Materials and Design 196 (2020) 109185

Contents lists available at ScienceDirect

Materials and Design

journal homepage: www.elsevier.com/locate/matdes

High-fidelity modelling of thermal stress for additive manufacturing by

linking thermal-fluid and mechanical models
Fan Chen a, Wentao Yan a,b,⁎
a
Department of Mechanical Engineering, National University of Singapore, 117575, Singapore
b
NUS Research Institute (NUSRI), Suzhou, Jiangsu 215123, China

H I G H L I G H T S G R A P H I C A L A B S T R A C T

• Modelling thermal stress in AM using

temperature profile from thermal-fluid
model.
• Rough surfaces and internal voids are
incorporated.
• Thermal stress concentrations due to
voids and rough surfaces are revealed.
• Simulation results of thermal stress can
rationally explain the cracks in
experiments.

a r t i c l e i n f o a b s t r a c t

Article history: The prediction of thermal stress and distortion is a prerequisite for high-quality additive manufacturing (AM).
Received 31 July 2020 The widely applied thermo-mechanical model using the finite element method (FEM) leaves much to be im-
Received in revised form 7 September 2020 proved due to their oversimplifications on material deposition, molten pool flow, etc. In this study, a high-
Accepted 22 September 2020
fidelity modelling approach by linking the thermal-fluid (computational fluid dynamics, CFD) and mechanical
Available online 26 September 2020
models (named as CFD-FEM model) is developed to predict the thermal stress for AM taking into account the in-
Keywords:
fluences of thermal-fluid flow. Profiting from the precise temperature profiles and melt track geometry extracted
Additive manufacturing from the thermal-fluid model as well as the remarkable flexibility of the quiet element method of FEM, this work
Finite element aims at simulating the thermal stress distribution by involving physical changes in the AM process, e.g., melting
Quiet element method and solidification of powder particles, molten pool evolution and inter-track inter-layer re-melting. Unlike the
Mapping conventional thermo-mechanical analysis, in this approach, thermal stress calculation is purely based on a me-
Thermal stress chanical model where the thermal loads are applied by using a linear interpolation function to spatially and tem-
Molten pool porally map the temperature values from the thermal-fluid model's cell centres into the FEM element nodes.
Thermal-fluid flow
With the proposed approach, the thermal stress evolution in the AM process of single track, multiple tracks
and multiple layers are simulated, where the rough surfaces and internal voids can be well incorporated. More-
over, a conventional thermo-mechanical simulation of two tracks with predefined track geometry is conducted
for cross comparison. Finally, the simulated thermal stress distribution can rationally explain the crack distribu-
tion observed in the experiments.
© 2020 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license (http://
creativecommons.org/licenses/by/4.0/).

1. Introduction
⁎ Corresponding author at: Department of Mechanical Engineering, National University
of Singapore, 117575, Singapore. Additive manufacturing (AM) is a promising technology capable of
E-mail address: [email protected] (W. Yan). producing complex-shaped parts without the need for special tooling

https://doi.org/10.1016/j.matdes.2020.109185
0264-1275/© 2020 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
F. Chen and W. Yan Materials and Design 196 (2020) 109185

