You are on page 1of 10

Finite Elements in Analysis and Design 86 (2014) 5160

Contents lists available at ScienceDirect

Finite Elements in Analysis and Design


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

Modeling metal deposition in heat transfer analyses of additive


manufacturing processes
Panagiotis Michaleris
Department of Mechanical and Nuclear Engineering, The Pennsylvania State University, Pan Computing LLC, United States

art ic l e i nf o

a b s t r a c t

Article history:
Received 5 July 2013
Received in revised form
27 March 2014
Accepted 1 April 2014
Available online 28 April 2014

Additive Manufacturing (AM) processes for metallic parts using both laser and electron beam heat
sources are becoming increasingly popular due to their potential of producing near net shape structural
components. The thermal history generated by additive manufacturing is essential in determining the
resulting microstructure, material properties, residual stress, and distortion.
In this work nite element techniques for modeling metal deposition heat transfer analyses of
additive manufacturing are investigated in detail. In particular, both quiet and inactive element
activation are reviewed in detail and techniques for minimizing errors associated with element
activation errors are proposed. 1D and 3D numerical examples are used to demonstrate that both
methods can give equivalent results if implemented properly. It is also shown that neglecting surface
convection and radiation on the continuously evolving interface between active and inactive elements
can lead to errors. A new hybrid quiet inactive metal deposition method is also proposed to accelerate
computer run times.
& 2014 Elsevier B.V. All rights reserved.

Keywords:
Additive manufacturing
Metal deposition
Element activation
Heat transfer

1. Introduction
Additive Manufacturing (AM) processes for metallic parts using
both laser and electron beam heat sources have the potential of
producing near net shape structural components over a range of
sizes (mm3 through m3). Applications range from customized
medical implants to aerospace components. Both laser and electron beam based deposition systems have been utilized to deposit
a range of materials. All Additive Manufacturing processes are
similar in that a three dimensional part represented by a CAD le
is sliced into layers (build plan) which in turn dene scan
trajectories of the heat source. A high power energy source (laser
or electron beam) is used to heat and melt metal powder or wire,
which solidies to form a fully dense layer. The powder may be fed
to the heat source through nozzles or may be raked into at layers
in powder bed systems. The addition of multiple layers can
produce a three dimensional fully dense part.
Signicant research has been devoted over the past ten years
into investigating the effects of processing parameters on the
resulting microstructure during additive manufacturing [17].
Another concern in additive manufacturing is distortion and
residual stress [813]. The thermal history generated by additive
manufacturing is essential in determining the resulting microstructure, material properties, residual stress, and distortion.
Modeling the thermal history of the additive manufacturing
process is similar to modeling multi-pass welding [4,1416].
Thermo-mechanical modeling of welding has been an active research
http://dx.doi.org/10.1016/j.nel.2014.04.003
0168-874X/& 2014 Elsevier B.V. All rights reserved.

area since the late 1970s [1719]. A detailed review of nite element
modeling for welding is available in references [2027]. Typically,
transient heat conduction analyses are performed in Lagrangian
reference frames as opposed to Eulerian transport analyses which
most often used to predict the temperature eld and physical shape
of the melt pool [2832], buoyancy, surface tension, and magnetohydro-dynamic effects [33], or powder melt pool interaction [34].
Building a complex part with additive manufacturing may require
depositing hundreds or thousands of layers of material which compared to multi-pass welding introduces signicant computational cost.
Therefore, computational efciency becomes paramount. Notable
work in the thermal and microstructural modeling in additive
manufacturing is available in Refs. [15,7]. Residual stress and deformation modeling is investigated in Refs. [1416,35,36].
The material deposition in additive manufacturing is modeled by
using inactive or quiet elements which are activated as the added
material (powder or wire) solidies. Two metal deposition methods
are reported for modeling material deposition: (1) the use of quiet
or (2) inactive elements [23,37,38]. In the quiet approach, the
elements are present in the analysis but are assigned properties
so they do not affect the analysis. In the inactive element approach,
elements are not included in the analysis until the corresponding
material has been added. A variety of general purpose commercial
codes have been reportedly used to model metal deposition: Zhu
et al. [39] use Ansys et al. [40,41] use Adina et al. [42] and Ye et al.
[43] use Abaqus et al. [44] use Comsol, and Lundback and Lindgren
[14] use Marc. However, the numerical approach implemented

52

P. Michaleris / Finite Elements in Analysis and Design 86 (2014) 5160

during element activation and potential sources of errors or


numerical efciency are not investigated in these references. In
welding, the size ller metal is usually small compared to the base
metal, and errors introduced by element activation may be negligible. However, in additive manufacturing, the size of the deposited
material is signicantly larger and thus element activation errors
could be signicant.
The objective of this work is to investigate nite element
techniques for modeling metal deposition in heat transfer analyses
of additive manufacturing and demonstrate potential sources of
error. In particular, both quiet and inactive element activation are
reviewed in detail and techniques for minimizing errors associated
with element activation errors are proposed. 1D and 3D numerical
examples are used to demonstrate that both methods can give
equivalent results if implemented properly. Is also shown that
neglecting surface convection and radiation on the continuously
evolving interface between active and inactive elements can lead
to errors. A new hybrid quiet inactive metal deposition method is
also proposed to accelerate computer run times.

