PROCEEDINGS, Twenty-ThirdWorkshop on GeothermalReservoir Engineering

Stanford University, Stanford, California, January 26-28, 1998

SGP-TR-158

A GEOMETRIC MODELING FRAMEWORK FOR THE NUMERICAL

ANALYSIS OF GEOTHERMAL RESERVOIRS
Brian Hardeman
Daniel Swenson
Kansas State University
302 Rathbone Hall
Manhattan, KS 66506
http://www.mne.ksu.edu/-geocrack
not met, the mesh or grid must be regenerated,
incurring a substantial cost in time and effort.

ABSTRACT
In the past decade, advancements in automatic mesh
generation and topological data structures have made
possible the use of a more general and abstract
geometric model for the description of an analysis
problem. Using a geometric model for describing a
reservoir results in a conceptual model of the
geothermal system, rather than a simple numerical
simulation. The analyst builds the geometric model
using true features of a reservoir, such as well bores
and known fracture locations. Material properties
and boundary conditions are then assigned to these
features, not to the underlying finite element mesh or
finite difference grid. This independent storage of
the problem description makes it simple to run
multiple analyses, potentially using different solution
schemes. Definition of both 2-d and 3-d models can
be greatly simplified through this more interactive,
intuitive model creation process. State-of-the-art
visualization and manipulation methods assist in
conveying the conceptual model of the reservoir and
the assumptions made for performing a simulation.

Relying on the input grid or mesh for problem

information can also cause difficulty when attributes
(material properties, boundary conditions, analysis
assumptions) or geometry change during the course
of an analysis. Placing all of the problem description
in the mesh requires a change to the mesh definition.
For example, to change attribute data, one must only
change a numeric value stored with the mesh
elements, but this change must occur for every
element in the region of the change.
While
interactive tools for querying and manipulating these
values can help, the task is still tedious, particularly
for meshes with a great level of detail and a fine
discretization. Even if the interface allows the user to
specify the new value and lets the software determine
which elements require updating, the program must
still perform repeated searches to find the affected
elements.
A change in geometry is even more serious and user
intensive. Because of the close link between the
location and size of elements and the physical
boundaries of the problem, a change in geometry
requires regeneration of the mesh. Such changes
cause problems for finite element meshes, which
must conform to boundaries, and can also affect finite
difference grids, where the grid density may be
closely related to high solution gradients caused by
boundary conditions or boundaries of different
materials. For many analysis systems based on input
files, this would involve recreating the entire grid and
reassigning all attributes. Even in systems which
allow interactive modification of the mesh, these
changes can be very labor intensive, particularly for
large or detailed grids.

MOTIVATION
Current reservoir analysis software (GEOCRACK,
TOUGH2, TETRAD) requires the user to have indepth knowledge of the numerical solution scheme
used. The problem description, in the form of a finite
element mesh or finite difference grid, serves as the
fundamental carrier of information into the
simulation. This mesh must satisfy not only the
restraints dictated by the problem at hand, such as
geometric boundaries and features (reservoir extent,
fractures, wellbores), but also must comply with
numeric restrictions of the solution method, such as
maximum or minimum element sizes, grid density,
and any special input parameters. This approach
requires the analyst to think ahead when creating the
mesh, because if any of the desired characteristics are

While modifying a two-dimensional solution grid can

be tedious, it is relatively straightforward. However,

147

direct manipulation of a three-dimensional grid or

mesh becomes nearly impossible.
Spatial
visualization is difficult, even when using interactive
and graphical modification tools. In addition, the
number of elements or cells increases greatly in three
dimensions, further complicating modification tasks.
In order to avoid such low-level manipulations, the
user needs a more intuitive and automated way of
incorporating geometric changes.