[1,2]. However, the heat source scanning usually induces steep temper- deposition process. Huang [42] et al. applied the progressive element
ature gradients, and hence causes high thermal stress. Such high ther- birth to simulate the material deposition of the wire and arc additive
mal stress would cause lots of defects, e.g. cracks, delamination, manufacturing. The inactive elements are attributed with ordinary ma-
shrinkage, warpage etc. [3–8], and has been considered to be a bottle- terial properties but excluded from the structural stiffness at the begin-
neck problem. The effects of different processing parameters on the ning of the calculation [41], so that the track forming procedure can be
thermal stress of the final products have been investigated experimen- represented by sequentially activating the designated element sets.
tally and numerically by a number of scholars and researchers around However, since the temperature of the interface nodes is obtained by
the world. the interpolation between active and inactive elements, the interface
Experimentally tracing the thermal stress evolution during the melt- nodal temperature will be lower than the desired value as the temper-
ing process is inexpedient under current circumstances due to the high ature for an inactive element is set to be zero by default(shown in Ap-
local temperature and the fast heating/cooling rate. Post-process exper- pendix Fig. 14). Such problem had been described in detail by
imental characterizations of thermal stress, e.g., hole-drilling method Michaleris et al. [43] who built a 1D model to analyse the temperature
[9,10], Neutron diffraction [11,12], and x-ray diffraction [13], can pro- calculation scheme on the node shared by two elements, active and in-
vide some useful information. However, the parameter optimization active respectively. For the pure mechanical analysis proposed in this
via trial-and-error experiments and post-process characterization is study, the heat transfer mechanism is superseded by the direct loading
quite expensive given the large number of processing parameters, and of the temperature, thus the inactive element method is not applicable
the spatial distribution of thermal stress, temperature and absorbed en- since the temperature compensation for the interface nodes is quite
ergy is also difficult to experimentally measure. knotty.
Companies and scientists have devoted a lot of money and efforts in In this work, we develop a high-fidelity modelling approach for ther-
developing efficient numerical tools to analyse the thermal stress and mal stress by mapping the temperature profiles from the thermal-fluid
distortion in the large scale, e.g., 3DSIM [14] acquired by ANSYS, model [8,44] to the mechanical model, to ensure the accuracy of the
3DEXPERIENCE Make platform [15,16] acquired by Dassault Systèmes, temperature profile and to incorporate the realistic geometry including
and Pan Computing LLC [17,18] acquired by Autodesk. Most of the cur- both rough surfaces and internal pores and voids. As illustrated in Fig. 1,
rent softwares use the inherent strain model to ensure the computation the coupling of the CFD and FEM is one-way, and the temperature pro-
efficiency, which simplifies the melting-solidification process along files extracted from the CFD model function in the FEM model in two as-
scan paths to a pure mechanical simulation of sequentially accumulat- pects, i.e., temperature load and material assignment. In the mechanical
ing representative elements of single tracks with inherent strains. model using the FEM, the quiet element method (QEM) [45,46] is
Liang et al. [19,20] put a lot of efforts in constructing the inherent strain employed to incorporate the complex changes of geometry due to
model to predict residual distortion. Agglomeration approaches is also melting/re-melting, solidification, and molten pool flow. The thermal-
commonly applied for the part-scale modelling of the AM products, fluid model, temperature mapping scheme, thermal stress model and
where a large computational layer is constructed by lumping many parameter settings are described in Section 2, the conventional
sub-layers together [21]. However, there is still much room for improve- thermo-mechanical model is built for cross comparison in Section 3,
ment of the applicability and predictability of these simulations, which and the simulation results and discussion are in Section 4.
are based on assumptions, theoretical estimation or thermal and geo-
metrical simplifications.
2. Model construction
Many researchers attempted to improve the predictability of FEM
simulation by considering the heating, melting and solidification factors
2.1. Thermal-fluid model
but omitted the effects of thermal-fluid flow. In most cases of thermo-
mechanical simulation, the heat conduction equations are solved di-
The thermal-fluid simulation that resolves individual powder parti-
rectly on the FEM model and the entire track or layer is set to be active
cles is by the CFD software FLOW-3D and has been experimentally val-
at the beginning of the melting process, regardless of the powder melt-
idated to be accurate in a variety of AM cases [8,44,47,48]. In the
ing process and the track forming procedure [22–26] [27,28]. However,
thermal-fluid simulation, the melted material flow is assumed to be in-
due to the sequentiality of the material deposition process, the elements
compressible laminar flow. The governing equations are mass continu-
being deposited should not be affected by the prospectively deposited
ity, momentum conservation and energy conservation given in Eq. 1 [8].
elements, and the prospectively deposited elements should be either
deactivated or suppressed [29–32]. Cheon et al. [33,34] analysed the
thermal-metallurgical-mechanical behaviour of the welding process 8
> ∇⋅ðρvÞ ¼ 0,
>
>
based on a CFD-FEM framework [35] with the consideration of the >
<∂
phase dependent thermal strains; the mechanical history reset of the ðρvÞ þ ∇⋅ðρv⊗vÞ ¼ ∇⋅ðμ∇vÞ−∇p þ ρg þ f B , ð1Þ
> ∂t
liquid state was however not considered. Bailey et al. [36] incorporated >
>
: hskip  150 ∂ ðρhÞ þ ∇⋅ðρvhÞ ¼ q þ ∇⋅ðk∇T Þ,
>
the influence of the molten pool, track re-melting and layer formation ∂t
process into laser direct deposition modelling by deactivating the
upper tracks and liquid state elements. Bailey's model was validated
by experiments in part scale, but it cannot locally reveal the stress con- where v is the velocity vector, μ is the viscosity, p is the pressure, k is the
centration on the individual tracks as the cross section of all the depos- thermal conductivity, and T is the temperature. In the energy conserva-
ited tracks are simplified to be smooth and uniform without defects. The tion equation, h represents the specific enthalpy calculated by h = cT +
defects generated during the manufacturing process, like high surface (1 − fs)L, where c, fs and L denote the specific heat, solid fraction and la-
roughness and porosity, can potentially change the stress distribution tent heat of melting, respectively, thus the latent is linearly released
or perhaps cause local stress concentration. Foteinopoulos et al. [37] in- with the temperature. The momentum equation incorporates gravity
volved the factors of the defects in the FEM by inducing the volumetric (g), buoyancy (fB, Boussinesq approximation), and viscosity (μ). The
fraction of porosity, to simplify the tracks into uniform-shaped blocks. boundary conditions for the momentum conservation equation incor-
However, the involvement of the small defects in thermo-mechanical porate surface tension, recoil pressure and Marangoni forces, and the
simulation seems impractical due to the intricate track geometry thermal boundary conditions incorporate the surface radiation and en-
approximation. ergy loss by evaporation.
The inactive element method [38–41] is well-known as one of the The Volume of Fluid (VOF) method [49] is applied to re-construct the
commonly applied numerical approaches to simulate the material free surface at each time increment, given in Eq. 2.

2
F. Chen and W. Yan Materials and Design 196 (2020) 109185

Fig. 1. Schematic of linking thermal-fluid and mechanical models, consisting of (1) mapping the temperature profiles from the thermal-fluid model to the mechanical model, and
(2) assigning the corresponding material states.

volume (CV) of the CFD model onto the node of the Lagrangian element
∂F of the FEM model, as shown in Fig. 2. Given the Eulerian structural mesh
þ ∇⋅ðFvÞ ¼ 0: ð2Þ
∂t in the CFD model, we utilize the shape function of the brick element in
the FEM as the interpolation function, given as Eq. 4 [52].
where F denotes the fluid fraction. Further details about the model can
be referred to [8,47]. 8 8
< T ¼ ∑i¼1 Ni ðξ, η, ζ ÞT i
ð4Þ
2.2. Thermal stress calculation in the FEM model : N ðξ, η, ζ Þ ¼ 1 ð1−ξξ Þ1−ηη ð1−ζζ Þ, i ¼ 1, 2, . . . , 8
i i i i
8
As the coupling between CFD and FEM is one-way, the thermal-fluid where ξ, η and ζ are the local coordinates in the CFD coordinate system,
simulation is entirely independent of the FEM computation and the ξi, ηi and ζi are the natural coordinates of the CV centres surrounding the
temperature history can be imposed directly onto the FEM model. FEM node and Ti(ξ, η, ζ), i = 1, 2, …, 8 denote the temperature values of 8
Therefore, no transient heat transfer analysis is involved and the FEM CV centers surrounding the element node, respectively. Moreover, T is
simulation is a pure mechanical analysis. Due to the lack of reliable ma-
terial parameters for the thermal-elastic-plastic constitutive mode, we
hereby assume the material to be linear elastic. The governing equations
applied to correlate the thermal strain, local temperature and thermal
stress are shown in Eq. 3 [50,51]..
(      
I
εth ¼ α T, f β T−T 0 −α T I , f β T I −T 0
ð3Þ
σ ¼ Dε

where α(T, fβ) represents the thermal expansion coefficient,

T is the current temperature, TI is the initial temperature, fβ represents
the current material state, fIβ represents the initial material state, and
T0 represents the reference temperature for the thermal expansion coef-
ficient. Additionally, σ denotes the stress matrix, D denotes the elastic-
ity tensor, and ε is the elastic strain matrix.

2.3. Mapping and interpolation

The temperature profiles are loaded based on the 3D interpolation Fig. 2. Schematic of the temperature interpolation between the CFD CV centres and FEM
that maps the temperature stored in the centre of Eulerian control nodes.

