FACULTY OF ENGINEERING

Department of Mechanical Engineering

Fluid Mechanics and Thermodynamics Research Group

Development of a
Scientific Visualization System
Thesis submitted in fulfillment of the requirements for the award of the degree of
Doctor in de ingenieurswetenschappen (Doctor in Engineering) by

Dean Vuini
July 2007
Advisor(s):

Prof. Dr. Ir. Chris Lacor

Prof. Dr. Ir. Charles Hirsch

CFView - Computational Flow Field Visualization System

to my beloved wife and children

Ester, Petra and Stefan
and to the memory of my mother Alice

PhD Committee
President:
Prof. Dr. Jacques Tiberghien
Vrije Universiteit Brussel (VUB)
Department of Electronics and Informatics (ETRO)
Vice President:
Prof. Dr. Ir. Rik Pintelon
Vrije Universiteit Brussel (VUB)
Department of Fundamental Electricity and Instrumentation (ELEC)
Secretary:
Prof. Dr. Ir. Jacques De Ruyck
Vrije Universiteit Brussel (VUB)
Head of the Mechanical Engineering Department (MECH)
Advisors:
Prof. Dr. Ir. Chris Lacor
Vrije Universiteit Brussel (VUB)
Department of Mechanical Engineering (MECH)
Head of the Fluid Mechanics and Thermodynamics Research Group
Prof. em. Dr. Ir. Charles Hirsch
Vrije Universiteit Brussel (VUB)
President of NUMECA International
Members:
Prof. Dr. Ir. Jan Cornelis
Vrije Universiteit Brussel (VUB)
Vice-rector for Research and Development
Head of the Electronics and Informatics department (ETRO)
Prof. Dr. Ir. Herman Deconinck
von Karman Institute for Fluid Mechanics (VKI)
Head of Aeronautics and Aerospace Department
Dr. Franois-Xavier Josset
Thales Research & Technology (TRT) France
Department of Cognitive Solutions
Prof. Dr. Arthur Rizzi
Royal Institute of Technology (KTH) Sweden
Head of Aerodynamics Division

Private defense held in Brussels on July 2, 2007 at 15:00

Public defense to be held in Brussels on September 5, 2007 at 17:00

TABLE OF CONTENTS
ABSTRACT........................................................................................................................................................... III
ACKNOWLEDGMENTS .......................................................................................................................................... IV
NOMENCLATURE ................................................................................................................................................. VI
LIST OF FIGURES ............................................................................................................................................... VIII
LIST OF TABLES ................................................................................................................................................. XII

Introduction ........................................................................................................ 1
SCIENTIFIC VISUALIZATION MODEL..................................................................................................................... 4
SCIENTIFIC VISUALIZATION ENVIRONMENT ......................................................................................................... 6
SCIENTIFIC VISUALIZATION SOFTWARE - STATE OF THE ART ............................................................................... 8
OBJECT ORIENTED METHODOLOGY ................................................................................................................... 13
R&D PROJECTS HISTORY .................................................................................................................................... 16
The flow downstream of a bluff body in a double annular confined jet......................................................... 18
In-cylinder axisymmetric flows...................................................................................................................... 19
Flow pattern of milk between 2 corrugated plates ........................................................................................ 20
TOWARDS AN INTEGRATED MODELING ENVIRONMENT ..................................................................................... 22
THESIS ORGANIZATION ...................................................................................................................................... 24

Modeling Concepts and Fundamental Algorithms .................................. 25

1.1
1.1.1
1.1.2
1.1.3
1.1.4
1.2
1.2.1
1.2.2
1.2.3
1.2.4
1.2.5
1.2.6
1.3
1.3.1
1.3.2
1.4
1.4.1
1.4.2
1.4.3
1.4.4
1.4.5

DATA MODELING ................................................................................................................................ 28

Cell class ....................................................................................................................................... 29
Zone class ...................................................................................................................................... 36
Coherence and Tolerance Concept ............................................................................................... 41
Zone classification and Surface class............................................................................................ 42
COMPUTATIONAL TOPOLOGY ............................................................................................................. 44
Topological space.......................................................................................................................... 44
Structured and Unstructured Topology ......................................................................................... 46
Cell Connectivity ........................................................................................................................... 49
Node Topology............................................................................................................................... 54
Domains Connectivity.................................................................................................................... 57
Cell Intersection Pattern ............................................................................................................... 61
CONTINUITY ASSUMPTION .................................................................................................................. 72
Interpolation Method..................................................................................................................... 73
Derived Quantity ........................................................................................................................... 85
EXTRACTION ALGORITHMS ................................................................................................................. 90
Marching Cell Algorithm............................................................................................................... 90
Threshold Regions ......................................................................................................................... 96
Curve Extraction: section and isoline algorithm........................................................................... 98
Point Extraction: local value algorithm ...................................................................................... 100
Particle Trace Algorithm............................................................................................................. 113