numerical analysis problem, such as material

properties and boundary conditions. Combining the
geometric model with the parameters of the analysis
creates a problem description unique to the problem
at hand, but independent of any given analysis
method.
The geometric model serves as a means of
communication between an analyst and the analysis
software. This model is independent of the solution
method, yet contains all of the necessary parameters,
making it a natural way to convey information about
a given problem to the simulation software. The
geometric model contains information about the
nature of the problem and can store the results
obtained from simulations of the problem. In this
way, the analyst need only interact with the abstract
description of a reservoir, not with the underlying
solution mesh or grid.
This simplifies and
streamlines the job of the analyst and provides an
organizational tool for defining, performing, and
interpreting reservoir simulations. The chore of
developing input files and translation file formats can
be greatly simplified and automated by the modeling
software because all of the problem information is
centralized in the geometric model.

The above arguments also apply to situations where

the solution mesh needs modification independently
of the defining geometry. Such situations occur as a
result of the numerical techniques used in the
solution. For instance, as the analysis of a particular
reservoir evolves through time, areas of high
temperature or pressure gradients may move to
different parts of the reservoir, requiring smaller
elements in a new area of the mesh. Again, it can be
tedious for the analyst to manually modify the mesh
to introduce new areas of refinement and assign
proper attribute information.
Adaptive meshing, where automatic mesh generation
is used in conjunction with an estimation of the error
in the calculated solution, is a powerful approach to
addressing appropriate mesh refinement (Lewis
199 1). But if the mesh provides the only source of
problem information, data such as geometry and
material properties, must be inferred from the current
mesh before a new one may be generated and
assigned the proper information.
Storing the
geometry and attributes independent of the mesh
greatly simplifies the implementation of an adaptive
meshing scheme.

Several different ways exist to implement a

geometric model. Each method has different levels
of representational capability and introduces a variety
of benefits and drawbacks. The three primary types
of geometric models are decomposition models,
constructive solid geometry, and boundary
representations (Mantayla 1988).
Topology
Boundary representation, a form of geometric
modeling where the explicit storage of boundary
information describes an object, provides the
underlying support for the geometric modeling
framework discussed herein. The boundary of a
model consists of adjacency information between
point sets known as regions, faces, edges, and
vertices. This adjacency information is commonly
referred to as topology.

Another reason for the use of geometric modeling

arises when the analyst desires to change the
numerical method of the solution, for instance from a
finite difference to a finite element or control-volume
finite element solution. This requires the creation of
another mesh with all of the proper data attached. A
higher-level description of the problem, independent
of the actual solution method, could allow the analyst
to describe the problems geometry and input
parameters only once and provide this data in an
automated way to the solution mesh or grid actually
used for the solution. The user would only have to
provide additional parameters specific to the chosen
method, as opposed to redefining the entire problem
to use the new method.

A region defines a separate portion of threedimensional space that may be either a bounded
(finite) or unbounded (infinite) subset of R3. A face
is a bounded, two-dimensional subset of R3 that
corresponds to a surface, and an edge is a onedimensional subset corresponding to a curve. A
vertex is a zero-dimensional entity that represents a
unique point in space.

GEOMETRIC MODELlNG
A geometric model is a high-level, geometry-based
representation of an object. This representation
provides a natural and convenient location to store all
of the necessary information associated with a

The boundary of a region is comprised of a set of

elements of dimension less than three. A typical

148

Non-Manifold Topology
Topology may be divided into two major categories manifold and non-manifold. Manifold refers to the
boundaries of the point set elements. A manifold
boundary is one that is of dimension one less than the
entity that it bounds at every location on the
boundary. For instance, a traditional solid model has
a manifold boundary if, at every spot on the
boundary, the boundary is two-dimensional. This is
equivalent to stating that every point on the boundary
is topologically identical to an open disk in R2.
Because the boundary must be two-dimensional
everywhere, each edge must have two and only two
faces incident upon it. These properties have been
used to optimize data structures for manifold models,
such as the winged-edge data structure (Baumgart,
1974). Every physical object has a manifold
boundary, so a manifold assumption does not limit
the real objects that a boundary representation can
model.

region' boundary would be simply a set of faces.

