Sunteți pe pagina 1din 25

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/322688604

A Computationally Efficient Finite Element Framework to Simulate Additive


Manufacturing Processes

Article  in  Journal of Manufacturing Science and Engineering · January 2018


DOI: 10.1115/1.4039092

CITATIONS READS
10 804

2 authors:

Shiyan Jayanath Ajit Achuthan


Clarkson University Clarkson University
4 PUBLICATIONS   11 CITATIONS    49 PUBLICATIONS   380 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Additive Manufacturing View project

All content following this page was uploaded by Shiyan Jayanath on 25 September 2018.

The user has requested enhancement of the downloaded file.


Journal of Manufacturing Science and Engineering. Received October 25, 2017;
Accepted manuscript posted January 24, 2018. doi:10.1115/1.4039092
Copyright (c) 2018 by ASME

A Computationally Efficient Finite Element


Framework to Simulate Additive Manufacturing
Processes

Shiyan Jayanath

d
Research Assistant
Dep. of Mech. & Aero. Eng.

ite
Clarkson University
Potsdam, NY-13699,USA

ed
Email: wewalas@clarkson.edu

py
Ajit Achuthan∗
Associate Professor
Dep. of Mech. & Aero. Eng.

Co
Clarkson University
Potsdam, NY-13699,USA
Email: aachutha@clarkson.edu

ot
tN
ABSTRACT
Macro-scale Finite Element (FE) models, with their ability to simulate Additive Manufacturing (AM) processes
rip

of metal parts and accurately predict residual stress distribution, are potentially powerful design tools. However,
these simulations require enormous computational cost, even for a small part only a few orders larger than the
sc

melt pool size. Existing adaptive meshing techniques to reduce computational cost substantially by selectively
coarsening are not well suited for AM process simulations due to the continuous modification of model geometry as
material is added to the system. To address this limitation a new FE framework is developed. The new FE framework
nu

is based on introducing updated discretized geometries at regular intervals during the simulation process, allowing
greater flexibility to control the degree of mesh coarsening than a technique based on element merging recently
reported in the literature. The new framework is evaluated by simulating Direct Metal Deposition (DMD) of a
Ma

thin-walled rectangular and a thin-walled cylindrical part, and comparing the computational speed and predicted
results with those predicted by simulations using the conventional framework. The comparison shows excellent
agreement in the predicted stress and plastic strain fields, with substantial savings in the simulation time. The
ed

method is then validated by comparing the predicted residual elastic strain with that measured experimentally by
neutron diffraction of the thin-walled rectangular part. Finally, the new frameworks capability to substantially
reduce the simulation time for large-scale AM parts is demonstrated by simulating a one-half foot thin-walled
pt

cylindrical part.
ce

1 Introduction
Additive Manufacturing (AM) is a layer by layer manufacturing process using the digital data of the part geometry
Ac

[1]. Originally developed as a rapid prototyping technology, the AM gradually found its way into the manufacturing of
many products in a wide range of industries, especially biomedical and aerospace [2] due to its numerous advantages over
traditional manufacturing techniques, such as the nearly unlimited freedom to design and build geometrically complex parts
[3], [4]. However, tapping the full potential of AM as an advanced manufacturing tool with the ability to realize novel designs
for parts and to engineer the micro-structure of materials requires addressing many technical challenges.
Two of the most critical challenges that impede the advancement of current metal AM processes into new applications
are residual stress formation and part distortion [5]. Typically, in metal AM, a powder bed, a powder jet or a wire feed

∗ Corresponding author, aachutha@clarkson.edu

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 01/29/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Journal of Manufacturing Science and Engineering. Received October 25, 2017;
Accepted manuscript posted January 24, 2018. doi:10.1115/1.4039092
Copyright (c) 2018[6]
system byisASME
utilized to deposit the feedstock material. The deposited material is sintered or melted with a laser or electron
beam heat source to form a layer of the target part according to the digital data that describes the geometric details of that
layer. The process is repeated on a layer by layer basis to fully build the part. As the AM process of a part progresses, the
region of the part that has already been built is continuously subjected to non-uniform thermal contraction and a series of
intermittent expansions. While the non-uniform thermal contraction is due to the natural cooling of the material at different
rates at different locations, the intermittent expansions are due to the reheating at material points as the heat source moves
following the scanning strategy. The non-uniform thermal contraction and the intermittent expansions produce large local
stress, often causing plastic yielding of material. The yielding of material results in the evolution of non-uniform plastic
strain in the part. The non-uniformity of the plastic strain also depends on the mechanical restriction to deformation from the
substrate. The non-uniform plastic and thermal strains accumulated in the part as the part cools down to room temperature
are the primary causes for residual stress formation and part distortion. The residual stress and part distortion result in

d
poor part quality and significantly affect the part’s ability to meet functional and other design specifications. In addition,
undesirable residual stress distribution could lead to the premature mechanical failure of a part, especially under cyclic

ite
loading. Therefore, the development of effective methods to mitigate the residual stress formation and part distortion is
imperative for the successful advancement of metal AM technology.

ed
Several methods to mitigate residual stress formation and part distortion have been reported in the literature. Post-build
treatments, such as hot isostatic pressing (HIP), have been used successfully [7]. However, post-build treatments typically
increase the production cost and part turnover time. They are also limited to the parts with relatively simple geometry. In

py
many advanced applications, the geometric precision and the extent to which the evolved mechanical properties match the
target properties are so critical that the post-build treatments could be either impractical or insufficient. Another approach to
mitigate residual stress formation and part distortion is determining optimal process parameters and using those parameters to

Co
manufacture a part [8], [9]. Typically, the optimal process parameters are determined through a combination of experimental
trial runs and engineering judgment. Methods to mitigate residual stress formation and part distortion based on the real-time
sensing of various parameters that provide a measure of part distortion and the real-time modification of process parameters

ot
based on these measurements have also been developed recently [6].
Computational models to simulate the AM process of a part can be quite beneficial for the development and the effective
tN
implementation of the aforementioned methods to mitigate residual stress formation and part distortion. For instance, the
determination of optimal process parameters based on experimental trial runs and engineering judgment can be exorbitantly
expensive and challenging, depending on many factors such as the complexity of the part. Rather than relying heavily on
rip

experimental trail runs, a computational model that can accurately predict the residual stress formation and part distortion as
a function of the process parameters can be used to determine these optimal process parameters. Similarly, a computational
model can also serve as the core component of the feedback loop in the residual stress mitigation method based on the real-
sc

time modification of process parameters. However, to be effective as an AM process design tool, the computational model
should account for all the essential physics of the processes and provide a solution with a satisfactory spatial and temporal
resolution using a reasonable amount of computational resources and simulation time.
nu

Among many different computational models to simulate AM processes reported in the literature, the macro-scale FE
models are often considered the most appropriate to predict part distortion and residual stress [10]. The characteristic length
Ma

scale of this computational model described by the size of the finite elements and the material constitutive model, though
too large to capture the complex melt-pool physics, is sufficiently small to accurately describe: 1) the geometry of the part
irrespective of its complexity, and 2) the gradients in temperature and mechanical deformation in bulk, except in a region
very close to the melt-pool. Lagrangian based finite element methods [11, 12, 13, 14, 15, 16, 17, 18] are commonly used in
ed

macro-scale modeling. For example, Mukherjee et al. [16] implemented a fluid flow model in a transient thermal analysis to
improve residual stress and distortion prediction of AM parts. Apart from the Lagrangian approach, some researchers [19]
have implemented the Eulerian approach to improve the computational efficiency of simulation of AM process.
pt

The main challenge in using the macro-scale FE models is the enormous computational cost to simulate real size AM
parts. The framework of these models requires a uniform discretization of the part geometry with a very large element density
ce

in order to predict results accurately. Uniform discretization with large element density increases the number of degrees of
freedom (DoF) substantially with an increase in the size of the part, which increases the computational cost exponentially.
Ac

Denlinger et al. [12] addressed this challenge by introducing an adaptive meshing strategy and demonstrated the model’s
ability to simulate large parts of size in the order of a meter. The adaptive meshing was realized through a layer by layer
mesh coarsening by progressively merging the elements in the lower layers of the FE mesh. The model was validated by
comparing the predicted distortion results with the experimental data for a Ti-6Al-4V part manufactured with a large electron
beam. Although a substantial increase in the computational efficiency was achieved, the merging of elements to realize mesh
coarsening drastically limits the freedom on controlling the degree of mesh coarsening. Thus, the ability to harness the full
potential of mesh coarsening based adaptive meshing for improving the computational efficiency of the FE model is severely
limited.
In the present study, a new FE framework that can improve the computational efficiency of the AM process simulation
by allowing better control over mesh adaptation is developed. The new framework relies on regularly introducing updated

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 01/29/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Journal of Manufacturing Science and Engineering. Received October 25, 2017;
Accepted manuscript posted January 24, 2018. doi:10.1115/1.4039092
Copyright (c) 2018 by geometries
discretized ASME for the already completed region of the build part as the simulation of the AM process continues to
progress. The regular introduction of the newly discretized geometries after the completion of a set of layers of the AM
part provides an avenue to adapt the mesh independent of the refinement of the discretized geometries used previously.
This freedom in discretization enhances the efficiency of mesh adaptation substantially. The continuity of the process is
maintained by mapping the solution from the previously discretized geometry to the newly discretized geometry. The new
framework is demonstrated by simulating a Direct Metal Deposition (DMD) AM process. The new framework is also
applicable to powder bed additive manufacturing processes, such as selective laser melting (SLM) or electron beam melting
(EBM).

