International Journal of Aeronautical and Space Sciences

https://doi.org/10.1007/s42405-020-00283-6

ORIGINAL PAPER

Thermal Analysis for Simulation of Metal Additive Manufacturing

Process Considering Temperature- and History-Dependent Material
Properties
Seon Ho Jeong1 · Eun Gyo Park1 · Jae Won Kang1 · Jin Yeon Cho1 · Jeong Ho Kim1 · Kichul Kim2

Received: 24 November 2019 / Revised: 16 April 2020 / Accepted: 27 April 2020

© The Korean Society for Aeronautical & Space Sciences 2020

Abstract
Additive manufacturing (AM) technology is increasingly being used in the aerospace industry due to its advantages for
aerospace components such as reduction of weight. A deep understanding of the behavior and properties of additively man-
ufactured materials or parts is required to effectively carry out the certification process which is inevitable for aerospace
components. However, since AM has so many parameters that affect the performance of products, the help of high-fidelity
process simulation techniques is essential to fully analyze and understand their effects. In this research, we propose a new
method to effectively implement the thermal analysis for process simulations of laser powder-bed fusion technique, a rep-
resentative AM technique for metal materials, using existing commercial finite element analysis software. Thermal analysis
for simulations of AM process is performed and the melt pool size is compared with test results to verify the accuracy of
the simulation. In AM process simulations, material properties may vary significantly with temperature, and they are also
dependent on the temperature history of the material because whether the current state is a powder or solid state is determined
by the maximum temperature value in the past temperature history. Therefore, in this paper, user-defined subroutines and
field variables are implemented so that the temperature history of each integration point for the finite element analysis can
be properly tracked and appropriate material properties can be assigned accordingly. Using the proposed methods, thermal
analysis for AM process simulations can be performed successfully with good accuracy compared with the existing test
results.

Keywords Additive manufacturing · Laser powder-bed fusion · Process simulation · Thermal analysis · Certification

1 Introduction According to the data of Wohlers Associates 2018 [1], AM

is mainly used in the machinery, aerospace and motor vehi-
Additive manufacturing (AM) is an emerging technology of cles as shown Fig. 1. Weight reduction, one of the biggest
the future and there are many attempts to utilize it in vari- advantages of AM, is more effective in the aerospace industry
ous fields. Since AM technique can easily form complicated than any other industries. Therefore, efforts are being made
shapes, complex shaped products which should be assembled actively to apply AM technology to the aerospace industry.
from multiple simple shaped parts by conventional manufac- However, there is a big obstacle to the wide application of
turing process can be manufactured as one part by additive AM technology to the aerospace industry. It is the difficulty in
manufacturing technique and advanced product design tech- the certification process required for all aviation parts. Since
niques based on topology optimization or lattice structures there are so many parameters that affect the performance of
for light weight or other desirable properties can be applied. additively manufactured materials or parts, it is very diffi-
cult to fully understand the effects of these parameters on the
B Jeong Ho Kim behavior and properties of additively manufactured materi-
[email protected]
als or parts, which is essential for the required certification
1 Department of Aerospace Engineering, Inha University, 100 process. This information or understanding on AM technol-
Inha-ro, Michuhol-Gu, Incheon 22212, Republic of Korea ogy cannot be obtained only with a test method, and the help
2 Korea Institute of Aviation Safety Technology, 155-11, Robot of high-fidelity process simulation techniques is necessarily
land-ro, Seo-gu, Incheon 22851, Republic of Korea required.

