Int. J. Materials and Product Technology, Vol. x, No.

x, xxxx 1

Modelling Microstructures with OOF2

Andrew C.E. Reid

Information Technology Laboratory,
National Institute of Standards and Technology,
100 Bureau Drive, Stop 8910,
Gaithersburg, MD, 20899-8910, USA
E-mail: [email protected]

Rhonald C. Lua
Materials Science and Engineering Laboratory,
National Institute of Standards and Technology,
100 Bureau Drive, Stop 8555,
Gaithersburg, MD, 20899-8555, USA
E-mail: [email protected]

R. Edwin García
School of Materials Engineering,
Purdue University,
West Lafayette, IN, 47907-2044, USA
E-mail: [email protected]

Valerie R. Coffman
Materials Science and Engineering Laboratory,
National Institute of Standards and Technology,
100 Bureau Drive, Stop 8555,
Gaithersburg, MD, 20899-8555, USA
E-mail: [email protected]

Stephen A. Langer*
Information Technology Laboratory,
National Institute of Standards and Technology,
100 Bureau Drive, Stop 8910,
Gaithersburg, MD, 20899-8910, USA
Fax: 301 975 3553 E-mail: [email protected]
*Corresponding author

Copyright © 200x Inderscience Enterprises Ltd.

2 A.C.E. Reid et al.

Abstract: OOF2 is a program designed to compute the properties and local

behaviour of material microstructures, starting from a two-dimensional
representation, an image, of arbitrary geometrical complexity. OOF2 uses the
finite element method to resolve the local behaviour of a material, and is
designed to be used by materials scientists with little or no computational
background. It can solve for a wide range of physical phenomena and can be
easily extended. This paper is an introduction to some of its most basic and
important features.

Keywords: finite element; microstructure.

Reference to this paper should be made as follows: Reid, A.C.E., Lua, R.C.,
Edwin García, R., Coffman, V.R. and Langer, S.A. (xxxx) ‘Modelling
Microstructures with OOF2’, Int. J. Materials and Product Technology, Vol. x,
No. x, pp.xxx–xxx.

Biographical notes: Andrew C.E. Reid has been an active developer of

computational modelling tools in the field of materials science for over
ten years, including six years with the OOF project at NIST. He holds a PhD in
Condensed-Matter Physics from Queen’s University in Kingston, Canada.

Rhonald C. Lua developed computational tools for studying knots in lattices,

proteins and curves in his PhD research at the University of Minnesota before
joining the OOF group at NIST. He also did analytical and computational work
on first passage times in diffusion.

R. Edwin García is an Assistant Professor in Materials Engineering at Purdue

University in West Lafayette, Indiana, USA. He earned his PhD in Materials
Science and Engineering at the Massachusetts Institute of Technology in 2003.
His research interests include the theoretical and numerical modelling of
materials of complex microstructural features in batteries, actuators, and
nanowires.

Valerie R. Coffman is an NRC post-doctoral researcher at NIST. She earned

her PhD in Physics from Cornell University in 2006, studying the macroscopic
effects of atomic scale defects in crystals.

Stephen A. Langer is a Physicist in the Mathematical and Computational

Sciences Division of the Information Technology Laboratory at the National
Institute of Standards and Technology. He received his PhD in Physics from
Cornell University in 1989.

1 Introduction

There are two general approaches to modelling the properties of materials with complex
microstructures. One is to use mean-field approximations and compute the average
properties of a statistically representative microstructure. The second is to accurately
model a specific microstructure (or set of microstructures), using all available
geometrical data. While the mean field method provides useful bounds to calculate the
effective properties and response of materials, the second method is useful in situations in
which
 Modelling Microstructures with OOF2 3

• the macroscopic properties are a non-linear function of the underlying properties