qp x; t h T  T 1 sT 4  T 41

where, h is the convection coefcient, and T 1 is the room


temperature. is the emissivity and s is the StefanBoltzmann
constant.
The energy ux q is expressed as a function of temperature T
using the nonlinear isotropic Fourier heat ux constitutive relation:
q  kTT

dH dH dT
dT

C p
dt
dT dt
dt

This section provides a brief summary of the nite element


formulation for transient conductive heat transfer. For more
details see Ref. [45]. Heat transfer through mass transport in the
melt pool is not directly simulated in this study. Its effects are
introduced into the simulation by a distributed heat input model
[46]. Enforcing energy balance in a Lagrangian reference frame x
for a domain of volume V (see Fig. 1) results into the following
partial differential equation:
dH
x; t   qx; t 0
dt
1

where x is spatial coordinate and t is time. T is the temperature, q


is the heat ux vector, Q is the body heat source, and H is the
enthalpy.
The initial temperature eld is given by
Tx; 0 t 0 T x

Q x; t  C p

dT
 kTT 0
dt
8

Using an implicit formulation the temporal derivatives at time


t are approximated by the backward nite difference:

dn T n T  n  1 T
Cn
dn t
t  n  1t

Aq element

dR

dn T

on the surface AT

qs x; t qp x; t

on the surface Aq

where T p x; t and qp x; t represent the prescribed temperature


and temperature-dependent surface ux, respectively. For surface
convection and radiation, qp x; t is dened as follows:

#
1
B TN  N
N N N C p n
B kB B
dV
T
T
t  n  1t
V element
"
#
Z
C p n T  n  1 T
N
NT N

dV
T n t  n  1 t
V element
Z
q
11
NT N dA

T
Aq element

where 0 T x is the prescribed initial temperature. The following


boundary conditions are applied on the surface,
Tx; t T p x; t

where, n T and n  1 T are the temperatures at times n t and n  1 t ,


respectively.
Using the Galerkin nite element discretization and the NewtonRaphson solution scheme, Eqs. (8) and (4) result into the
following element residual R and Jacobian dR=dn T:
(
)
Z
n
T  n  1T
BT kBn T  NT Q NT NC p n
R
dV
t  n  1t
V element
Z
NT qp dA
10

in the entire volume V

where is the density of the owing body, Cp is the specic heat.


Substitution of Eqs. (6) and (7) into Eq. (1) results into the
following:

in the entire volume V of the material

where, k is the thermal conductivity. The rate of the enthalpy can


also be rewritten as

in the entire volume V of the material

2. Transient conductive heat transfer of additive


manufacturing

Q x; t 

"

T k n

T Q

where, T is the element temperature nodal vector, and N and B are


the operators that compute the temperature and temperature
gradient as follows:
T NT

12

T BT

13

3. Material deposition modeling

Aq

Two metal deposition methods are reported in the literature for


modeling material deposition: (1) the quiet and (2) the inactive
element method [14,23,37,38]. In this section the methods are
described in further detail.
3.1. Quiet element method

AT
Fig. 1. A body with volume V, prescribed temperature on surface AT, and prescribed
surface ux on surface Aq.

In the quiet element method, the elements representing metal


deposition regions are present from the start of the analysis.
However, they are assigned properties so they do not affect the

P. Michaleris / Finite Elements in Analysis and Design 86 (2014) 5160

analysis. For heat transfer analyses, the thermal conductivity k is


set to a lower value to minimize conduction into the quiet
elements, and the specic heat Cp is set to a lower value to adjust
energy transfer to the quiet elements:
kquiet sk k

14

C p quiet sC p C p

15

where, kquiet and C p quiet are the thermal conductivity and specic
heat used for quiet elements, and sk and sC p are the scaling factors
used for the thermal conductivity and specic heat, respectively.
The quiet element method has the following advantages:

53

T2

T1

T3

Fig. 2. Illustration of interpolated temperature during element activation in 2 node


1D elements.

 It is easy to implement and can be applied to commercial nite


element codes via user subroutines.

 Since the number of elements does not change, the number of


equations is constant through the analysis and no additional
equation renumbering and solver initialization is needed.
The quiet element method has the following disadvantages:

 If the scaling factors sk and sC are not small enough, thermal


p




energy conducts into the inactive elements resulting to errors.


If the scaling factors sk and sC p are too small, the Jacobian
resulting from Eq. (11) may be ill-conditioned.
Implementation of the quiet element method in modeling additive
manufacturing, where most of the analysis domain is composed of
quiet elements, may result into long computer runs.

3.2. Inactive element method


In the inactive element method, the elements representing metal
deposition regions are removed from the analysis and only nodal
degrees of freedom corresponding to active elements are considered.
Numerical implementation involves computing the element residual
(Eq. (10)) and Jacobian (Eq. (11)) for the active elements only and
solving for the active nodal degrees of freedom only.
The inactive element method has the following advantages:

 There are no errors or ill-conditioning introduced by scaling


factors as is the case of the quiet element method.

 Element residual and Jacobian calculations are performed for


active elements only.