123
 International Journal of Aeronautical and Space Sciences

[5] investigated the effect of laser power and scan speed on

the cooling rate and temperature gradient in the melt pool
depth direction using the finite element method. From these
results, laser power and scan speed for appropriate melt pool
size were determined. Keller et al. [6] determined param-
eters of heat source and material absorption coefficients by
comparing the melt pool size of experiments and simulations
using thermal finite element analysis. Zhang et al. [7] deter-
mined the processing parameters to reach the user-specified
melt pool depth by simulations and Dai et al. [8] used sim-
ulations to predict the melt pool size depending on the laser
energy density.
As such, many studies have been conducted to estimate
Fig. 1 AM industry applications appropriate material parameters or optimal process parame-
ters from the melt pool size. However, these previous studies
rarely deal with simulation methods such as techniques for
For the laser powder-bed fusion (LPBF) technique which tracking the temperature history of the material despite their
is frequently used in the aerospace industry for production importance in the LPBF process with extreme temperature
of metal parts, the characteristics of products are determined changes. Therefore, in this research, thermal analysis tech-
by key process parameters such as power, scan speed, point niques for LPBF process simulations using ABAQUS were
distance, exposure time, hatch spacing stripe width, stripe proposed and described in detail.
overlap, beam offset and layer thickness [2]. Studies on the In the LPBF process, the current state of the material is
effects of these key process parameters are very important determined by the temperature history as well as the current
because the use of an inappropriate process parameter set temperature since the current state at the same temperature
may cause unacceptable quality of manufactured parts for can be a powder or solid state, depending on the maximum
certification due to poor sintering between layers or a boil- temperature value from the past temperature history, and the
ing phenomenon and so on. Therefore, the effects of these density or thermal conductivity varies greatly depending on
parameters for AM process need to be analyzed systemically the material state and the temperature of the material. To deal
by numerical simulations to control and guarantee the qual- with these conditions properly, ABAQUS field variables and
ity of manufactured products. Since the manufacturing cost user subroutines were used, and LPBF process simulations
for metal AM is high including the material cost, numerical could be performed with the appropriate material properties
simulations, which can predict the properties of manufac- depending on the temperature and the material state of each
tured parts without manufacturing of prototypes, are more element. The melt pool size calculated in the simulations was
useful and important for metal AM. compared with the test results to verify the proposed analysis
Process simulations for AM generally consist of thermal method.
analysis to predict the temperature history of the powder-bed
and structural analysis to predict residual stress and distor-
tion of the manufactured parts. In this paper, thermal analysis 2 Modeling for LPBF Simulation
techniques for LPBF process simulations were studied. In
thermal analysis for LPBF process simulations, heat transfer To perform LPBF simulations, laser heat source modeling,
and phase change due to laser heat source in single powder material state model, etc. are required. The thermal analysis
layer is analyzed and the melt pool size of powder bed is usu- flow chart of the overall LPBF process is shown in Fig. 2.
ally checked to confirm that the current process parameters This section describes modeling of LPBF simulations.
are suitable for the manufacture of the product [3–8]. Zhang
et al. [3] compared the melt pool size of the experiments and 2.1 Moving Heat Source Model
simulations, and then selected a suitable heat source model
for finite element method among eight heat source models. In the simulation of the LPBF process, the heat source must
Afterward, anisotropic material properties were applied to be distributed over a specific area and moved at a desired
increase the accuracy of the simulation. Foroozmehr et al. speed. This is called a moving heat source model and many
[4] applied a heat source model considering optical pen- researches on moving heat source have already been made in
etration depth (OPD) to finite element analysis, and then the welding simulation [9–13]. The papers studying AM sim-
compared the melt pool size of the experiments and simu- ulations use various heat source distribution models. Zhang
lations to perform model calibration and validation. Li et al. et al. [3] divided the types of heat source models into roughly