• the underlying properties are spatially non-uniform
• it is not clear which properties should be averaged (do all grains in an ‘average’
microstructure have the same size, shape, orientation, and modulus, or do they have
a distribution of all or some of those properties?), or
• the properties of interest arise from the extremes of a property distribution (a failure
mechanism might depend on the presence of a single point of extreme stress).
The goal of the Object Oriented Finite (OOF) element analysis project at NIST
(http://www.ctcms.nist.gov/oof/) is to model microstructures using the second, or
direct-computation, approach. With OOF, the user assigns material properties to the
features in an experimental or simulated micrograph, generates a finite element mesh,
and performs virtual experiments. OOF is used to visualise the microscopic response of
the microstructure to external conditions, to compute effective macroscopic material
properties, or to design material microstructures of optimal (tailored) performance.
Historically, OOF was developed to perform linear elasticity and thermal conductivity
calculations on arbitrary two-dimensional microstructures (Langer et al., 2001).
As additional physics was added to OOF, it became clear that extending the OOF1
framework would be a complex and cumbersome task that would make the code difficult
to maintain. OOF2, a complete re-write of OOF1, was designed to have a more powerful
and much more easily extendible software infrastructure while retaining OOF1’s
user-friendly Graphical User Interface (GUI). It differs from OOF1 in its object-oriented
modular design, which makes use of the Python and C++ programming languages.
OOF2 currently solves linear elasticity, thermal conductivity, electric polarisation,
and piezoelectricity problems in two dimensions, and can easily be extended to include
any problem that can be expressed in the schematic form
Ψ j = ∑ K ij ∇ϕi
i

∇•Ψj = fj

where Ψj is a flux, ϕi is a field, Kij is a modulus or coupling constant, and fj is a

generalised applied force. For example, to describe the mechanical equilibrium of
a material subjected to an external load, the field is displacement, the modulus is the
elastic stiffness tensor, the ‘gradient’ of the field is the strain, and the ‘flux’ is the stress.
In general, the flux is a sum over contributions from many thermodynamic fields.
OOF2 material properties are specified in three-dimensional form, and after assigning
them to an image they are reduced to two dimensions by specifying either that fields have
no out-of-plane gradients (e.g., plane-strain) or that the fluxes have no out-of-plane
components (e.g., plane-stress).
This paper briefly discusses some of the most important and fundamental features of
OOF2. For more details, the reader is referred to the OOF2 manual, which may be found
at the OOF website (http://www.ctcms.nist.gov/oof/), along with the source code and
installation instructions. OOF2 is distributed at no charge. It will run on any Unix system,
including Linux, Mac OS X, and, with modifications, on Windows with the Cygwin
environment (http://www.cygwin.com/). Installation requires the presence of a C++
compiler, a Python interpreter (http://www.python.org/), the X11 window system,
4 A.C.E. Reid et al.

and a small number of additional free, open-source libraries. What follows is a brief
description of using OOF2 to analyse a particular microstructure.

2 Using OOF2

OOF2 starts with a micrograph, such as the one shown in Figure 1. While this particular
micrograph happens to be of a silicon nitride ceramic, Si3N4, the topology of the
images that OOF2 can analyse is arbitrary, and is only limited by the resolution and
quality of the input. By using simple-point-and-click operations, OOF2 can assign any
desired (real or hypothetical) single-crystal material properties to each of the components
of the microstructure, and rapidly examine the effects of those properties on the local
microstructural fields or the macrosopic response.

Figure 1 A Si3N4 microstructure, courtesy of Paul et al. (1998). The dark regions are a crystalline
phase, and light regions are glassy amorphous interfacial regions

3 Microstructures

The fundamental abstract entity used by OOF2 is a Microstructure (with a capital

M), just as the fundamental object manipulated by a word processor is a document.
A Microstructure is a rectangular array of pixels, where each pixel carries
information about the local material properties. A Microstructure may (but does not
have to) contain one or more Images. An Image (an experimental micrograph or the
output of a simulation) is the usual starting point for an OOF2 calculation, and OOF2 can
create a Microstructure object directly from an Image file. Most common image
file formats are accepted. If different image processing techniques are needed to bring out
different features of a (small m) microstructure, the (capital M) Microstructure may
contain more than one Image, as long as they all have the same pixel size.
When an Image is loaded into OOF2, its physical width and height can be specified,
in any type of units. OOF2 contains no fixed system of units. The only requirement is that
all numerical quantities must be entered using a consistent set of units. The width and
height do not have to be commensurate with the Image’s size in pixels; that is, the pixels
do not have to be physically square.
 Modelling Microstructures with OOF2 5

Microstructures can also be created without using an Image, by simply

specifying their sizes both in physical units and in pixels.