Adaptation of Visualization Tools ........................................................... 121

2.1
2.2
2.3
2.4
2.4.1
2.4.2
2.4.3
2.4.4
2.4.5
2.4.6
2.5
2.6
2.7
2.8

INPUT DATA MODEL ......................................................................................................................... 124

SURFACE MODEL ............................................................................................................................... 129
GEOMETRY REPRESENTATIONS ......................................................................................................... 132
QUANTITY REPRESENTATIONS .......................................................................................................... 134
Isolines......................................................................................................................................... 135
Quantity Fields and Thresholds................................................................................................... 135
Point based numerical probes ..................................................................................................... 138
Curve based numerical probes .................................................................................................... 139
Surface based numerical probes.................................................................................................. 143
Vector line numerical probes....................................................................................................... 146
USER-CENTERED APPROACH ............................................................................................................ 148
INTERACTION MODELING .................................................................................................................. 150
VIEWING SPACE AND NAVIGATION ................................................................................................... 155
VISUALIZATION SCENARIOS .............................................................................................................. 159

Object-Oriented Software Development ................................................. 167

3.1
3.2
3.2.1
3.2.2
3.2.3
3.2.4
3.3
3.4
3.4.1
3.4.2
3.4.3
3.4.4
3.4.5
3.5
3.6
3.7
3.7.1
3.7.2
3.7.3
3.8
3.8.1
3.8.2
3.8.3

SOFTWARE ENGINEERING MODEL ..................................................................................................... 167

OBJECT ORIENTED CONCEPTS ........................................................................................................... 170
Objects and Classes ..................................................................................................................... 171
Messages and Methods ................................................................................................................ 173
Inheritance................................................................................................................................... 174
Polymorphism .............................................................................................................................. 176
OBJECT ORIENTED PROGRAMMING IN C++ ....................................................................................... 177
EXAMPLE OF OBJECT-ORIENTED SOFTWARE DEVELOPMENT ............................................................ 179
Requirements, Analysis and Design Example.............................................................................. 180
Data Abstraction and Encapsulation........................................................................................... 184
Inheritance and Polymorphism.................................................................................................... 187
Templates and Exception Handling ............................................................................................. 189
Dynamic Memory Management ................................................................................................... 189
MAPPING ENTITY-RELATIONSHIP MODEL TO OBJECT-ORIENTED MODEL .......................................... 191
MODEL-VIEW-CONTROLLER DESIGN ................................................................................................ 192
VISUALIZATION SYSTEM ARCHITECTURE.......................................................................................... 196
Model Layer CFD, Continuum and Geometry classes ............................................................. 198
View Layer 3D Graphics Category........................................................................................... 199
Controller Layer 3D and GUI input processing ....................................................................... 202
SOFTWARE DEVELOPMENT ENVIRONMENT ....................................................................................... 205
Software Reusability .................................................................................................................... 205
Software Portability ..................................................................................................................... 205
Integrated Development Environment ......................................................................................... 205

Conclusion and Future Developments ......................................................... 208

PASHA PARALLEL CFVIEW IMPROVING THE PERFORMANCE OF VISUALIZATION ALGORITHMS ................. 210
ALICE QFVIEW TOWARDS THE TRANSPARENT VISUALIZATION OF NUMERICAL AND EXPERIMENTAL DATA
SETS .................................................................................................................................................................. 212
QNET-CFD ON QUALITY AND THRUST IN INDUSTRIAL APPLICATIONS .......................................................... 214
LASCOT VISUALIZATION AS DECISION-MAKING AID .................................................................................... 215
SERKET SECURITY SITUATION AWARENESS................................................................................................. 218
FUTURE DEVELOPMENTS .................................................................................................................................. 219
Web3D for Collaborative Analysis and Visualization ................................................................................. 219
European Scientific Computing Organization............................................................................................. 220
Development trends in Interactive Visualization systems............................................................................ 220

References ...................................................................................................... 225

Appendixes ..................................................................................................... 234
LOOKUP TABLE AND ITS C++ IMPLEMENTATION FOR THE PENTAHEDRON CELL ............................................... 234
RESEARCH PROJECTS TIMELINE ....................................................................................................................... 239
AUTHORS PUBLICATIONS AT VUB .................................................................................................................. 240
RESEARCH DEVELOPMENTS TIMELINE ............................................................................................................. 242
SUMMARIZING THE CURRENT STATE-OF-THE-ART........................................................................................... 244
PERFORMANCE ANALYSIS ................................................................................................................................ 246
Theoretical Test Cases................................................................................................................................. 246
Hardware Environment ............................................................................................................................... 247
Algorithms Parameters................................................................................................................................ 248
Measurements Timing.................................................................................................................................. 250
Measurements Results ................................................................................................................................. 250