 Only the active nodal degrees of freedom are considered at a


time resulting into smaller algebraic systems by the Newton
Raphson linearization.
The inactive element method has the following disadvantages:

 The method cannot be easily incorporated into general purpose


commercial codes using user subroutines.

 The equation numbering and solver initialization have to be

repeated every time elements are activated. This may negate


the computational advantage of solving for a reduced active
number of degrees of freedom.
When elements are activated, nodes shared by active elements
may not be at the initial temperature which may result into
articial energy being introduced into the model. The effect is
illustrated in Fig. 2. Node 2 is shared by active element 1 and
inactive element 2. If the temperature at node 2 is not equal to
initial temperature 0 T , when element 2 is activated, the initial
temperature across the element as interpolated by Eq. (12) is
not equal to the initial temperature 0 T which results into
articial numerical thermal energy introduced during element
activation. A remedy to this problem is to set n  1 T 0 T in

Fig. 3. Illustration of interface between active and inactive (or quiet) elements.

Eqs. (10) and (11) for the inactive elements before their
activation. n  1 T is not reset for the active elements, otherwise,
energy would be lost.
4. Interface between active and inactive elements
Fig. 3 illustrates a nite element mesh for the simulation of
building a thin wall by powder fed metal deposition additive
manufacturing. The active element region is shown as contour shaded
with temperature results and the inactive (or quiet) region as translucent. The interface between active and inactive (or quiet) elements is
also illustrated in the gure. Surface convection and radiation conditions (Eq. (5)) need to be applied to this interface since it is an external
free surface. The interface continuously changes as the metal is
deposited during processing. Therefore, algorithms need to be developed to calculate the location of the evolving interface and apply
surface convection and radiation as needed.
As seen in Fig. 3, the interface is an internal surface in the mesh.
When using general purpose FEA codes, it is difcult to compute
the interface between active and inactive (or quiet) elements using
user subroutines. Thus, it is common practice to neglect the
surface convection and radiation on this interface. This simplication may be appropriate in weld modeling since the ller metal
and therefore the size of the interface is negligible compared to
the base metal. However, in additive manufacturing, the size of the
deposited material could be signicant compared to the substrate
and thus neglecting surface convection and radiation on the
interface between active and inactive elements can be a source
of error as demonstrated in the examples section.

5. Numerical examples
5.1. 1D Examples
The 1D bar problem of Fig. 4 is used to illustrate and quantify
potential errors of the implementation of the quiet and inactive

P. Michaleris / Finite Elements in Analysis and Design 86 (2014) 5160

5.1.1. Quiet element method results


Table 1 lists the analyses cases performed using the quiet
element method. The analyses are performed by implementing
(Eqs. (10)15) into a Matlab code for 100 time increments of 0.1 s.
Results for the rst and last time increments are illustrated in
Figs. 5 and 6, respectively. Temperature results of the center node
are also listed in Table 1. Results at the rst time increment
evaluate the effectiveness of the method to compute the discontinuous temperature transition from active to quiet elements,
while, results at the last time increment indicate any articial
(numerical) heat loss/generation by the method.
In cases q1q4, the thermal conductivity scaling factor sk is
varied from 1 to .000001. As seen in Table 1 and Fig. 5, the error at
time t 0.1 s decreases from  50% to  11.5739% as sk is set to
lower values. However, the improvement from sk .0001 (case q3)
to sk .000001 (case q4) is minimal, and further reduction of sk
risks introducing ill-conditioning in the system Jacobian. The error
at time t 10 s for cases q1q4 is less than 10  10%.
In cases q5 and q8, the thermal conductivity scaling factor sk is
set to sk .0001 and the specic heat scaling factor sC p is varied
from .1 to .01 respectively. As seen in Table 1 and Fig. 5, the error at
time 0.1 s decreases to 1.3296% and  0.1653% as sC p is set to
lower values. However, the error at the last time increment
increases to 5.2786% and 7.2180%, indicating that there is articial
(numerical) heat generation.
In case q7, the number of elements is increased tenfold and
sk .0001 and sC p :01. As the element length is decreased, the
error at t .1 s decreases to 0.0984%, but at t 10 s is still at
6.4596%.
In case q8, sk .0001 and sC p :01 but the previous increment
temperature n  1 T during the activation of the quiet elements is
reset to the initial temperature 0 T , enforcing the discontinuity of
the solution at the previous time increment. The error now at
Q

t.1 s is reduced to  0.0729%. Indicating the articial energy loss/


generation is diminished.
5.1.2. Inactive element method results
Table 2 lists the analyses cases performed using the inactive
element method. Results for the rst (t .1 s) and last (t 10 s) time
increments are illustrated in Figs. 7 and 8, respectively. Temperature
results of the center node are also listed in Table 2. In the rst time
increment, the inactive element method is 100% accurate over the
active element region. However, on the last time increment, there
is articial energy generation resulting to 10% error for 10 elements
(case i1) and 1% for 100 elements (case i2). When the previous
increment temperature n  1 T during the activation of the inactive
elements is reset to the initial temperature 0 T , the error at the last
time increment is diminished (cases i3 and i4).
100
q1:sk=1,scp=1
q2:sk=.01,scp=1
q3:sk=.0001,scp=1
q5:sk=.0001,scp=.1
q6:sk=.0001,scp=.01
q7:sk=.0001,scp=.01,100elmts
exact

80
60
Temperature (C)

element methods. The length of the bar is set to 10 mm and the


cross section to 1 mm2. The thermal conductivity k is set to .05
W/mm/1C, the specic heat Cp to 1000 kJ/kg/1C, and the density
to 10  6 kg/mm3. The half left of the bar is set to be active at all
time, while the right half is set to be inactive or quiet for times
0.1 s, and then switched to active. A heat source Q of 1 W/mm3 is
applied to the left half of the bar for times 0.1 s.