2 THEORY

d
2.1 Conventional FE Framework for AM simulation
Generally, the macro-scale computational models for simulating AM process are formulated on a decoupled thermo-

ite
mechanical analysis framework assuming that the effect of mechanical deformation on the thermal field is negligible. Ac-
cordingly, these models typically consist of two decoupled modeling steps; a transient thermal analysis modeling step and a

ed
quasi-static structural analysis modeling step. The transient thermal analysis and the quasi-static structural analysis model-
ing steps will be referred to as thermal analysis and structural analysis in the rest of the article for the sake of convenience.
Both the analysis steps are FEA based where the spatial and temporal domains are discretized into finite elements and time

py
steps, respectively. The spatial domain is then analyzed by applying the fundamental balance laws, constitutive models and
boundary conditions at every time step, as the time progresses from the initial state.
In the thermal analysis, as the AM part is built, the temperature evolution is determined by simulating the melting and

Co
solidification of powder or wire feed material under a moving heat source. The material phase of individual elements is
determined at every time step by tracking the elements’ temperature history. The change of state of the material phase from
a solid as powder or wire feed to liquid during melting and back to solid during solidification are simulated by changing

ot
their temperature dependent material properties appropriately ( [11,14,20,16]). The temperature field and the material phase
information are then exported to the structural analysis. In the structural analysis, the evolution of stress and strain fields in
tN
the AM part is determined at every time step based on the thermal and material phase information obtained at the same time
step in the thermal analysis.
In both the thermal and structural modeling steps, the material volume is added as illustrated in the schematic shown in
rip

Fig. 1. Two-time steps of a typical conventional analysis are shown. Note that though the schematic is illustrated specifically
for powder jet deposition technology, the schematic is applicable to other AM technologies such as powder bed and wire
feed, as well. Both the thermal and structural analyses use the same base geometry that consists of the part-to-be-built and
sc

the build plate. For the simulation of powder bed technology, the whole powder volume that would eventually be deposited
forms the base geometry even if a large portion of this powder will be left unused. Typically, the same FE mesh, ie. the
discretized geometry into elements of appropriate sizes, is used in both analysis steps. Initially, in both the analysis steps,
nu

all the elements, except those for the build plate, are set to a dormant state. The elements are then progressively activated in
subsequent time steps, thus simulating the building process. The activation of elements follows the feed material deposition
Ma

strategy in the thermal analysis step. In the structural analysis step, only the elements representing the solid material phase
are activated based on the phase information obtained for that given time step from the thermal analysis. The elements
representing the melt pool or the feed material do not have any stress-bearing capacity and therefore are left in the dormant
state in the structural analysis until they solidify. Generally, thermal and structural steps are performed as an uncoupled
ed

system, where thermal analysis of the whole part is performed first, and then the temperature field and material phase
information are transferred to the structural analysis to perform the stress analysis.
The setting of elements to the dormant state is typically employed following one of two commonly used methods: the
pt

quiet element method or the in-active element method. In the quiet element method, the dormant state of an element is
realized by diluting its influence on the solution by assigning appropriate values for material properties rather than real
ce

values. Since the temperature field is the primary solution determined in the thermal analysis, an element can be set to the
dormant state by assigning very low values, a couple of orders lower than their actual values, for conductivity and specific
Ac

heat. Similarly, an element’s influence on the stress field in the structural analysis can be diluted by assigning a low stiffness
value [21]. Subsequently, the activation of these elements when needed is a simple process of setting their material properties
back to their real temperature dependent values. Since the dormant elements are not actually removed in the quiet element
method, the DoF associated with the dormant elements are always present in the system of equations. In the in-active
element method, the setting of the elements to dormant state is implemented by completely removing their DoF from the
system of equations. Elements are activated as the simulation progresses by adding the corresponding DoF into the system of
equations. Though the total DoF are relatively small in this method, thus potentially increasing the computational efficiency
substantially, the associated overhead cost for the renumbering of nodes and elements can increase quickly with its repeated
usage in an analysis.
A hybrid method that uses the in-active method for a layer-by-layer addition of the material and the quiet method for

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 01/29/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Journal of Manufacturing Science and Engineering. Received October 25, 2017;
Accepted manuscript posted January 24, 2018. doi:10.1115/1.4039092
Copyright (c) 2018 by ASME
Thermal analysis Structural analysis
Melt pool
---------- ----------

LAYER
LAYER
(n+2)
4 LAYER
LAYER
(n+2)
4
LAYER
LAYER
(n+1)
3 LAYER
LAYER
(n+1)
3
Import Active region in the thermal analysis
t=t1 LAYER n Temp. Field LAYER n

Solid Fraction In-active / Quiet elements

d
Activated elements

ite
In the thermal analysis
(Deposited material)

ed
---------- ----------
Activated elements
In the structural analysis
LAYER
LAYER
(n+2)
4 LAYER
LAYER
(n+2)
4

py
(Solid Fraction)
LAYER
LAYER 3
(n+1) Import LAYER (n+1)
Temp. Field Solid Fraction

Co
LAYER n LAYER n
t=t2 Solid Fraction

ot Active region in the structural analysis


tN
Fig. 1: Schematic illustrating the conventional framework for two-time steps
rip

the addition of material within a layer has also been proposed [21]. The hybrid method is realized by initially setting all
sc

the elements in the entire geometry as in-active, and then progressively switching the elements in individual layers to quiet
elements as the building progresses. Michaleris et al. have shown that the hybrid method is more computationally efficient
for the thermal analysis step of AM process than using quiet or in-active methods individually.
nu

The major drawback of the conventional framework is that the discretization of the full part geometry is performed
before the analysis and remains the same throughout the analysis. As a consequence, the element density of the already built
Ma

region of the part continues to remain dense. Therefore, the total DoF of the system increases substantially as the simulation
continues with the addition of more and more layers. A recently reported method of merging existing elements to coarsen
the mesh can address this issue to a limited extent [12]. The new framework developed in this study is a lot less-cumbersome
alternative approach that offers more freedom to coarsen the mesh, thus yielding higher computational efficiency.
ed

2.2 New FE Framework for AM simulation


pt

Rather than using a single discretized geometry and merging elements, the new FE framework proposes to attain higher
computational efficiency by enriching the implementation of discretized geometries. While the conventional framework
ce

employs a single discretized geometry, the new framework proposes to coarsen the mesh appropriately at regular intervals
through introducing new discretized geometries.
Ac

The new FE framework is illustrated in Fig. 2. As shown in Fig. 2 there are two stages in the proposed new framework
which have been given the names main analysis and fine mapping. The stage called main analysis illustrated on the left of
the figure is similar to the conventional framework discussed above, but with additional steps to introduce new discretized
geometries.
Therefore, the total simulation in the main stage can be considered as grouped into a number of discretization steps,
with each step consisting of the simulation of a set of a few layers. At the end of each discretization step, a new discretized
geometry with a coarser mesh for the region in the already built part that is relatively far away from the layer to be built in
the following time step is deployed. The basic premise behind this approach is that the gradient of the incremental solution
for that time step is relatively smaller for that region since the heat source and the rapidly cooling region is relatively far
away. Within each discretization step, the thermal and the structural analysis simulation steps at various time increments are

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 01/29/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Journal of Manufacturing Science and Engineering. Received October 25, 2017;
Accepted manuscript posted January 24, 2018. doi:10.1115/1.4039092
Copyright (c) 2018 by ASME
Main Analysis Fine Mapping
Pre-defined mesh 1 (Parallel/post processing step)
Discretization Temp.
Thermal Structural Compute solution on ^ (t1 )
step 1
analysis Solid Fraction analysis S(t1 ) original config. S
^ (t1 )
S
Discretization Coarse
step 2 mapping
Pre-defined ~(t ) ~(t )
T2 0
S2 0

mesh 2

Deformed conf. 2

d
Thermal Temp. Structural Compute solution on ^ (t2 )
analysis Solid Fraction analysis S(t2 ) original config. S

ite
^ (t2 )
S
Discretization Coarse
step 3 mapping

ed
~(t ) ~(t )
Pre-defined T3 0
S3 0
Fine
mesh 3 mapping

py
Deformed conf. 3
Solution on
fine mesh

Co
Thermal Temp. Structural
S(t3 ) Compute solution on ^ (t3 )
S
analysis Solid Fraction analysis original config.