3
F. Chen and W. Yan Materials and Design 196 (2020) 109185

the interpolated temperature value for the FEM node inside CV, and Ni 2.4. Material state
(ξ, η, ζ) is the interpolation function. It should be noted that this map-
ping scheme including the interpolation function is generally applicable The QEM remedies the shortcomings of the inactive element method
to any 3D FEM element shapes, and the 8-node linear brick element in simulating the evolution of the molten pool and the remelting of pre-
(C3D8R) is used in the FEM model of this study. After obtaining the tem- vious tracks. The idea of the QEM is to assign null material properties to
perature T(t1) and T(t2) by Eq. 4 on two adjacent instants t1 and t2 spec- non-deposited elements whose presence should not affect the depos-
ified in CFD simulation, the temperature value T(t') in between (t1 ≤ t' ≤ ited elements, as Fig. 3 shows. In the QEM, the material properties of
t2) can be approximated by the 1D interpolation, given by Eq. 5. the elements should change with the corresponding state, i.e., air, liquid
or solid. As illustrated in Fig. 4, the elements' material states transfer to
T ðt 2 Þ−T ðt 1 Þ 0 the liquid state from the solid state where the powders or the previous
T ðt 0 Þ ¼ T ðt 1 Þ þ ðt −t 1 Þ ð5Þ tracks and substrate are melted; elements transfer from the liquid state
t 2 −t 1
to the solid state due to solidification. Besides, with the molten pool evo-
lution, the elements material states may transfer from liquid state to the
The time step size in the FEM model, Δt', is different from the air state when the liquid flows away, or from the air state to the liquid
data output time interval of the CFD model, Δt. To fully exploit the state when it flows in. It should be noted that there is no state transfer
CFD temperature profiles output, Δt' is smaller than Δt to allow a tem- between solid and gas. In addition, since the effect of the unmelted pow-
perature interpolation between any two time-adjacent CFD tempera- ders on the stress distribution can be ignored, the unmelted powders
ture profiles. are treated to be the air state.

Fig. 3. Schematic of the quiet element model.

Fig. 4. The material state changes during the melting and solidification process, including ① the substrate and powders are molten into liquid metal, ② the liquid material flows away and
the state transfers to air and ③ the state transfers to liquid when liquid flows in, ④ the solidified material may be re-melted into liquid, and ⑤ the liquid metal solidifies.

4
F. Chen and W. Yan Materials and Design 196 (2020) 109185

Therefore, the material state of each element depends on not only gas phase needs to be considered. More details for the thermal-fluid
the current temperature but also the temperature history. Within a modelling are given in references [8, 44, 47].
time increment Δt, the interpolated temperature value for a node is Tb In the FEM model, three material states, i.e. solid, liquid and air,
at the beginning and becomes Te after Δt, so the temperature change change depending on the temperature profile as shown in Fig. 4. The
is ΔT = Te − Tb. The melting point (the liquidus temperature for an material properties for different states are listed in Table 2, in which
alloy) and initial temperature (i.e., the preheating temperature) are the solidus elements are assigned with the material properties of Ti-
constant and denoted by Tl and Tp, respectively, and then the tempera- 6Al-4 V. In practise, the quiet elements still count for parts of the global
ture criteria for the material state transformation are defined as follows: stiffness but do not affect the solidus phase if setting the stiffness of all
• melting: if ΔT > 0, Tb < Tl and Te > Tl, the node undergoes melting the quiet elements as zero. Considering the convergence, assigning a
process and the material state for the node changes from solid to liquid; zero Young's modulus to the air and molten liquid metal will lead to se-
• solidification: if ΔT < 0, Tb > Tl and Te < Tl, the node is solidifying rious element deformation near the interface with the solid phases.
underway and the material state changes from liquid to solid; Therefore, the air and liquid elements are assigned with a small Young's
• if Te < Tp, it is the environmental temperature and the material modulus 1 × 104 Pa and Poisson ratio 1 × 10−3, which are large enough
state is set as air. to avoid numerical divergence for the non-solid elements but small
enough to be” quiet” for the stress analysis. It has been tested that the
Poisson ratio values assigned to the air and liquid states do not make dif-
2.5. Material parameters ference to the thermal stress distribution of the solid if the Young's
modulus is sufficiently small.
The material is Ti-6Al-4 V. For the thermal-fluid simulation, the ma-
terial properties are listed in Table 1 [8]. The simulation case is the elec-
tron beam melting process under the vacuum environment where no 2.6. Simulation cases and computational cost

It is worth mentioning that our modelling approach is applicable to a

Table 1 wide variety of fusion-based manufacturing processes, e.g., selective
The material parameters for CFD simulation. laser melting, electron beam melting, directed energy deposition, and
Material property Value
wire arc additive manufacturing. Here we conduct an example case of
electron beam melting. We run a simulation case of 2-layer 2-track elec-
Specific heat (cp) 872 J/(K ⋅ kg)
tron beam melting process. The electron beam diameter, scanning
Latent heat of melting (Lm) 2.86 × 105J/kg
Latent heat of evaporation(Lv) 9.7 × 106J/kg speed, power and layer thickness are 0.4 mm, 0.5 m/s, 60 W and 0.05
Density(ρl) 4000 kg/m3 mm, respectively. The preheating temperature is 873 K. In the case,
Surface radiation coefficient (αb) 0.4 the geometrical features, e.g., rough surfaces and voids, are incorpo-
Surface tension coefficient(σ) 1.68 N/m
rated, and the re-melting of the previously deposited track and layer is
Viscosity (μ) 0.005 Pa ⋅ s
Solidus thermal conductivity (k) 16 W/(m ⋅ K)
also incorporated, which is representative of such AM processes. The
Liquidus thermal conductivity (k) 32 W/(m ⋅ K) temperature profiles are output from the thermal-fluid model in an in-
Temperature sensitivity of surface tension coefficient (σTs ) 0.00026 terval of 0.0002 s, while the time step size of the FEM model is set to be
Solidus temperature (Ts) 1878 K 0.0001 ∼ 0.0002 s. The mesh size in the thermal-fluid model is 0.005
Liquidus temperature (Tl) 1928 K
mm, as detailed in our previous work [8]. As Fig. 5 shows, in the FEM
model, there are 922,752 elements and 969,612 nodes; the element di-
mension is 0.006 × 0.006 × 0.008 mm3 in the substrate and 0.006 ×
Table 2 0.006 × 0.006 mm3 in the region above the substrate. A mesh conver-
Material parameters for different material states in the FEM model. gence test has been conducted to exclude mesh sensitivity.
Material Young's modulus Poisson Thermal The computation time is around 700 h for the thermal-fluid simula-
states (Pa) ratio expansion tion on an Intel Core i7–2600 CPU (4 cores, 8 threads and 3.40 GHz for
coefficient(1/K) the Processor Base Frequency) and 330 h for the FEM simulation on
an Intel X5650 CPU (6 cores, 12 threads and 2.66 GHz for the Processor
Solid 1.32 × 1011 0.31 9.2 × 10−6 Base Frequency). The computational cost for FEM model can be greatly
Liquid 1 × 104 0.001 9.2 × 10−6 reduced by optimizing the temperature loading scheme, which is under
Air 1 × 104 0.001 0
further investigation and will be introduced in future publications.