40
20
0
20
40

10

10

Xcoordinate (mm)

Fig. 5. 1D quiet element method results at t 0.1 s.

54
53.5
53
Temperature (C)

54

52.5
q1:sk=1,scp=1
q2:sk=.01,scp=1
q3:sk=.0001,scp=1
q5:sk=.0001,scp=.1
q6:sk=.0001,scp=.01
q7:sk=.0001,scp=.01,100elmts
q8:sk=.0001,scp=.01,(n1)T reset
exact

52
51.5
51
50.5

50
49.5

Xcoordinate (mm)

Fig. 4. 1D Test problem geometry and boundary conditions.

Fig. 6. 1D quiet element method results at t 10 s.

Table 1
1D quiet element method test cases and results.
Case #

# of elements

sk

sC p

n1

Exact
q1
q2
q3
q4
q5
q6
q7
q8

10
10
10
10
10
10
100
10

1
.01
.0001
.000001
.0001
.0001
.0001
.0001

1
1
1
1
.1
.01
.01
.01

No
No
No
No
No
No
No
Yes

T reset

T center (% err)

100
50.0000 (  50.0000)
85.7959 (  14.2041)
88.3958 (  11.6042)
88.4261 (  11.5739)
98.6704 (  1.3296)
99.8347 (  .1653)
99.9016 (  .0984)
99.8347 (  .1653)

100

T center (% err)

50
50.0000 (2.8422e  14)
50.0000 (2.8422e  14)
50.0000 (1.8943e  11)
50.0000 (1.8915e  11)
52.6393 (5.2786)
53.6090 (7.2180)
53.2298 (6.4596)
49.9635 (  .0729)

P. Michaleris / Finite Elements in Analysis and Design 86 (2014) 5160

Table 2
1D inactive element method test cases and results.
Case # # of elements

n1

Exact
i1
i2
i3
i4

No
No
Yes
Yes

10
100
10
100

T reset

T center (% err)

100
100
100
100
100

100

T center (% err)

50
55.0000
50.5000
50.0000
50.0000

(0)
(0)
(0)
(0)

(10.0000)
(1.0000)
(  4.2633e  14)
(3.4106e  13)

100
i1:10elmts
i2:100elmts
i3:10elmts,(n1)T reset
i4:100elmts,(n1)T reset
exact

90
80

Temperature (C)

60
50
40
30
20
10
0

10

Xcoordinate (mm)

Fig. 7. 1D inactive element method results at t .1 s.

56
55

Temperature (C)

54

i1:10elmts
i2:100elmts
i3:10elmts,(n1)T reset
i4:100elmts,(n1)T reset
exact

53
52
51
50
49

Process description: A 3 mm thick wall is built using the


Optomec LENS system. The process uses a laser beam as heat
source and metal powder is fed into the beam path. The powder
particles under the laser beam melt and fall onto the partially
build part and solidify upon cooling. A substrate is used to start the
metal deposition. The wall length is 38.1 mm and it is build by 62
layers of 0.2032 mm height each. Each layer is generated by a
single straight laser scan. The travel speed is 8.466 mm/s, and the
laser power is 425 W. The substrate is 6.35 mm thick, 38.1 mm
wide, and 76.2 mm long.
Heat input model: The double ellipsoid [46] volumetric source
heat input model is used to simulate the heat input onto the part.
The heat source distribution Q is given by
Q

70

10

Xcoordinate (mm)

Fig. 8. 1D inactive element method results at t 10 s.

5.2. 3D Models of the LENS process


In this section, the quiet and inactive element methods are
evaluated in 3D models of building a thin wall using the directed
powder process by the Optomec LENS system. The process conditions and material properties are set to correspond to typical
experimental settings. However, the intention of this study is model
verication rather model validation which involves a systematic
and detailed experimental effort as well. (Eqs. (10)15) are programmed into the CUBIC code by Pan Computing LLC. The code is
parallelized using the Open Multi Processing (OMP) interface.
Results are presented for 8 node hex (hex8) elements with full
numerical integration and consistent heat capacity matrix. The
maximum number of threads is set to the number of physical
processors (cores).