123
International Journal of Aeronautical and Space Sciences

Fig. 3 Exponentially decaying method

 
2P x 2 + y2
I (x, y, z)  exp −2 · f (z),
πr 2 r2
 
β |z|
f (z)  · exp − . (1)
H H

2.2 Material Properties

SS17-4PH and mild carbon steel are the materials used in

this research. SS17-PH is used as a powder layer and mild
Fig. 2 Flow chart of the LPBF simulation
carbon steel is used as a solid substrate part. Figure 4 shows
the finite element analysis model. According to the study
eight categories and compared and analyzed. According to by Roberts et al. [14], if the size of the finite element is one-
Zhang et al. [3], they confirmed that the exponentially decay- fourth of the laser-beam diameter, it can sufficiently simulate
ing equation model gives more accurate simulation results the steep temperature gradient. In the test and simulation, the
than other heat source models. Therefore, the exponentially laser-beam diameter was 100 µm, so the length and width of
decaying equation model is used in this research. The distri- the scanned region elements were appropriately determined
bution shape of the heat source is shown in Fig. 3 and can be to be 20 µm. The element size in the thickness direction
expressed by Eq. (1). P is the power of the stationary laser of the powder layer was determined by dividing the powder
source, r is the radius of the laser beam, (x, y, z) are the coor- layer into four equal parts. On the other hand, the scanned
dinates of the center of the heat source, β is the laser-beam region of the solid substrate was set up to 150 µm by properly
absorptivity and H is regarded as the powder layer thick- considering the melt pool depth for the thickness direction,
ness. The moving heat source model is implemented using and the thickness of the element size was determined by
ABAQUS DFLUX user subroutine. dividing it into eight equal parts. Large elements were used
outside the scanned region to reduce calculation time. The

123
 International Journal of Aeronautical and Space Sciences

Fig. 4 Finite element analysis

model for thermal analysis

Fig. 5 Temperature-dependent thermal material properties a density of SS17-4PH; b thermal conductivity of SS17-4PH; c heat capacity of SS17-
4PH; d material properties of mild carbon steel

properties of the materials used in the simulation are shown to be in a liquid state. However, at the temperature below
in Fig. 5. the melting temperature, the state of the powder layer can be
In the LPBF process, some of the powder layer undergoes either a powder state that is not yet melted or a solid state that
a phase change by a heat source and become a liquid state. has cooled down after going through a liquid state. Figure 6
Thereafter, as the material cools, the melted parts change indicates that the material below the melting temperature
to a solid state. If the temperature of the material exceeds (1714 K) in the LPBF process may be in a powder or solid
the melting point, the state of the material can be assumed state. Since the difference of material properties between the

123
International Journal of Aeronautical and Space Sciences

Fig. 6 Material state in LPBF

Table 1 ABAQUS UEXTERNALDB subroutine code

powder and the solid state is very large, the current state of 3000 1.5
Temperature
the material should be identified correctly according to the 2500 Flag
Temperature [K]

phase change history so that appropriate properties can be 1

2000
used.

Flag
To keep track of the phase change history of the mate- 1500
rial, we created a global array of size equal to the number 1000
0.5
of integration points and set the initial value to zero which
500
indicates a powder state. In this research, the total number of
0
elements is 30,243 and each element has 8 integration points, 0
0 0.5 1 1.5
so a global array of 32,043 × 8 components is created. In Time[sec] -4
10
ABAQUS, global arrays can be created using the UEXTER-
NALDB user subroutine. Table 1 shows the subroutine code. Fig. 7 Flag change of an arbitrary integration point with respect to tem-
In the simulation stage, the temperature of all integration perature
points is checked at every increment and if the temperature
is higher than the melting temperature, the value of the cor-
responding global array component to the integration point The boundary conditions of the simulation model are fixed
is changed to one which represents that the integration point at 293 K on the side and bottom surfaces, and convection
already has melted once. Figure 7 shows the component value and radiation are considered on the top surface. The detailed
of the global variable as a flag at an arbitrary integration point parameter values for the simulation are listed in Table 2.
in the simulation. The flag changes to one as the temperature In addition to the global array for the UEXTERNALDB
of the point rises above the melting point. Afterwards, as the user subroutine to distinguish between powder and solid state
material cools, the flag value remains one even though the at the temperature below the melting point, two methods
temperature is lower than the melting point. using ABAQUS were considered to assign appropriate mate-

123
 International Journal of Aeronautical and Space Sciences

Table 2 The parameter values for thermal analysis

Initial temperature 353 K Ambient 293 K
temperature

Convective heat 15 W/mK Emissivity coef 0.5

transfer coef
Radius of the laser 100 µm Absorptivity of 0.5665
beam laser beam
Laser power 195 W Scan speed 800 mm/s

rial properties depending on the temperature and the history

of the material state. One method is using field variables and
the other method is using the UMATHT user subroutine.

