Deepankar Pal1

Assistant Professor
Department of Mechanical Engineering,
J.B. Speed School of Engineering,
University of Louisville,
Louisville, KY 40292
e-mail: d0pal001@louisville.edu

Nachiket Patil
Senior Research Engineer
3DSIM, LLC,
201 E. Jefferson Street,
Louisville, KY 40202
e-mail: nachiket.patil@3dsim.com

Kai Zeng
Department of Industrial Engineering,
J.B. Speed School of Engineering,
University of Louisville,
Louisville, KY 40292
e-mail: k0zeng01@louisville.edu

Brent Stucker
Clark Chair Professor
J. B. Speed School of Engineering,
University of Louisville,
Louisville, KY 40292
e-mail: brent.stucker@louisville.edu

An Integrated Approach
to Additive Manufacturing
Simulations Using Physics
Based, Coupled Multiscale
Process Modeling
The complexity of local and dynamic thermal transformations in additive manufacturing
(AM) processes makes it difficult to track in situ thermomechanical changes at different
length scales within a part using experimental process monitoring equipment. In
addition, in situ process monitoring is limited to providing information only at the
exposed surface of a layer being built. As a result, an understanding of the bulk
microstructural transformations and the resulting behavior of a part requires rigorous
postprocess microscopy and mechanical testing. In order to circumvent the limited
feedback obtained from in situ experiments and to better understand material response, a
novel 3D dislocation density based thermomechanical finite element framework has been
developed. This framework solves for the in situ response much faster than currently used
state-of-the-art modeling software since it has been specifically designed for AM
platforms. This modeling infrastructure can predict the anisotropic performance of
AM-produced components before they are built, can serve as a method to enable in situ
closed-loop process control and as a method to predict residual stress and distortion in
parts and thus enable support structure optimization. This manuscript provides an
overview of these software modules which together form a robust and reliable AM
software suite to address future needs for machine development, material development,
and geometric optimization. [DOI: 10.1115/1.4028580]

Introduction
AM is a way of converting a 3D computer aided design (CAD)
file into a useful physical object using the sequential application
of material and energy (such as light, ultrasonic vibration, lasers,
or electron beams) to bond or otherwise join materials. Depending
on the technology and the material(s) used for creation of the final
part, the strength and type of these bonds vary. Evaluation of these
bonds as a function of process parameters and materials is a critical step for process reliability and quality control.
In AM, the layer-by-layer addition of materials to fabricate the
final geometry enables complex internal features such as channels
and embedment of prefabricated components such as fibers and
sensors. The inclusion of these internal features and the highly
dynamic nature of these processes lead to residual stresses and
dimensional warping, which may lead to failure during the build
(such as when the blade distributing powder crashes into a distorted part being build) or premature failure of these parts in service. This calls for evaluation of residual stresses and dimensional
warping of the parts as a function of process parameters, materials,
and build geometry. In addition, the surface finish and geometrical
accuracy of AM-produced parts are not necessarily good enough
for precision applications and thus postprocessing operations are
often required. Therefore, evaluation of surface finish and geometrical accuracy as a function of slice information, part orientation,
material, and process characteristics is also important.
To fully capitalize on the future potential of AM for precision
engineered components, a theoretical understanding of the physics
1
Corresponding author.
Contributed by the Manufacturing Engineering Division of ASME for publication
in the JOURNAL OF MANUFACTURING SCIENCE AND ENGINEERING. Manuscript received
April 21, 2014; final manuscript received September 12, 2014; published online
October 24, 2014. Assoc. Editor: Joseph Beaman.

involved in AM processing and validated software tools with the

capabilities to predict process effects are needed both in academia
and industry. These software tools will enable future machines
to be more efficient, scientists and researchers to develop new
material alloys specifically tweaked for cooling rates experienced
during AM, and designers to more fully explore the geometric
freedom and functionality desired by the end-user while still
maintaining the required strength, fatigue, corrosion, or other
attributes.

Broader Objectives
In order to efficiently simulate AM, new software tools, which
can simulate highly dynamic, coupled manufacturing, and metallurgical processes, including the high strain and cooling rates
exhibited by AM, are required. Some of the key aspects of a
successful AM simulation software toolset include
(a) Process map response surfaces, which correlate parameters
and outcomes such as beam power and scan speed to grain
size and shapes and melt pool geometry. This enables optimization of materials and geometry a priori. Other predictive outputs can include residual stress, dynamic warping,
and mechanical properties.
(b) Enabling of closed loop control where on-the-fly compensation for out-of-spec thermal or microstructural defects
can be achieved. A possible path for doing so includes
using optical sensors, which record layer-by-layer information such as temperatures, geometry, distortion, porosity,
etc., and their spatiotemporal locations so that this information can be fed for an instantaneous next layer feed forward
simulation using the real, current build geometry. This
could enable, for instance, a local tweak of the process
parameters in the vicinity of a defect in the previous layer

Journal of Manufacturing Science and Engineering

C 2014 by ASME
Copyright V

DECEMBER 2014, Vol. 136 / 061022-1

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 05/03/2015 Terms of Use: http://asme.org/terms

Fig. 1 Meshing configuration for a point energy source located within the fine mesh region. The point source
could be a Gaussian heat source such as a concentrated laser or electron beam and the fine mesh region moves as
the point energy source moves.

to counter-act or fix the defect. With the increasing speed

of simulations achievable by porting codes to a general
purpose graphical processing unit (GP-GPU), these defects
may even be fixable within a layer by changing, for
instance, beam power and scan speed in the next scan
vector.
(c) Accurate predictions of the mechanical properties of
parts as a function of process parameters, material(s), and
geometry.
(d) Formulations of better scan strategies and optimized support
structure placement, leading to lower residual stresses and
less postprocessing of the fabricated part.