Fig. 5. Geometry dimension, meshing and boundary conditions of the FEM model.

5
F. Chen and W. Yan Materials and Design 196 (2020) 109185

3. Thermo-mechanical model for cross comparison of the rough surface can be problematic for the FEM model to generate
mesh and to apply the heat source, the surfaces of the tracks have to be
A thermo-mechanical simulation is conducted for cross comparison modified to be smooth, as shown in Fig. 6 (b). The mesh contains
with the CFD-FEM simulation. To ensure a fair cross-comparison, we 209,304 hexahedral elements and 234,866 nodes and the computa-
constructed a curved track shape, mimicking that from the thermal- tional cost is about 200 h on an Intel X5650 CPU. The elements type is
fluid simulation as shown in Fig. 6. Since the fine geometrical features the 8-node thermally coupled brick element (C3D8T). The scanning

Fig. 6. (a) Thermal-fluid simulation results. (b) The geometrical shape of the thermo-mechanical model mimicking the middle segment of the first layer in the thermal-fluid model.

6
F. Chen and W. Yan Materials and Design 196 (2020) 109185

parameters and the material properties are identical to those listed in the geometrical features and the temperature distribution of the FEM
section 2.5, and the input energy obeys Gaussian distribution as given and CFD models show a good agreement, except the terraced surface
in Eq. 6 [44], caused by the voxel mesh. It is worth mentioning that the void has
also been mapped, which causes the stress concentration especially in
2 0 h i13
C⋅ ðx−xb Þ2 þ ðy−yb Þ2 the z direction, as shown in Fig. 7 (b). The fidelity of the FEM tempera-
C
qðx, yÞ ¼ Q  η  4 2 exp @− A5 ð6Þ ture distribution can be improved by refining the mesh and reducing
πRb R2b the time step size of the FEM simulation.

where no electron penetration has been considered because the ele- 4.2. Stress distribution
ment size is larger than the maximum penetration depth of the electron
beam. In Eq.6, C is the concentration coefficient, Rb is the electron beam The temperature gradient along the x-y plane in the cross section of
radius, (xb, yb) represents the coordinate of the electron beam centre, the 1st layer is plotted at the end of the fabrication of the 1st layer (t =
and Q is the electron beam power and η is the energy absorptivity. Q 0.008 s) as shown in Fig. 8. The temperature gradient in the header
× η is equal to the absorbed power applied in thermal-fluid simulation. part of the track after heating is larger than that of the middle region,
More details for Eq. 6 are available in reference [44, 47]. which will cause a higher level of compression as displayed in Appendix
Fig. 16.
4. Results and discussion As the simulated time interval for the two-track two-layer example
is not long enough for the fully cooling down, we only observe the ther-
4.1. Temperature distribution and morphology mal stress history during the heating and cooling process. The stress
curves are plotted based on the stress datum recorded on 7 designated
Fig. 7 (a) provides two sectional views partially showing the voids, nodes, which are located on the vertical line that passes through the
rough surface and molten pool morphology of the mechanical and center region of the first and third tracks, as illustrated in Fig. 9. Nodes
thermal-fluid simulation results, while more detailed comparative A and B are inside the third track, node C is inside the third track but
snapshots of the FEM simulation results and thermal-fluid simulation quite close to the top surface of the first track, node D is within the
results are shown in the Appendix Fig. 15. The powders and the liquid first track, and nodes E, F and G are within the substrate. Their corre-
state elements that can hardly affect the stress distribution are not visu- sponding stress component curves in x, y and z directions, respectively
alized in the FEM results. Thus the molten pool shape looks slightly dif- denoted as X-stress, Y-stress and Z-stress. The heat source scans over
ferent, but the shape of the formed track is quite similar. The results of the observation region from t1′ = 0.00155 s to t2′ = 0.00255 s; during

Fig. 7. (a) Comparison of the geometrical features and the temperature field between the simulation results of the coupled CFD-FEM model and thermal-fluid model. (b) Stress
concentrations in z direction caused by the voids.

7
F. Chen and W. Yan Materials and Design 196 (2020) 109185

Fig. 8. Temperature and temperature gradient in the cross section of the 1st layer at t = 0.008 s along the x-y plane.