55

p
6 3P f  3x2 =a2 3y2 =b2 3z vw t2 =c2 
p e
abc

16

where P is the power; is the process efciency; x, y, and z are the


local coordinates of the double ellipsoid model; a, b, and c are the
transverse, through the depth, and longitudinal ellipsoid axes.
Usually, a is set to the deposition half width, and b to the melt pool
depth. Typically, different values are used in front and behind the
melt pool for the longitudinal ellipsoid axis c. However, here the
same value is used here since the melt pool is nearly spherical in
the LENS process. f is a scaling factor; vw is the heating spot travel
speed; and t is time. The values used here are: P 425 W, .45,
a 1.5 mm, b .9 mm, c 1.5 mm, and f 1. Eq. (16) is applied on
the entire model and beyond the edges of the ellipsoid (dened at
the 5% value [46]) and thus no correction for cut-off is needed.
FEA mesh: Fig. 9 shows a symmetric model of the substrate only
and Fig. 10 shows a symmetric model of the substrate and wall. The
substrate model (Fig. 9) is composed of 8480 hex8 elements and
10,086 nodes. The substrate and wall model contains 14,432 hex8
elements and 19,200 nodes. The mesh density is targeted towards
obtaining temperature results for follow on mechanical analyses for
computing residual stress and distortion. Finer meshes would be
required for solidication studies.
Time incrementation: During heating, the time increment is set
to 0.0885827 s, which results into a 0.75 mm (quarter of deposition width) travel distance per time increment. When there is no
heating, the time increments are increased.
Element activation criterion: In the double ellipsoid model, the
heat source at the edges of the ellipsoid reduces to 5% of the peak
value at the center. Since the edges of the ellipsoid correspond to
the edge of the molten region, quiet or inactive elements are
activated when the following is valid at any Gauss point of an
element:
p
2
2
6 3f
2
2
2
2
p e  3x =a 3y =b 3z vw t =c  Z :05
abc

17

Other boundary conditions: The free surface convection set to


h.00001 W/mm2/1C. The gas and powder ow near the heating
zone is simulated as a forced convection region moving along with
the heat source with a spherical shape with 5 mm radius and a
forced convection of H .000210 W/mm2/1C. The emissivity for
radiation is set to .5. The initial temperature 0 T and ambient
temperature T 1 are set to 30.5 1C.
Material properties. Material properties for TI64 are used. The
thermal conductivity k and specic heat Cp are listed in Table 3
[47]. Material properties are linearly interpolated between the
values listed on the table, and kept constant beyond the minimum
and maximum listed values. The density is 4.43  10  6 kg/mm3.
The latent heat of fusion 365 kj/kg and is spread over a temperature range from 1600 1C to 1670 1C.

56

P. Michaleris / Finite Elements in Analysis and Design 86 (2014) 5160

Fig. 9. Substrate heating model (1C, case s1).

Fig. 10. Substrate heating and quiet element wall model (1C, case sq1).

P. Michaleris / Finite Elements in Analysis and Design 86 (2014) 5160

Table 3
Thermal properties of TI64.
Temperature (1C)

Thermal conductivity (W/mm/1C)

Specic heat (kJ/kg/1C)

23
93
205
315
426
537
649
760
871

6.6E  3
7.3E  3
9.1E  3
10.6E  3
12.6E  3
14.6E  3
17.5E  3

565
574
603
649
699
770
858
959

2000
s1:substrate only, reference
sq1:sk=.000001,scp=.01
si1

1500

Temperature (C)

1000

500

500

1000
10

10

15

Zcoordinate (mm)
Fig. 11. Comparison of wall build temperature results.

Table 4
Substrate test cases and results.
Case #

Description

Method

sk

sC p

n1

s1
sq1
si1

Substrate only
Substrate and wall
Substrate and wall

Quiet
Inactive

.000001

.01

T reset

max % err
Reference
.0390
6.375e  5

57

minimization of the norm of the global residual by 7 orders of


magnitude.
Tables 5 and 6 list the analysis cases performed for the wall build
test and the respective wall CPU run times on a 16 3.1 GHz core
computer using either all or a single core. Cases wq1 and wq2 use
the quiet element method with thermal conductivity and specic
heat scaling factors at the quiet elements set as sk .000001 and
sC p :01. Initially, all elements on the wall are set to quiet, then,
when Eq. (17) is true at any Gauss point of an element, it is switched
to active. The scheme results into a continuous element activation
emulating the metal deposition of the process. The quiet element
cases use the same number of equations throughout the analysis
and require no additional equation renumbering and solver initialization. Run times are about 1009 s wall CPU using 16 cores
(7587 s using 1 core).
Cases wi1 and wi2 use the inactive element method. Initially,
the all elements on the wall are set to inactive, and then switched
to active based on the heat input similarly to the quiet element
method. Since elements are activated continuously, equation
numbering and solver initialization is repeated at nearly all time
increments resulting into slower computer run times of about
1875 s wall (8179 s using 1 core).
Cases wiq1wiq3 use a hybrid inactive/quiet element method
in which wall elements are initially set as inactive. Then, layer by
layer elements are switched to quiet, and then switched individually to active based on the heat input similarly to the quiet element
method implemented in cases wq1 and wq2. In this approach,
equation numbering and solver initialization is repeated only
when each layer is activated resulting into slightly faster computer
run times of about 976 s wall (6284 s using 1 core) per case.
In cases wq2, wi2, wiq2, and wiq3, the temperature of quiet or
inactive elements before activation n  1 T is set to the initial
temperature 0 T , while in cases wq1, wi1, wiq1, it is not. The
temperature result during the deposition of the 61st layer for case
wq2 is illustrated in Fig. 12. Fig. 13 illustrates the temperature
result for case wq2 at the same time instance with all quiet
element removed. The temperature result during the deposition of
the 61st layer for case wi2 is illustrated in Fig. 14. Temperature
results along line AA are extracted and plotted in Fig. 15. Results
from cases wq2 and wi3 are also plotted in the same gure. Using
Table 5
Wall build test cases and results.