^3
S

ot
tN
^ (n−1)
S
Discretization Coarse
step n mapping
~(t ) ~(t )
rip

Tn 0
Sn 0

Pre-defined
mesh n
sc

Deformed conf. n
nu

Temp.
Thermal Structural
Sn Compute solution on ^n
S
analysis Solid Fraction analysis original config.
Ma

Fig. 2: Schematic illustrating the new framework


ed

performed following the conventional FE framework discussed above (Fig. 1).


In Fig. 2, the block named marked in blue at the center of the main analysis stage and named as coarse mapping, which
pt

represents the mapping of the solution from the end of the previous discretization step to the newly introduced undeformed
discretized geometry, thus enforcing the continuity of the solution in the time domain. The coarse mapping primarily consists
ce

of two tasks: 1) interpolation of the solution, and 2) introducing the interpolated solution as the pre-existing condition
for the next discretization step. The mapping of displacement solution from the structural model allows the continuity of
Ac

deformation across the discretization steps. The mapping is implemented by simply setting the initial nodal locations of the
newly introduced discretized geometry to their deformed locations as calculated from the interpolated displacement solution.
The continuity of stress, plastic strain, and temperature field solutions across discretization steps are also ensured by mapping
these quantities, implemented by defining the interpolated values of their solution from the previous discretization step as
pre-existing quantities in the new discretization step. The mapping of the displacement solution is not necessary for the
thermal analysis since its coupling to the structural analysis is typically weak.
The second stage of the framework involves mapping the solution at the end of every discretization step to a reference
geometry that is discretized uniformly with very high element density. This stage is referred to as the fine mapping stage and
is represented on the right side of the schematic (Fig. 2). The fine mapping stage ensures that the resolution of a solution
obtained in a region of high element density in a discretization step is retained irrespective of whether that region is coarsened

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 01/29/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Journal of Manufacturing Science and Engineering. Received October 25, 2017;
Accepted manuscript posted January 24, 2018. doi:10.1115/1.4039092
Copyright (c) 2018
in the by ASMEdiscretization steps. The fine mapping stage can be performed in parallel with the main analysis stage or
subsequent
in series after the completion of the main analysis stage. The process can be described as,

j+k
Si = Sij + ∑ ∆Sl+1
i
(1)
l= j

where S is the solution (temperature, stress or plastic strain field), subscript i indicates the layer number and j indi-
cates the discretization step number in which the layer i is activated. ∆Sl+1 i indicates the incremental solution at (l + 1)
discretization step, where,

d
ite
i i i
∆Sl+1 = (Sl+1 − Ŝl+1 ) (2)

ed
i
Ŝl+1 is the mapped solution of Sli from l th mesh to (l + 1)th mesh; Sli 7→ Ŝl+1
i . The constant k determines the number of

discretization steps from which the solution is mapped. Note that the summation of each term involves interpolation from
a coarser discretization to a finer discretization of the layer. For maximum accuracy, the solution from all the discretization

py
steps following the j-th step can be added; i.e. k = n − j, where n is the total number of discretization steps. However, the
summation of the incremental solutions is typically required only for a few discretization steps, since the solution of a layer

Co
saturates significantly after the printing of a few more layers above it.
The main advantage of the new framework is that it offers a greater freedom to choose a mesh of appropriate distribution
of element sizes independent of any previous discretization. As a result, the total DoF of the system can be maintained to
remain relatively low as more layers being added, thus making substantial savings in computational cost. Accuracy of the

ot
solution can be controlled by varying: 1) the mesh density appropriately, 2) the number of layers per discretization step, and
3) the parameter k for the ‘fine mapping’ stage (Eqn. 1).
tN
2.3 Governing equations
rip

2.3.1 Thermal analysis


The governing equation for the thermal analysis is given below [10].
sc

dT
Qv (x,t) − ρC p + ∇.(k∇T ) = 0 (3)
dt
nu

Where Qv (x,t) is the volumetric heat source, x is the position vector, t is time, ρ is density, C p is the specific heat, T
Ma

is temperature and k is the heat transfer coefficient. The governing equation is non-linear due to the temperature dependent
material properties. Initial and boundary conditions of the above differential equation are as below.

Initial conditions,
ed
pt

T (x,t0 ) = T0 (x)
ce

Where t0 and T0 are the initial time and initial temperature field.

Following boundary conditions are appropriate for the scanning surface and the surfaces of the already built region of
Ac

the part,
Surface convection

qconv (x,t) = h(T − T∞ ) (4)

Surface radiation

qrad (x,t) = εσ(T 4 − T∞4 ) (5)

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 01/29/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Journal of Manufacturing Science and Engineering. Received October 25, 2017;
Accepted manuscript posted January 24, 2018. doi:10.1115/1.4039092
Copyright (c) 2018 by ASME Table 1: DMD parameters

Parameter Value/details
Number of layers 32 (smaller parts), 300 (larger part)
Layer thickness 0.5 mm
Laser Power, P 1550 W
Laser scan speed ≈14 mm/sec
Cooling time
10 sec
between layers
Zig-zag (plate)
Laser patten Random start,
alternating direction (cylinder)

d
ite
Where, qconv and qrad are the heat fluxes, h is the convection coefficient, T∞ is the room temperature, ε is the emissivity

ed
and σ is the Stefan-Boltzmann constant.

py
2.3.2 Structural Analysis
Governing equation for the quasi-static structural analysis is given as, [11]

Co
∇.σ + F = 0 (6)

ot
Where, σ is the stress tensor, F is the body forces. A simple thermo-elastic-plastic constitutive model is used to describe
tN
the material behavior as shown below,
rip

σ = C(ε − ε p − εth ) (7)


sc
nu

ε = εe + ε p + εth (8)
Ma

Where, ε, εe ,ε p ,εth are total strain, elastic strain, plastic strain and thermal strain, respectively, and C is a tensor that
describes temperature dependent elastic properties. The structural and thermal analysis are linked through εth .
ed

3 Model validation
pt

The validation of the model was carried out by performing the following three tasks: 1) simulating the AM process
for two relatively small parts with simple geometries (a rectangular part with length, l=42mm (1.65”) , height, h=16 mm
ce

(0.63”) , and thickness, t=3 mm (0.12”) and a cylindrical part with diameter, d=42mm (1.65”), height, h=16 mm (0.63”),
and thickness, t=3 mm (0.12”)) using the new FE framework and comparing the computational time and the predicted stress
Ac

and plastic strain fields to those predicted by a simulation using the conventional framework; 2) comparing the predicted
residual elastic strain with experimentally measured values for the small rectangular part; and 3) successfully simulating
the AM process for a relatively large part (a cylindrical part with diameter, d=42mm (1.65”), height, h=150 mm (5.91”),
and thickness, t=3 mm (0.12”)) using the new FE framework employing only a moderate level of computational resources.
Though the framework is applicable to powder jet, powder bed and wire-feed processes, only the powder jet process (DMD)
was considered for the purpose of validation in this study. The process parameters used in this study are listed in Table 1.
Note that the parts considered have thin-walled geometries and use a single scan line in the build plane.
The material used was 316L steel alloy. Temperature-dependent thermal and mechanical material properties of 316L
are listed in Table 2, Table 3 and Table 4, respectively [14]. Room temperature (298 K) was set as the initial temperature of
the build plate and newly added layers before the elements are activated.