Similarly, faces are bounded by a collection of edges.
Every edge is then bound at either end by a vertex.
Together, all of the regions, faces, edges, and vertices
of the model make up the entire model space of R3.
This hierarchical representation of an object is
depicted in Figure 1.

Region

Face

0
3
1

Manifold boundaries simplify some data structure

and manipulation tasks, while not limiting the class
of objects that can be represented, yet are still too
restrictive for a truly general geometric modeling
system. For example, it is very natural to model a
truss structure as a set of wireframe edges in space, or
a car fender as an open surface (face). Under
manifold boundary assumptions, these situations
would have to be modeled exactly as they are in real
life. The truss elements would have to be modeled
explicitly, each with their own cylindrical face
boundary, and the fender would have to be modeled
as a complete solid with faces for both the front and
back and for the thin area around the edges. A nonmanifold boundary allows the bounding elements to
be a mixture of elements of dimension one less that
the bounded region.
This extends the
representational capability of the boundary
representation from real objects to also handle useful
abstractions that arise in numerical analysis and
simulation.

Edge

Vertex

Figure 1

Topological elements in
representation modeling.

boundary

Storage of the boundaries of all entities in the model

explicitly maintains the adjacency information of the
model. Because the adjacencies are stored directly, a
boundary representation supports very efficient query
and traversal methods. This simplifies many tasks,
such as display and local modifications of the model.
Because the adjacency information simplifies local
modification of a boundary representation,
construction and modification operators are designed
around this property. These operators, known as the
Euler operators, were first introduced by Baumgart
(1974).
The Euler operators guarantee that
modifications keep the model in a consistent
topological state. While simple and robust, the Euler
operators are very low-level tools and are not suitable
as the sole means of creating a large and complex
model. However, the use of Boolean set operations
and feature-based modeling in conjunction with the
Euler operators provides a convenient and powerhl
geometric modeling interface.

The geometric modeling framework discussed here

utilizes a fully non-manifold topological database.
The data structure, called the Multi-Link data
structure, is a combination of ideas and techniques
from previous work on non-manifold boundary
representation (Weiler 1986, Choi 1989, Rossignac
1990).
Feature-Based Modeling
A geometric model is based upon a mathematical,
geometric description of an object. This high degree
of abstraction makes developing an interface to the
model based upon the features of the problem domain
natural. In models with a low degree of abstraction,

149

such as finite element meshes, the mesh must be

created with the geometry in mind. Nodes and
elements must be placed so that the final mesh
conforms to the material boundaries and allows
boundary conditions to be appropriately imposed.
For example, to represent a wellbore as a onedimensional feature, nodes would have to be placed
along the well and elements formed so that edges lie
exactly along the well.
Using a geometric model makes a higher level of
abstraction possible. A wellbore may be represented
by simply inserting an edge at the location of the well
in the model. Intersections with other edges and
faces create vertices, and all information can be
propagated to the solution mesh or grid.
To increase abstraction even further, a well may be
inserted simply by specifying the geometric
description of the wellbore and allowing the
geometric model to take care of the details of
representation. Similarly, the user would add a
fracture or fault to a reservoir model by locating it
within the model and supplying its strike/dip
specification. A geometric description of a planar
surface patch would be constructed from the input,
and a face would be intersected and added to the
geometric model using that geometry. Just as the
abstraction of a geometric model insulates a user
from solution method details, using features to
describe the problem insulates the analyst from the
implementation details of the geometric model.

Figure 2

Boundary of a reservoir.

Fractures are added to the model by inserting planar

polygons or infinite planes that are truncated by the
reservoir boundary. Two zones of high permeability
are observed at the Hijiori reservoir. A simple model
of this system that can be used for analysis of basic
behavior of the reservoir contains two infinite
fractures. Figure 4 shows the basic reservoir system
with two flowing fractures. The fractures are located
and oriented by the input of a point on the plane and
the known strike and dip information.