Novel dynamically adaptive meshing strategies have been formulated with mesh movement tied to the energy sources and sinks
employed in AM. Adaptive meshes are generally refined near the
energy source or energy dispersive mechanism(s), since high thermal or stress/strain gradients at these locations require fine-scale
accuracy to minimize the errors, which would occur if the mesh
remained at a coarser scale. The required refinement regimes are

Software Capabilities
Finite element analysis (FEA) software suites generally have
three components built into them, namely, preprocessing, the core
solver, and postprocessing capabilities. In this paper, we will
focus on the preprocessing and solver capabilities of our software.
It should be noted that all of these capabilities were written as
FORTRAN and/or MATLAB routines and can run independently from
commercial FEA software tools and that these capabilities are currently being ported to C# and C code for execution on a
CUDA GPU.
Three Dimensional Multiscale Dynamic Meshing
Capabilities (Preprocessing). FEA preprocessing and mesh generation strategies have been in existence for years [1,2] while the
meshing infrastructure required for carrying out efficient AM process modeling using adaptive mesh generation (primarily static)
were developed more recently [3,4]. A myriad of dynamic and
automatic adaptive automatic mesh generation tools [5] are available, but there exists a huge difference between various mesh generation communities, including computer scientists/graphic
modelers and simulation communities which primarily include
physicists, engineers, and geologists. In reference to AM predictive simulation literature specifically, most meshing strategies
available from commercial vendors do not combine the benefits of
innovative solution algorithms and preprocessing [615]. Several
offer submodeling options [16] such that problems involving
localized input of energy having global thermomechanical effects
can be solved in a globallocal simulation framework involving
transfer of boundary conditions from one region to another. It
should be noted that the mesh primitives described in this paper,
utilizing existing best practices, make it possible to take advantage
of the multispatial nature of the problem occurring in metal melting and solid state AM processes to generate a solution algorithm
with hierarchical and intelligent thermal stiffness generation,
seamless integration of the coarse and fine mesh regions, and
dynamic combined globallocal stiffness updates without node
renumbering. A brief description of the preprocessing tools created for tackling AM problems will be discussed in this section.
061022-2 / Vol. 136, DECEMBER 2014

Fig. 2 Multiscale mesh for UC showing: (a) the full 3D mesh

with selective adaptive refinement near the sonotrode and the
interfacial mating surfaces and (b) a close-up showing how the
hexahedral elements 2 are preserved throughout refinement.
The vertical refined region moves with sonotrode motion while
the horizontal fine mesh region is indexed up layer-by-layer as
new foils are added.

Transactions of the ASME

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 05/03/2015 Terms of Use: http://asme.org/terms

thermal and deformation fields experienced in AM. Figure 4 [18]

demonstrates a microstructurally informed hexagonal mesh for
predicting the mechanical properties of Ti6/4 parts produced by
EBM.
The preprocessing tools available from this software are also
applicable for solving thermomechanical fields in other closely
related joining and material processing phenomenon such as laser
joining and surface treatment [1922] or for using these tools in
already existing predictive algorithms [23] specially designed for
solving these phenomenon where the above-mentioned tools
could be used to provide a seamless transition from fine to coarse
scale simulation domain.

Fig. 3 1/8th view of a volumetric refined mesh of two secondary precipitates (at diagonally opposite corners with different
curvatures) diagonally opposite to each other placed in an
Inconel 718 matrix (in blue) with different curvatures [17]

multidimensional in nature. Some examples of energy sources

include Gaussian point energy sources (Fig. 1) for selective laser
melting (SLM) and electron beam melting (EBM) and line energy
ultrasonic vibration compressive forces (Fig. 2(a)) for ultrasonic
consolidation (UC). An example of superficial or area based
refinement occurs at the interface of mating surfaces of foils
(Fig. 2(b)) undergoing UC or friction stir welding. An example
of volume based refinements include the modeling of Inconel
alloys, since they contain secondary and tertiary precipitates
(Fig. 3 [17]), which respond differently than the bulk to the

Thermomechanical Finite Element Framework Including

Macroscopic Nonlinear Thermal and Deformation Behavior
of Materials. A general-purpose nonlinear coupled thermomechanical solver has been developed. SLM thermal contours
(Fig. 5) are shown for a slowly moving melt pool with a speed of
100 mm/s and 180 W for Ti6Al4V. The significance of this comparison is that the low temperature dimensions as predicted using
the simulation (627 lm and 970 lm) and observed using the infrared (IR) camera (635 lm and 976 lm) match with a maximum
error of <1%. A slow moving melt pool has been chosen to
accurately capture the thermal contours. The reason why the low
temperature regions have been selected for this study is due to the
fact that the IR camera emissivities are much more tractable in the
lower temperature regime in addition to the simulation error,
which increases with decreasing isothermal contours away from
the melt pool center. Since the simulation correctly matches the
low temperature contour, it provides a reasonable basis for
trusting that the high temperature melt pool predications are

Fig. 4 (a) Optical micrograph of an EBM-made Ti 6/4 microstructure showing hexagonal

prior b grain motifs and (b) microstructurally informed mesh, and (c) in-plane mesh
complexity [18]

Journal of Manufacturing Science and Engineering

DECEMBER 2014, Vol. 136 / 061022-3

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 05/03/2015 Terms of Use: http://asme.org/terms

Fig. 5 Match of thermal contours between nonlinear thermomechanical analysis (a) (temperature in K) using the mesh
provided in Fig. 1 and experiments (b) obtained using a forward looking infrared (FLIR) camera (temperature in Celsius)

accurate. Furthermore, the accuracy of the melt pools have been

indirectly validated for an EOS M 270 machine using scan parameters of 186 W (190 W was used as the input parameter in the
machine) and 1200 mm/s for Ti6Al4V material as shown in Fig. 6
using an optical micrograph of a build which was suddenly interrupted for the purposes of melt pool observations. Additional formulation details on this solver have been described in prior
publications along with its formulation [615,24] and efficiency
improvements [25,26].
For residual stress predictions, three cases have been considered
for force and moment balance for computing the threedimensional residual stresses and distortions, as shown in Fig. 7.
A simple single bead example has been provided for computing
the longitudinal residual stresses and distortions using the force
and moment balance as shown in Fig. 7(a).
Some of the previous work on thermal stresses and distortions
were conducted earlier in the literature assuming linear, nonlinear,

and large deformation constructs for welding scenarios and using

the profiles in AM to look at the part errors [2730].
The thermal contours, compressive nonlinear longitudinal
stresses (in MPa), distortions (in m), and the angle of curvature (in
radians) are shown in Fig. 8. In Fig. 8(a), the thermal contours are
shown. The melt pool thermal contour is specifically of interest
since it is responsible for the shrinkage shown in Fig. 7(a). This
shrinkage is further responsible for the longitudinal residual
stresses, distortions and the curvature angle w.r.t the horizontal
plane, as this shrinkage causes a force to act in the opposite direction to counteract the shrinkage, which leads to further force
imbalance such that the global translational force equilibrium is
not disturbed. The variables of interest here are the curvature and
the neutral axis. Hence, apart from the translational force equilibrium, the equilibrium of the first moment of forces is also applied
to obtain the above-mentioned variables, leading to a second set
of equations required for solving both of the above-mentioned