this period, the heated area around the molten pool tends to expand but 4.3. Comparison with the thermo-mechanical model
is constrained by the surrounding cooler elements, thus the stress com-
ponents of nodes D, E and F around the heated region are under com- The comparison of the thermal stress and temperature distribution
pression, while the stresses for nodes A, B and C are 0 because the between the thermo-mechanical and coupled CFD-FEM simulation are
third track remains” quiet”. Once material states of the nodes D and E shown in Fig. 10, where only the first layer is included. While the
change to the liquid at the moment the molten pool arrives, their stress absorbed power Q × η applied for the thermo-mechanical and
components become 0, because the compressive stresses are totally re- thermal-fluid model is equal, the peak temperature in the molten pool
leased in the liquid state. In the melting of the second track (0.004 s∼ region of the thermo-mechanical simulation (4228 K) is much higher
0.008 s), the heat source is away from the observation nodes, and the than the CFD-FEM simulation (2382 K) as the energy dissipation by
1st-track shrinks in the x direction but under the expansion force evaporation is omitted in the thermo-mechanical simulation. From
along the Y direction from the heating of the 2nd-track. Therefore, as Fig. 10, the molten pool size is relatively small in the thermo-
shown in Fig. 9, X-stress changes from compression to tension and Y- mechanical simulation, because the molten pool flow is not incorpo-
stress remains negative. Node G is relatively far from the heating area rated, particularly the Marangoni effect which usually enlarges the mol-
and still primarily affected by the material shrinkage. ten pool by enhancing convection. In the CFD-FEM modelling approach,
As the 3rd-track is manufactured during 0.008 s∼ 0.012 s, the 1st- the high-fidelity thermal-fluid simulation provides the physically-
track simultaneously experiences the re-melting treatment. From informed temperature profiles and authentic geometrical features for
Fig. 9, the molten pool once more moves through the observation posi- the mechanical simulation, while in the thermo-mechanical model,
tion with another layer of elements covering the 1st-track. Nodes A, B the heat source is directly applied on the predefined track surface.
and C are inside the newly deposited track. Since the top surface can ex- From Fig. 10 (b), there are some stress concentration regions on
pand more easily than the inner region, the stress level of the inner node the processed track of the CFD-FEM model, while the stress in the
C is larger than that of node A located near the top surface. Meanwhile, thermo-mechanical model is mostly released after the heating as
as the heat source scans forward from the starting point, the tensile the track surface is smooth. Furthermore, the three stress compo-
stress of nodes D, E and F will be counteracted by the expansion, then nents in x, y and z directions, namely X-stress, Y-stress and Z-stress,
fully released to 0 due to re-melting and changes to compression are compared on nodes D, E and F in Fig. 11. As the meshes for these
again, but then immediately recovers to tension in the subsequent two simulations are different, the coordinates of the observation
rapid cooling process. Node G is sufficiently far from the heating region nodes D, E and F in the thermo-mechanical model cannot be selected
and thus maintains a tensile stress, which even rises to a higher level identically but are as close as possible to those of the FEM-CFD model,
after heating. as illustrated in Fig. 11 (a). When the molten pool arrives at the ob-
Note that compression stresses magnitude in some nodes are quite servation position, node D is inside molten pool, nodes E is located
large, over 800 MPa for X-stress on the nodes A, B and C. It is because at the boundary of the molten pool and node F is below the molten
the thermo-elastic-plastic relationship is not considered and in reality, pool. From the stress curves shown in Fig. 11 (b), node D of the
the Young's modulus is not a constant but becomes lower at high tem- thermo-mechanical model undertakes the higher molten pool peak
peratures. The subsequent cooling-down pattern after t1′ = 0.016 s can temperature as well as the larger temperature gradient given the
not be observed based on the current outcome due to the simulation same temperature level in the cooling region, thereby the stress mag-
time limit. Yet the stress pattern on the upper layers can be hypothe- nitude is higher than the CFD-FEM simulation, while on node F it is
sized to be similar to the lower layers. lower. With the consideration of the liquid state, the stress evolution
for the CFD-FEM simulation is more sophisticated than the simplified

8
F. Chen and W. Yan Materials and Design 196 (2020) 109185

Fig. 9. The material deposition process specified at the observed position; the distribution of the observation nodes in different layers and their stress component curves in X, Y and Z
directions.

9
F. Chen and W. Yan Materials and Design 196 (2020) 109185

Fig. 10. The comparison of (a) the temperature and (b) stress distribution between the CFD-FEM and thermo-mechanical simulation.

thermo-mechanical simulation; at the moment that the molten pool (Z component) concentration occurs mainly around the interface be-
scans over the observation nodes, the compressive stress on nodes tween different layers, especially in the regions around the voids,
D and E will be released to 0 in the CFD-FEM model, while in the which may further lead to the delamination. It should be mentioned
thermo-mechanical model, the compression in the same region that we observe the Z-stress distribution at the instant when two layers
reaches the peak, as shown in Fig. 11 (b) and (c). are fully deposited so that the defects between two layers can be shown.
As the X and Y stress components mainly concentrate on the layer sur-
4.4. Example applications face, both of them are output at the instant when the 1st layer has been
fully deposited and almost cooled down.
We try to find the potential evidence for the mechanism of the crack Furthermore, we compared the single track CFD-FEM simulation re-
generation or other forms of defects [53,54] using the proposed high- sults with the experimental results by Vrancken et al. [55]. As shown in
fidelity modelling. Normally, the deposited elements bear large com- Fig. 13, the experimental cases conducted with different sets of laser
pressive thermal stress in the heated region; with the heat source mov- power and scan speed are shown at different instants, i.e., t1′, t2′, t3′
ing away, the elements starts to shrink. Thus, we focus on the cooling and t4′. At t1′, the laser reaches the left side; at t2′, the laser is in the mid-
region where the residual stress concentration may occur. The snap- dle of the track segment; at t3′, the laser just passes the observed region;
shots revealing the evolution of the stress component in the x direction at t4′, the laser passes far away, and cracking perpendicular to the track
in the multiple-track simulation are displayed in Appendix Fig. 16 in occurs. For the evolution of the X-stress from the CFD-FEM simulation,
chronological order. It reveals that t′ = 0.00915 s, the first layer has al- the track segment is under compression at t3′, so no cracks can be
most cooled down and the overall X-stress component state in the mid- found in the experimental results. At t4′ when the laser passes far
dle region changes to be tensile, together with the emergence of many away and the track segment is further cooling down and shrinking,
small stress concentration regions. the X-stress becomes tensile, with these stress concentration regions
As shown in Fig. 12, the red color of the stress distribution highlights perpendicular to the track, which rationally explains the cracks genera-
the tensile stress concentration regions where the cracks are more likely tion in the single track experiments. Moreover, our approach has also
to occur. As shown in Fig. 12 (a), the single-track experiments by the been applied in the investigation of high-density dislocations in addi-
SLM of the Fe-based metallic glass is conducted on Aconity Mini tively manufactured copper [56]. The residual stress distribution
(Aconity GmbH, Germany) with the heat power 100 W, scanning achieved by the CFD-FEM simulation rationally explained the disloca-
speed 200 mm/s and the layer thickness of 100 m; the propagation ori- tion accumulation mechanism.
entation of the surface crack is perpendicular to the scanning path; such The CFD-FEM simulation has shown its rationality for the thermal
cracks are often caused by the large stress component along the scan- stress distribution and evolution on the individual tracks with accurate
ning path; the stress (X component) concentration regions in the CFD- temperature fields and few geometry simplifications. It should be men-
FEM simulation are distributed primarily around the edge of the tioned that without incorporating accurate temperature-dependent
surface's small wrinkles, which possibly explains the mechanism for thermal expansion coefficient and thermo-elastic-plastic constitutive
crack generation vertical to the scanning path; it can be speculated relation, the current models are not able to provide accurate quantita-
that the long cracks vertical to the scanning path on the multiple tracks tive prediction of thermal stress. Moreover, the quantitative experimen-
surface may be generated by the propagation of these small cracks. In tal measurement of thermal stress on individual tracks is also difficult,
the thermo-mechanical simulation, the X-stress is however distributed making the quantitative validation very challenging. Although some
quite smoothly on the track surface, where no tensile stress concentra- nice experimental measurement results of residual stresses have been
tion region can be found. reported by US national labs using X-ray [21] or neutron diffraction
Moreover, the stress component distributions in y and z directions [57], such advanced large facilities are not accessible to us.
are shown in Fig. 12 (b) and (c), respectively. There is a long stress (Y Due to the large computational cost, the CFD-FEM model faces
component) concentration region along the inter-track gap, which difficulty to simulate the fabrication of a whole part. However, once
may cause the cracks parallel to the scanning path direction. The stress the thermos-elastic-plastic constitutive law and the temperature-