2.3 Assignment of Material Properties using Field

Variable

Use of field variable was considered as the one method to Fig. 8 Density assignment using field variable in ABAQUS
apply the appropriate properties depending on the current
material state. Field variables can be defined in ABAQUS
to change material properties depending on arbitrary field
variables. There are some restrictions on the variables that
can be used as field variables, but the components of global
variables declared in the UEXTERNALDB user subroutine
can be used as field variables, which can be used to assign
proper material properties depending on the current material value of the global variable ‘k_flag’ to 1 when the tempera-
state. ture of the integration point is higher than the melting point.
To use the field variables for this purpose, USDFLD user Figure 8 shows the density of the powder layer and the field
subroutine must be used. Table 3 shows the USDFLD user variable in ABAQUS. When the value of ‘Field 1’ is 0, the
subroutine code where ‘k_flag’ is the global variable to keep density of powder is used. On the other hand, when the value
track of the material state. In addition, the DFLUX user sub- of ‘Field 1’ is 1, the density of solid is used. Conductivity
routine must include the code that changes the component can also be set in the same way.

Table 3 USDFLD subroutine code

123
International Journal of Aeronautical and Space Sciences

Table 4 Conductivity calculation in ABAQUS UMATHT subroutine

2.4 Assignment of Material Properties using 4000

Field variable
UMATHT User Subroutine UMATHT subroutine
3000
Temperature[K]

Use of the UMATHT user subroutine was considered as the

other method to apply the appropriate properties depending 2000
on the current material state instead of using field variables.
The UMATHT user subroutine allows the user to arbitrarily
1000
set the conductivity and specific heat for thermal analysis.
In this research, the properties of SS17-4PH in Fig. 5 were
fitted as a function of temperature, and then the functions 0
0 1 2 3 4 5
were used in the UMATHT user subroutine. As shown in Time[sec] 10-4
Fig. 5, the conductivity, specific heat and density of the mate-
rial were approximated by four functions, six functions, and Fig. 9 Temperature comparison between using field variable and
five functions, respectively. Table 4 shows the source code UMATHT subroutine at arbitrary point
to calculate the conductivity function in the UMATHT user
subroutine. In this code, an appropriate function is selected
depending on the material state and the range of the temper- routine has been devised. The theoretical background of this
ature. method is as follows.
However, despite the fact that the density is largely depen- First, the basic energy balance can be expressed as Eq. (2)
dent on the current material state and the temperature of by Green and Naghdi [15].
the material, the density of the material cannot be directly   
modified in the UMATHT user subroutine. Therefore, an ρ U̇ dV  qdS + r dV , (2)
alternative method to properly reflect the effect of varying V S V

density depending on the current material state which is

determined by the temperature history is required. In this where V is a volume, S is a surface area, ρ is the density, U̇
research, a method using the value obtained by multiplying is the material time rate of the internal energy, q is the heat
the specific heat and density within the UMATHT user sub- flux per unit area and r is the heat supplied externally into
the body per unit volume. In uncoupled thermal analysis, the
internal energy is a function of temperature only. Equation (3)

123
 International Journal of Aeronautical and Space Sciences

Table 5 Internal thermal energy calculation in ABAQUS UMATHT subroutine

can be derived by applying the Fourier law and the standard θ is the temperature of the material, k is the conductivity
Galerkin approach to Eq. (2). matrix. Equation (4) can be derived by solving Eq. (3) with
the backward difference algorithm.
  
∂NN ∂θ 1
N N ρ U̇ d V + ·k· dV N N ρ (Ut+t − Ut ) dV
V V ∂x ∂x t V
  
∂NN ∂θ
 N N r dV + N N qdS, + ·k· dV
(3) ∂x ∂x
V sq V 
− N N r dV − N N qdS  0.
V Sq (4)

123
International Journal of Aeronautical and Space Sciences

Nonlinear Eq. (4) can be solved using the modified New-

ton method, and Eq. (5) represents the internal energy term
in the solved equation. In the Eq. (5), (dU /dθ )|t+t is the
specific heat.
 