Fig. 6 Indirect validation of the melt pool showing the melt pool asymmetry due to the existence of solid (conductor on one
side) and powder (insulator on the other side) of the melt pool

061022-4 / Vol. 136, DECEMBER 2014

Transactions of the ASME

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 05/03/2015 Terms of Use: http://asme.org/terms

Fig. 7 Force and moment balance scenarios to be considered

for residual stress and distortion predictions

variables. The distortion, stress, and curvature contours contain

rough edges due to the adherence of powder particles to the melt
pool and computation of the deformation variables every fine
mesh offset. As illustrated in Fig. 1, the fine mesh is fixed w.r.t
the coarse mesh location until the laser has crossed from 25% to
75% of the fine mesh extremas, and this coarse-fine mesh configuration is termed an offset. Since the laser traverses around 50% of
the fine mesh region for each offset, an appreciable number of
time steps are associated with each offset. An important aspect
included in this simulation framework is the runtime update of the
mesh coordinates such that the mesh geometry keeps track of
various aspects of deformation related to the thermomechanical
phenomenon such as shrinkage and curling along with the geometrically nonlinear thermal stiffness matrix, which is updated at
each time step to compute the thermal behavior in the next time
step. This leads to further mechanical deformation. Experiments
are underway to validate the accuracy of the longitudinal distortions and residual stresses (using neutron beams) along with angle
of curvature for single bead and other canonical geometry
predictions.

Dislocation Density Based Finite Element Method. The UC

problem (Fig. 2) as a function of normal force, weld speed and
surface roughness has been modeled using a dislocation density
based crystal plasticity finite element method (DDCPFEM)
[22,23,3135]. The model has been shown to accurately predict
average grain size at the interface. Figure 9 [36] illustrates the
ability of the model to predict trends for surface-finish-based
differences in subgrain refinement at the interface. Some recent
studies on dislocation density-based grain refinement modeling
have been also conducted recently [37,38].
The effect of microstructure (Fig. 4) on mechanical properties
has also been predicted using DDCPFEM. The analysis provides
both global and local scales results. Results in Figs. 10 and 11
Journal of Manufacturing Science and Engineering

Fig. 8 In situ single bead (a) thermal contours, (b) longitudinal

residual stress contours, (c) longitudinal distortion, and (d) distortion angle w.r.t horizontal plane as a function of dynamic
melt pool on the top layer of the build

illustrate the detailed stress/strain analysis using DDCPFEM in

regimes where commercial software packages fail to accurately
predict the grain specific response for plastic deformation. The
plastic strain evolution in DDCPFEM simulations is grainorientation specific in Fig. 11(a) whereas the plastic strain is
symmetric about the top and bottom centers of the simulated
model in Fig. 11(b), which is physically incorrect. The value of
the maximum plastic strain at the top center is 17.88% in both
the ANSYS and DDCPFEM simulations [18]. It should be noted that
parts fabricated using macroscopic manufacturing operations such
as forming, welding, wire drawing, and rolling could be better
simulated using phenomenological approaches such as the nonlinear structural approach used in ANSYS and classical crystal plasticity [3941] the required strength, fatigue, corrosion, or other
attributes.
Eigenmodal Solver (EigenSolver). A unique methodology
being implemented to more quickly solve the thermomechanical
SLM process problem [25]. All SLM process models are computationally expensive to solve using traditional FEM methods, as
they require fine-scale resolution across a part volume that is
many orders of magnitude larger than the fine-scale lengths. This
has led to efforts to solve the problem using asymptotic behaviors,
modal space methodologies, or beam theories. The intent behind
this new EigenSolver is to solve for dominant modes in the
DECEMBER 2014, Vol. 136 / 061022-5

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 05/03/2015 Terms of Use: http://asme.org/terms

Fig. 9 Variation of GND with surface roughness. The simulated average grain size at the mating interface for the rough sample was 1.3 lm whereas the experimentally obtained and weight averaged value was found to be 1.33 lm. The simulated
average grain size at the mating interface for the smooth sample was 2.43 lm which is in good agreement with Dr. K. Johnsons Ph.D. work [36].

macroscopic domain and local high frequency modes in the microscopic domain. These can be accurately applied at certain locations (e.g., five layers) below the melt pool.
Beam and plate theories are meant to take benefit of asymptotic
theories involved in dimensionally reducible structures. Beam theories for complicated geometries such as aircraft wings and ship
hulls and for material variations across the cross section and
length of a component have been an area of intense research. An
analytical method, namely, the variational asymptotic method
[42], has been used extensively to derive these beam theories. One
of the limitations of this methodology lies in the difficulty of
deriving beam theories for very complicated beams. To overcome
this problem, we have developed a new EigenSolver beam theory
derivation for use with FEA. This novel method can consider any
complicated shape and cross-sectional variation and derives beam
or plate theories involved in it. Further it has applications in

problems which do not fall in the categories of beams or plates

(e.g., prismatic bodies or any general structural configuration).
The importance of this methodology lies in calculating eigenvectors of the cross section or group of nodes in FEM matrices (such
as the stiffness matrix). It is well known that eigenvalues and
eigenvectors are computationally expensive to calculate. To
overcome this difficulty, any set of orthogonal vectors can be
employed for this purpose and their coupling is calculated to
derive simplified simultaneous equations. Various material and
geometrical nonlinearities can be incorporated in this beam theory
with the help of our other existing solver tools.
This methodology can be further generalized to work with
groups of nodes in banded FEM meshes for large size problems.
These can then be solved on desktop computers. In summary, this
methodology is capable of

Fig. 10 Volume averaged true stressstrain plot for a vertically

built EBM Ti6/4 sample computed by ANSYS and DDCPFEM.
These average stressstrain evolutions have been compared
against experiments [18].

061022-6 / Vol. 136, DECEMBER 2014

representing complicated geometry and material inhomogeneity in beams

inexpensive computations leading to globallocal coupled
response
generalization to be used for any problem with some hidden
unknown asymptotics
solving large size FEM problems (for example, 10 million
degrees of freedom or much larger in some specific cases) on
desktop computers
simplified and fast assembly of FEM matrices.

