• HYPELA2 is used for the thermomechanical user-defined solver and solid. 𝜔 is used as an indicator to account for these different
for different states of the material. This subroutine gives the tem- material states.1
peratures to enter temperature dependent mechanical constitutive
models for powder/liquid/solid phases. Different thermal and • powder (𝜔 = −1)
mechanical properties for the baseplate and build elements were • liquid (𝜔 = 0)
used. Similarly, zero stress values with low stiffness properties • solid (𝜔 = 1)
were used to define mechanical response of powder and liquid
Any property in the liquid–solid state transformation regime, that is
states. HYPELA2 is accessed during the simulations for the active
also referred as the mushy regime, is computed by the rule of mixtures.
elements only.
Eq. (26) shows the application of the rule of mixtures method to find
the property at the corresponding state by using the solid, 𝜌𝑠 , and
2.4. Mesh sensitivity
liquid, 𝜌𝑙 , phases to determine the mass density in the mushy regime,
A mesh sensitivity study is conducted to select a proper mesh size 𝜌 [54].
for the FE model in the scanned region. The mesh convergence study 𝜌 = 𝜔 𝜌(𝑠) + (1 − 𝜔) 𝜌(𝑙) (26)
is performed for single layer deposition using the process parameters
shown in Table 1. The reference mesh size value was 50 µm that
has the same size as the radius of the laser beam. Fig. 3(a) shows 1
The state of the properties are represented by the superscripts ‘(p)’, ‘(s)’,
the maximum and minimum temperatures with respect to the mesh and ‘(l)’ for the powder, solid, and liquid states, respectively. That is, 𝐶𝑝(𝑠) is
refinement level. A higher percentage indicates a finer mesh. Mesh the specific heat of the solid.

5
C. Bayraktar and E. Demir Additive Manufacturing 54 (2022) 102708

Fig. 3. Mesh sensitivity study for a. temperature, b. longitudinal stress and Von Mises stress. Percentage mesh refinements were computed using a reference mesh size of 50 µm.

The powder material properties are computed using the properties The area fraction of the surface, 𝐴(𝐻) , refers to the areal fraction of
of solid and the porosity of the powder bed, 𝜑. Direct proportionality is the voids in a porous medium (Eq. (32)).
used to calculate the density, 𝜌(𝑝) , and specific heat, 𝐶𝑝(𝑝) , of the powder
0.908 𝜑
from the corresponding solid properties, 𝜌(𝑠) and 𝐶𝑝(𝑠) , as in Eqs. (27) 𝐴(𝐻) = (32)
1.908 𝜑 − 𝜑2 + 1
and (28). A powder porosity, 𝜑, of 0.4 was used in the calculations [16].
The emissivity, 𝜀(𝐻) , in porous media is calculated using Eq. (33).
[ ( )2 ]
𝜌(𝑝) = (1 − 𝜑) 𝜌(𝑠) (27) 1−𝜑
𝜀 (𝑠) 2 + 3.082
𝜑
𝐶𝑝(𝑝) = (1 − 𝜑) 𝐶𝑝(𝑠) (28) 𝜀(𝐻) = [ ( )2 ] (33)
(𝑠)
1−𝜑
Specific heat or heat capacitance shall be modified to take into 𝜀 1 + 3.082 +1
𝜑
account the latent heat effects [55]. In this study, the latent heat effects
were not considered. The latent effect would help reduce the num-
ber of iteration cycles of thermal solution by reducing the maximum 2.6. Time stepping procedure
temperatures thereby decreasing the number of cycles required for a
temperature increment.
The thermal conductivity of the powder material properties is a Two different time intervals are used in the model: scan time and
function of solid material properties and the powder porosity in the re-coating time. The scan time is divided by time increments of 𝛥𝑡𝑠𝑐𝑎𝑛 .
powder bed, 𝜑, in accordance with the Ref. [56]. In Eq. (29), 𝑘(𝐴𝑟) ’s After the completion of a layer, the recoating time, 𝛥𝑡𝑟𝑒𝑐𝑜𝑎𝑡 , is applied
the conductivity of argon gas surrounding the powder particles and 𝑘(𝑠) to account for the cooling during spreading of new powder with a
is the conductivity of the solid metal. recoating blade, as in the actual process. Although the recoating is
usually complete within seconds in a real LPBF process, the re-coating
[ 𝑘(𝑠)
𝜑 𝑓0 + (1 − 𝜑 𝑓0 ) time in the simulations is selected to be much less, that is sufficient
𝑘 (𝐴𝑟)
𝑘(𝜑) = 𝑘(𝐴𝑟) (1 − 𝑎) + enough for the model to cool down to the baseplate temperature
𝑘(𝑠)
1 − 𝜑(1 − 𝑓0 ) + 𝜑(1 − 𝑓0 ) completely considering due to the high cooling rates ( 104 K/s).
𝑘(𝐴𝑟)
( (𝑠) )2 (29)
𝑘 𝑘(𝑠) ] The time step during the deposition, 𝛥𝑡𝑠𝑐𝑎𝑛 , is determined automat-
2 (1 − 𝜑) + (1 + 2𝜑) ically by setting the tolerance value for temperature increment to 50
𝑘(𝐴𝑟) 𝑘(𝐴𝑟)
𝑎 K. The solution algorithm reduces the time step, if necessary, starting
𝑘 (𝑠)
(2 + 𝜑) +1−𝜑 from a predefined magnitude down to a prescribed minimum value.
𝑘(𝐴𝑟)
𝑓0 is a function between 0–1 for scaling conductivity, which is given The maximum time step was selected such that the position change of
by Eq. (30). the heat source during the time step lies within the effective radius of
the laser beam, 𝑟, in accordance with Eq. (34), which is similar to that
𝑓0 = 0.8 + 0.1 𝜑 (30) in [32].
The scaling factor, 𝑎, is also a function of the powder bed porosity 𝑟
𝛥𝑡𝑠𝑐𝑎𝑛 < (34)
(Eq. (2.5)). 𝑣
⎧−4.898𝜑, 0 ≤ 𝜑 < 0.0827 The effective size of the region, 𝑟, is calculated from the beam size
⎪ using Eq. (35).
log(𝑎) = ⎨−0.405 − 3.154 (𝜑 − 0.0827), 0.0827 ≤ 𝜑 < 0.298
⎪ √
⎩−1.084 − 6.778 (𝜑 − 0.298), 0.298 ≤ 𝜑 < 0.580 𝑟 = 𝑟2𝑥 + 𝑟2𝑦 + 𝑟2𝑧 (35)
The conductivity of the liquid state along the scan direction (𝑘𝑥 )
is assumed to be 1.5 times higher than its actual magnitude to ac- The time step must also be selected considering the thermal shock
count for the enhanced heat transfer inside the melt-pool due to the effects; otherwise, negative temperatures can be observed in the simula-
Marangoni effect. Table A.1 shows the values of temperature-dependent tions. To avoid this, the restriction in Eq. (36) is used in the time step
conductivity for IN718. during scanning, 𝛥𝑡𝑠𝑐𝑎𝑛 , in which 𝐿 is the characteristic element size
The emissivity of the powder particles, 𝜀(𝑝) , is computed from the along the scan direction, 𝑘𝑥 is the conductivity along the scan direction,
emissivity of the same material in the solid phase, 𝜀(𝑠) , considering the 𝜌 is the density, and 𝐶𝑝 is the specific heat of the material [57].
radiation between the powder particles, Eq. (31).
𝐿2 𝜌 𝐶𝑝
𝜀 (𝑝)
= 𝐴 (𝐻) (𝐻)
𝜀 + (1 − 𝐴 (𝐻) (𝑠)
)𝜀 (31) 𝛥𝑡𝑠𝑐𝑎𝑛 > (36)
𝑘𝑥