4 Materials, pixel selections, and pixel groups

After creating a Microstructure, the next required task for the OOF2 user is to
assign Materials to each pixel in the Microstructure. Materials are defined
as sets of Properties. Most Properties are coefficients in physical constitutive
relations, such as elastic modulus, thermal conductivity, piezoelectric coupling tensor, or
mass density. Other Properties, such as Orientation, indirectly affect the
constitutive equations, while others, such as Colour, are purely decorative. Figure 2
shows a possible ‘material map’ of the Microstructure from Figure 1, in which each
pixel has been coloured by the Colour Property assigned to its Material.

Figure 2 A ‘Material map’ constructed from the Microstructure shown in Figure 1.

The colour of the pixels indicates what Material has been assigned to them:
(a) shows the entire Microstructure and (b) is a detail demonstrating unphysical
jagged edges

(a) (b)

When assembling a Material from Properties, the Properties are chosen from
a hierarchical list. The Properties can be parameterised, copied, named, and saved
for future reference. New Properties can be added to the hierarchy. At the current
time, the hierarchy includes mechanical properties (linear elasticity, stress-free strain, and
force density), thermal properties (thermal conductivity and heat sources), electric
properties (dielectric permittivity and space charge density), couplings (thermal
expansion, piezoelectricity, and pyroelectricity), and the non-constitutive properties
orientation and colour. All of the moduli and coupling constants have both isotropic
and anisotropic variants, appropriate for any crystal symmetry. Although OOF2 only
solves two dimensional problems, crystal symmetries and orientations have their full
three-dimensional form.
The pixels to which the Materials are assigned correspond to microstructural
constituents (grains, grain boundaries, precipitate phases, etc.). These sets of pixels are
identified and grouped by the user using OOF2’s pixel selection tools. This operation in
general requires some judgement on the part of the user to physically interpret the image
6 A.C.E. Reid et al.

data, although there may be special cases where fully automatic pixel grouping is
possible.
The OOF2 pixel selection tools include both graphical (operated by mouse clicks on
an Image) and non-graphical operations, and tools to create and modify named groups
of pixels, as well as tools for modifying the underlying Images. The graphical pixel
selection tools include methods for selecting all pixels within a given colour range of a
clicked pixel, and various methods for choosing a contiguous set of pixels. Non-graphical
selection tools include methods for expanding or contracting the currently selected set, or
selecting all pixels within a given absolute colour range.
Selected pixels can be added to or subtracted from named pixel groups. Creating pixel
groups is not required, but is a convenient way to recover a previous set of selected
pixels. Furthermore, Materials can be assigned to a named pixel group, as well as to
whichever pixels are currently selected. Figure 2 was constructed in one of the simplest
possible ways. First, the Image was thresholded; i.e., each pixel was set to either black
or white, depending on whether its original grey level was below or above a certain
predetermined value. Then two pixel groups named ‘grains’ and ‘matrix’ were created.
The black pixels were selected and added to the ‘grains’ group, and the white pixels were
selected and added to the ‘matrix’ group. Two Materials, also called ‘grains’ and
‘matrix’ were created, and assigned to the pixels in the corresponding pixel groups.
Finally, Colour Properties were added to the two Materials.

5 Skeletons

After creating a Microstructure and assigning Materials to its pixels, the next
step in using OOF2 is to create a finite element mesh Skeleton. A Skeleton
is not quite a complete finite element mesh – it specifies the positions and shapes
of the elements, but contains no information about finite element interpolation functions,
which fields are defined, or which equations are being solved. Like Images and
Materials, Skeletons are components of an OOF2 Microstructure object.
One Microstructure can be home to many Skeletons. After a Skeleton has
been created, it can be adapted via node motion and element refinement so that
it provides a good geometrical representation of its associated microstructure.
A Skeleton is a good representation if each of its elements is homogeneous, meaning
that its area consists of only one type of Material in the Microstructure. In the
end, the Material assigned to each Skeleton element, whether or not the element is
homogeneous, will be the Material assigned to the majority of pixels within the
element (pixels at element edges can make fractional contributions to this calculation).
There are three ways to create a Skeleton from a Microstructure in OOF2.
The simplest creates a single quadrilateral element or two triangular elements from each
pixel. While this process is straightforward and requires no user intervention, in most
cases it almost certainly will create an unnecessary number of elements, which will slow
down further computations. Additionally, it will introduce internal jagged edges, which
are unphysical and do not represent the topology of the analysed image. For example, the
jagged edges of the grains in Figure 2(b) are just a result of the pixelisation of the original
image, and it makes no sense for a mesh to follow the stairstep boundaries pixel by pixel.
 Modelling Microstructures with OOF2 7