In addition, an analytical method using a multidimensional,

multiresolution wavelet based eigentheory is under development.
This methodology can solve both linear and nonlinear problems
provided a space transformation between the cross sections is
available. The theory has been extended to account for inhomogeneous material distributions, which are very common among powder based AM technologies where solid and powder continuously
interact in a nonlinear mode with the applied thermal and resulting
deformation based boundary conditions.
Transactions of the ASME

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 05/03/2015 Terms of Use: http://asme.org/terms

Fig. 11 Plastic strain distribution at 10% total average strain for the stress/strain curves
from Fig. 10: (a) DDCPFEM simulations and (b) ANSYS anisotropic multilinear continuum
plasticity model

Fig. 12 Schematic diagram illustrating the concept of an eigensolver

A brief overview of the basic eigenvalue solver is shown in

Fig. 12. In Fig. 12(a), first, a Gaussian heat flux is applied to the
material followed by the split of the flux in orthogonal modes in
Fig. 12(b). These orthogonal modes are a function of geometry
and the material (in-homogeneity and nonlinearity), which
traverse forward through the cross sections shown in Fig. 12(c)
followed by their temperature counterparts, which traverse backwards toward the source as the flux modes hit the fixed temperature boundary. In Fig. 12(d), the stiffness modal coupling of flux
and temperature modes between changing cross sections is shown.
These matrices will assume an identity matrix or integral multiples of the identity matrix along the matrix diagonal in cases
where consecutive cross-sectional material properties/geometry
are assembled in a fashion such that they are integral multiples
w.r.t one another. If the material properties/geometry continuously
change, this will lead to off-diagonal terms comprising an interaction of one cross section w.r.t another.
Methodology. A novel solution strategy for prediction of thermomechanical variables away from the point of laser exposure
has been formulated. The involved strategy has been borrowed
from structural vibrations and image analysis algorithms where
modal contributions (eigenmodes) are computed in terms of orthogonal functions. The typical approach for determination of
modes using sine or cosine functions has been extended to fit a
Journal of Manufacturing Science and Engineering

finite element framework where an eigenvalue problem of the

thermomechanical propagator matrices (these matrices compute
the propagation of the thermomechanical field from one crosssectional layer to another along the z direction) is computed in

Fig. 13 Nodal point by point match of a finite element solution

for a thermal field with a modal reconstructed (eigensolver) solution for a Gaussian point energy source. The error is negligibly small (0.01%), and the results lie right on top of each other.

DECEMBER 2014, Vol. 136 / 061022-7

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 05/03/2015 Terms of Use: http://asme.org/terms

Fig. 14 Nodal point by point comparison of a finite element solution with a modal reconstructed solution for a constant line
energy source. The match is good (1% error) but not excellent
when compared to Fig. 13.

order to determine the orthogonal basis functions (eigenvectors)

for thermomechanical fields. It has been observed that very near
to the point of laser exposure the number of modes required to
reconstruct the solutions is very high (the number of modes is limited by the number of nodal points in each cross section) and as
the distance from the point of laser exposure increases, the number of modes required to accurately predict the solution drops
very quickly. In the case of Ti6Al4V, accurate thermomechanical
solutions can be calculated almost instantaneously four or more
layers away from the point of laser exposure using a limited number of modes. The error introduction using this approach for a
point energy source is negligibly small (see Fig. 13) while the

Fig. 15 Nodal point by point comparison of a finite element solution with a modal reconstructed solution for a constant area
energy source. The modally reconstructed solution matches
the trend (7% error) but not the magnitude of the solution due
to unrefined orthogonal modes. This result is irrelevant for
moving point energy problems associated with SLM or EBM,
but should be taken into account when seeking the correct solution for area energy sources.

speed improvement is substantial. For cases of a line energy

source (see Fig. 14), the match is good but not excellent. For cases
of a constant area source, due to unrefined orthogonal modes the
trend of solutions is replicated but the error is much worse when
compared to a point energy-source solution (see Fig. 15). The use
of Arnoldi and other methods for refinement of modes [43] for
this scenario, which is relevant for manufacturing processes in
which point sources of energy are not used. For focused energy
sources such as SLM and EBM, this approach completely

Fig. 16 First four cross-sectional eigenmodes of the prismatic thermal problem

061022-8 / Vol. 136, DECEMBER 2014

Transactions of the ASME

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 05/03/2015 Terms of Use: http://asme.org/terms

Fig. 17 First fifth and sixth cross-sectional eigenmodes of the prismatic thermal problem

eliminates the need to recompute thermomechanical fields using

fine-scale FFDAMRD for the problem domains, which are more
than a few layers away from the laser, thus eliminating the vast
majority of the computational time necessary to simulate real
parts (which are made up of thousands or tens of thousands of
layers).
It should be noted that so far eigenvalues have been identified
and reconstructed for the thermal case only. A parallel architecture is required for solving the deformation problem using our dislocation density based crystal plasticity (DDCP) finite element
solver. For DDCP, the eigenvalue problem has orders of magnitude higher complexity than for the FFDAMRD thermal problem
and will thus require a very fine refinement to compute the dispersion matrix and self-containment matrices correctly in order to
obtain the propagator matrix for its eigenvalue determination.

Basic Eigenmodes for Prismatic Structural Problems. The

structural problem is orders of magnitudes more complex (3
degrees of freedom per node) compared to the thermal problem in
terms of determination of eigenmodes and reconstruction of the
nodal degrees of problem. Nevertheless, eigenvalues have been
obtained for this problem and plotted in Fig. 19. The reconstruction
to obtain nodal degrees for this problem is still under development.

Basic Eigenmodes for Prismatic Thermal Problems. The

first few cross-sectional eigenmodes for reconstructing the nodal
degrees of freedom for prismatic thermal problems are shown in
Figs. 16 and 17.
These and other eigenmodes with increasingly higher eigenvalues are used for reconstructing the nodal degrees of freedom for a
variety of energy sources as shown in Figs. 13-15. Cross-sectional
eigenvalues are shown in Fig. 18 for up to 3600 degrees of
freedom.