6
C. Bayraktar and E. Demir Additive Manufacturing 54 (2022) 102708

Table 1
Default process parameters for selective laser melting of EOS-M209 LPBF system
[31].
Laser power Scan speed Beam radii Hatch spacing Layer thickness Efficiency
𝑃 𝑣 𝑟 𝑥 , 𝑟𝑦 , 𝑟𝑧 𝛥𝑑 𝛥ℎ 𝜂
[W] [mm/s] [μm] [μm] [μm] –
285 960 50/50/125 110 40 0.3

2.7. Process parameters

Table 1 lists the process parameters for selective laser melting,

which include laser power, scan speed, beam radii, hatch spacing, and
layer thickness.
The efficiency factor, 𝜂, is different for the powder and solid states.
A larger efficiency is used for the powder state, 𝜂 (𝑝) = 0.6, than for the
solid state, 𝜂 (𝑠) = 0.3.

3. Inherent strain method

Fig. 4. Schematic sketch of stress–strain diagram during cooling of a single material
The inherent strain method (ISM) is used to calculate stresses in- point.
stead of strains from temperature distributions in the current study.
Inherent strains, 𝜖𝐼𝑆 , are computed based on the maxima of the temper-
atures [41]. ISM computes the strains based on the certain temperature in the formation of residual compressive strains and tensile stresses
limiting values. Eq. (37) shows the criteria based on the temperature after the cooling down. Therefore, the amount of plastic straining
limits 𝑇1 and 𝑇2 . 𝛽 represents the magnitude of rigidity constraints occurs in both heating and cooling stages depending on a critical
along each direction, and it takes a value between 0 and 1. temperature, 𝑇𝑚𝑎𝑥 = 𝑇2 . The temperature limits 𝑇1 and 𝑇2 are also
dependent on rigidity constraints that are similar to spring stiffness.
⎧0, 𝑇𝑚𝑎𝑥 < 𝑇1
⎪ Because the constraint in the longitudinal direction is greater than that
𝜖𝐼𝑆 = ⎨−𝛼 𝑇𝑚𝑎𝑥 + 𝜎𝑦 ∕𝐸∕𝛽 𝑇1 < 𝑇𝑚𝑎𝑥 < 𝑇2 (37) of the transverse direction, 𝑇1𝑦 is larger than 𝑇2𝑥 . Hence, even though
⎪
⎩−𝜎𝑦 ∕𝐸∕𝛽 𝑇2 < 𝑇𝑚𝑎𝑥 the maximum temperature within the material exceeds 𝑇1𝑥 , only the
The temperature limits are related to the material properties and compressive plastic strain occurs in the longitudinal direction up to
rigidity constraints with Eqs. (38) and (39). 𝑇𝑚𝑎𝑥 = 𝑇2𝑥 , and it is a function of 𝑇𝑚𝑎𝑥 . In case 𝑇𝑚𝑎𝑥 exceeds 𝑇2𝑥 ,
further plastic strain at both the heating (compressive) and cooling
𝑇1 = 𝜎𝑦 ∕𝐸∕𝛼∕𝛽 + 𝑇0 (38) (tensile) stages, the residual stresses will retain their magnitude, the
𝑇2 = 2 𝑇1 (39) same as the yield stress along the longitudinal direction. In this regime,
the plastic strain becomes proportional to 𝑇𝑚𝑎𝑥 , as shown in Eq. (43)
In the original inherent strain formulation [41], the longitudinal along the transverse direction. 𝐶 is a constant that is computed as
direction (𝑥-direction) was aligned with the weld or velocity vector. 𝐶 = 𝑇2𝑥 from the continuity condition of Eqs. (43) and (45). 𝐴 is
Therefore, in the previous studies, welding directions were always an unknown constant that needs to be determined by calibration. A
along the longitudinal direction, which did not require any transforma- detailed description of ISM can be found in previous studies [28].
tion [45,58] which is the same in the current study. However, to keep
generality for AM processes rotation, 𝐑, has to be defined to transform ⎧0, 𝑇𝑚𝑎𝑥 < 𝑇1𝑥
⎪
the stresses, 𝐒, at the reference frame defined with scan direction to the 𝜖𝑥𝑥 = ⎨−𝛼 𝑇𝑚𝑎𝑥 + 𝜎𝑦 ∕𝐸∕𝛽𝑥 𝑇1𝑥 < 𝑇𝑚𝑎𝑥 < 𝑇2𝑥 (42)
sample reference, Eq. (40). ⎪
⎩−𝜎𝑦 ∕𝐸∕𝛽𝑥 𝑇2𝑥 < 𝑇𝑚𝑎𝑥
{
𝝈 = 𝐑 𝐒 𝐑𝑇 (40) 0, 𝑇𝑚𝑎𝑥 < 𝑇2𝑥
𝜖𝑦𝑦 = (43)
Transformation involves rotations, 𝐑, in which 𝜃 is the rotation −𝐴𝛼(𝑇𝑚𝑎𝑥 − 𝐶) 𝑇2𝑥 < 𝑇𝑚𝑎𝑥
angle about the build direction (z-axis of the sample reference), as in ISM predicts stresses by using only the thermal solution, more
Eq. (41). The direction of the scan in the plane, 𝜃, needs to be known precisely the maximum temperatures, 𝑇𝑚𝑎𝑥 . The maximum stress in the
or entered for each layer in this case. In the current study the scan longitudinal direction is assumed to be equal to the yield stress if 𝑇𝑚𝑎𝑥
direction is the same as 𝑥-direction of the sample giving 𝜃 = 0◦ . exceeds 𝑇2𝑥 . On the other hand, the transverse component of stress is
⎡ cos(𝜃) sin(𝜃) 0⎤ assumed to be a function of 𝑇𝑚𝑎𝑥 , as shown in Fig. 4.
𝐑 = ⎢− sin(𝜃) cos(𝜃) 0⎥ (41) The constant 𝐴𝑠 needs to be calibrated using the results of thermo-
⎢ ⎥
⎣ 0 0 1⎦ mechanical simulations as a function of the maximum temperatures,
𝑇𝑚𝑎𝑥 . Note that calibration can also be done experimentally. The cal-
In the case of inherent strain method, only the thermal problem
ibration was performed using only for a set of process parameters.
was solved, and the maximum temperature values are stored for each
The magnitudes of 𝑇𝑚𝑎𝑥 and 𝜎𝑦𝑦 were collected from thermomechanical
integration point of each finite element. The stresses are then computed
simulations for three different laser power levels which allowed to
by subsequent addition of a layer with inherent strains and quasi-static
investigate the effect of different 𝑇𝑚𝑎𝑥 values.
mechanical solutions for the displacement field [41,42,45,58].
{
Fig. 4 illustrates the thermal cycle of a material during deposi- 0, 𝑇𝑚𝑎𝑥 < 𝑇2𝑥
tion schematically. Eqs. (42) and (43) relate temperatures to inherent 𝜎𝑥𝑥 = (44)
𝜎𝑦 , 𝑇2𝑥 < 𝑇𝑚𝑎𝑥
strains. As the material heats up, the displacement constraint prevents {
thermal expansion of the build layer that leads to compressive strains. If 0, 𝑇𝑚𝑎𝑥 < 𝑇2𝑥
𝜎𝑦𝑦 = (45)
the stresses exceed the yield stress, plastic straining starts which results 𝐴𝑠 𝐸𝛼(𝑇𝑚𝑎𝑥 − 𝐶), 𝑇2𝑥 < 𝑇𝑚𝑎𝑥