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 01/29/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Journal of Manufacturing Science and Engineering. Received October 25, 2017;
Accepted manuscript posted January 24, 2018. doi:10.1115/1.4039092
Copyright (c) 2018 by ASME Table 2: Thermal conductivity of bulk material [14]

k
T(K)
(×10−6W /m.K )
298 13
458 14
612 16
771 20
920 23
1077 25
1234 28
1388 30

d
1543 33

ite
1617 27
1708 27

ed
py
Co
Table 3: Enthalpy change with time [14]

Enthalpy change
T(K)
(J/g )

ot
298 0
473 83
tN
873 312
1073 436
1473 696
rip

1658 834
1723 1131
sc

1773 1176
1873 1255
nu
Ma

Table 4: Temperature depended mechanical properties [14]


ed

T E α YTS UTS
ν %
(K) (GPa) (×10−6 K −1 ) (MPa) (MPa)
pt

298 199.0 0.3 14.3 290.0 627.0 55


473 187.0 0.3 15.6 - - -
ce

477 - - - 241.0 558.0 51


589 - - - 214.0 538.0 48
673 172.0 0.3 16.9 - - -
Ac

700 - - - 190.0 524.0 47


811 - - - 165.0 483.0 44
873 157.0 0.3 17.7 - - -
922 - - - 145.0 393.0 40
973 141.0 0.3 18.5 - - -
1033 - - - 124.0 241.0 37
1073 106.0 0.3 19.1 - - -
1144 - - - 110.0 165.0 44
1573 10.0 3 19.5 - - -

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 01/29/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Journal of Manufacturing Science and Engineering. Received October 25, 2017;
Accepted manuscript posted January 24, 2018. doi:10.1115/1.4039092
Copyright (c)
3.12018 by ASME of old and new frameworks for the small parts
Comparison
Both the new FE framework and the conventional FE framework were implemented on the ABAQUS (standard, 6.14.1)
FE package [22]. ABAQUS Automatic time step option was selected for the simulations. The selective mesh coarsening was
implemented using the PYTHON programming language. The coarse mapping was implemented using ABAQUS mesh-
to-mesh solution mapping technique except for displacements. The mapping of displacement solution for coarse mapping
was implemented using PYTHON griddata interpolation tool. The PYTHON griddata interpolation tool was also used for
mapping all the solutions in the ‘fine mapping’ stage. The automation of various steps in both the new and conventional FE
frameworks was made using PYTHON and UNIX shell scripts. The model change element activation method [22] available
in ABAQUS was used for setting elements to a dormant state and then activating them as needed.
All the simulations were performed on a desktop PC, having OpenSUSE 42.3 OS, 1.2 GHz, 12 cores, Intel Xeon CPU
and 32 GB RAM system configuration. In the conventional method, the entire printing process was simulated using single FE

d
mesh. The hybrid element activation method was used for the thermal analysis and the in-active element activation method
was used for the structural analysis (section 2.1).

ite
In the new framework, separate discretization steps (FE meshes) were used for 4 sets of subsequent layers (i.e total 8
discretization steps). Similar to the conventional method, the new framework was implemented using the hybrid method

ed
for the thermal simulation and the inactive method for the structural analysis. Thermal and structural analysis of each layer
were performed sequentially as per the schematic in Fig. 2. Selective mesh coarsening was implemented in four stages as
illustrated in Fig. 3 for both the parts. Grayscale represents the relative mesh density of each mesh region. Implementation

py
of each mesh region of the part is shown in Table 5.

Co
Element size Element size
0.5 mm, F1 (4 layers) 0.5 mm,
2 elements per F2 (4 layers) 1 element per

ot
layer thickness tN layer thickness

F: fine
C1 C: coarse
rip

Max. element
sc

size 1.1 mm
nu
Ma

Fig. 3: Selective mesh coarsening scheme for small parts


ed
pt

Table 5: Application of selective mesh coarsening at each discretization step for small parts
ce

Additive Newly added Mesh coarsening


step no. layers regions
Ac

1 1-4 F1
2 5-8 F1,F2
3-8 9-32 F1,F2,C1

The discretized geometries used for the smaller parts for both the new and conventional FE frameworks are shown in
Fig. 4 (at the last discretization step). Fig. 4 (a) and (b) are the FE meshes used for the conventional analysis without any
selective mesh coarsening. Fig. 4 (c) and (d) are the FE meshes used in the new framework. 8-node hexagonal elements,
DCC3D8 for thermal analysis and C3D8 for structural analysis, were used for both the frameworks [22].

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 01/29/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Journal of Manufacturing Science and Engineering. Received October 25, 2017;
Accepted manuscript posted January 24, 2018. doi:10.1115/1.4039092
Copyright (c) 2018 by ASME

# nodes= 122,258
# nodes= 72,985

d
ite
(a) (b)

ed
# nodes= 37,200 # nodes= 66,153

py
Co
(c)
ot (d)
tN
Fig. 4: FE meshes of (a) thin-walled rectangular part without selective mesh coarsening (b) thin-walled cylindrical part
without selective mesh coarsening (c) thin-walled rectangular part with selective mesh coarsening (d) thin-walled cylindrical
rip

part with selective mesh coarsening


sc

Laser heat source was modeled using universal cylindrical-involution-normal (CIN) [23] model.
nu

kKz ηL P
Q(r, z) = exp − (kr2 + Kz z) (9)
π(1 − exp(Kz s))
Ma

r = x2 + y2
ed

Where ηL is the laser efficiency, x, y, z are the local coordinates, KZ = 3/s represents heat source power exponent, s is
pt

the laser beam penetration depth, k = 3/r02 is beam focus coefficient and r0 is the radius of the laser beam. ηL , s, r0 were
selected as 45%, 0.5 mm, 2.5 mm respectively. (the parameters for other values are listed in Table 1)
ce

The area covered by the laser beam was considered when activating quiet elements (depositing materials) in thermal
analysis. Effect of forced convection and radiation was also considered as per the Equation 4 and Equation 5. A value of 100
W /m2 K was selected for convection coefficient.
Ac

3.2 Experimental Validation


Rectangular and cylindrical samples were built using the DMD AM process, following the same process parameters
used for the simulations. Experimental validation was done by comparing residual elastic strain and overall part distortion.
The neutron diffraction method was used to measure the residual elastic strain. These measurements were performed with the
High Flux Isotope Reactor (HFIR) at the 2nd Generation Neutron Residual Stress Facility (NRSF2) at Oak Ridge National
Laboratory. At each measurement point, d-spacing change in build directions was measured and the strain was calculated
by taking the ratio of d-space change/d-space in the stress-free material. The residual strain was taken only for the thin-wall
rectangular part along three lines as illustrated in Fig. 15.

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 01/29/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Journal of Manufacturing Science and Engineering. Received October 25, 2017;
Accepted manuscript posted January 24, 2018. doi:10.1115/1.4039092
Copyright (c)
3.32018 by ASME
Simulation of a large part

The discretization scheme used for the large part is illustrated in Fig. 5 and Table 6. The corresponding conventional
analysis was not performed for a comparison due to the enormous amount of simulation time that would have been required.
Each discretization step consists of the deposition of 4 subsequent layers. Six levels of selective mesh coarsening were
implemented as illustrated. The same process parameters used for the simulation of the small parts discussed in section 3.1
were used for the simulation of the large part as well.

d
ite
Element size Element size
0.5 mm, F1 (4 layers) 0.5 mm,
2 elements per 1 element per

ed
layer thickness
F2 (4 layers) layer thickness

C1

py
Max. element
size 1.5 mm F: fine
C: coarse

Co
C2 Max. element
size 2.5 mm

ot
tN
Max. element Element size
size 1.5 mm
C3 0.5 mm,
1 element per
rip

F3 (4 layers) layer thickness


sc
nu

Fig. 5: Selective mesh coarsening scheme for the large part


Ma
ed
pt
ce
Ac

Table 6: Application of selective mesh coarsening at each discretization step for the large part

Additive Newly added Mesh coarsening


step no. layers regions
1 1-4 F1
2 5-8 F1,F2
3 9-12 F1,F2,F3
4-10 13-40 F1,F2,C1,F3
11-75 41-300 F1, F2, C1, C2, C3, F3