A better way to create a Skeleton is to use OOF2’s ‘Auto’ feature, which requires
the user to specify the minimum and maximum sizes of the features of interest in the
Microstructure. OOF2 will then create an initial Skeleton just fine enough to
resolve the maximum sized features, and will locally refine and adapt the Skeleton so
that it can resolve the minimum sized features. Such a Skeleton is shown in Figure 3.
Note that most, but not all, of the elements have good shapes (containing no highly
obtuse or acute angles, and having aspect ratios near 1.0) and that most, but not all, of the
boundaries between Materials coincide with element edges.

Figure 3 (a) A Skeleton created from the Microstructure shown in Figure 2(a), using the Author: Please
Auto Skeleton feature. In this and the following figures, the jagged edges in the element check if Figure 3
boundaries are display artifacts; (b) of the Skeleton, corresponding to the region and 6 captions are
shown in Figure 2(b) and (c) with elements coloured according to their dominant ok.
Material

(a) (b) (c)

The third way to create a Skeleton is to do it by hand, first creating a regular array of
large elements, and then using various Skeleton modification tools to adapt the
Skeleton to the Microstructure. It is also possible to use the same tools to
improve a Skeleton that was initially created automatically. The tools include ways of
refining, or subdividing, inhomogeneous elements, various methods for moving nodes to
improve element homogeneity or shape, and methods for merging elements and removing
extremely badly shaped elements. It is possible to select a portion of the
Microstructure and to apply the tools only to elements in that area, or even to select
an individual Skeleton node and specify its desired position exactly. Figure 4 shows
the result of using some of these tools on a portion of the Skeleton shown in Figure 3.
The new Skeleton follows the Material boundaries better, but in places comes
perilously close to resolving the individual pixels.
When creating a Skeleton, it is important to remember that the original micrograph
is an approximation to the actual microstructure, the material map is an approximation to
the micrograph, and the Skeleton is an approximation to the material map.
Furthermore, the finite element solution that will be obtained is yet another
approximation. In this context, there is nothing to be gained by over refining a
Skeleton.
8 A.C.E. Reid et al.

Figure 4 A detail (a) of the Skeleton from Figure 3, with additional refinement
and (b) the same detail, with elements coloured according to their dominant Material

(a) (b)

6 Meshes, fields, equations, boundary conditions and solutions

Once a satisfactory representation is obtained, an actual finite element Mesh can be

created from a Skeleton. The Mesh adds physics and math to the Skeleton
geometry to create a fully functional finite element mesh. Many Meshes may be made
from a single Skeleton. In each Mesh, one Mesh element is created for each
Skeleton element, and one Mesh node is created for each Skeleton vertex.
Additional nodes are created on Mesh edges and within elements if required by the finite
element interpolation order specified during Mesh construction. OOF2 currently supports
three- and six-noded triangular elements and four-, eight-, and nine-noded quadrilateral
elements.
After creating a Mesh, the next step is to define and activate Fields. In OOF2,
Fields are the spatially distributed quantities which determine the local physical state
of a system. Fields in this sense are distinct from the usual fields (with a lowercase f ).
For example, stress is a field in the sense that it has values everywhere in the continuum,
but it is not a Field, because in OOF2 it is a computed secondary characteristic and not
a fundamental quantity.
OOF2 currently contains displacement, temperature, and voltage Fields. A Field
is said to be defined on a Mesh if the values of its components are stored in the Mesh
nodes. A defined Field can be evaluated at any point within a Mesh by using the finite
element interpolation functions. Field values can be initialised to fixed values, to
Python functions of x and y, or by copying from another Mesh (and interpolating if
necessary). Fields are said to be active if their values are to be obtained from the
finite element solution. Finally, Fields are in-plane if they are constrained to have no
out-of-plane derivatives. This generalises the concept of plane-strain.
OOF2 can solve two classes of Equations: divergence equations and plane-flux
(generalised plane-stress) equations. Divergence equations specify that the divergence of
some flux equals the applied forces. The built-in divergence equations are the heat
equation (Fick’s second law), the force balance (mechanical equilibrium) equation, and
Coulomb’s equation in its differential form. The exact form of the equations depends on
the Properties defined in the Microstructure, and on which Fields are
defined and active. The moduli and body forces at any point will be zero unless the local
 Modelling Microstructures with OOF2 9