7
C. Bayraktar and E. Demir Additive Manufacturing 54 (2022) 102708

Fig. 5. Melt-pool images obtained for laser power magnitudes of (a) 160 W, (b) 190 W, and (c) 220 W at a scan speed of 960 mm/s.

Fig. 6. Experimental characterization of a test sample with multi layer builds that was manufactured with the default process parameters of 285 W, 960 mm/s, 80 μm of laser
power, scan speed, and layer thickness, respectively. (a) SEM images with different magnification, (b) EBSD map, (c) texture pole figures (RD is along the build direction).

In this study, the stress components along longitudinal and trans- the cross-sectioned surfaces. After that, surfaces were etched with a
verse directions were obtained using Eqs. (44) and (45). Finally, the equal mixture of HCl, CH3COOH and HNO3. Finally, melt-pool images
magnitudes and distributions of these stresses were predicted using were obtained by Nikon LV1000ND optical microscope as shown in
only the 𝑇𝑚𝑎𝑥 value at each integration point of every element, which Fig. 5. Dimensions of melt-pools were measured using a raster graphics
were determined from the thermal simulations for any given process editor. Each optical measurement was performed ten times to quantify
parameters. measurement uncertainties. Depth of the melt-pools were measured
with respect to the baseplate level using a similar method as in the
4. Experimental
Ref. [59].
Another cubic sample was prepared with 10 mm edge length using
Experimental studies include analysis of cross-sections of deposited
the standard manufacturing parameters of EOS-M290 LPBF system. The
specimens that were manufactured with different levels of laser power.
process parameters were 285 W, 960 mm/s, 80 μm for laser power,
The manufacturing of the samples were performed using EOS-M290
LPBF system. Single tracks of IN718 material were deposited on SS316 scan speed, and layer thickness, respectively. Sample was wet ground
base plate. Three different laser power values were used as 160 W, and polished with a final solution of colloidal silica. The analysis
190 W, and 220 W with the default scan speed of 960 mm/s. 40 mm was performed using ZEISS-Cross beam scanning electron microscopy
length was used in the tracks and each track was transversely posi- (SEM) and electron backscattered diffraction (EBSD) system. Fig. 6
tioned 3 mm away from each other to prevent any interaction between shows results of the micro scale characterization of a test sample. SEM
the build single tracks. Cross-section of build layers of single track images shows a uniform fusion of the melt zones with each other.
samples were investigated to find the dimensions of the melt-pool size The melt zone structures consisted of needle-like dendritic structures,
to validate the proposed thermal model. Fig. 6(a). High resolution EBSD analysis of the cross-section shows large
The samples were cut in the middle using abrasive water-jet cutting grains with growth along the build direction, Fig. 6(b). In addition,
process. Next, epoxy resin and hardener used to apply cold mount- the texture pole figures reveals the presence of a cube texture in the
ing. Then, wet polishing was applied using SiC grinding papers to material, Fig. 6(c). Therefore, the build material had proper fusion

8
C. Bayraktar and E. Demir Additive Manufacturing 54 (2022) 102708

Fig. 7. (a) Temperature distribution and (b) state of the material phase during deposition of first layer at 𝑡 = 0.00899 s. Process parameters: 𝑃 = 160 W, 𝑣 = 1000 mm/s, 𝛥𝑑
= 90 μm, 𝛥ℎ = 40 μm, and unidirectional scan pattern. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Table 2 Table 3
Process parameters that were used in the simulations. Melt-pool dimensions for altering process parameters.
Laser power Scan speed Hatch spacing Layer thickness Scan pattern Laser Scan speed Hatch spacing Layer Width Depth Length
𝑃 𝑣 𝛥𝑑 𝛥ℎ – power thickness
[W] [mm/s] [μm] [μm] – [W] [mm/s] [μm] [μm] [μm] [μm] [μm]
160 600 75 30 Unidirectional 160 1000 90 40 124 77 276
190 1000 90 40 Zigzag 190 1000 90 40 130 86 333
220 1200 100 50 – 220 1000 90 40 149 93 402
160 600 90 40 160 95 284
160 1200 90 40 121 68 272
160 1000 75 40 122 81 274
of deposited layers having both crystallographic and morphological 160 1000 100 40 127 78 281
160 1000 90 30 121 72 265
texture. Although a less sharper texture is expected in case lower laser 160 1000 90 50 123 76 283
power levels were used for IN718 [60]. The presence of significant
amount of texture challenges the isotropic assumption used in the
model and suggest the use of a texture-based calculation of thermal Table 4
Comparison between melt-pool dimensions gathered from experiments and simulation
and mechanical properties of the material such as thermal conductivity,
results.
elastic modulus, and yield stress.
Laser Scan Layer Width Width err. Depth Depth err.
power speed thick. exp. sim. exp. sim.
5. Results and discussion [W] [mm/s] [μm] [μm] [μm] [%] [μm] [μm] [%]
160 960 40 116 ± 2 122 +5.2 68 ± 2 74 +8.8
Fig. 7 shows the instantaneous distributions of temperatures and 190 960 40 125 ± 4 128 +2.4 88 ± 4 81 −7.9
material states during the deposition of the first layer of the thermo- 220 960 40 138 ± 3 144 +4.3 95 ± 2 89 −6.3
mechanical model. The laser power, scan speed, hatch spacing, and
layer thickness were 160 W, 1000 mm/s, 90 μm, 40 μm, respectively
and a uni-directional scan pattern was used in the simulations. The decreases and does not become influenced by the scan speed as much
melt region is indicated with a gray colored contour in Fig. 7(a). The as the other directions. Variations in layer thickness and hatch spacing
conductivity of the material in the powder phase is much lower than do not have a significant effect on the melt-pool size.
that of the solid or liquid phases that leads to the asymmetric tem- Fig. 8 shows the temperature distribution within the melt-pool for
perature distribution. Therefore, higher temperatures were computed the following process parameters: laser power of 160 W, scan speed of
on the solidified region in comparison to powder region with much 1000 mm/s, hatch spacing of 90 μm, and layer thickness of 40 μm. The
lower thermal conductivity. Fig. 7(b) shows the instantaneous states melt-pool dimensions were calculated based on the width and depth
of powder, solid and liquid during deposition. dimension indicated in Fig. 8(a) and (b), respectively.
Table 2 shows the values of the process parameters that were used in Melt-pool dimensions from the cross-sections were also obtained
the simulations; laser power, scan speed, hatch spacing, layer thickness, from single track simulations and compared to the experimental find-
and scan pattern. ings to identify the simulation accuracy. Fig. 9 shows the image of the
Table 3 shows the melt-pool dimensions with respect to different melt-pool obtained from the experimental study and from the simula-
process parameters that were obtained from the thermomechanical tions for the same process parameters (𝑃 = 190 W, 𝑣 = 960 mm/s, and
simulations. The simulation results were collected at the instant when 𝛥ℎ = 40 μm.). The track lengths were 1 mm long, the same as in the
the laser was in the middle of the fifth scan vector, after the deposition simulations which was selected for simulation efficiency.
of one half of the hatch, or at the center of the printed layer. As In Table 4 melt dimensions that were measured from experimental
the laser power increases, the melt-pool expands in all directions at cross-sections are compared with the values obtained from the simu-
approximately at the same rate. Increasing the scan speed decreases lations. The difference in the experiments and simulations (error) was
the amount of heat energy input per unit volume of material and con- less than 10% in both width and depth directions for three different
sequently, both the depth and width of the melt-pool become smaller. laser power values. This clearly shows accurate predictive capacity of
Similarly, the melt-pool slightly shortens along the scan direction; the melt-pool dimensions with the proposed model.
however, at the same time, increasing speed tends to stretch the melt Fig. 10 shows the variation of the maximum temperature with
zone longitudinally. As a result, the length of the melt-pool barely respect to process parameters. As expected, the maximum temperature