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 01/29/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Journal of Manufacturing Science and Engineering. Received October 25, 2017;
Accepted manuscript posted January 24, 2018. doi:10.1115/1.4039092
Copyright (c) 2018 by ASME

d
ite
ed
Fig. 6: FE mesh of large part at last discretization step

py
Co
4 RESULTS AND DISCUSSION
4.1 Performance evaluation of the new framework using the simulation of small-sized parts
The computational efficiency of the new FE framework was determined by comparing the total computational time of

ot
the simulation based on the new FE framework with that of the conventional framework. Then, the residual stress and plastic
strain predicted by the simulation based on the new framework was compared with those predicted by the simulation based
tN
on the conventional framework. Results are first discussed for the rectangular part, followed by cylindrical part. The parts
being thin-walled, only the dominant components of stress and the corresponding plastic strain components were used for the
comparison. The dominant components of stress are in-plane normal stress components in the scanning and build directions.
rip

The predicted results after removing the build plate were also studied.
sc

4.1.1 Computational efficiency


The computational efficiency of the new framework is demonstrated first by comparing the increase in the total DoF of
the system between the two frameworks and then by comparing the total simulation time between them.
nu

The increase in the total DoF in the system as the simulation progresses is shown as a function of the number of layers
in the system in Fig. 7. In the conventional framework, the total DoF of the system increased linearly since the mesh density
Ma

was uniform and the part had a uniform cross-section in the build direction. On the contrary, with selective mesh coarsening,
the total DoF in the new framework remained relatively low as indicated by the very small slope of the DoF curve. Note
that the top four layers were discretized with the same fineness in all the discretized geometries used in the new framework.
Therefore, as seen in the results, the total DoF in the system up to 4th layer in the new framework and the conventional
ed

framework were equal. For the next few layers, a slight deviation can be seen due to moderate selective mesh coarsening
implemented in the bottom layers in the new framework. Subsequently, the deviation between the new framework and the
conventional framework progressively enlarged as the selective mesh coarsening increased in the new framework. The area
pt

under the curve, which is cumulative of the total DoF in the system, is a first cut approximation of the size of the problem.
Accordingly, the decrease in the size of the problem in terms of the cumulative of the total DoF was estimated for the
ce

thin-walled rectangular part and the cylindrical part as 1.5 million and 4.2 million, respectively.
A considerable saving in simulation time was achieved for the new framework (Table 7). The total simulation time shown
Ac

is the sum of the simulation times of the thermal analysis step, the structural analysis step and the analysis performed to
simulate removal of build plate, although it is very small. For the thin-walled rectangular part, the savings in total simulation
time was well over 50% (Table 7) ,with the total simulation time for the new framework and conventional framework being
6.5 and 14.0 hours, respectively, In the simulation of the thin-walled cylindrical part using the conventional method, the
analysis failed at the 20th layer due to a convergence issue caused by excessive element distortion. However, in the simulation
using the new framework, all 32 layers were simulated without any numerical error, which indicates the robustness of the
new framework. The total simulation time to complete 32 layers for the new framework was 26.7 hours compared to 45
hours to complete only 20 layers using the conventional framework (Table 7). In both the frameworks, since the structural
analysis step consumed significantly more simulation time compared to thermal analysis, a large portion of the savings in
the total simulation time was contributed from the structural analysis.

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 01/29/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Journal of Manufacturing Science and Engineering. Received October 25, 2017;
Accepted manuscript posted January 24, 2018. doi:10.1115/1.4039092
Copyright (c) 2018 by ASME

1,466,775
4,231,395

d
ite
ed
py
(a) (b)

Co
Fig. 7: The comparison of the variation of total DoF in the system with the addition of layers between new and conventional
frameworks (a) thin-walled rectangular part (b) cylindrical part

ot
tN
Table 7: Comparison of the total simulation time for thin-walled rectangular plate and cylinder
rip

Thin-walled Thin-walled
Framework rectangular plate cylinder
sc

(hr) (hr)
Thermal 5.4 15.2 **
nu

Conventional Structural 8.6 29.8 **


Total 14.0 45.0 **
Thermal 2.8 11.8
Ma

SMC Structural 3.7 15.8


Total 6.5 26.7
**Conventional model stopped at 20th layer due to a convergence issue
ed
pt
ce
Ac

The consequence of the progressively increasing total DoF of the system due to continued addition of layers in any
AM simulation is that the computational time for the simulation of the newly added layer increases. In Fig. 8, the increase
in simulation time with increase in number of layers are shown for the thermal and structural analysis steps in the new
framework. The simulation time per layer remained almost constant due to selective mesh coarsening. Typically, simulation
time is expected to increase with increase in the total DoF. However, a small decrease in computational time immediately
after the introduction of a new discretized model with a relatively larger total DoF can be seen at a few locations. For
example, the computational time for the simulation of the printing of 5th layer in 2nd discretization step was lower than the
4th layer in 1st discretization step. This can be attributed to the dependence of computational time on other factors, such as
the number of iterations it takes to achieve convergence. The convergence may vary with local mesh density, especially in
the case of simulations with temperature dependent material properties and plastic behavior.

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 01/29/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Journal of Manufacturing Science and Engineering. Received October 25, 2017;
Accepted manuscript posted January 24, 2018. doi:10.1115/1.4039092
Copyright (c) 2018 by ASME

d
ite
(a) (b)

ed
Fig. 8: The comparison of the simulation time required for each layer between new and conventional frameworks (a) thin-

py
walled rectangular part (b) cylindrical part

Co
4.1.2 Stress and plastic strain distribution comparison for rectangular part
The distribution of residual stress after the part cooled down to room temperature is shown in Fig. 9. The normal
component in the scan direction σxx and in the build direction σzz are shown in Fig. 9(a) and Fig. 9(b), respectively. The

ot
distribution of the normal stress components shows a region near at the top with large tensile stress in the scan direction and
nearly zero stress in the build direction. The tensile stress in the scan direction decreased gradually as it got closer to the build
tN
plate. The decrease was very large toward the bottom region of the part, though it remained tensile. The stress component
in the build direction was compressive in the lower region of the part near the interface with the build plate. The gradients
of both stress components were high near the interface due to the higher constraint to deformation at the interface. The
rip

comparison of the predicted stress fields by the simulations based on the new framework and the conventional framework
showed excellent agreement. The transition of the stress component in the scan direction from a higher value in the top
region to a lower value near the interface was captured fairly well by the new framework. High compressive stress in the
sc

build direction near the interface was also well captured by the new framework, although a slight increase in the compressive
stress can be seen near the interface between the part and the build plate.
Once the part has been built, it will be removed from the build plate which redistributes some of the stress. Redistribution
nu

of stress as a result of the removal of the build plate is shown in Fig. 10. The low tensile stress in the scan direction observed
at the very top region of the part became compressive. The region immediately below the very top regions where a high
Ma

tensile stress in the scan direction was observed decreased substantially. Instead of a single high tensile region, two regions
with high tensile stress had formed on both sides of this top region as shown in the figure. Right below these two tensile
regions, regions of high compressive stress had also formed. In addition, two high tensile stress regions near the interface
moved from the sides toward the middle as shown in the figure. Similar to the stress in the scan direction, the stress in the
ed

build direction also changed substantially upon the removal of the build plate. The high compressive region at the middle
redistributed to two regions more toward the sides of the part as shown in Fig. 10(b). The comparison of stress distributions
both in the scan and build directions show that the predicted results by the simulation based on the new framework are in
pt

very good overall agreement with those predicted by the simulations based on the conventional framework.
The distribution of the in-plane normal components of plastic strain is shown for the part after the removal of the build
ce

plate in Fig. 11. Since the removal of the build plate does not affect the plastic strain distribution, the plastic strain distribution
shown is the same for the part before the removal of the build plate as well. Overall, a uniform tensile in-plane plastic strain is
Ac

developed in the scanning direction in the middle region of the part. However, a large gradient in the plastic strain distribution
was seen toward the sides: changing from tensile in the interior to compressive toward the vertical edges. The component of
plastic strain in the build direction also showed a uniform distribution in the middle region of the part, but with a compressive
strain. A gradient in plastic strain can also be observed for the strain component in the build direction near the vertical edges.
However, it varies from the uniform compressive strain in the middle to a large tensile strain toward the vertical edges. In
addition to the gradient near the vertical edges, the component of plastic strain in the build direction has compressive strain
accumulated near the top and bottom edges. The comparison of plastic strain distribution shows that the simulation based on
the new framework agrees quite well with that predicted by the simulation based on the conventional framework, especially
in the middle region where the plastic strain distribution is uniform. The predicted results also agree fairly well near the
edges where the gradient is very high. Minor differences in the results between the conventional framework and the new

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 01/29/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Journal of Manufacturing Science and Engineering. Received October 25, 2017;
Accepted manuscript posted January 24, 2018. doi:10.1115/1.4039092
Copyright (c) 2018 by ASME

σ xx

Conventional New
(a)

σ zz

d
ite
ed
Conventional (b) New

py
-400 -300 -200 -100 0 100 200 300 400 500 600 MPa

Co
Fig. 9: The comparison of the residual stress distributions between simulations based on conventional and new frameworks
in thin-walled rectangular plate before the removal of the build plate; (a) along the scan direction, (σxx ), (b) along the build

ot
direction, (σzz )
tN
rip

σ xx
sc

Conventional New
nu

(a)
Ma

σ zz
ed

Conventional New
(b)
pt

-200 -155 -110 -65 -20 25 70 115 160 205 250 MPa
ce

Fig. 10: The comparison of the residual stress distributions between simulations based on conventional and new frameworks
Ac

in thin-walled rectangular plate after the removal of the build plate; (a) along the scan direction, (σxx ), (b) along the build
direction, (σzz )

framework are mainly due to the difference in the mesh as a result of the mesh coarsening. Better accuracy can be achieved
by increasing the mesh density and by increasing the number of layers per discretization step, but at the cost of increased
overall computational time.