Large Matrix Eigenvalue Problems. With increasing number

of degrees of freedom, it becomes increasingly hard to solve the
eigenvalue problem due to the complexity in solving the corresponding characteristic equation as the order of the polynomial
equation increases tremendously. The eigenvalue evolution as a
function of increasing number of degrees of freedom has been
shown in Fig. 20. The powder bed in metal laser sintering requires
a huge amount of discretization to populate the entire powder bed.
In order to solve the problem efficiently, it is required to predict
the evolution of eigenmodes with increasing number of degrees of
freedom in the cross section. It was observed that with increasing
number of degrees of freedom a 10 parameter Fourier expansion
could be used for fitting the evolution of eigenvalues. The reason
why a 10 parameter Fourier expansion could fit this evolution is
due the fact that there are 10 parameters required to extract the
eigenvalues, namely, the thermal conductivity, discretizations,
and the physical dimensions of the problem in the x-, y-, and
z-directions along with a constraint that the first eigenvalue essentially needs to be 1 (meaning the flux travels straight through all

Fig. 18 Cross-sectional eigenvalues of the thermal prismatic

problem

Fig. 19 Cross-sectional eigenvalues of the structural prismatic

problem

Results and Discussions

Journal of Manufacturing Science and Engineering

DECEMBER 2014, Vol. 136 / 061022-9

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 05/03/2015 Terms of Use: http://asme.org/terms

Fig. 20 Eigenvalues as a function of cross-sectional degrees

of freedom

the cross sections in the corresponding mode). The fit using a 10

parameter Fourier expansion is shown with increasing number of
nodal density in the cross section in Fig. 21. In the case of metal
laser sintering, with combinations of unmelted powder and
regions of solidified material, the number of parameters will be 13
(an additional 3 for powder thermal conductivities).
A number of simplified cases for cross sections containing powder and solidified material are shown in Fig. 22. Figure 23 shows
regions where a 10 parameter Fourier expansion could be used to
fit the eigenvalues to different cases shown in Fig. 22. The nonlinear effects close to the melt pool when solved using the
FFD-AMRD algorithm [25] do not affect the linear solution space
outside fine-scale domain. Due to the existence of only linear
solutions far from the melt pool, the macroscopic coarse-scale
powder bed eigenvalues should correctly be obtained using only
13 parameters. The change in eigenvalues in the coarser problem
with transformation of a given volume from powder to solid could
be accommodated by addition of two more parameters correlated

Fig. 21 (a)(d) Eigenvalue fitting as a function of a seven parameter Fourier expansion

061022-10 / Vol. 136, DECEMBER 2014

Transactions of the ASME

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 05/03/2015 Terms of Use: http://asme.org/terms

Fig. 22 Different cases (16) with solidified portion (in black) and powder portion (in
white) in a given cross section of the powder bed

to the location and volume fraction change from powder to the

solidified domain, thus increasing the total number of parameters
to 15 to accurately predict the powder bed transitions at a coarser
length scale.
Banded Vectorization Solver. This solver is based on the fact
that any sparse stiffness matrix from a finite element solution is
subdivided into lower triangular and upper triangular matrices

Fig. 23 Eigenvalue evolution in cross sections for cases

shown in Fig. 22. The trend is exactly the same as the completely solidified bed and has been observed to be fit individually with a 10 parameter Fourier expansion. The evolution using
a 10 parameter Fourier expansion can be used to solve large
eigenvalue problems.

Journal of Manufacturing Science and Engineering

[26]. The lower triangular matrices (lower layers) act as wave

propagators from one cross section to another cross section in the
direction of energy propagation from the applied boundary condition(s) and involve in-plane wave scattering as well. This wave
propagator generally solves for equilibrium whereas the upper
triangular wave coming back from a fixed boundary condition at
the base of a matrix redistributes the response variable (such as
deformation or temperature). Thus, the lower triangular matrix is
of primary importance for solving FEA problems and its importance is increased for AM problems since the layer-by-layer build
cross sections are normal to the applied thermomechanical boundary conditions. A new methodology has been formulated to compute the required number of essential lower triangular values,
which preserves the accuracy of the solution as well as the matrix
values required to solve the problem while eliminating a large
number of flops. The addresses of these values are stored in a
banded vectorized fashion for quasi-static and dynamic problems
to solve for deformation and/or temperatures using the solvers
described above.
In Fig. 24, the MATLAB CHOL (cholesky) algorithm has been
used to compute the lower triangular matrix of a thermal stiffness
matrix of a 3D prismatic thermal problem with a Gaussian flux
applied on its top surface. The base of the domain has been kept

Fig. 24 MATLAB generated lower triangular matrix for a sample

thermal problem with periodic boundary conditions

DECEMBER 2014, Vol. 136 / 061022-11

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 05/03/2015 Terms of Use: http://asme.org/terms

tracked for solution and flops as a function of discrete values of

logarithm of L  LT capping threshold.

Fig. 25 Preserved solution optimized lower triangular matrix

by modified CHOL (Cholesky factorization) method. Only 6.7%
of the values were required to efficiently solve the problem with
solution quality preservation.

at a temperature of 353 K with opposing faces perpendicular to x

and y directions comprising periodic boundary conditions. The
periodic boundary conditions involve constraints, which affect the
bandwidth of the thermal stiffness matrix, and therefore the lower
triangular matrix in Fig. 24 is highly populated. When performing
banded vectorization using L(row number1,:) *L(row number2,:)
capping for calculation of updated L(row number1,:) with increasing thresholds, and checking the resulting changes in quality of
the solution vector in terms of error compared to the original solution using first and second norms of the difference vector, it has
been found that the solution vector has not changed beyond
0.026% while the number of the values required to compute the
lower triangular matrix has decreased to only 6.7% of that
required by the MATLAB CHOL algorithm. The optimized matrix at
104 is shown in Fig. 25. In Fig. 26, the number of flops calculated with respect to the logarithm of L  LT capping threshold is
shown. It has been found that for the matrix represented in Fig. 25
the number of flops required to calculate that matrix is around
1.57% of the flops required for calculating the lower triangular
matrix by the Cholesky algorithm. Also, the efficiency and speed
statistics of the modified Cholesky method has been completely