9
C. Bayraktar and E. Demir Additive Manufacturing 54 (2022) 102708

Fig. 8. Melt-pool during deposition (a)top view and (b) cross-sectional view. 𝑃 = 160 W, 𝑣 = 1000 mm/s, 𝛥𝑑 = 90 μm, 𝛥ℎ = 40 μm, and uni-directional scan pattern.

Fig. 9. Melt-pool cross sections from (a) experimental study and (b) simulation from a single track deposition. 𝑃 = 190 W, 𝑣 = 960 mm/s, and 𝛥ℎ = 40 μm.

Fig. 10. Effect of laser power, scan speed and layer thickness on the maximum temperature.

was directly proportional to the laser power and inversely proportional transverse stresses were directly proportional to the layer thickness.
to the scan speed similar to literature findings [61]. Increasing the This is attributed to the constant thermal load applied over a larger
layer thickness increases the unit powder volume that absorbs the thickness, and thereby lowering the stresses with increasing layer thick-
energy, thereby slightly decreasing the maximum temperatures, that nesses which has been also concluded in several studies [29,63]. In
is in agreement with the previous studies [62]. In addition, increasing addition, the average transverse stresses was directly proportional to
the hatch spacing from 75 μm to 100 μm had a negligible effect on the
the laser power and an inverse proportionality with the scan speed that
maximum temperatures. All four process parameters had a different
is consistent with previous articles [26,64]. However, the influence of
effect on the maximum temperatures of which the laser power and scan
scan speed on transverse stresses was not significant, which has been
speed was the most dominant.
Fig. 11 shows the effect of process parameters on the average also confirmed by Shiomi et al. [28]. The hatch spacing did not have
longitudinal and transverse normal stresses in the build layer after a considerable effect on stress. Residual stresses along the longitudinal
cooling down. The maximum value of the longitudinal stress was very direction were approximately 150% greater than the stresses along the
close to the yield strength of the material and it was not affected transverse direction for all of the parameters which is in agreement
by any of the process parameters. Both the average longitudinal and with the similar experimental findings in the literature [30,65].

10
C. Bayraktar and E. Demir Additive Manufacturing 54 (2022) 102708

Fig. 11. Effect of laser power, scan speed and layer thickness on average longitudinal, average transverse and average Von-Mises stresses after cooling down to ambient temperature.

Fig. 12. Thermomechanical simulation results for different scan patterns: a. 𝜎𝑥𝑥 [MPa] for unidirectional, (a) 𝜎𝑥𝑥 [MPa] for unidirectional, (b) 𝜎𝑥𝑥 [MPa] for zigzag, (c) 𝜎𝑦𝑦 [MPa]
for unidirectional, and (d) 𝜎𝑦𝑦 [MPa] for zigzag. 𝑃 = 160 W, 𝑣 = 1000 mm/s, 𝛥𝑑 = 90 μm, 𝛥ℎ = 40 μm.

Fig. 12 shows the effect of unidirectional and zigzag scan patterns was much higher in comparison to the unidirectional pattern and
on the resultant stress field using thermomechanical simulations. Both hence causing higher thermal gradients and higher stresses. In contrast,
the average and maximum stresses for the two components were almost each scan track starts near the end of the previously scanned track,
the same in terms of magnitude, while the stress results were slightly thus causing less temperature differences between the two consecutive
different. The temperature difference at the starting point of a track tracks for the zig-zag pattern. However, the longer distance traveled

11
C. Bayraktar and E. Demir Additive Manufacturing 54 (2022) 102708

Fig. 13. Stress distributions after cooling down of the single layer. (a) 𝜎𝑥𝑥 [MPa], (b) 𝜎𝑦𝑦 [MPa], (c) 𝜎𝑧𝑧 [MPa], (d) 𝜏𝑥𝑦 [MPa], (e) 𝜏𝑦𝑧 [MPa], (f) 𝜏𝑥𝑧 [MPa]. 𝑃 = 160 W, 𝑣 =
1000 mm/s, 𝛥𝑑 = 90 μm, 𝛥ℎ = 40 μm, and unidirectional scan pattern.