10
F. Chen and W. Yan Materials and Design 196 (2020) 109185

Fig. 11. (a) Selected nodes, D, E and F, in the CFD-FEM and thermo-mechanical simulations. The stress component curves in x, y and z direction at 3 nodes, (b) D, (c) E, and (d) F.
11
F. Chen and W. Yan Materials and Design 196 (2020) 109185

12
F. Chen and W. Yan Materials and Design 196 (2020) 109185

Fig. 12. (a) The crack occurring on the single-track sample; the X-stress distribution in the thermo-mechanical and CFD-FEM simulation at t1″ = 0.008s, t1′ = 0.00855s, respectively;
(b) the Y-stress distribution at t1′ = 0.00835 s; (c) the Z-stress distribution at t1′ = 0.01595 s.

Fig. 13. The evolution of the X-stress by CFD-FEM simulation compared with the observed cracking in the single track experiments [55]. (The experimental figures are reused under CC-BY
4.0 license.)

13
F. Chen and W. Yan Materials and Design 196 (2020) 109185

dependent material properties are accurately measured in experiments, CRediT authorship contribution statement
the more accurate thermal strain and stress in melt tracks can be
achieved. Subsequently, with proper multi-scale modelling approaches, Chen Fan: Methodology, Software, Validation, Formal analysis, In-
e.g., macro-scale modelling of the sequential accumulation of melt tracks vestigation, Data curation, Writing - original draft. Wentao Yan: Con-
with the accurate thermal strains from this meso-scale CFD-FEM model, a ceptualization, Methodology, Software, Resources, Writing - review &
whole part can be simulated in an efficient and accurate way, and the ex- editing, Supervision, Project administration, Funding acquisition.
perimental validation in part scale can be further conducted.

5. Conclusion Declaration of Competing Interest

A high-fidelity modelling approach for the thermal stress prediction The authors declare that they have no known competing financial
of the powder bed fusion process has been constructed. At first, the interests or personal relationships that could have appeared to influ-
thermal-fluid and mechanical models are linked by the temperature in- ence the work reported in this paper.
terpolation that maps the temperature profiles from the thermal-fluid
model into the FEM model. Then, by loading the temperature instead Acknowledgments
of the heat flux, the FEM simulation is simplified as a pure mechanical
analysis. Unlike the traditional thermo-mechanical applying the heat The authors acknowledge the support of the National Natural Sci-
source directly on the predefined track surface, the temperature profiles ence Foundation of China (Grant No. 51975393).
of the thermal-fluid model have enabled the reappearance of the so-
phisticated physical phenomena of the melting procedure in the FEM Appendix A. Appendix
model. Therefore, based on the quiet element method and the accurate
temperature field, the defects such as the voids and the rough surface In Fig. 15, the first two tracks are formed during 0 ∼ 0.008 s to be the
morphology have been revealed in the simulation results. Furthermore, first layer which is subsequently overlaid by the deposition of the rest
as illustrated in the cross comparison with the thermo-mechanical two tracks during 0.008 ∼ 0.016 s. Due to the different iterations of the
model. With the consideration of the melting and re-melting procedure, two simulations, the displayed time frames for the results of the FEM
the stress pattern for the CFD-FEM model in the molten pool region is simulation is 0.0005 s earlier than that of the thermal-fluid simulation.
more complicated; while in the unmelted region of the two methods, It reveals that the temperature field shows a remarkable agreement be-
the trend of stress curves is similar but the magnitude differs. Finally, tween two models in the whole deposition process of 4 tracks, and the
the overall stress distribution in the cooling region of the CFD-FEM FEM simulation manages to inherit the geometrical feature of the
model reveals the possible reasons for the generation of the cracks. thermal-fluid model.

Fig. 14. Schematic of the inactive elements model and the relatively lower temperature on the interface nodes between inactive elements and active elements.

14
F. Chen and W. Yan Materials and Design 196 (2020) 109185

Fig. 15. The comparison of the simulation results between the FEM model and the CFD model within 4 heating intervals which denote the deposition process of the 1st-layer 1st-track dur-
ing 0 ∼ 0.004 s, the 1st-layer 2nd-track during 0.004 ∼ 0.008 s, the 2nd-layer 1st-track during 0.008 ∼ 0.012 s, and the 2nd-layer 2nd-track during 0.012 ∼ 0.016 s, respectively.

15
F. Chen and W. Yan Materials and Design 196 (2020) 109185

Fig. 16. Picture frames showing the evolution process of the X-stress distribution on the first layer.