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 01/29/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Journal of Manufacturing Science and Engineering. Received October 25, 2017;
Accepted manuscript posted January 24, 2018. doi:10.1115/1.4039092
Copyright (c) 2018 by ASME

p
ϵ xx

Conventional New
(a)
p
ϵzz

d
ite
Conventional New
(b)

ed
-50 -35 -20 -5 10 25 40 55 70 85 100 x 10-3

py
Fig. 11: The comparison of the plastic strain distributions between simulations based on conventional and new frameworks
in thin-walled rectangular plate before the removal of the build plate; (a) along the scan direction, (εPxx ), (b) along the build

Co
direction, (εPzz )

ot
4.1.3 Stress and plastic strain distribution comparison for cylindrical part
In the cylindrical part, stress and plastic strain have an axisymmetric distribution. The choice of arbitrary starting
tN
locations for the heat source in individual layers does not have any significant effect. Therefore, the results are shown on a
cross-section of the thin wall.
The residual stress distribution in the hoop direction and the build direction after the part cooled down to room temper-
rip

ature is shown in Fig. 12. Residual stress in the hoop direction was tensile in a region close to the top of the part, with a
positive stress gradient towards the inner wall. Closer to the build plate, the outer region of the part showed a tensile stress
sc

and the inner region showed a compressive stress. Hoop residual stress is mostly a low tensile stress in the middle of the
part. The predicted residual stress distribution in the hoop direction by the simulation based on the new framework agreed
quite well with the overall behavior predicted by the simulation based on the conventional framework. The residual stress
nu

in the build direction was mostly a low uniform tensile stress in the top region of the part. Near the interface, with the build
plate, a large gradient in the stress distribution was found in the thickness direction of the wall. The stress was mostly com-
pressive near the inner wall and mostly tensile near the outer wall. The simulation based on the new framework predicted
Ma

the distribution of stress in the build direction quite accurately in the whole part, even near the interface where the gradient
of stress distribution was very high.
The removal of the build plate introduced redistribution of the residual stress field in order to maintain static equilibrium.
ed

The residual stress distribution after the removal of the build plate is shown in Fig. 13. The hoop stress near the top edge
changed from a relatively small tensile stress to a compressive stress. In the middle region of the part, the hoop-stress, in
general, showed a transition from a low tensile stress to a low compressive stress. Closer to the build plate, the hoop stress
pt

remained tensile in the outer region, but with a relatively lower magnitude. The low compressive stress in the inner wall
region transitioned into a low tensile stress. The change in the stress distribution due to the removal of the build plate was
ce

relatively more dominant in the build direction. A large stress release was observed near the part to the build-plate interface,
with the two large regions of compressive and tensile stress near the part vanishing almost completely, leaving only two very
Ac

small regions of concentrated tensile and compressive stress. The comparison of stress distribution after the removal of the
build plate, for both stress components, showed very good agreement between the simulations based on the new framework
and the conventional framework.
The plastic strain distribution showed that the plastic strain component in the hoop direction is overall in a tensile state,
with higher values in the regions close to the top and bottom of the part (Fig. 14). The tensile stress is highest in the bottom
region towards the inner wall of the part. In the middle region, the plastic strain in the hoop direction is close to zero. The
plastic strain in the build direction is very low and tensile in general, except in the region near the bottom edge, close to the
inner wall where the significant compressive strain was formed. There is also a region with large tensile stress near the outer
edge. The predicted plastic strain distribution agrees quite well between the simulations based on the new framework and
the conventional framework.

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 01/29/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Journal of Manufacturing Science and Engineering. Received October 25, 2017;
Accepted manuscript posted January 24, 2018. doi:10.1115/1.4039092
Copyright (c) 2018 by ASME
σ θθ σ zz

Outer Inner

Conventional New Conventional New

Z
(a) (b)

d
R
θ

ite
-600 -480 -360 -240 -120 0 120 240 360 480 600 MPa

ed
Fig. 12: The comparison of the residual stress distributions between simulations based on conventional and new frameworks
in thin-walled cylindrical part before the removal of the build plate; (a) along the circumference (hoop) direction, (σθθ ), (b)
along the build direction, (σzz )

py
σ θθ σ zz

Co
ot
tN
rip

Conventional New Conventional New


Z
(a) (b)
sc

R
θ
-300 -240 -180 -120 -60 0 60 120 180 240 300 MPa
nu

Fig. 13: The comparison of the residual stress distributions between simulations based on conventional and new frameworks
in thin-walled cylindrical part after the removal of the build plate; (a) along the circumference (hoop) direction, (σθθ ), (b)
Ma

along the build direction, (σzz )

p p
ed

ϵθ θ ϵzz
pt
ce
Ac

Conventional New Conventional New


Z (a) (b)
R
θ

-40 -32 -24 16 -8 0 8 16 24 32 40 x 10-3


Fig. 14: The comparison of the plastic strain distributions between simulations based on conventional and new frameworks
in thin-walled cylindrical part before the removal of the build plate; (a) along the circumference (hoop) direction, (εPθθ ), (b)
along the build direction, (εPzz )

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 01/29/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Journal of Manufacturing Science and Engineering. Received October 25, 2017;
Accepted manuscript posted January 24, 2018. doi:10.1115/1.4039092
Copyright (c)
4.22018 by ASME validation
Experimental
The elastic strain distribution predicted in the thin-walled rectangular part after its removal from the build plate was com-
pared with the experimental results obtained from neutron diffraction measurements (Fig. 15). The elastic strain component
in the build direction was measured along two vertical paths (B1 and B3) and one horizontal path (B2) as shown.
The experimental results showed a large tensile strain near the interface along the vertical path B1. The large tensile
strain decreased to a small value as the distance from the build plate increased, and then became a large compressive strain.
The compressive strain appeared to be decreasing toward the top region. The experimental result agreed well with predictions
from both the simulations, especially in the middle region, However, the high tensile strain close to the interface and the high
compressive strain close to the top region were not accurately predicted.
The experimental results along the longitudinal direction (line B2) showed a high tensile elastic strain close to the edges
of the part while going toward the center of the part it showed rapid fluctuations between compressive and tensile states. The

d
simulations based on both the conventional and the new frameworks predicted similar behavior along line B2. In comparison
with the experimental results, the predicted results followed similar trend, although the predictions showed a relatively lower

ite
level of fluctuations in the interior region.
The elastic strain measurements along the path B3 showed large tensile strain in the middle region, then decreasing to

ed
very low values in the regions close to interface and the top free surface. As in the case of paths B1 and B2, the predictions
of both simulations for path B3 also showed very good agreement with the trend exhibited by the measurements.
Overall, the predicted results showed fairly good agreement with the trends shown by the experimental measurements

py
for all three different paths. Although the results also agreed quantitatively in some regions, there were other regions where
the difference between the predicted and experimental results was significant. This disagreement could be attributed to the

Co
selection of gage volume based on standard test data obtained from conventionally manufactured material samples. The
microstructure of the same material manufactured with AM could be quite different, requiring a very different size for the
gage volume. It is important to note that the predictions by both the simulations agreed extremely well between them.

ot
tN
rip
sc
nu
Ma
ed
pt
ce

(+)
Ac

B2

B1
(+)