Material contains the corresponding Properties. For example, an elastic and

thermal problem may have both temperature and displacement Fields, and elasticity
and thermal-conductivity properties. This is by itself a well-posed problem, but there will
be no contribution to the stress from changes in temperature unless a thermal-expansion
property, with associated modulus, is also included in the local Material.
Plane-flux equations specify that the out-of-plane component of a Flux (stress, heat
flux, etc.) is zero. For mechanical problems, plane stress is often imposed by constructing
effective in-plane moduli which do not contribute to out-of-plane stresses, and then
solving an effective two-dimensional finite element problem (Zienkiewicz and Taylor,
2000). OOF2 does not do this, because the form of the effective moduli depends upon
which couplings are present, and these are not known until run-time. Instead, OOF2
solves plane-flux equations along with the divergence equations within the finite-element
scheme, which achieves the same effect in a much more general context. This means that
the out-of-plane Fluxes will in general not be pointwise identically zero, but they will
tend towards zero as the size of the Mesh elements goes to zero and the finite element
approximation improves.
OOF2 supports five kinds of boundary conditions: Dirichlet, Neumann, floating,
generalised force, and periodic. It also supports two kinds of boundaries: point and edge.
Point boundaries are named collections of element nodes, and edge boundaries are named
collections of ordered contiguous element edges. Boundaries are defined in Skeletons
and implicitly defined in Meshes derived from the Skeletons. Each Skeleton has
eight predefined boundaries: edge boundaries at the top, left, bottom, and right edges, and
point boundaries each consisting of a single node at each of the four corners of the
Microstructure. New boundaries can be defined, and need not lie along the actual
external edges of the Microstructure.
Dirichlet boundary conditions specify the values of Field components at
boundaries. Neumann conditions specify the normal components of Fluxes. Floating
boundary conditions specify Field values, but only up to an unspecified offset, which
will be determined as part of the solution. Generalised force conditions specify the
divergence of a Flux at each node of a point boundary. Periodic boundary conditions, as
expected, enforce the equality of Field values at paired boundaries. They can only be
applied to Meshes constructed from Skeletons that were declared to be periodic
when they were built.
All of the boundary conditions, except periodic, can be specified in terms of an
arbitrary ‘profile’ function, which determines how the Field or Flux value varies
along the boundary. Profiles can be specified in terms of arbitrary Python functions of
position, absolute arc length, or fractional arc length along the boundary.
Figure 5 shows the Si3N4 microstructure from Figure 1 after assigning fictional
material parameters and equilibrating. The dark grains are isotropically elastic with
Young’s modulus E = 10 and Poisson’s ratio ν = 0.3, and have an isotropic thermal
expansion coefficient alpha = 1.0. (The units are arbitrary, as mentioned above. In these
units, the size of the whole undistorted microstructure is 200 × 200.) The matrix between
the grains is softer than the grains, with E = 0.1 and ν = 0.45, and has no thermal
expansion. Dirichlet boundary conditions were applied to the bottom boundary, setting
the x component of displacement to zero and constraining y to follow a parabolic
profile. All other boundaries were free. The temperature was defined and initialised to
0.1 everywhere, and the displacement was computed. OOF2 solved an 8966 × 8966
10 A.C.E. Reid et al.

matrix equation, which converged to a relative tolerance of 10–13 in 2191 iterations of

a conjugate gradient solver. The results are shown in Figure 5.

