Data Structures and Workflows for ICME
23
using reusable libraries [10, 11]. The Multiphysics Object Oriented Simulation
Environment (MOOSE) is a similar package that relies on a generic software architecture, utilizing Jacobian-free Newton-Krylov methods [12, 13]. Other approaches
besides FEM exist for solving systems of partial differential equations. For example,
problems may be recast into convolutional forms, allowing for solutions using
spectral (Fourier-based) solvers. Examples include simulating the elastic response
of composite materials [14, 15], eigenstrains in thermal barrier coatings [16],
and the viscoplastic response of polycrystals [17, 18]. The Düsseldorf Advanced
Materials Simulation Kit (DAMASK) implements the spectral approach for solving
the polycrystalline elasto-viscoplastic problem in an open-source format [19].
First principle and small-scale simulation tools also have a wide use within
research and development. The Vienna Ab initio Simulation Package (VASP) is
a broad toolset for electronic structure calculations, with capabilities for computing
energy functionals, optical properties, and many-body problems [20]. The Largescale Atomic/Molecular Massively Parallel Simulator (LAMMPS) is an opensource tool developed by Sandia National Laboratories for molecular dynamics
problems [21]. For investigating dislocation dynamics, the open-source ParaDiS tool
is available from Lawrence Livermore National Laboratory [22].
While the above software packages are only a small sampling of the toolsets
available to a materials researcher, their outputs produce highly disparate data
streams. For example, data from an FEM simulation are topologically organized
onto a mesh, which may consist of a variety of unit element types (triangles,
quadrilaterals, tetrahedra, hexahedra, etc.). However, a spectral solver requires
data on a regular grid. Data may even exist on line segments, as in dislocation
dynamics, or points, as in molecular dynamics. Additionally, data may be scalar
(e.g., temperature fields from a DEFORM ® forging simulation), vector (e.g.,
atomic displacement vectors from LAMMPS), or tensorial (e.g., strain tensors
from polycrystalline viscoplasticity evaluated using DAMASK). Simulation data
are also typically time dependent; this results in an additional dimension, which,
in certain models, may also result in a change in geometry. The same variety
of data is observed for characterization information. For example, atom probe
tomography yields information about points in space (i.e., atoms), while computed
tomography produces volumetric images. Electron backscatter diffraction (EBSD)
scans yield orientation data on regular grids, which can be represented in only three
numbers. However, the original Kikuchi patterns, which are of significant interest in
applications such as high-resolution strain imaging, may be images that are upwards
of 1024×1024 in dimension. If storage of these original patterns is desired, then the
EBSD scan would store a pattern image at each grid location. The diversity of data
forms generated by simulation packages and characterization techniques presents a
unique integration challenge for downstream analytic tools.
3.1 Analytic Tools
Unlike the wealth of tools available for materials simulation, software specifically
for materials analytics is a relatively nascent field. Historically, processing data
23
using reusable libraries [10, 11]. The Multiphysics Object Oriented Simulation
Environment (MOOSE) is a similar package that relies on a generic software architecture, utilizing Jacobian-free Newton-Krylov methods [12, 13]. Other approaches
besides FEM exist for solving systems of partial differential equations. For example,
problems may be recast into convolutional forms, allowing for solutions using
spectral (Fourier-based) solvers. Examples include simulating the elastic response
of composite materials [14, 15], eigenstrains in thermal barrier coatings [16],
and the viscoplastic response of polycrystals [17, 18]. The Düsseldorf Advanced
Materials Simulation Kit (DAMASK) implements the spectral approach for solving
the polycrystalline elasto-viscoplastic problem in an open-source format [19].
First principle and small-scale simulation tools also have a wide use within
research and development. The Vienna Ab initio Simulation Package (VASP) is
a broad toolset for electronic structure calculations, with capabilities for computing
energy functionals, optical properties, and many-body problems [20]. The Largescale Atomic/Molecular Massively Parallel Simulator (LAMMPS) is an opensource tool developed by Sandia National Laboratories for molecular dynamics
problems [21]. For investigating dislocation dynamics, the open-source ParaDiS tool
is available from Lawrence Livermore National Laboratory [22].
While the above software packages are only a small sampling of the toolsets
available to a materials researcher, their outputs produce highly disparate data
streams. For example, data from an FEM simulation are topologically organized
onto a mesh, which may consist of a variety of unit element types (triangles,
quadrilaterals, tetrahedra, hexahedra, etc.). However, a spectral solver requires
data on a regular grid. Data may even exist on line segments, as in dislocation
dynamics, or points, as in molecular dynamics. Additionally, data may be scalar
(e.g., temperature fields from a DEFORM ® forging simulation), vector (e.g.,
atomic displacement vectors from LAMMPS), or tensorial (e.g., strain tensors
from polycrystalline viscoplasticity evaluated using DAMASK). Simulation data
are also typically time dependent; this results in an additional dimension, which,
in certain models, may also result in a change in geometry. The same variety
of data is observed for characterization information. For example, atom probe
tomography yields information about points in space (i.e., atoms), while computed
tomography produces volumetric images. Electron backscatter diffraction (EBSD)
scans yield orientation data on regular grids, which can be represented in only three
numbers. However, the original Kikuchi patterns, which are of significant interest in
applications such as high-resolution strain imaging, may be images that are upwards
of 1024×1024 in dimension. If storage of these original patterns is desired, then the
EBSD scan would store a pattern image at each grid location. The diversity of data
forms generated by simulation packages and characterization techniques presents a
unique integration challenge for downstream analytic tools.
3.1 Analytic Tools
Unlike the wealth of tools available for materials simulation, software specifically
for materials analytics is a relatively nascent field. Historically, processing data