(+)

B3

Fig. 15: The comparison of elastic strain (εe zz ) predicted by the new and conventional frameworks for the rectangular
part after the removal of the build plate with those determined experimentally by neutron diffraction measurements. The
measurements were made along paths B1, B2 and B3 as shown

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 01/29/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Journal of Manufacturing Science and Engineering. Received October 25, 2017;
Accepted manuscript posted January 24, 2018. doi:10.1115/1.4039092
Copyright (c) 2018
The by ASME of part distortions is illustrated for the rectangular and the cylindrical parts in Fig. 16. Experimental
comparison
measurements were taken at several locations on each part using a digital caliper. Overall, the predicted distortions agreed
very well with the measured values for both the new and the conventional frameworks. The distortion predicted at the middle
of the vertical edges of the part was slightly lower than that measured. The maximum error observed at the center of the
rectangular sample is ≈ 4.0%. In the thin-wall cylindrical part, the prediction by the simulation based on the new framework
was slightly higher than the experimental results, especially at the center of the horizontal edge of the part (Fig. 16(b)).
The maximum error observed at the center of the cylindrical sample is ≈ 18.0 %. Apart from those minor disagreements,
the predictions were fairly close to the experimental results. Note that the predictions based on the conventional framework
were not available for the thin-walled cylindrical part since the simulation failed after building 20 layers. Although the
repeated failures of the simulation based on conventional framework for relatively large parts due to numerical issues were
not investigated further, the fact that the simulation based on the new framework for the same parts under exactly same

d
conditions except for the difference in discretization never encountered such an issue indicates the robustness of the new
framework owing to the smaller size of the system.

ite
ed
Un-deformed

py
Conventional
New
Experimental

Co
Outer Inner

ot
tN
rip

(a) (b)
sc

Fig. 16: The comparison of the part distortion predicted by the new and conventional frameworks for the rectangular and
cylindrical part after the removal of the build plate with those determined experimentally; (a) rectangular part, (b) cylindrical
nu

part
Ma
ed
pt
ce
Ac

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 01/29/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Journal of Manufacturing Science and Engineering. Received October 25, 2017;
Accepted manuscript posted January 24, 2018. doi:10.1115/1.4039092
Copyright (c)
4.32018 by ASME
Simulation of the large-scale model
The change in total DoF of the system with the addition of more layers is shown for the large-scale part in Fig. 17.
While the DoF for the simulation based on the conventional framework increased rapidly with increase in the number of
layers, the total DoF remained almost constant throughout for the simulation based on the new framework. The simulation
time for completion of each discretization step is shown in Fig. 18. The plot shows that the simulation time, consistent with
the total DoF, also remained constant throughout the analysis. The variation of simulation time for the new framework is
consistent with the variation of DoFs (the variation of DoF is not clearly visible in Fig. 17 due to rapid growth in DoFs for
the conventional framework, but can be validated by changing the scale of the graph).

d
ite
ed
Total DoF

py
reduction:
56,015,253

Co
ot
tN
rip

Fig. 17: The comparison of the variation of total DoF in the large part with the addition of layers between new and conven-
tional frameworks
sc
nu
Ma
ed

Simulation time (hr)


pt

Thermal Structural Total


ce

32.6 81.4 113.6


Ac

Fig. 18: The variation of the simulation time required for each layer in the simulation of the large part based on new
framework

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 01/29/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Journal of Manufacturing Science and Engineering. Received October 25, 2017;
Accepted manuscript posted January 24, 2018. doi:10.1115/1.4039092
Copyright (c) 2018
The by ASME of residual stress component along the build direction, σ , is shown in Fig. 19 as an example result for
distribution zz
the large part. According to the distribution, the variation of σzz along the length is not very significant, but a large variation
from a compressive stress on the inside surface of a tensile stress on the outside surface can be seen.

d
ite
ed
py
-600 -480 -360 -240 -120 0 120 240 360 480 600 MPa

Co
Fig. 19: The distribution of the residual stress component in the build direction, σzz predicted by the simulation based on
new framework for the large part

ot
5 CONCLUSIONS
tN
In this article, the development of a new computationally efficient FE framework for problems involving a continuous
change in the model geometry is reported. The new framework allows introduction of updated discretized geometries with
selectively coarsened mesh at regular intervals during the simulation process. At every discretization step, solutions were
rip

mapped from the previous step to the new step to use it as the initial state for the new step. Similarly, the solution from every
step was also mapped to a reference geometry of the full part with uniform fine discretization. The basic premise behind
the new framework was that because the gradient of the incremental solution for a region far away from the heat source is
sc

small, it is possible to introduce an updated discretized geometry which allows greater flexibility to coarsen the mesh in this
region, thus maintaining total degrees of freedom of the system and accuracy of the solution within a desirable range. As a
consequence, the computational cost of the simulation decreases substantially.
nu

The new framework was evaluated by simulating DMD of a thin-walled rectangular and a thin-walled cylindrical part and
then comparing the computational speed and the predicted results with those predicted by simulations using the conventional
Ma

framework. The mesh coarsening allowed the total DoF of the system to remain almost a constant when more and more
layers were added, in contrast to the linear growth of DoF in the simulation based on the conventional framework. This
resulted in a substantial reduction in the cumulative of the total DoF of the system for the whole simulation, resulting in
large savings in computational time. While the saving in computational time was about 50% for the rectangular part, the
ed

cylindrical part with more volume showed much larger savings.


The comparison of stress and plastic strain distributions both in the scan and build directions, both with and without
build plate, showed that the predicted results by the simulation based on the new framework were in very good agreement
pt

with those predicted by the simulations based on the conventional framework.


The new framework was then validated by comparing the predicted residual elastic strain with that measured experi-
ce

mentally by neutron diffraction measurements of the thin-walled rectangular part. The elastic strain was measured using a
neutron diffraction technique along three characteristic lines of the thin-walled part. The corresponding prediction by the
Ac

new framework showed fairly good agreement qualitatively. In addition, the prediction of the measurement of overall part
distortion of both parts was also in good agreement.
Finally, the DMD of a one-half foot thin-walled cylindrical part was simulated using the new framework to demonstrate
its capability to simulate large-scale AM parts using a realistic simulation time. The comparison of the total DoF of the
system as a function of the total number of layers showed that it remained almost a constant even during the addition of
the last (300th ) layer for such a large part. A remarkable reduction of about 56 million in the cumulative of the total DoF
for the whole simulation was achieved. The tracking of the simulation time required for each layer showed that it remained
almost a constant throughout the simulation, which is consistent with the total DoF of the system. The demonstration of the
simulation of the large part reveals the remarkable ability of the new framework for simulating relatively large parts using
realistic computational time. In conclusion, the new FE framework with selective mesh coarsening has greater flexibility to

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 01/29/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Journal of Manufacturing Science and Engineering. Received October 25, 2017;
Accepted manuscript posted January 24, 2018. doi:10.1115/1.4039092
Copyright (c) 2018 by ASME
substantially decrease the total DoF of the system for a typical AM process simulation and achieve significant simulation
time-saving in contrast to the conventional framework and the frameworks based on an element merging technique.

Acknowledgment
This research was funded in part by a grant from the New York State Energy Research and Development Authority
(NYSERDA). The authors would like to thank Dr. Jeffery Bunn (Oak Ridge National Laboratory) for the neutron diffraction
measurements.