between the start and the end of a single zig-zag loop leads to a higher The computational time was reduced by approximately six times in
temperature gradient. Consequently, the stress distribution of zigzag comparison to the thermomechanical solution.
showed a relatively higher alternating behavior with as can be observed Table 5 presents a comparison of the simulation results of the
in Fig. 12(b). This result is in agreement with the findings of Parry ISM and thermomechanical solution for different process parameters.
et al. [32]. The stress values were obtained by taking the average values over
Fig. 13 shows all six stress components at the end of the thermo- the surface of the deposited region/area. The resulting distributions
mechanical simulation after cooling down to baseplate temperature. showed the negligible effect of the process parameters on the average
Longitudinal and transverse stress components were dominant in terms of longitudinal stress component. On the contrary, the ISM successfully
of magnitudes in comparison to the others. The maximum of the normal predicts longitudinal stress distributions with a reasonable accuracy.
stress along the build direction and shear stresses were at least and ISM predicted stresses within approximately 15% variation with re-
order of magnitude lower than the in-plane stresses. The corners at the spect to the thermomechanical results by using only the maximum
boundaries of the deposition area act as stress concentrations, hence
temperature information. The only exceptional parameter was the layer
local stresses were greater in those regions. The powder region, on the
thickness, which had the largest mismatch between longitudinal and
other hand, had relatively low stresses. Since a relatively low elastic
transverse stresses of 13.2% and 15.1%, respectively. This is actually an
modulus (approximately 0.1 GPa) was used in the simulations for
expected result since ISM does not take into account the layer thickness.
powder, elements at powder state experience large deformations with
The volume of the domain changes with varying layer thickness. But
relatively lower stresses.
the thermal loads remain the same for the same power and scan speed.
Fig. 14 shows distributions of longitudinal and transverse stresses
of the thermomechanical and ISM solutions. These figures show the Stress as being force over a unit area becomes inversely proportional
stresses after the cool down of build layer to a steady-state base- to the layer thickness which was consistent with the findings in the
plate temperature completely. The process parameters were 160 W, literature [29,63]. The ISM shall further be improved by using a
1000 mm/s, 90 μm, and 40 μm for laser power, scan speed, hatch correction factor to account for the layer thickness.
spacing, and layer thickness, respectively. The distributions of both Fig. 15 shows the comparison of the current numerical to simulation
longitudinal and transverse stresses of the ISM results were consistent and experimental findings in the literature for IN718. The average
with those of the thermomechanical solution. The region that had the stresses were compared only to be able compare to the simulation
largest longitudinal stresses for the ISM results were slightly greater findings in the literature. However, the stresses are over predicted
than that for the thermomechanical solution. Besides, the maximum by the simulations in comparison the experiments. The model needs
magnitude of the transverse stresses was slightly over-predicted for the further improvement such as incorporation of latent heat effect to
ISM method in comparison to that for the thermomechanical solution. reduce the maximum temperatures hence stresses.

12
C. Bayraktar and E. Demir Additive Manufacturing 54 (2022) 102708

Fig. 14. Results of stress components for different methods: (a) 𝜎𝑥𝑥 [MPa] of ISM, (b) 𝜎𝑥𝑥 [MPa] of thermomechanical solution, (c) 𝜎𝑦𝑦 [MPa] of ISM, and (d) 𝜎𝑦𝑦 [MPa] of
thermomechanical solution. 𝑃 = 160 W, 𝑣 = 1000 mm/s, 𝛥𝑑 = 90 μm, 𝛥ℎ = 40 μm.

Table 5
Comparison between average stresses computed by thermomechanical solution and ISM results.
laser power scan speed hatch spacing layer thick. thermomechanical ISM err. thermomechanical ISM err.
[W] [mm/s] [μm] [μm] 𝜎𝑥𝑥 [MPa] 𝜎𝑥𝑥 [MPa] [%] 𝜎𝑦𝑦 [MPa] 𝜎𝑦𝑦 [MPa] [%]
160 1000 90 40 1085 1100 +1.4 661 654 −1.1
190 1000 90 40 1082 1100 +1.7 703 704 +0.1
220 1000 90 40 1066 1100 +3.2 736 733 −0.4
160 600 90 40 1082 1100 +1.7 669 717 +7.2
160 1200 90 40 1060 1100 +3.8 658 632 −4.0
160 1000 75 40 1082 1100 +1.7 660 671 +1.7
160 1000 100 40 1080 1100 +1.9 689 635 −7.8
160 1000 90 30 1174 1100 −6.3 763 651 −14.7
160 1000 90 50 972 1100 +13.2 578 665 +15.1
260 1500 90 40 1103 1100 −0.3 788 718 −8.9

Simulation findings of Cheng et al. [66] for line scan pattern reveals reason, the experimental stresses were significantly lower than the
average longitudinal stress magnitude that was close to the current theoretically predicted stresses. In Refs. [68,69], the direction of the
study. Similarly, the numerical study of Denlinger et al. captured measured stress component were not indicated besides, the stresses
approximately the same trends as in the current study for the av- were obtained by averaging over the build region for comparison. In
erage of the longitudinal stress component [17]. However, in both the Ref. [67], although the effect of overlap was investigated, it is
references the transverse stresses were over predicted in comparison possible to conclude from the unidirectional patterns where the stresses
the current study. On the other hand, Chen et al. predicted a lower reaching to a steady-state magnitude. Accordingly the experimental
average for the transverse stress component [67]. The experimental stress results for the long line builds that were obtained by contouring
works were obtained using an average over a large build region that method were provided for comparison [30]. In general, the averaging
was necessary to carry out experimental stress measurements. For this of the experimental results revealed lower stress magnitudes than the

13
C. Bayraktar and E. Demir Additive Manufacturing 54 (2022) 102708

Fig. 15. Comparison of residual stress result with various simulation (sim.) and experimental (exp.) findings from literature for IN718.

numerical predictions although the simulations were consistent with CRediT authorship contribution statement
the similar studies in the literature.
Thermomechanical process simulations have numerical challenges Can Bayraktar: Data curation, Validation, Visualization, Software,
due to the existence of high temperature gradients leading to large Writing – original draft. Eralp Demir: Conceptualization, Methodology,
thermal strains and translations between the very soft powder and the Software, Supervision, Visualization, Writing – review & editing.
other states. Therefore, the ISM provides an easy alternative to find the
stresses directly from a thermal solution. However, ISM method fails to Declaration of competing interest
correlate to layer thickness. In addition, further studies are necessary to
check its effectiveness in multiple layer simulations, and its predictive The authors declare that they have no known competing finan-
capability at part scale through part scale experimental measurements. cial interests or personal relationships that could have appeared to
influence the work reported in this paper.
6. Conclusions
Acknowledgments
In this study, a numerical methodology was developed in a commer-
cial FE environment with user-subroutines to simulate thermomechan- We greatly acknowledge the scholarship awarded by Sabanci Uni-
ical LPBF process with temperature- and state-dependent properties. In versity - Faculty of Natural Sciences and Engineering, which is used
addition, an inherent strain method was implemented and its compar- to fund the graduate student (C.B.) to perform the research activities
ison to the thermomechanical model reveals the following important in this study. We thank M.Sc. Kerem Dortkasli at Sabanci University
conclusions: and Dr. Kemal Davut at Izmir Institute of Technology for the electron
microscopy images.
• The thermomechanical model predicts experimental melt-pool
dimensions accurately, by considering the temperature and state- Appendix A. Material properties
dependent material properties.
• Expression of surface heat losses in terms of a volumetric heat Note that the latent heat effect, that causes a jump in the heat
loss terms simplifies the need for a definition of free surfaces and capacitance at the solid to liquid transition state, can be incorporated
their update after deposition of each layer. to the heat capacitance as in the Refs. [55] (see Table A.2).
• The process parameters do not significantly change the residual
stresses along the scanning direction which are close to the yield
stress. Table A.1
• The ISM revealed nearly six times faster solution than that of the Constant properties of stainless steel (SS316) baseplate material (SS316) [71].
thermomechanical approach within approximately 10% accuracy. 𝜌 𝐶𝑝 𝑘𝑠 𝛼 𝐸 𝜎𝑦 𝜈
[kg/m3 ] [J/K/kg] [W/m/kg] [K−5 ] [GPa] [MPa]
• Thermomechanical model reveals residual stresses that are in-
versely proportional to layer thickness. The ISM procedure needs 7700 450 14 1.17 198.5 282 0.3

further improvement to consider the effect of layer thickness.