Enhanced Contact Solver. A matrix based self-updating

solver has also been created to solve for adhesion/delamination
and crack growth problems relevant to solid state metal AM, support snapping in melting based metal AM and fatigue crack propagation in additively manufactured parts [44]. A simple problem
has been solved to illustrate the feasibility of this method using
linear elastic blocks of Stainless Steel 304 held on top of each
other via normal compressive force and subject to unidirectional
simple shear. The blocks have a dimension of 1 mm  1 mm
 500 lm (100 grain diameters in rolled SS 304 sheets in the z
direction). The objective is to obtain a macroscopic shear resistanceshear slip law and use this law at the interface while solving
contact FEM problems using anisotropic plasticity based constitutive laws or DDCPFEM. The shear resistance as a function of
macroscopic slip has been plotted in Fig. 27, and the deformed
configuration and reference configuration are shown in Fig. 28. It
can be clearly observed that after a certain number of slip steps,
the blocks lose their resistance in shear at the interface and the
block on top slides without much difficulty.
Effective Thermal Conductivity of Support Structure. Support
structures [4547] are designed to reduce deformation due to
residual stress for most metal AM processes such as metal powder
sintering processes. Support structures are experimentally
optimized based on trial and error strategies introduced by the
machine manufacturer and/or utilizing user experience. Several
types of geometrical variations for supports are available when
using support generation tools such as the MATERIALISE MAGICS
software tool. In order to incorporate the effects of support structures in the current thermomechanical solver, a study on their
effective thermal conductivity was undertaken. The objective of
this study is to help the support structure design process by
informing the coarser continuum scale model with the effective
thermal conductivity of a converged representative volume
element. An example support structure is shown in Fig. 29 where
the geometrical attributes of the support structure can be defined
using wall thickness and the interwall distance. For future studies,
these two sets of parameters will be modified and updated in order
to satisfy prescribed constraints. Once the geometrical attributes
of these support structures are optimized numerically and validated concurrently using experimentations, the final goal of this
study and module is to investigate topologically/geometrically
optimized organic-looking support structures which ensure faster
builds and easy removal.

Fig. 26 Efficiency and speed statistics for modified CHOL method

061022-12 / Vol. 136, DECEMBER 2014

Transactions of the ASME

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 05/03/2015 Terms of Use: http://asme.org/terms

Fig. 27 Shear resistance as a function of macroscopic slip

A brief description of the software module is presented here to

understand the methodology and implementation of this module.
In a nutshell, Fig. 30 shows the schematic representation of the
support software modules. The FFDAMRD software tool is
designed to solve thermomechanical problems in AM processes.
It is able to simulate the thermal history and residual stress for a
given CAD file along with known build process parameters and
material properties. As shown, a support structure and scan pattern
generation module has been incorporated into the current thermomechanical process solver. Thermal distributions vary as a function of scan strategies and the scan pattern generation module
provides real scan strategies from various AM processes to the
FFDAMRD tools. The thermal history and residual stress information generated from the machine build parameters are used for
efficient support structure design.
The goal of the support structure generation module is to create
a force/strength map on the downward facing surfaces of a part to
be built that guides the fitting of an optimized support structure to
restrain the generated force/strength map. One approach currently
under investigation includes placing an initial homogenous
support structure into the FFDAMRD model along with the part
geometry. An example of this initially assumed support structure
could be a block structure with powder in the intercellular spaces
as shown in Fig. 29. This block structure will be used to attach the
fabricated part to the substrate. The solidified support and the
powder trapped inside the support structure will be modeled using
appropriate thermophysical properties. A cross section of the
mesh with solidified struts and trapped powder inside the intercellular space has been shown in Fig. 31. This model will be solved
using a thermomechanical eigenmodal strategy layer-by-layer
with a contact model to account for snapping of supports at their

Fig. 28 Reference and deformed configurations during macroscopic slip

Journal of Manufacturing Science and Engineering

Fig. 29 3D model with support structure

interface with the baseplate and/or the part. The dimensions of the
solidified struts will be periodically updated as a function of layerby-layer buildup of the fabricated structure resulting in updated
element properties from powder to solid with an updated total
stiffness matrix.
Accumulated residual stress/thermal strain will be recorded
until the end of fabrication and a force/strength map will be
created in three dimensions that represent the force needed in a
particular location to hold the part in place to a prescribed
accuracy/deformation level. That force/strength map will then be

Fig. 30 Schematic of incorporating support structure and

scan pattern generation modules to the current FFDAMRD
model

DECEMBER 2014, Vol. 136 / 061022-13

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 05/03/2015 Terms of Use: http://asme.org/terms

Fig. 31 Meshing in the z-direction

used to create a second, efficient support structure. This efficient

truss structure will be used as the real support structure to be
made in the machine. Early on in this codes development, we
anticipate a possible need to iterate by running the FFDAMRD
model again using the updated support structure to make sure it
meets the anticipated loading requirements, as the change in
thermal history between the initial support structure and the
efficient support structure may be significant.
Three
Dimensional
Surface
Reconstruction
Software. Geometrical errors during 2D slicing and thermomechanical deformation during processing are two sources of inaccuracies in the parts produced using AM technologies [48]. In
most AM processes, the CAD model is represented by the STL
file format which is sliced into layers. It has been assumed that the
Delaunay triangulation in an STL file is an accurate representation
of the 3D CAD geometry. The common layer interface (CLI) and
SLI files are two common file formats containing the layer information. We have developed a software program to access the
cross-sectional information from the CLI/SLI files and stack the
extracted layers to reconstruct the 3D geometry. The geometrical

errors in the prefabrication stage can be obtained by comparing

the STL file against the stacked model. The scan strategy
including build orientation, scan angle, hatch space, and beam offsets can be also extracted from the stacked model. A sample
comparison of the original CAD model of a semisphere with the
stacked model has been shown in Fig. 32. This information is further used in thermomechanical analysis of the SLM process using
our framework. The predicted deformation using this analysis is
compared against the actual STL file, and the error will be used as
a mechanism to offset the original STL file or as a corrector for
feedforward control of the SLM process.

Conclusions
In response to an increasing need for optimization of parameters, materials and geometries for AM-fabricated parts, an integrated approach to accurately simulate the physics and materials
science aspects of AM has been undertaken. Since AM is a particularly time and resource-consuming problem to solve numerically, based upon moving energy sources and material additions
over time and space, new numerical algorithms and software

Fig. 32 Comparison of the original CAD model of a semi sphere with the stacked model. (a)
Solid CAD model and (b) stacked model from layers of cross sections with stair-step-effect
errors. The magnitude of the error increases from top to bottom.

061022-14 / Vol. 136, DECEMBER 2014

Transactions of the ASME

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 05/03/2015 Terms of Use: http://asme.org/terms

tools must be developed. Homogenization strategies applied to

some of these tools may lead to better machine architectures in
the near future, such as for modeling the multibeam capabilities
of electron beam-based and solid state AM. The numerical
algorithms and tools are generic for highly dynamic processes
with high strain and thermal gradient environments and can be
flexibly adapted for relatively quasi-static and macroscopic traditional manufacturing processes as well. The process of integration of these software tools to result in a fully functional
software suite for AM processes is underway through the
start-up company 3DSIM, LLC.