EXAMPLE
A simple example demonstrates the expressive power
and intuitive interface provided by a geometric
modeling system. A basic representation of a
geothermal system requires only a few operations.
The hot dry rock reservoir in Hijiori, Japan provides
a useful example.
The first step is to provide the dimensions of the
overall boundaries of the reservoir. At Hijiori, the
active region of circulation covers approximately
500m by lOOOm in plan view and extends IOOOm in
depth. Figure 2 shows the bounding volume of the
reservoir as represented in the geometric modeling
framework.
Next, the wells are added to the model. At Hijiori,
three wells of interest extend into the fractured
region, HDR-1, HDR-2a, and HDR-3. For a simple
approximation, the wells are represented as straight
lines. The wells may be input as a sequence of line
segments or as a spline curve if desired for a more
accurate representation. A wireframe view of the
model is shown in Figure 3.

Figure 3

Geometric model including boundary and

wells.

To assist in visualization and conceptualization of the

reservoir geometry, the software supports real-time
manipulation and rotation of the model. The
reservoir can be viewed as either a wireframe or solid

150

of a geothermal reservoir. Mesh-based approaches

are simply too detailed and dependent on the
associated numerical method to be generally useful.
The specification of a geometric model and
associated problem data are logical first steps in an
analysis and can serve as the medium of
communication between an analyst and the analysis
software.

model, and cutting planes can be used to strip away

portions and look inside, as shown in Figure 5.

The use of geometric modeling, combined with

advanced visualization and manipulation tools, can
greatly simplify the task of preparing complex
problems for analysis. These advances have the
potential to allow reservoir engineers perform
calculations previously limited to scientists and
computer programmers.
ACKNOWLEDGEMENTS
We thank Mark James for long and insightful
discussions on non-manifold geometric modeling.
This work was supported by the Office of
Geothermal Technologies, U.S. Department of
Energy, and perfornied in the Applied Computation
Laboratory at Kansas State University.

Figure 4

Twofiacture model of Hijiori.

REFERENCES
Baumgart, B. Geometric Modeling for Computer
Vision. Ph.D. Thesis, Stanford University,
1974.
Choi, Young. Vertex-Based Representation of NonManifold Geometric Models. Ph.D. Thesis,
Carnegie-Mellon University, August 1989.
Lewis, R. W., et. al. Finite Element Analysis of
Heat Transfer and Flow Problems Using
Adaptive Remeshing Including Application to
Solidification Problems. International Journal
for Numerical Methods in Engineering, Vol. 32,
pp. 767-781, 1991.
Mantyla, Martti. An Introduction to Solid Modeling.
Computer Science Press, 1988.
Rossignac, Jaroslaw R. and Michael A. OConnor.
SGC: A dimension-independent model for
pointsets with internal structures and incomplete
boundaries. Geometric Modeling for Product
Engineering. Eds. Wozny, M. J., J. U. Turner,
and K. Preiss. Amsterdam: North-Holland,
1990. 145-180
Swenson, Daniel, et. al. GEOCRACK: A Coupled
Fluid Flow / Heat Transfer / Rock Deformation
Program for Analysis of Fluid Flow in Jointed
Rock. Mechanical Engineering Department,
Kansas State University, Manhattan, KS.
Weiler, Kevin. Topological Structures for
Geometric Modeling. Ph.D. Thesis, Rensselaer
Polytechnic Institute, August 1986.

Figure 5

Use of a cutting plane to view the interior

of a reservoir.

After creation of the geometry, attribute information

representing rock material properties, fracture
opening and permeability, and operating conditions
will be assigned to the elements of the model.
Ultimately, this information will be used to generate
an appropriate solution grid or mesh. All data is then
available to perform an analysis of the reservoir or
create an input file for an external analysis package.
CONCLUSIONS

Geometric modeling provides a flexible and efficient

analysis interface to serve as a high-level description

151