• In this study latent heat effect is not considered that has led to
the over prediction of temperatures and stresses. It is essential to
Table A.2
incorporate the latent heat effect for more accurate temperature Evaporation heat loss constants in Eq. (7) [50].
and stress predictions. 𝛥𝐻𝜈∗ 𝑀 𝑅 𝑇𝜈 𝑃0
[J/g] [g/mol] [J/mol/K] [K] [Pa]
The proposed ISM method and its comparison to thermomechanical 6.4 × 103 59.75 8.314 3188 101 325
solution for a single layer was the initial attempt complete this study.
This method will be employed to simulate multi-layered parts to predict
distortions and residual stresses at part scale accurately.

14
C. Bayraktar and E. Demir Additive Manufacturing 54 (2022) 102708

Fig. A.1. Temperature dependent thermal properties of IN718 [70]: (a) solid volume fraction, 𝑓 (𝑠) , (b) emissivity, 𝜖, (c) heat capacitance, 𝜌𝐶𝑝 , and (d) thermal conductivity, 𝑘.
The black, blue, and red lines indicate the different fits for solid, mushy, and liquid states. (For interpretation of the references to color in this figure legend, the reader is referred
to the web version of this article.)

Fig. A.2. Temperature-dependent mechanical properties of IN718 [72]: (a) thermal expansion coefficient, 𝛼, (b) Poisson’s ratio, 𝜈, (c) elastic modulus, 𝐸, and (d) yield strength,
𝜎𝑦 .

Appendix B. Supplementary data compatible with MSC MARC® software. We also provide MATLAB® m-
files to generate the input files for the finite element analysis and the
procedure file to construct the finite element model file.
Supplementary data contains the sofware developed within this
study. We provided the user-subroutines for MSC MARC® for the
thermomechanical modeling of PBF process under fortran folder. Note Supplementary material related to this article can be found online
that the main runner file is ’SUPBF_TU.f’. These subroutines are only at https://doi.org/10.1016/j.addma.2022.102708.