Acknowledgment
The authors would like to thank Dr. Khalid Rafi at Nanyang
Technological University, Mr. Hengfeng Gu at North Carolina
State University, Dr. Haijun Gong, and Mr. Ashabul Anam at University of Louisville for their insightful discussions with the
authors.
The authors would also like to gratefully acknowledge funding
support from the Office of Naval Research (N000141110689 and
N000140710633), the Air Force Research Laboratory (as a subcontractor to Mound Laser & Photonics Center on three SBIR
projects), the National Institute of Standards and Technology
(70NANB12H262), and the National Science Foundation
(CMMI-1234468).

References
[1] Wordenweber, B., 1984, Finite Element Mesh Generation, Comput. Aided
Des, 16(5), pp. 285291.
[2] Ho-Le, K., 1988, Finite Element Mesh Generation Methods: A Review and
Classification, Comput. Aided Des., 20(1), pp. 2738.
[3] Caendish, J. C., Field, D. A., and Frey, W. H., 1985, An Approach to Automatic Three-Dimensional Finite Element Mesh Generation, Int. J. Numer.
Methods Eng., 21(2), pp. 329347.
[4] Staten, M. L., Woodbury, A. C., Benzley, S. E., and Shepherd, J. F., 2012, U.S.
Patent No. 8,194,068.
[5] Yerry, M. A., and Shephard, M. S., 1983, A Modified Quadtree Approach to
Finite Element Mesh Generation, IEEE Trans. Comput. Graphics Appl., 3(1),
pp. 3946.
[6] Matsumoto, M., Shiomi, M., Osakada, K., and Abe, F., 2002, Finite Element
Analysis of Single Layer Forming on Metallic Powder Bed in Rapid Prototyping by Selective Laser Processing, Int. J. Mach. Tools Manuf., 42(1), pp.
6167.
[7] Shiomi, M., Yoshidome, A., Abe, F., and Osakada, K., 1999, Finite Element
Analysis of Melting and Solidifying Processes in Laser Rapid Prototyping of
Metallic Powders, Int. J. Mach. Tools Manuf., 39(2), pp. 237252.
[8] Verhaeghe, F., Craeghs, T., Heulens, J., and Pandelaers, L., 2009, A Pragmatic
Model for Selective Laser Melting With Evaporation, Acta Mater., 57(20), pp.
60066012.
[9] Zaeh, M. F., and Branner, G., 2010, Investigations on Residual Stresses and
Deformations in Selective Laser Melting, Prod. Eng., 4(1), pp. 3545.
[10] Dai, K., and Shaw, L., 2004, Thermal and Mechanical Finite Element Modeling of Laser Forming From Metal and Ceramic Powders, Acta Mater., 52(1),
pp. 6980.
[11] Kolossov, S., Boillat, E., Glardon, R., Fischer, P., and Locher, M., 2004, 3D
FE Simulation for Temperature Evolution in the Selective Laser Sintering Process, Int. J. Mach. Tools Manuf., 44(2), pp. 117123.
[12] Roberts, I. A., Wang, C. J., Esterlein, R., Stanford, M., and Mynors, D. J., 2009,
A Three-Dimensional Finite Element Analysis of the Temperature Field During Laser Melting of Metal Powders in Additive Layer Manufacturing, Int. J.
Mach. Tools Manuf., 49(12), pp. 916923.
[13] 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, Mater. Des., 52,
pp. 638647.
[14] Childs, T. H. C., Hauser, C., and Badrossamay, M., 2005, Selective Laser Sintering (Melting) of Stainless and Tool Steel Powders: Experiments and Modelling, Proc. Inst. Mech. Eng., Part B, 219(4), pp. 339357.
[15] Jhabvala, D., Boillat, E., Antignac, T., and Glardon, R., 2010, On the Effect of
Scanning Strategies in the Selective Laser Melting Process, Virtual Phys. Prot.
J., 5(2), pp. 99109.
[16] Zeng, K., Pal, D., Patil, N., and Stucker, B. E., 2013, A New Dynamic Mesh
Method Applied to the Simulation of Selective Laser Melting, Proceedings of
the Solid Freeform Fabrication Symposium, Austin, TX, Aug. 1214, pp.
549559.
[17] Pal, D., Behera, S. K., and Ghosh, S., 2009, Crystal Plasticity Modeling of
Creep and Microtwinning in Nickel Based Superalloys, 10th United States
National Congress on Computational Mechanics, Columbus, OH, July 1519.