5.2.1. Substrate heating test


In this test, heating of the substrate only with no metal deposition
is modeled. A model of the substrate only (Fig. 9, case s1) is used as
reference to evaluate the accuracy of models with active elements on
the substrate and quiet or inactive elements on the wall (cases sq1
and si1). For the quiet element model (shown in Fig. 10), the
conductivity and specic heat scaling factors at the quiet elements
are set as sk .000001 and sC p :01. Temperature results along line
AA as illustrated in Figs. 9 and 10 are extracted and plotted in Fig. 11.
Convection and radiation are also applied on the interface between
active and inactive (or quiet) elements. The maximum error for the
quiet element and inactive methods is .0390% and 6.375e 5%,
respectively (Table 4). It is noted that no elements are switched from
quiet to active in this test.

5.2.2. Wall build test


Fig. 10 shows the symmetric model of the substrate and wall
used for this test. A time period of 1000 s is simulated. The metal
deposition lasts 286.95 s and requires 3223 time increments.
Further cool down for the remaining of the simulation requires
9 additional time increments. An average of 6 iterations per time
increment is required to achieve convergence, which is dened as

Case #

Interface
conv & rad

Method

sk

sC p

n1

wq1
wq2
wi1
wi2
wiq1
wiq2
wiq3

Yes
Yes
Yes
Yes
Yes
Yes
No

Quiet
Quiet
Inactive
Inactive
Inactive/quiet
Inactive/quiet
Inactive/quiet

.000001
.000001

.01
.01

.000001
.000001
.000001

.01
.01
.01

No
Yes
No
Yes
No
Yes
Yes

T reset

max % err

4.1234
.3734
5.3056
Reference
3.5323
.5242
7.5832

Table 6
Wall build run times (s).
Case #

Wall CPU (16 cores)

Wall CPU (1 core)

wq1
wq2
wi1
wi2
wiq1
wiq2
wiq3

1029
1009
2383
1875
1062
976
1014

7864
7587
8102
8179
6313
6284
6296

58

P. Michaleris / Finite Elements in Analysis and Design 86 (2014) 5160

Fig. 12. Wall build using quiet elements (1C, case wq2).

Fig. 13. Wall build not showing quiet elements (1C, case wq2).

P. Michaleris / Finite Elements in Analysis and Design 86 (2014) 5160

59

Fig. 14. Wall build using inactive element layers, quiet elements within layer (1C, case wiq2).

example, the maximum error is about 7.5832% since the size of


the interface is small compared to the total surface of the part.
However, the error may be higher when the size of the interface is
larger such as in multi-layer cladding and manufacturing of
complex 3D parts.

2500

Temperature (C)

2000

wq2: quiet,sk=.000001,scp=.01
wi2: inactive, reference
wi3:inactive, no interface bcs

1500

6. Conclusions
1000

500

500
10

10

15

Zcoordinate (mm)
Fig. 15. Comparison of wall build temperature results.

case wiq2 as reference, the maximum % error along the line is


listed in Table 5. As seen in Table 5, neglecting to reset n  1 T upon
element activation leads to errors (3.53235.3056%). The error is
higher in the inactive element method. When n  1 T is reset, all
methods result into nearly equivalent results (maximum
error .5%).
In all cases, except case wi3, surface and convection and
radiation are applied on the interface between active and inactive
(or quiet) elements. Neglecting the application of convection and
radiation on the interface between active and inactive (or quiet)
elements results into higher computed temperatures. In this

Finite element techniques for modeling metal deposition in


heat transfer analyses of additive manufacturing are investigated
in detail. A popular method used in general purpose commercial
codes through user subroutines is the quiet element method,
where elements are present in the analysis but quiet elements
are assigned properties rendering them irrelevant until their
activation. Another approach is the inactive element method,
where elements are removed from the analysis until their
activation.
1D and 3D numerical examples are used to demonstrate that
when the thermal conductivity and specic heat are scaled by
.000001 and .01 respectively, both active and inactive element
method results are practically identical (less than .5% difference).
However, both methods result in articial heating generation
(more than 5% error) unless, the temperature of the active (or
quiet) elements is reset to the initial temperature during activation
to enforce the discontinuity of the temperature eld in the
transition from active to inactive (or quiet) elements.
Although the inactive element method results into a reduced
number of degrees of freedom at the initial stages of the analysis,
it requires longer computer run times than the quiet element
method due to the continuous equation renumbering and solver
initialization.
A new hybrid inactive/quiet element method is proposed for
modeling additive manufacturing. In this approach, elements