15
C. Bayraktar and E. Demir Additive Manufacturing 54 (2022) 102708

References [25] B. Vrancken, R. Wauthle, J.P. Kruth, J. Van Humbeeck, Study of the influence
of material properties on residual stress in selective laser melting, in: 24th
International SFF Symposium - An Additive Manufacturing Conference, SFF 2013,
[1] M.K. Thompson, G. Moroni, T. Vaneker, G. Fadel, R.I. Campbell, I. Gibson, A.
2013, pp. 393–407.
Bernard, J. Schulz, P. Graf, B. Ahuja, et al., Design for additive manufacturing:
Trends, opportunities, considerations, and constraints, CIRP Ann. 65 (2) (2016) [26] T. Simson, A. Emmel, A. Dwars, J. Böhm, Residual stress measurements on aisi
737–760. 316l samples manufactured by selective laser melting, Addit. Manuf. 17 (2017)
183–189.
[2] J.-P. Kruth, M.-C. Leu, T. Nakagawa, Progress in additive manufacturing and
[27] J.L. Bartlett, X. Li, An overview of residual stresses in metal powder bed fusion,
rapid prototyping, CIRP Ann.-Manuf. Technol. 47 (2) (1998) 525–540.
Addit. Manuf. 27 (January) (2019) 131–149, http://dx.doi.org/10.1016/j.addma.
[3] T. DebRoy, H. Wei, J. Zuback, T. Mukherjee, J. Elmer, J. Milewski, A.M.
2019.02.020.
Beese, A. Wilson-Heid, A. De, W. Zhang, Additive manufacturing of metallic
[28] M. Shiomi, K. Osakada, K. Nakamura, T. Yamashita, F. Abe, Residual stress
components–process, structure and properties, Prog. Mater. Sci. 92 (2018)
within metallic model made by selective laser melting process, CIRP Ann.
112–224.
- Manuf. Technol. 53 (1) (2004) 195–198, http://dx.doi.org/10.1016/S0007-
[4] D. Bourell, J.P. Kruth, M. Leu, G. Levy, D. Rosen, A.M. Beese, A. Clare, Materials
8506(07)60677-5.
for additive manufacturing, CIRP Ann. 66 (2) (2017) 659–681.
[29] L.S. Anderson, A.M. Venter, B. Vrancken, D. Marais, J.V. Humbeeck, Investigating
[5] C. Lindemann, U. Jahnke, M. Moi, R. Koch, Analyzing product lifecycle costs for the residual stress distribution in selective laser melting produced Ti-6Al-4V using
a better understanding of cost drivers in additive manufacturing, in: 23th Annual neutron diffraction, Mech. Stress Eval. Neutron Synchrotron Radiat. 4 (2018)
International Solid Freeform Fabrication Symposium–an Additive Manufacturing 73–78, http://dx.doi.org/10.21741/9781945291678-11.
Conference. Austin Texas USA 6th-8th August, 2012. [30] J. Robinson, I. Ashton, P. Fox, E. Jones, C. Sutcliffe, Determination of the
[6] M. Galati, L. Iuliano, A literature review of powder-based electron beam melting effect of scan strategy on residual stress in laser powder bed fusion additive
focusing on numerical simulations, Addit. Manuf. 19 (2018) 1–20. manufacturing, Addit. Manuf. 23 (2018) 13–24.
[7] J. Smith, W. Xiong, W. Yan, S. Lin, P. Cheng, O.L. Kafka, G.J. Wagner, J. Cao, [31] E. GmbH, Corporate website, 2020, URL http://www.eos.info/en (Accessed:
W.K. Liu, Linking process, structure, property, and performance for metal-based 17.05.2020).
additive manufacturing: computational approaches with experimental support, [32] L. Parry, I. Ashcroft, R.D. Wildman, Understanding the effect of laser scan
Comput. Mech. 57 (4) (2016) 583–610. strategy on residual stress in selective laser melting through thermo-mechanical
[8] M.M. Francois, A. Sun, W.E. King, N.J. Henson, D. Tourret, C.A. Bronkhorst, simulation, Addit. Manuf. 12 (2016) 1–15.
N.N. Carlson, C.K. Newman, T.S. Haut, J. Bakosi, et al., Modeling of additive [33] S.D. Bagg, L.M. Sochalski-Kolbus, J.R. Bunn, The effect of laser scan strategy
manufacturing processes for metals: Challenges and opportunities, Curr. Opin. on distortion and residual stresses of arches made with selective laser melting,
Solid State Mater. Sci. 21 (LA-UR-16-24513; SAND-2017-6832J) (2017). 2016.
[9] W. King, A.T. Anderson, R.M. Ferencz, N.E. Hodge, C. Kamath, S.A. Khairallah, [34] C. Li, J. Liu, Y. Guo, Efficient multiscale prediction of cantilever distortion by
Overview of modelling and simulation of metal powder bed fusion process selective laser melting, in: Proceedings of the 27th Annual International Solid
at lawrence livermore national laboratory, Mater. Sci. Technol. 31 (8) (2015) Freeform Fabrication Symposium, 2016, pp. 236–246.
957–968. [35] S. Afazov, W.A. Denmark, B.L. Toralles, A. Holloway, A. Yaghi, Distortion
[10] A. Bauereiß, T. Scharowsky, C. Körner, Defect generation and propagation prediction and compensation in selective laser melting, Addit. Manuf. 17 (2017)
mechanism during additive manufacturing by selective beam melting, J. Mater 15–22.
Process. Technol. 214 (11) (2014) 2522–2528. [36] N. Patil, D. Pal, B. Stucker, et al., A new finite element solver using numer-
[11] L.-E. Lindgren, A. Lundbäck, M. Fisk, R. Pederson, J. Andersson, Simulation of ical eigen modes for fast simulation of additive manufacturing processes, in:
additive manufacturing using coupled constitutive and microstructure models, Proceedings of the Solid Freeform Fabrication Symposium, 2013, pp. 12–14.
Addit. Manuf. 12 (2016) 144–158. [37] Y. Zhang, G. Guillemot, M. Bernacki, M. Bellet, Macroscopic thermal finite
[12] J. Liu, B. Jalalahmadi, Y. Guo, M.P. Sealy, N. Bolander, A review of com- element modeling of additive metal manufacturing by selective laser melting
putational modeling in powder-based additive manufacturing for metallic part process, Comput. Methods Appl. Mech. Engrg. 331 (2018) 514–535.
qualification, Rapid Prototyp. J. (2018). [38] R.J. Williams, C.M. Davies, P.A. Hooper, A pragmatic part scale model for
[13] B. Schoinochoritis, D. Chantzis, K. Salonitis, Simulation of metallic powder bed residual stress and distortion prediction in powder bed fusion, Addit. Manuf.
additive manufacturing processes with the finite element method: A critical 22 (2018) 416–425.
review, Proc. Inst. Mech. Eng. B 231 (1) (2017) 96–117. [39] R.K. Ganeriwala, N.E. Hodge, J.M. Solberg, Towards improved speed and accu-
[14] Z. Luo, Y. Zhao, A survey of finite element analysis of temperature and thermal racy of laser powder bed fusion simulations via multiscale spatial representations,
stress fields in powder bed fusion additive manufacturing, Addit. Manuf. 21 Comput. Mater. Sci. 187 (2021) 110112.
(2018) 318–332. [40] E. Neiva, M. Chiumenti, M. Cervera, E. Salsi, G. Piscopo, S. Badia, A.F. Martín, Z.
[15] N.W. Klingbeil, J.L. Beuth, R. Chin, C. Amon, Residual stress-induced warping in Chen, C. Lee, C. Davies, Numerical modelling of heat transfer and experimental
direct metal solid freeform fabrication, Int. J. Mech. Sci. 44 (1) (2002) 57–77. validation in powder-bed fusion with the virtual domain approximation, Finite
Elem. Anal. Des. 168 (2020) 103343.
[16] I.A. Roberts, C. Wang, R. Esterlein, M. Stanford, D. Mynors, A three-dimensional
[41] Y. Luo, H. Murakawa, Y. Ueda, Prediction of welding deformation and residual
finite element analysis of the temperature field during laser melting of metal
stress by elastic fem based on inherent strain (report i): mechanism of inherent
powders in additive layer manufacturing, Int. J. Mach. Tools Manuf. 49 (12–13)
strain production (mechanics, strength & structure design), Trans. JWRI 26 (2)
(2009) 916–923.
(1997) 49–57.
[17] E.R. Denlinger, M. Gouge, J. Irwin, P. Michaleris, Thermomechanical model
[42] D. Camilleri, T. Comlekci, T.F. Gray, Computational prediction of out-of-plane
development and in situ experimental validation of the laser powder-bed fusion
welding distortion and experimental investigation, J. Strain Anal. Eng. Des. 40
process, Addit. Manuf. 16 (2017) 73–80.
(2) (2005) 161–176.
[18] M. Chiumenti, E. Neiva, E. Salsi, M. Cervera, S. Badia, J. Moya, Z. Chen, C. Lee,
[43] A.M. Korsunsky, G.M. Regino, D. Nowell, Variational eigenstrain analysis of
C. Davies, Numerical modelling and experimental validation in selective laser
residual stresses in a welded plate, Int. J. Solids Struct. 44 (13) (2007)
melting, Addit. Manuf. 18 (2017) 171–185.
4574–4591.
[19] P. Michaleris, Modeling metal deposition in heat transfer analyses of additive [44] T.-S. Jun, A.M. Korsunsky, Evaluation of residual stresses and strains using
manufacturing processes, Finite Elem. Anal. Des. 86 (2014) 51–60, http://dx. the eigenstrain reconstruction method, Int. J. Solids Struct. 47 (13) (2010)
doi.org/10.1016/j.finel.2014.04.003. 1678–1686.
[20] R. Andreotta, L. Ladani, W. Brindley, Finite element simulation of laser additive [45] J. Ding, P. Colegrove, J. Mehnen, S. Ganguly, P.S. Almeida, F. Wang, S. Williams,
melting and solidification of inconel 718 with experimentally tested thermal Thermo-mechanical analysis of wire and arc additive layer manufacturing process
properties, Finite Elem. Anal. Des. 135 (2017) 36–43. on large multi-layer parts, Comput. Mater. Sci. 50 (12) (2011) 3315–3322.
[21] C. Fu, Y. Guo, Three-dimensional temperature gradient mechanism in selective [46] L. Cheng, A. To, Part-scale build orientation optimization for minimizing resid-
laser melting of ti-6al-4V, J. Manuf. Sci. Eng. 136 (6) (2014). ual stress and support volume for metal additive manufacturing: Theory and
[22] B. Schoinochoritis, D. Chantzis, K. Salonitis, Simulation of metallic powder bed experimental validation, Comput. Aided Des. 113 (2019) 1–23.
additive manufacturing processes with the finite element method: A critical [47] P. Promoppatum, V. Uthaisangsuk, Part scale estimation of residual stress
review, Proc. Inst. Mech. Eng. B 231 (1) (2017) 96–117, http://dx.doi.org/10. development in laser powder bed fusion additive manufacturing of inconel 718,
1177/0954405414567522. Finite Elem. Anal. Des. 189 (January) (2021) 103528, http://dx.doi.org/10.
[23] L. Scime, J. Beuth, Melt pool geometry and morphology variability for the 1016/j.finel.2021.103528.
inconel 718 alloy in a laser powder bed fusion additive manufacturing process, [48] M. Bugatti, Q. Semeraro, Limitations of the inherent strain method in simulating
Addit. Manuf. 29 (August) (2019) 100830, http://dx.doi.org/10.1016/j.addma. powder bed fusion processes, Addit. Manuf. 23 (2018) 329–346.
2019.100830. [49] X. Liang, L. Cheng, Q. Chen, Q. Yang, A.C. To, A modified method for estimating
[24] T. Mukherjee, W. Zhang, T. DebRoy, An improved prediction of residual stresses inherent strains from detailed process simulation for fast residual distortion
and distortion in additive manufacturing, Comput. Mater. Sci. 126 (2017) http: prediction of single-walled structures fabricated by directed energy deposition,
//dx.doi.org/10.1016/j.commatsci.2016.10.003. Addit. Manuf. 23 (2018) 471–486.