1 N dU 

N ρ N M dV . (5)
t V dθ t+t

According to Eq. (5), the specific heat and density are

multiplied to calculate internal energy. Therefore, in this
research, the functions of specific heat and the functions
of the density depending on the material state and temper-
ature of the material are inserted into the UMATHT user
subroutine. And then, using the multiplied value of the two
functions, the internal energy is calculated. This allows the
Fig. 10 Comparison of melt pool dimensions between experiment and
simulation user to set the density of the material as desired, although the
density cannot be controlled in the UMATHT subroutine. At
this time, the density of the ABAQUS input file should be set
to 1.

Fig. 11 The melt pool size of

simulation using solid density
only, a width; b depth

123
 International Journal of Aeronautical and Space Sciences

Fig. 12 The melt pool size of

simulation using powder density
only, a width; b depth

Figure 9 shows the temperature change over time at an case with 195 W of laser power and 800 mm/s of scan speed
arbitrary point. Comparing the results of the simulation which had the smallest error between the experiment and
between using the field variable and using the UMATHT simulation has been selected for verification of simulation
user subroutine, the average relative error is about 0.01%, results.
which means that the temperature results are identical. Table
5 shows the source code for calculating the internal thermal 3.1 Thermal Analysis with Density for Single State
energy of the material in the UMATHT subroutine. Only

To estimate the effect of density change due to the material

3 Thermal Analysis state on the analysis using the field variables or the UMATHT
user subroutine, thermal analysis without considering the
In this research, the simulation results were verified by refer- density change due to the material state at the temperature
ring to the test results by Zhang et al. [3] that was tested below the melting point during the process simulation has
under the same conditions as the simulation performed in this been performed.
research. Figure 10 shows the results of the tests and simu- The first simulation considers the density of the solid state
lations by Zhang et al. [3] which represent the melt pool size only. This means that only the density values corresponding
of only solid substrate except powder layer. In this research, to ➀, ➁, ➂ in Fig. 5a are used. Figure 11 shows the result
among the test and simulation cases by Zhang et al.[3], the of the simulation. The simulation result shows that the width

123
International Journal of Aeronautical and Space Sciences

Fig. 13 The melt pool size of

simulation using density of
current material state, a width;
b depth

of the melt pool is about 106.9 µm and the depth is about 3.2 Thermal Analysis with Density for Current
72.3 µm. The next simulation considers the density of the Material State
powder state only. This means that only the density values
corresponding to ➃, ➄, ➁, ➂ in Fig. 5a are used. Figure 12 Figure 13 is the result of a simulation using the appropriate
shows the result of the simulation and the result shows that density depending on whether it is in powder or solid state.
the width of the melt pool is about 114.3 µm and the depth The width of the melt pool is about 121.4 µm and the depth
is about 81.4 µm. is about 80.3 µm. Table 6 compares the experimental results
By comparing these two results, one can see that the melt from the reference and the simulation results of this research.
pool size of the simulation result using only the solid density Table 7 summarizes the relative errors of all simulations. In
is smaller than that using only the powder density. This means Table 7, ‘solid’ represents the simulation using the density
that by using the solid density which has larger value than the of solid state only, ‘powder’ represents the simulation using
density of the powder state, more energy is needed to raise the density of powder state only, and ‘both’ represents the
the temperature. Conversely, if only the density of the powder simulation considering the density depending on the current
is used, the energy required to increase the temperature is material state (solid or powder). The results show that the
relatively low and the melt pool size becomes large. average error is the largest when solid density is used, and the
least error occurs when the proper density value depending
on the current material state is used.