60

P. Michaleris / Finite Elements in Analysis and Design 86 (2014) 5160

corresponding to metal deposition are initially inactive, then, they


are switched to quiet on a layer-by-layer basis. The approach
results into equivalent results within faster run times.
Numerical examples also demonstrate that neglecting the surface convection and radiation on the continuously evolving interface between active and inactive (or quiet) elements can lead into
over predicting the resulting temperature.
References
[1] S.M. Kelly, S.L. Kampe, Metall. Mater. Trans. Phys. Metall. Mater. Sci. 35A
(2004).
[2] S.M. Kelly, Thermal and microstructure modeling of metal deposition processes with application to Ti-6Al-4V (Ph.D. thesis), Virginia Tech, 2004.
[3] P. Peyre, P. Aubry, R. Fabbro, R. Neveu, A. Longuet, Analytical and numerical
modelling of the direct metal deposition laser process, J. Phys. D Appl. Phys. 41
(025403) (2008) 10.
[4] X. Xiong, W.H. Hofmeister, Z. Cheng, J.E. Smugeresky, E.J. Lavernia,
J.M. Schoenung, in situ thermal imaging and three-dimensional nite element
modeling of tungsten carbidecobalt during laser deposition, Acta Mater. 57
(2009) 54195429.
[5] I.A. Roberts, C.J. Wang, R. Esterlein, M. Stanford, D.J. Mynors, A threedimensional nite element analysis of the temperature eld during laser
melting of metal powders in additive layer manufacturing, Int. J. Mach. Tools
Manuf. 49 (2009) 916923.
[6] G.P. Dinda, A.K. Dasgupta, J. Mazumder, Laser aided direct metal deposition of
inconel 625 superalloy: microstructural evolution and thermal stability, Mater.
Sci. Eng. A 509 (1) (2009) 98104.
[7] B. Baufeld, O. Van der Biest, R. Gault, Additive manufacturing of ti-6al-4v
components by shaped metal deposition: microstructure and mechanical
properties, Mater. Des. 31 (2010) S106S111.
[8] Alireza Fathi, Amir Khajepour, Ehsan Toyserkani, Mohammad Durali, Clad
height control in laser solid freeform fabrication using a feedforward PID
controller, Int. J. Adv. Manuf. Technol. 35 (3) (2007) 280292.
[9] Almir Herali, Anna-Karin Christiansson, Bengt Lennartson, Height control of
laser metal-wire deposition based on iterative learning control and 3d
scanning, Opt. Lasers Eng. (2012).
[10] Lijun Song, Vijayavel Bagavath-Singh, Bhaskar Dutta, Jyoti Mazumder, Control
of melt pool temperature and deposition height during direct metal deposition process, Int. J. Adv. Manuf. Technol. 58 (1) (2012) 247256.
[11] N.W. Klingbeil, J.L. Beuth, R.K. Chin, C.H. Amon, Residual stress-induced
warping in direct metal solid freeform fabrication, Int. J. Mech. Sci. 44 (1)
(2002) 5777.
[12] K. Dai, L. Shaw, Distortion minimization of laser-processed components
through control of laser scanning patterns, Rapid Prototyp. J. 8 (5) (2002)
270276.
[13] Aditad Vasinonta, Jack L. Beuth, Michelle Grifth, Process maps for predicting
residual stress and melt pool size in the laser-based fabrication of thin-walled
structures, J. Manuf. Sci. Eng. 129 (1) (2007) 101109.
[14] A. Lundback, L.-E. Lindgren, Modelling of metal deposition, Finite Elem. Anal.
Des. 47 (2011) 11691177.
[15] A. Anca, V.D. Fachinotti, G. Escobar-Palafox, A. Cardona, Computational
modelling of shaped metal deposition, Int. J. Numer. Methods Eng. 85 (2011)
84106.
[16] M. Chiumenti, M. Cervera, A. Salmi, C.A. de Sraacibar, N. Dialami, K. Matsui,
Finite element modeling of multi-pass welding and shaped metal deposition
processes, Comput. Methods Appl. Mech. Eng. 199 (2010) 23432359.
[17] H. Hibbitt, P.V. Marcal, A numerical, thermo-mechanical model for the welding
and subsequent loading of a fabricated structure, Comput. Struct. 3 (1973)
11451174.
[18] J.H. Argyris, J. Szimmat, K.J. Willam, Computational aspects of welding stress
analysis, Comput. Methods Appl. Mech. Eng. 33 (1982) 635666.
[19] V.J. Papazoglou, K. Masubuchi, Numerical analysis of thermal stresses during
welding including phase transformation effects, J. Press. Vessel Technol. 104
(1982) 198203.
[20] L.-E. Lindgren, Finite element modelling and simulation of welding, part 1
increased complexity, J. Therm. Stresses 24 (2001) 141192.
[21] L.-E. Lindgren, Finite element modelling and simulation of welding, part 2
improved material modelling, J. Therm. Stresses 24 (2001) 195231.