Figure 5 (a) A Microstructure deformed by Dirichlet boundary conditions on the bottom

edge, and thermal expansion of the grains; (b) the xx component of the total strain.
Values range from black (–0.423, tension) through red and yellow to white
(0.368, compression); (c) the trace of the geometric strain. Black = –0.239
and white = 0.31 and (d) the trace of the stress tensor. Black = –5.51 and white = 0.31

(a) (b)

(c) (d)

7 Output and analysis

OOF2 contains a number of methods for examining the results of a virtual experiment.
Figure 5 displays a few of the ways of visualising an equilibrated Mesh. OOF2
can plot the components and invariants of all Fields and Fluxes, as well as energy
densities and strains, using either original, actual deformed, or enhanced coordinates.
(Enhanced coordinates exaggerate displacements.) It can plot cross sections along
arbitrary straight lines drawn on the Microstructure, and can compute averages and
integrals over Microstructures, cross sections, subregions, and boundaries of a
Mesh. This allows it to be used to compute effective macroscopic moduli.
 Modelling Microstructures with OOF2 11

8 GUI and scripting

OOF2 can be driven either by a GUI, or via scripts. Figure 6 illustrates some features
of the user interface.

Figure 6 (a) The main OOF2 control window. The window has many separate pages,
for performing the various required tasks. The page shown creates and modifies
Skeletons and (b) the OOF2 graphics window, which displays microstructures
and handles interactive mouse-based input. Different tasks are handled by different
toolboxes on the left hand side of the window. The toolbox shown is showing the
attributes of the Mesh element under the mouse in the display area

(a)

(b)
12 A.C.E. Reid et al.

Every action in the GUI that has a consequential outcome is recorded as a text command,
which can be saved in a log file. The file can be read back into the program to repeat a
calculation or part of one. Log files are actually Python scripts, so they can
be edited to include arbitrary Python variables, functions, and control statements.
For example, a script can be saved for doing a calculation on one image, and then
modified to run the calculation on a whole series of images, or with a range of material
parameters. Scripts can be loaded either with or without the GUI running, allowing
background batch computation.

9 OOF2 extensions

Many of the features of OOF2 are designed to be easily extendible by end users.
It’s possible to add new Fields, Fluxes, or Equations with a few lines of code.
New Material Properties, output functions, and finite element types can be added
with a bit more effort. The details, including sample code, are given in the on-line OOF2
manual. Extensions can either be compiled into the main OOF2 code base, or built as
separate modules that can be loaded at run-time if and when they are needed.

10 Current and future development

At the current time, OOF2 development is concentrated on adding time dependent and
non-linear properties, such as plasticity. OOF2 will soon be able to directly
importorientation imaging microscopy (OIM) data. Finally, OOF2 is being extended to
work on three dimensional micrographs. The long-term goal is to make OOF a general
purpose platform for the computation of physical properties of two and three dimensional
microstructures.

Acknowledgements

The authors would like to thank Seung-Ill Haan for his contributions to the OOF2 code.
Rhonald Lua is supported by the Penn State University MatCASE project, which is
funded by NSF ITR research grant DMR-0205232.

Disclaimer

Certain commercial equipment, instruments, or materials are identified in this paper to

specify adequately the experimental procedures and conditions. Such identification does
not imply recommendation or endorsement by the National Institute of Standards and
Technology, nor does it imply that the materials or equipment identified are necessarily
the best available for the purpose.
 Modelling Microstructures with OOF2 13

References
Langer, S.A., Fuller Jr., E.R. and Carter, W.C. (2001) ‘OOF: an image-based finite-element
analysis of material microstructures’, Computing in Science and Engineering, Vol. 3, No. 3,
p.15.
Paul, F.B., Sun, E.Y., Plucknett, K.P., Alexander, K.B., Hsueh, C-H., Lin, H-T., Waters, S.B.,
Westmoreland, C.G., Kang, E-S., Hirao, K. and Brito, M.E. (1998) Journal of the American
Ceramic Society, Vol. 81, No. 11, pp.2821–2830. AUTHOR PLEASE SUPPLY ARTICLE
TITLE.
Zienkiewicz, O.C. and Taylor, R.L. (2000) The Finite Element Method, 5th ed., Vol. 1, Butterworth
Hienemann, Oxford, p.91.

Websites
Cygwin may be found at http://www.cygwin.com/
Python may be found at http://www.python.org/
The OOF website is located at http://www.ctcms.nist.gov/oof/