[14] K. Zeng, D. Pal, H. Gong, N. Patil, B. Stucker, Comparison of 3dsim thermal modelling
of selective laser melting using new dynamic meshing method to ansys, Mater. Sci.
Technol. 31 (8) (2015) 945–956.
[15] L. Gasman, Additive aerospace considered as a business, Additive Manufacturing for
References the Aerospace Industry, Elsevier 2019, pp. 327–340.
[16] Y. Yang, M. Allen, T. London, V. Oancea, Residual strain predictions for a powder bed
[1] J. Smith, W. Xiong, W. Yan, S. Lin, P. Cheng, O.L. Kafka, G.J. Wagner, J. Cao, W.K. Liu, fusion inconel 625 single cantilever part, Integrating Mater. Manuf. Innov. 8 (3)
Linking process, structure, property, and performance for metal-based additive (2019) 294–304.
manufacturing: computational approaches with experimental support, Comput. [17] J. Heigel, P. Michaleris, E.W. Reutzel, Thermo-mechanical model development and
Mech. 57 (4) (2016) 583–610. validation of directed energy deposition additive manufacturing of ti–6al–4v,
[2] C. Li, Z. Liu, X. Fang, Y. Guo, Residual stress in metal additive manufacturing, Proc. Addit. Manuf. 5 (2015) 9–19.
Cirp 71 (2018) 348–353. [18] E.R. Denlinger, P. Michaleris, Effect of stress relaxation on distortion in additive
[3] E.R. Denlinger, J.C. Heigel, P. Michaleris, Residual stress and distortion modeling of manufacturing process modeling, Addit. Manuf. 12 (2016) 51–59.
electron beam direct manufacturing ti-6al-4v, Proc. Inst. Mech. Eng. B J. Eng. [19] X. Liang, L. Cheng, Q. Chen, Q. Yang, A. C. To, A modified method for estimating in-
Manuf. 229 (10) (2015) 1803–1813. herent strains from detailed process simulation for fast residual distortion predic-
tion of single-walled structures fabricated by directed energy deposition, Addit.
[4] Y. Lu, S. Wu, Y. Gan, T. Huang, C. Yang, L. Junjie, J. Lin, Study on the microstructure,
Manuf. 23 (2018) 471–486.
mechanical property and residual stress of slm inconel-718 alloy manufactured by
differing island scanning strategy, Opt. Laser Technol. 75 (2015) 197–206. [20] X. Liang, Q. Chen, L. Cheng, D. Hayduke, A. C. To, Modified inherent strain method for
efficient prediction of residual deformation in direct metal laser sintered compo-
[5] V. Cain, L. Thijs, J. Van Humbeeck, B. Van Hooreweder, R. Knutsen, Crack propagation
nents, Comput. Mech. 64 (6) (2019) 1719–1733.
and fracture toughness of ti6al4v alloy produced by selective laser melting, Addit.
[21] R. Ganeriwala, M. Strantza, W. King, B. Clausen, T.Q. Phan, L.E. Levine, D.W. Brown, N.
Manuf. 5 (2015) 68–76.
Hodge, Evaluation of a thermomechanical model for prediction of residual stress
[6] N.J. Harrison, I. Todd, K. Mumtaz, Reduction of micro-cracking in nickel superalloys
during laser powder bed fusion of ti-6al-4v, Addit. Manuf. 27 (2019) 489–502.
processed by selective laser melting: a fundamental alloy design approach, Acta
[22] Y. Zhang, J. Zhang, Finite element simulation and experimental validation of distor-
Mater. 94 (2015) 59–68.
tion and cracking failure phenomena in direct metal laser sintering fabricated com-
[7] A. Bauereiß, T. Scharowsky, C. Körner, Defect generation and propagation mecha- ponent, Addit. Manuf. 16 (2017) 49–57.
nism during additive manufacturing by selective beam melting, J. Mater. Process.
[23] P. Prabhakar, W.J. Sames, R. Dehoff, S.S. Babu, Computational modeling of residual
Technol. 214 (11) (2014) 2522–2528.
stress formation during the electron beam melting process for inconel 718, Addit.
[8] W. Yan, W. Ge, Y. Qian, S. Lin, B. Zhou, W.K. Liu, F. Lin, G.J. Wagner, Multi-physics Manuf. 7 (2015) 83–91.
modeling of single/multiple-track defect mechanisms in electron beam selective [24] A. Hussein, L. Hao, C. Yan, R. Everson, Finite element simulation of the temperature
melting, Acta Mater. 134 (2017) 324–333. and stress fields in single layers built without-support in selective laser melting,
[9] D. Peral, J. De Vicente, J. Porro, J. Ocana, Uncertainty analysis for non-uniform resid- Mater. Des. 52 (2013) (1980-2015) 638–647.
ual stresses determined by the hole drilling strain gauge method, Measurement 97 [25] R. Xie, G. Chen, Y. Zhao, S. Zhang, W. Yan, X. Lin, Q. Shi, In-situ observation and nu-
(2017) 51–63. merical simulation on the transient strain and distortion prediction during additive
[10] A. Salmi, E. Atzeni, L. Iuliano, M. Galati, Experimental analysis of residual stresses on manufacturing, J. Manuf. Process. 38 (2019) 494–501.
alsi10mg parts produced by means of selective laser melting (slm), Procedia CIRP 62 [26] X. Song, S. Feih, W. Zhai, C.-N. Sun, F. Li, R. Maiti, J. Wei, Y. Yang, V. Oancea, L.R.
(2017) 458–463. Brandt, et al., Advances in additive manufacturing process simulation: residual
[11] P. Rangaswamy, M. Griffith, M. Prime, T. Holden, R. Rogge, J. Edwards, R. Sebring, Re- stresses and distortion predictions in complex metallic components, Mater. Des.
sidual stresses in lens® components using neutron diffraction and contour method, 108779 (2020).
Mater. Sci. Eng. A 399 (1–2) (2005) 72–83. [27] P. Foteinopoulos, A. Papacharalampopoulos, K. Angelopoulos, P. Stavropoulos, De-
[12] B. Vrancken, V. Cain, R. Knutsen, J. Van Humbeeck, Residual stress via the contour velopment of a simulation approach for laser powder bed fusion based on scanning
method in compact tension specimens produced via selective laser melting, Scr. strategy selection, Int. J. Adv. Manuf. Technol. 108 (9) (2020) 3085–3100.
Mater. 87 (2014) 29–32. [28] S.H. Nikam, N. Jain, Modeling and prediction of residual stresses in additive layer
[13] M. Kabiri, Nonuniform residual-stress measurement by hole-drilling method, Exp. manufacturing by microplasma transferred arc process using finite element simula-
Mech. 24 (4) (1984) 328–336. tion, J. Manuf. Sci. Eng. 141 (6) (2019).