[22] L.-E. Lindgren, Finite element modelling and simulation of welding, part 3
efciency and integration, J. Therm. Stresses 24 (2001) 305334.
[23] L.E. Lindgren, P. Michaleris, Modeling of welding for residual stresses, in:
Jian Lu (Ed.), Handbook on Residual Stress, vol. 2, SEM, USA, 2005, pp. 4767.
[24] L.-E. Lindgren, Numerical simulation of welding, Comput. Methods Appl.
Mech. Eng. 195 (2006) 67106736.
[25] L.-E. Lindgren, Computational Welding Mechanics. Thermomechanical and
Microstructural Simulations, Woodhead Publishing, Cambridge, England,
2007.
[26] J.A. Goldak, M. Akhlaghi, Computational Welding Mechanics, Springer, New
York, NY, USA, 2005.
[27] P. Michaleris (Ed.), Minimization of Welding Distortion and Buckling. Modeling and Implementation, Woodhead Publishing, Cambridge, England, 2011.
[28] S. Mishra, T. DebRoy, A heat-transfer and uid-ow model to obtain a specic
weld geometry using various combinations of welding variables, J. Appl. Phys.
98 (044902) (2005).
[29] Z. Yang, N. Chen, H.W. Ludewig, Z. Cao, Virtual welded-joint design by
coupling thermal-metallurgical-mechanical modeling, in: 6th International
Trends in Welding Research Conference Proceedings, ASM International, Pine
Mountain, GA, 2003, pp. 861866.
[30] G. Taylor, M. Hughes, N. Strusevich, K. Pericleous, Finite volume methods
applied to the computational modelling of welding phenomena, in: 2nd
International Conference on CFD in the Minerals and Process Industries,
CSIRO, Melbourne, Australia, 1999, pp. 405410.
[31] T. Zacharia, S.A. David, J.M. Vitek, T. DebRoy, Weld pool development during
GTA and laser beam welding of type 304 stainless steel, part 1-theoretical
analysis, Weld. Res. Suppl. S (1989) 499509.
[32] C. Kim, W. Zhang, T. Debroy, Modeling of temperature eld and solidied
surface prole during gas-metal arc llet welding, J. Appl. Phys. 94 (4) (2003)
26672679.
[33] T. Zacharia, A. Eraslan, D. Aldun, S.A. David, Three-dimensional transient
model for arc-welding process, Metall. Trans. B 20 (1989) 645659.
[34] Lijun Han, Kaushik M. Phatak, F.W. Liou, Modeling of laser deposition and
repair process, J. Laser Appl. 17 (2005) 89.
[35] S. Ghosh, J. Choi, Three-dimensional transient nite element analysis for
residual stresses in the laser aided direct metal/material deposition process,
J. Laser Appl. 17 (2005) 144.
[36] H. Tersing, J. Lorentzon, A. Francois, A. Lundback, B. Babu, J. Barboza, V. Backer,
L.-E. Lindgren, Simulation of manufacturing chain of a titanium aerospace
component with experimental validation, Finite Elem. Anal. Des. 51 (2012)
1021.
[37] L.-E. Lindgren, H. Runnemalm, M.O. Nasstrom, Simulation of multipass
welding of a thick plate, Int. J. Numer. Methods Eng. 44 (9) (1999) 13011316.
[38] L.-E. Lindgren, E. Hedblom, Modelling of addition of ller material in large
deformation analysis of multipass welding, Commun. Numer. Methods Eng. 17
(9) (2001) 647657.
[39] Gangxian Zhu, Anfeng Zhang, Dichen Li, Yiping Tang, Zhiqiang Tong,
Qiaopan Lu, Numerical simulation of thermal behavior during laser direct
metal deposition, Int. J. Adv. Manuf. Technol. 55 (912) (2011) 945954.
[40] R. Jendrzejewski, G. liwiski, M. Krawczuk, W. Ostachowicz, Temperature and
stress elds induced during laser cladding, Comput. Struct. 82 (7) (2004)
653658.
[41] Rafa Jendrzejewski, Gerard liwiski, Investigation of temperature and stress
elds in laser cladded coatings, Appl. Surf. Sci. 254 (4) (2007) 921925.
[42] S. Ghosh, J. Choi, Modeling and experimental verication of transient/residual
stresses and microstructure formation in multi-layer laser aided DMD process,
J. Heat Transf. 128 (7) (2006) 662679.
[43] Riqing Ye, John E. Smugeresky, Baolong Zheng, Yizhang Zhou, Enrique
s
J. Lavernia, Numerical modeling of the thermal behavior during the lens
process, Mater. Sci. Eng. A 428 (1) (2006) 4753.
[44] P. Peyre, P. Aubry, R. Fabbro, R. Neveu, Arnaud Longuet, Analytical and
numerical modelling of the direct metal deposition laser process, J. Phys. D
Appl. Phys. 41 (2) (2008) 025403.
[45] J. Song, J. Shanghvi, P. Michaleris, Optimization of thermo-elasto-plastic
manufacturing processes, Comput. Methods Appl. Mech. Eng. 193 (2004)
45414566.
[46] J. Goldak, A. Chakravarti, M. Bibby, A new nite element model for welding
heat sources, Metall. Trans. B 15B (1984) 299305.
[47] Rodney F. Boyer, E.W. Collings, Materials Properties Handbook: Titanium
Alloys, ASM International, Materials Park, OH, USA (1994) 514.