Journal of Manufacturing Science and Engineering

[18] Pal, D., Patil, N., and Stucker, B. E., 2012, Prediction of Mechanical Properties of Electron Beam Melted Ti6Al4V Parts Using Dislocation Density Based
Crystal Plasticity Framework, Proceedings of the Solid Freeform Fabrication
Symposium, Austin, TX, Aug. 68, pp. 507525.
[19] Brandal, G., Satoh, G., Yao, Y. L., and Naveed, S., 2013, Beneficial Interface
Geometry for Laser Joining of NiTi to Stainless Steel Wires, ASME J. Manuf.
Sci. Eng., 135(6), p. 061006.
[20] Tan, H., and Yao, Y. L., 2010, Laser Joining of Continuous Glass Fiber Composite Preforms, ASME J. Manuf. Sci. Eng., 135(1), p. 011010.
[21] Satoh, G., Huang, X., Ramirez, A. G., and Yao, Y. L., 2012, Characterization
and Prediction of Texture in Laser Annealed NiTi Shape Memory Thin Films,
ASME J. Manuf. Sci. Eng., 134(5), p. 051006.
[22] Wang, H., Kongsuwan, P., Satoh, G., and Yao, Y. L., 2012, Femtosecond LaserInduced Simultaneous Surface Texturing and Crystallization of a-Si:H Thin Film:
Absorption and Crystallinity, ASME J. Manuf. Sci. Eng., 134(3), p. 031006.
[23] Wang, H., Hsu, S. T., Tan, H., Yao, Y. L., Chen, H., and Azer, M. N., 2013,
Predictive Modeling for Glass-Side Laser Scribing of Thin Film Photovoltaic
Cells, ASME J. Manuf. Sci. Eng., 135(5), p. 051004.
[24] Deng, D., and Murakawa, H., 2006, Numerical Simulation of Temperature
Field and Residual Stress in Multi-Pass Welds in Stainless Steel Pipe and Comparison With Experimental Measurements, Comput. Mater. Sci., 37(3),
pp. 269277.
[25] Patil, N., Pal, D., and Stucker, B. E., 2013, A New Finite Element Solver
Using Numerical Eigen Modes for Fast Simulation of Additive Manufacturing
Processes, Proceedings of the Solid Freeform Fabrication Symposium, Austin,
TX, Aug. 1214, pp. 535548.
[26] Nikoukar, M., Patil, N., Pal, D., and Stucker, B. E., 2013, Methods for Enhancing the Speed of Numerical Calculations for the Prediction of the Mechanical
Behavior of Parts Made Using Additive Manufacturing, Proceedings of the
Solid Freeform Fabrication Symposium, Austin, TX, Aug. 1214, pp. 525534.
[27] Asadi, M., and Goldak, J. A., 2014, An Integrated Computational Welding
Mechanics With Direct-Search Optimization for Mitigation of Distortion in an
Aluminum Bar Using Side Heating, ASME J. Manuf. Sci. Eng., 136(1),
p. 011007.
[28] Afazov, S. M., Becker, A. A., and Hyde, T. H., 2012, Mathematical Modeling
and Implementation of Residual Stress Mapping From Microscale to Macroscale Finite Element Models, ASME J. Manuf. Sci. Eng., 134(2), p. 021001.
[29] Pal, R., Anand, S., and Gerner, F., 2014, Effect of Thermal Deformation on
Part Errors in Metal Powder Based Additive Manufacturing Processes, ASME
J. Manuf. Sci. Eng., 136(3), p. 031009.
[30] McCune, R. W., Murphy, A., Price, M., and Butterfield, J., 2012, The Influence of Friction Stir Welding Process Idealization on Residual Stress and Distortion Predictions for Future Airframe Assembly Simulations, ASME J.
Manuf. Sci. Eng., 134(3), p. 031011.
[31] Pal, D., and Stucker, B. E., 2012, Time Homogenization of Al3003 H-18 Foils
Undergoing Metallurgical Bonding Using Ultrasonic Consolidation, Proceedings of the Solid Freeform Fabrication Symposium, Austin, TX, Aug. 68, pp.
373384.
[32] Pal, D., and Stucker, B. E., 2011, Some Studies on Dislocation Density Based
Finite Element Modeling of Ultrasonic Consolidation, Proceedings of the
5th International conference on Virtual and Rapid Prototyping, Leiria, Portugal,
Sept. 28Oct. 1, pp. 667676.
[33] Pal, D., and Stucker, B. E., 2011, Dislocation Density Based Finite Element
Modeling of Ultrasonic Consolidation, Proceedings of the Solid Freeform Fabrication Symposium, Austin, TX, Aug. 810, pp. 375404.
[34] Pal, D., and Stucker, B. E., 2010, Dislocation Density Based Finite Element
Modeling of Ultrasonic Consolidation, Proceedings of the Solid Freeform Fabrication Symposium, Austin, TX, Aug. 911, pp. 583591.
[35] Pal, D., and Stucker, B. E., 2012, Modeling of Ultrasonic Consolidation Using
a Dislocation Density Based Finite Element Framework, Virtual and Physical
Prototyping Journal, 7(1), pp. 6579.
[36] Johnson, K. E., 2008, Interlaminar Subgrain Refinement in Ultrasonic Consolidation, Ph.D. thesis, Loughborough University, Loughborough, UK.
[37] Ding, H., and Shin, Y. C., 2014, Dislocation Density-Based Grain Refinement
Modeling of Orthogonal Cutting of Titanium, ASME J. Manuf. Sci. Eng.,
136(4), p. 041003.
[38] Ding, L., Zhang, X., and Liu, C. R., 2014, Dislocation Density and Grain Size
Evolution in the Machining of Al6061-T6 Alloys, ASME J. Manuf. Sci. Eng.,
136(4), p. 041020.
[39] Hammami, W., Gilles, G., Habraken, A. M., and Duch^ene, L., 2011,
Phenomenological and Crystal Plasticity Approaches to Describe the Mechanical Behaviour of Ti6Al4V Titanium Alloy, Int. J. Mater. Form., 4(2), pp.
205215.
[40] Bridier, F., McDowell, D. L., Villechaise, P., and Mendez, J., 2009, Crystal
Plasticity Modeling of Slip Activity in Ti6Al4V Under High Cycle Fatigue
Loading, Int. J. Plast., 25(6), pp. 10661082.
[41] Barton, N. R., Bernier, J. V., Lebensohn, R. A., and Rollett, A. D., 2009,
Direct 3D Simulation of Plastic Flow From EBSD Data, Electron Backscatter
Diffraction in Materials Science, Springer, New York, pp. 155167.
[42] Berdichevskii, V. L., 1979, Variational-Asymptotic Method of Constructing a
Theory of Shells, J. Appl. Math. Mech., 43(4), pp. 711736.
[43] Saad, Y., 1992, Numerical Methods for Large Eigenvalue Problems, Vol. 158,
Manchester University Press, Manchester, UK.
[44] Patil, N., Pal, D., and Stucker, B. E., 2013, An Energy Dissipative Constitutive
Model for Multi-Surface Interface at Weld Defect Sites in Ultrasonic Consolidation Process, Proceedings of the Solid Freeform Fabrication Symposium,
Austin, TX, Aug. 1214, pp. 684696.

DECEMBER 2014, Vol. 136 / 061022-15

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 05/03/2015 Terms of Use: http://asme.org/terms

[45] Byun, H. S., and Lee, K. H., 2006, Determination of the Optimal Build Direction for Different Rapid Prototyping Processes Using Multi-Criterion Decision
Making, Rob. Comput. Integr. Manuf., 22(1), pp. 6980.
[46] Vayre, B., Vignat, F., and Villeneuve, F., 2012, Metallic Additive
Manufacturing: State-of-the-Art Review and Prospects, Mech. Ind., 13(2),
pp. 8996.

061022-16 / Vol. 136, DECEMBER 2014

[47] Gibson, I., Rosen, D. W., and Stucker, B., 2010, Additive Manufacturing Technologies, Springer, New York.
[48] Zeng, K., Patil, N., Gu, H., Gong, H., Pal, D., Starr, T., and Stucker, B. E.,
2013, Layer by Layer Validation of Geometrical Accuracy in Additive Manufacturing Processes, Proceedings of the Solid Freeform Fabrication
Symposium, Austin, TX, Aug. 1214, pp. 7687.

Transactions of the ASME

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 05/03/2015 Terms of Use: http://asme.org/terms