16
F. Chen and W. Yan Materials and Design 196 (2020) 109185

[29] M. Yi, B.-X. Xu, O. Gutfleisch, Computational study on microstructure evolution and [44] W. Yan, S. Lin, O.L. Kafka, Y. Lian, C. Yu, Z. Liu, J. Yan, S. Wolff, H. Wu, E. Ndip-Agbor,
magnetic property of laser additively manufactured magnetic materials, Comput. et al., Data-driven multi-scale multi-physics models to derive process–structure–
Mech. 64 (4) (2019) 917–935. property relationships for additive manufacturing, Comput. Mech. 61 (5) (2018)
[30] R. Paul, S. Anand, F. Gerner, Effect of thermal deformation on part errors in metal 521–541.
powder based additive manufacturing processes, J. Manuf. Sci. Eng. 136 (3) (2014). [45] T.K. Ales, An Integrated Model for the Probabilistic Prediction of Yield Strength in
[31] W. Yan, W. Ge, J. Smith, S. Lin, O.L. Kafka, F. Lin, W.K. Liu, Multi-scale modeling of Electron-Beam Additively Manufactured Ti-6al-4v, 2018.
electron beam melting of functionally graded materials, Acta Mater. 115 (2016) [46] L.-E. Lindgren, E. Hedblom, Modelling of addition of filler material in large deforma-
403–412. tion analysis of multipass welding, Commun. Numer. Methods Eng. 17 (9) (2001)
[32] T. Mukherjee, W. Zhang, T. DebRoy, An improved prediction of residual stresses and 647–657.
distortion in additive manufacturing, Comput. Mater. Sci. 126 (2017) 360–372. [47] W. Yan, Y. Qian, W. Ge, S. Lin, W.K. Liu, F. Lin, G.J. Wagner, Meso-scale modeling of
[33] J. Cheon, S.-J. Na, Prediction of welding residual stress with real-time phase transfor- multiple-layer fabrication process in selective electron beam melting: inter-layer/
mation by cfd thermal analysis, Int. J. Mech. Sci. 131 (2017) 37–51. track voids formation, Mater. Des. 141 (2018) 210–219.
[34] J. Cheon, D.V. Kiran, S.-J. Na, Cfd based visualization of the finger shaped evolution in [48] S.M.H. Hojjatzadeh, N.D. Parab, W. Yan, Q. Guo, L. Xiong, C. Zhao, M. Qu, L.I. Escano,
the gas metal arc welding process, Int. J. Heat Mass Transf. 97 (2016) 1–14. X. Xiao, K. Fezzaa, et al., Pore elimination mechanisms during 3d printing of metals,
[35] J. Cheon, D.V. Kiran, S.-J. Na, Thermal metallurgical analysis of gma welded ah36 Nat. Commun. 10 (1) (2019) 1–8.
steel using cfd–fem framework, Mater. Des. 91 (2016) 230–241. [49] C.W. Hirt, B.D. Nichols, Volume of fluid (vof) method for the dynamics of free
[36] N.S. Bailey, C. Katinas, Y.C. Shin, Laser direct deposition of aisi h13 tool steel powder boundaries, J. Comput. Phys. 39 (1) (1981) 201–225.
with numerical modeling of solid phase transformation, hardness, and residual [50] B. Cheng, S. Shrestha, K. Chou, Stress and deformation evaluations of scanning strat-
stresses, J. Mater. Process. Technol. 247 (2017) 223–233. egy effect in selective laser melting, Addit. Manuf. 12 (2016) 240–251.
[37] P. Foteinopoulos, A. Papacharalampopoulos, P. Stavropoulos, On thermal modeling [51] D. Dassault Systèmes, Abaqus Analysis User's Guide, Tech. rep., Technical Report
of additive manufacturing processes, CIRP J. Manuf. Sci. Technol. 20 (2018) 66–83. Abaqus 6.14 Documentation Simulia Corp, 2016.
[38] A. Lundbäck, L.-E. Lindgren, Modelling of metal deposition, Finite Elem. Anal. Des. 47 [52] J.N. Reddy, An Introduction to the Finite Element Method, New York 27, 1993.
(10) (2011) 1169–1177. [53] I. Yadroitsev, I. Yadroitsava, Evaluation of residual stress in stainless steel 316l and
[39] E.R. Denlinger, J. Irwin, P. Michaleris, Thermomechanical modeling of additive ti6al4v samples produced by selective laser melting, Virtual Phys. Prototy. 10 (2)
manufacturing large parts, J. Manuf. Sci. Eng. 136 (6) (2014). (2015) 67–76.
[40] M. J. Dantin, W. M. Furr, M. W. Priddy, Towards an Open-Source, Preprocessing [54] B. Vrancken, R. Wauthlé, J.-P. Kruth, J. Van Humbeeck, Study of the influence of ma-
Framework for Simulating Material Deposition for a Directed Energy Deposition terial properties on residual stress in selective laser melting, Proceedings of the Solid
Process. Freeform Fabrication Symposium 2013, pp. 393–407.
[41] A. Krishnakumar, K. Suresh, A. Chandrasekar, Towards Assembly-Free Methods for [55] B. Vrancken, R.K. Ganeriwala, M.J. Matthews, Analysis of laser-induced
Additive Manufacturing Simulation, in: ASME 2015 International Design Engineer- microcracking in tungsten under additive manufacturing conditions: experiment
ing Technical Conferences and Computers and Information in Engineering Confer- and simulation, Acta Mater. (2020) 464–472.
ence, American Society of Mechanical Engineers Digital Collection 2015. [56] G. Wang, H. Ouyang, C. Fan, Q. Guo, Z. Li, W. Yan, Z. Li, The origin of high-density dis-
[42] H. Huang, N. Ma, J. Chen, Z. Feng, H. Murakawa, Toward large-scale simulation of re- locations in additively manufactured metals, Mater. Res. Lett. 8 (8) (2020) 283–290.
sidual stress and distortion in wire and arc additive manufacturing, Addit. Manuf. [57] K. An, L. Yuan, L. Dial, I. Spinelli, A.D. Stoica, Y. Gao, Neutron residual stress measure-
(2020) 101248. ment and numerical modeling in a curved thin-walled structure by laser powder
[43] P. Michaleris, Modeling metal deposition in heat transfer analyses of additive bed fusion additive manufacturing, Mater. Des. 135 (2017) 122–132.
manufacturing processes, Finite Elem. Anal. Des. 86 (2014) 51–60.

17