123
 International Journal of Aeronautical and Space Sciences

Table 6 Comparison of experiment and simulation results [unit: µm] 2. Yakout M, Cadamuro A, Elbestawi MA, Veldhuis SC (2017)
The selection of process parameters in additive manufacturing for
Experiment Simulation aerospace alloys. Int J Adv Manuf Technol 92:2081–2098
Maximum Minimum Average 3. Zhang Z, Huang Y, Kasinathan AR, Shahabad SI, Ali U, Mah-
moodkhani Y, Toyserkani E (2019) 3-Dimensional heat transfer
Width 130.49 113.85 122.145 112.84 modelling for laser powder-bed fusion additive manufacturing with
volumetric heat sources based on varied thermal conductivity and
Depth 76.35 71.32 73.835 75.56
absorptivity. Opti Laser Technol 109(2019):297–312
4. Foroozmehr A, Badrossamay M, Foroozmehr E, Golabi S (2016)
Table 7 Relative error comparison [unit: %] “Finite element simulation of selective laser melting process con-
sidering optical penetration depth of laser in powder bed. Mater
Density Width Depth Average Des 89:255–263
5. Li Y, Dongdong Gu (2014) Parametric analysis of thermal behavior
Solid 12.49 2.04 7.26 during selective laser melting additive manufacturing of aluminum
Powder 6.39 10.25 8.32 alloy powder. Mater Des 63:856–867
6. Keller N, Ploshikhin V (2014) New method for fast predictions of
Both 7.62 2.33 4.97
residual stress and distortion of AM parts. In: Conference: solid
freeform fabrication symposium, Austin, Texas, USA, vol 25, pp
1229–1237
4 Conclusion 7. Zhang D, Cai Q, Liu J, Zhang L, Li R (2010) Select laser melting
of W-Ni-Fe powders: simulation and experimental study. Int J Adv
Manuf Technol 51(5):649–658
In this research, thermal analysis techniques for LPBF pro- 8. Dai D, Gu D (2014) Thermal behavior and densification mecha-
cess simulations using ABAQUS were proposed and thermal nism during selective laser melting of copper matrix composites:
analysis has been performed using proposed methods. Since simulation and experiments. Mater Des 55:482–491
the simulation of the LPBF process includes phase changes, 9. Goldak J, Chakravarti A, Bibby M (1984) A new finite element
model for welding heat sources. Metall Trans B 15(2):299–305
appropriate physical properties should be used depending on 10. Lee J-S (2010) Welding deformation analysis of plates using
the history-dependent material state as well as the tempera- the inherent strain-based equivalent load method. J Weld Joi
ture of the material. To simultaneously consider the material 28(2):39–46
state and temperature conditions, we used the field variables 11. Standardization for Temperatur (2005) Standardization for temper-
ature distribution prediction of the arc weld using FEA. J Weld Join
and the user subroutines in ABAQUS. In particular, since the 23(6):1–7
user subroutine cannot directly modify the density value, the 12. Bang H-S, Chong-In Oh, Ro C-S, Park C-S, Bang H-S (2007)
density change was modeled by simultaneously considering Analysis of thermal and welding residual stress for hybrid welded
the input file and the user subroutine code. joint by finite element method. J KWJS 25(6):565–570
13. Öberg TT (1991) Computation of temperature distribution due
By comparing the simulation results and the test results, it to welding in piping systems. In: Mechanical effects of welding,
is confirmed that the error is about 5%. This error is less than international union of theoretical and applied mechanics (IUTAM)
those from the cases using the density value of single state symposium, Luleå, Sweden, 10–14 June 1991
only and confirmed the validity of the simulation method 14. Roberts IA, Wang CJ, Esterlein R, Stanford M, Mynors DJ (2009)
A three-dimensional finite element analysis of the temperature field
proposed in this research. during laser melting of metal powders in additive layer manufac-
turing. Int J Mach Tools Manuf 49:916–923
Acknowledgements This research is supported by a Grant (17CHTR- 15. ABAQUS (2012) ABAQUS documentation. ABAQUS, Provi-
C128889-01) from Establishment of Design and Manufacturing Certi- dence
fication Infrastructure on Rotorcraft Certification funded by Ministry
of Land, Infrastructure and Transport of Korean government and Korea
Agency for Infrastructure Technology Advancement.
Publisher’s Note Springer Nature remains neutral with regard to juris-
dictional claims in published maps and institutional affiliations.

References
1. Wohlers TT, Wohlers Associates, Campbell I, Caffrey T, Diegel O,
Kowen J, Wohlers Report (2018) 3D printing and additive manu-
facturing state of the industry. Annual Worldwide Progress Report,
Wohlers Associates

123