References

d
[1] Gibson, I., Rosen, D., and Stucker, B., 2015. Additive Manufacturing Technologies.
[2] Gibson, I., Rosen, D. W., and Stucker, B., 2010. Additive manufacturing technologies: Rapid prototyping to direct

ite
digital manufacturing.
[3] Sames, W. J., List, F. A., Pannala, S., Dehoff, R. R., and Babu, S. S., 2016. “The metallurgy and processing science of

ed
metal additive manufacturing”. International Materials Reviews, 6608(March), pp. 1–46.
[4] Gu, D. D., Meiners, W., Wissenbach, K., and Poprawe, R., 2012. “Laser additive manufacturing of metallic compo-
nents: materials, processes and mechanisms”. International Materials Reviews, 57(3), pp. 133–164.

py
[5] Oropallo, W., and Piegl, L. A., 2016. “Ten challenges in 3D printing”. Engineering with Computers, 32(1), pp. 135–
148.
[6] Frazier, W. E., 2014. “Metal additive manufacturing: A review”. Journal of Materials Engineering and Performance,

Co
23(6), pp. 1917–1928.
[7] Atkinson, H., and Davies, S., 2000. “Fundamental aspects of hot isostatic pressing: an overview”. Metallurgical and
Materials Transactions A, 31A(December), pp. 2981–3000.

ot
[8] Denlinger, E. R., Heigel, J. C., Michaleris, P., and Palmer, T., 2015. “Effect of inter-layer dwell time on distortion and
residual stress in additive manufacturing of titanium and nickel alloys”. Journal of Materials Processing Technology,
tN
215, jan, pp. 123–131.
[9] Bikas, H., Stavropoulos, P., and Chryssolouris, G., 2016. “Additive manufacturing methods and modelling approaches:
a critical review”. The International Journal of Advanced Manufacturing Technology, 83(1), pp. 389–405.
rip

[10] Megahed, M., Mindt, H.-W., N’Dri, N., Duan, H., and Desmaison, O., 2016. “Metal additive-manufacturing process
and residual stress modeling”. Integrating Materials and Manufacturing Innovation, 5(1), p. 4.
[11] Denlinger, E. R., Heigel, J. C., and Michaleris, P., 2014. “Residual stress and distortion modeling of electron beam
sc

direct manufacturing Ti-6Al-4V”. Proceedings of the Institution of Mechanical Engineers, Part B: Journal of Engi-
neering Manufacture, 229(10), pp. 0954405414539494–.
[12] Denlinger, E. R., Irwin, J., and Michaleris, P., 2014. “Thermomechanical Modeling of Additive Manufacturing Large
nu

Parts”. Journal of Manufacturing Science and Engineering, 136(6), p. 061007.


[13] Hodge, N. E., Ferencz, R. M., and Solberg, J. M., 2014. “Implementation of a thermomechanical model for the
Ma

simulation of selective laser melting”. Computational Mechanics, 54(1), pp. 33–51.


[14] Hussein, A., Hao, L., Yan, C., and Everson, R., 2013. “Finite element simulation of the temperature and stress fields in
single layers built without-support in selective laser melting”. Materials and Design, 52, pp. 638–647.
[15] Li, C., Liu, J. F., and Guo, Y. B., 2016. “Prediction of Residual Stress and Part Distortion in Selective Laser Melting”.
ed

Procedia CIRP, 45, pp. 171–174.


[16] Mukherjee, T., Zhang, W., and DebRoy, T., 2017. “An improved prediction of residual stresses and distortion in additive
manufacturing”. Computational Materials Science, 126, jan, pp. 360–372.
pt

[17] Papadakis, L., Loizou, A., Risse, J., and Schrage, J., 2014. “Numerical computation of component shape distortion
manufactured by Selective Laser Melting”. Procedia CIRP, 18, pp. 90–95.
ce

[18] Zaeh, M. F., and Branner, G., 2010. “Investigations on residual stresses and deformations in selective laser melting”.
Production Engineering, 4(1), pp. 35–45.
Ac

[19] Ding, J., Colegrove, P., Mehnen, J., Ganguly, S., Almeida, P. M. S., Wang, F., and Williams, S., 2011. “Thermo-
mechanical analysis of Wire and Arc Additive Layer Manufacturing process on large multi-layer parts”. Computational
Materials Science, 50(12), jul, pp. 3315–3322.
[20] Amine, T., Newkirk, J. W., and Liou, F., 2014. “An investigation of the effect of direct metal deposition parameters on
the characteristics of the deposited layers”. Case Studies in Thermal Engineering, 3, pp. 21–34.
[21] Michaleris, P., 2014. “Modeling metal deposition in heat transfer analyses of additive manufacturing processes”. Finite
Elements in Analysis and Design, 86, pp. 51–60.
[22] abaqus, 2014. ABAQUS user’s manual : version 6.14. Pawtucket, RI : ABAQUS.
[23] Kubiak, M., Piekarska, W., and Parkitny, R., 2015. “Theoretical study on thermal and structural phenomena in thin
elements heated by a laser beam”. Archives of Mechanics, 67(1), pp. 3–24.

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 01/29/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Journal of Manufacturing Science and Engineering. Received October 25, 2017;
Accepted manuscript posted January 24, 2018. doi:10.1115/1.4039092
Copyright (c)
List2018 by ASME
of Figures
1 Schematic illustrating the conventional framework for two-time steps . . . . . . . . . . . . . . . . . . . . 4
2 Schematic illustrating the new framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3 Selective mesh coarsening scheme for small parts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
4 FE meshes of (a) thin-walled rectangular part without selective mesh coarsening (b) thin-walled cylindrical
part without selective mesh coarsening (c) thin-walled rectangular part with selective mesh coarsening (d)
thin-walled cylindrical part with selective mesh coarsening . . . . . . . . . . . . . . . . . . . . . . . . . . 10
5 Selective mesh coarsening scheme for the large part . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
6 FE mesh of large part at last discretization step . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
7 The comparison of the variation of total DoF in the system with the addition of layers between new and
conventional frameworks (a) thin-walled rectangular part (b) cylindrical part . . . . . . . . . . . . . . . . . 13

d
8 The comparison of the simulation time required for each layer between new and conventional frameworks
(a) thin-walled rectangular part (b) cylindrical part . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

ite
9 The comparison of the residual stress distributions between simulations based on conventional and new
frameworks in thin-walled rectangular plate before the removal of the build plate; (a) along the scan direction,

ed
(σxx ), (b) along the build direction, (σzz ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
10 The comparison of the residual stress distributions between simulations based on conventional and new
frameworks in thin-walled rectangular plate after the removal of the build plate; (a) along the scan direction,

py
(σxx ), (b) along the build direction, (σzz ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
11 The comparison of the plastic strain distributions between simulations based on conventional and new frame-
works in thin-walled rectangular plate before the removal of the build plate; (a) along the scan direction, (εPxx ),

Co
(b) along the build direction, (εPzz ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
12 The comparison of the residual stress distributions between simulations based on conventional and new
frameworks in thin-walled cylindrical part before the removal of the build plate; (a) along the circumference

ot
(hoop) direction, (σθθ ), (b) along the build direction, (σzz ) . . . . . . . . . . . . . . . . . . . . . . . . . . 17
13 The comparison of the residual stress distributions between simulations based on conventional and new
tN
frameworks in thin-walled cylindrical part after the removal of the build plate; (a) along the circumference
(hoop) direction, (σθθ ), (b) along the build direction, (σzz ) . . . . . . . . . . . . . . . . . . . . . . . . . . 17
14 The comparison of the plastic strain distributions between simulations based on conventional and new frame-
rip

works in thin-walled cylindrical part before the removal of the build plate; (a) along the circumference (hoop)
direction, (εPθθ ), (b) along the build direction, (εPzz ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
15 The comparison of elastic strain (εe zz ) predicted by the new and conventional frameworks for the rectan-
sc

gular part after the removal of the build plate with those determined experimentally by neutron diffraction
measurements. The measurements were made along paths B1, B2 and B3 as shown . . . . . . . . . . . . . 18
16 The comparison of the part distortion predicted by the new and conventional frameworks for the rectangular
nu

and cylindrical part after the removal of the build plate with those determined experimentally; (a) rectangular
part, (b) cylindrical part . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Ma

17 The comparison of the variation of total DoF in the large part with the addition of layers between new and
conventional frameworks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
18 The variation of the simulation time required for each layer in the simulation of the large part based on new
framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
ed

19 The distribution of the residual stress component in the build direction, σzz predicted by the simulation based
on new framework for the large part . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
pt
ce
Ac

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 01/29/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Journal of Manufacturing Science and Engineering. Received October 25, 2017;
Accepted manuscript posted January 24, 2018. doi:10.1115/1.4039092
Copyright (c)
List2018 by ASME
of Tables
1 DMD parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2 Thermal conductivity of bulk material [14] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3 Enthalpy change with time [14] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
4 Temperature depended mechanical properties [14] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
5 Application of selective mesh coarsening at each discretization step for small parts . . . . . . . . . . . . . 9
6 Application of selective mesh coarsening at each discretization step for the large part . . . . . . . . . . . . 11
7 Comparison of the total simulation time for thin-walled rectangular plate and cylinder . . . . . . . . . . . . 13

d
ite
ed
py
Co
ot
tN
rip
sc
nu
Ma
ed
pt
ce
Ac

DownloadedViewFrom:
publicationhttp://manufacturingscience.asmedigitalcollection.asme.org/
stats on 01/29/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

S-ar putea să vă placă și