16
C. Bayraktar and E. Demir Additive Manufacturing 54 (2022) 102708

[50] Y. Lee, W. Zhang, Modeling of heat transfer, fluid flow and solidification [61] B.K. Panda, S. Sahoo, Thermo-mechanical modeling and validation of stress field
microstructure of nickel-base superalloy fabricated by laser powder bed fusion, during laser powder bed fusion of alsi10mg built part, Results Phys. 12 (2019)
Addit. Manuf. 12 (2016) 178–188. 1372–1381, http://dx.doi.org/10.1016/j.rinp.2019.01.002.
[51] M. Marc, Volume A: Theory and User Information, MSC. Software Corporation, [62] M.F. Zaeh, G. Branner, Investigations on residual stresses and deformations in
2010. selective laser melting, Prod. Eng. 4 (1) (2010) 35–45, http://dx.doi.org/10.
[52] R.D. Cook, et al., Concepts and Applications of Finite Element Analysis, John 1007/s11740-009-0192-y.
wiley & sons, 2007. [63] L. Van Belle, G. Vansteenkiste, J.C. Boyer, Investigation of residual stresses
[53] Y.-c. Fung, P. Tong, X. Chen, Classical and Computational Solid Mechanics, Vol. induced during the selective laser melting process, in: Key Engineering Materials,
2, World Scientific Publishing Company, 2017. Vol. 554, Trans Tech Publ, 2013, pp. 1828–1834.
[54] X. Zhao, A. Iyer, P. Promoppatum, S.-C. Yao, Numerical modeling of the thermal [64] G. Vastola, G. Zhang, Q.X. Pei, Y.W. Zhang, Controlling of residual stress in
behavior and residual stress in the direct metal laser sintering process of titanium additive manufacturing of Ti6Al4V by finite element modeling, Addit. Manuf.
alloy products, Addit. Manuf. 14 (2017) 126–136. 12 (2016) 231–239, http://dx.doi.org/10.1016/j.addma.2016.05.010.
[55] P. Promoppatum, V. Uthaisangsuk, Part scale estimation of residual stress [65] A.V. Gusarov, M. Pavlov, I. Smurov, Residual stresses at laser surface remelting
development in laser powder bed fusion additive manufacturing of inconel 718, and additive manufacturing, Physics Procedia 12 (PART 1) (2011) 248–254,
Finite Elem. Anal. Des. 189 (2021) 103528. http://dx.doi.org/10.1016/j.phpro.2011.03.032.
[56] G.R. Hadley, Thermal conductivity of packed metal powders, Int. J. Heat Mass [66] B. Cheng, S. Shrestha, K. Chou, Stress and deformation evaluations of scanning
Transfer 29 (6) (1986) 909–920. strategy effect in selective laser melting, Addit. Manuf. 12 (2016) 240–251.
[57] Y. Zhang, G. Guillemot, M. Bernacki, M. Bellet, Macroscopic thermal finite [67] C. Chen, J. Yin, H. Zhu, Z. Xiao, L. Zhang, X. Zeng, Effect of overlap rate and
element modeling of additive metal manufacturing by selective laser melting pattern on residual stress in selective laser melting, Int. J. Mach. Tools Manuf.
process, Comput. Methods Appl. Mech. Engrg. 331 (2018) 514–535, http://dx. 145 (2019) 103433.
doi.org/10.1016/j.cma.2017.12.003. [68] Y. Lu, S. Wu, Y. Gan, T. Huang, C. Yang, L. Junjie, J. Lin, Study on the
[58] Q. Chen, X. Liang, D. Hayduke, J. Liu, L. Cheng, J. Oskin, R. Whitmore, A.C. To, microstructure, mechanical property and residual stress of slm inconel-718 alloy
An inherent strain based multiscale modeling framework for simulating part-scale manufactured by differing island scanning strategy, Opt. Laser Technol. 75
residual deformation for direct metal laser sintering, Addit. Manuf. 28 (2019) (2015) 197–206.
406–418. [69] J.L. Bartlett, X. Li, An overview of residual stresses in metal powder bed fusion,
[59] Z. Zhang, Y. Huang, A. Rani Kasinathan, S. Imani Shahabad, U. Ali, Y. Addit. Manuf. 27 (2019) 131–149.
Mahmoodkhani, E. Toyserkani, 3-dimensional heat transfer modeling for laser [70] K.C. Mills, Recommended Values of Thermophysical Properties for Selected
powder-bed fusion additive manufacturing with volumetric heat sources based Commercial Alloys, Woodhead Publishing, 2002.
on varied thermal conductivity and absorptivity, Opt. Laser Technol. 109 (July [71] D. Deng, S. Kiyoshima, Numerical simulation of residual stresses induced by laser
2018) (2019) 297–312, http://dx.doi.org/10.1016/j.optlastec.2018.08.012. beam welding in a SUS316 stainless steel pipe with considering initial residual
[60] V. Popovich, E. Borisov, A. Popovich, V.S. Sufiiarov, D. Masaylo, L. Alzina, stress influences, Nucl. Eng. Des. 240 (4) (2010) 688–696, http://dx.doi.org/10.
Functionally graded inconel 718 processed by additive manufacturing: Crystallo- 1016/j.nucengdes.2009.11.049.
graphic texture, anisotropy of microstructure and mechanical properties, Mater. [72] H.T. Metals, [link], 2015. URL http://www.hightempmetals.com/techdata/
Des. 114 (2017) 441–449. hitempInconel718data.php (Accessed: 01.09.2021).

17