ii

Abstract
This thesis describes a novel approach to create, develop and utilize software tools for visualization in scientific
and engineering applications. These Scientific Visualization (SV) tools are highly interactive visual aids which
allow analysis and inspection of complex numerical data generated from high-bandwidth data sources such as
simulation software, experimental rigs, satellites, scanners, etc... The data of interest typically represent physical
variables -- 2- and 3-dimensional scalar, vector and tensor fields on structured and unstructured meshes in
multidomain (multiblock) configurations. The advanced SV tools designed during the course of this work permit
data extraction, visualization, interpretation and analysis at a degree of interaction and effectiveness that was not
available with previous visualization techniques.
The Object-Oriented Methodology (OOM), which is the software technology at the basis of the approach
advocated in this thesis, is very well adapted for large-scale software development: OOM makes SV tools
possible and makes them a usable, innovative investigation instrument for the engineer and the researcher in all
areas of pure and applied research. The advanced SV tools that we have developed allow the investigator to
examine qualitatively and quantitatively the details of a phenomenon of interest, in a unified and transparent
way. Our SV tools integrate several well-known algorithms -- such as the cutting plane, iso-surface and particle
trace algorithms -- and enhance them with an ergonomic graphical user interface. The resulting SV system
implements the reusability and encapsulation principles in its software components, which support both space
discretization (unstructured and structured meshes) and continuum (scalar and vector fields) unconstrained by
the grid topology. New implementation mechanisms applied to the class hierarchies have been developed
beyond existing object-oriented programming methods to cover a broader range of interactive techniques. A
solution was found to the problem of developing, selecting and combining classes as reusable components. The
object-oriented software development life-cycle was mastered in the development of these classes, which were
finally packaged in a set of original class libraries.
A main outcome of our approach was to deliver one of the first frameworks, which integrates 3D graphics and
windowing behavior based on the software components implemented in C++ only. This framework ensures
maximal portability to different hardware platforms and establishes the basis for reusable software in industrial
applications, such as an integrated Computational Fluid Dynamics (CFD) environment (pre-post processing and
the solver). The important outcome of this work is an integrated set of VS tools -- the Computational Field
Visualization System (CFView) -- available for investigators in field physics in general, and specifically for CFD
researchers and engineers. Several CFD examples are presented and discussed to illustrate the new development
techniques for scientific visualization tools.

iii

Acknowledgments
I would like to express my gratitude to my supervisor, Prof. Charles Hirsch, for taking the risk of introducing an
innovative software-construction methodology to the CFD community and for giving me the opportunity to
master it during all these years. His probing remarks, his continuous support and advice spurred many thoughts
and were always an incentive for me to pursue excellence in this work. Without Prof. Chris LACOR, this thesis
would not have come to completion, and I deeply thank him for his support and encouragement in the
finalization phase of this thesis.
Thanks go to all my colleagues, former or present members of the Department of Mechanical Engineering, the
former Department of Fluid Mechanics of the Vrije Universiteit Brussel, who contributed to the development of
CFView: Michel Pottiez, Vincent Sotiaux, Cem Dener, Marc Tombroff, Jo Decuyper, Didier Keymeulen, Jan
Torreele, Chris Verret, as well as to the developers at NUMECA: Jorge Leal Portela, Etienne Robin, Guy
Stroobant and Alpesh Patel, who pursued the development of CFView and made it an advanced scientific
visualization product for turbo-machinery applications.
Special thanks go to the many researchers and engineers who used CFView: Shun Kang, Peter Segaert, Andreas
Sturmayr, Marco Mulas, Prasad Alavilli, Zhongwen Zhu, Peter Van Ransbeeck, Erbing Shang, Nouredine
Hakimi, Eric Lorrain, Benoit Lonard, Francois Schmitt, Evgeni Smirnov, Andrei Khodak, Martin Aube, Eric
Grandry, for their valuable comments and suggestions. Since 1988, when I joined the Department, I have had
many useful and challenging interactions with Francisco Alcrudo, Antoine van Marcke de Lummen, Guido Van
Dyck, Peter Grognard, Luc Dewilde, Rudy Derdelinckx, Steven Haesaerts, Famakan Kamissoko, Hugo Van
Bockryck, Karl Pottie, Wim Teugels, Wim Collaer, Kris Sollie and Pascal Herbosch; these exchanges of views
contributed to making this work richer and better. I happily acknowledge the inputs of the younger group of
scientists at our Department and their ideas on novel visualization aspects in acoustics, combustion, medicine
and environment domains: Stephan Geerts, Jan Ramboer, Tim Broeckhoven, Ghader Ghorbaniasl, Santhosh
Jayaraju, Mark Brouns, Mahdi Zakyani and Patryk Widera. I am indebted to Jenny Dhaes, Alain Wery and
Michel Desees, without whom my daily work at the Department would not be possible.
I am pleased to mention the contribution of Prof. Jacques de Ruyck on PLOT83, elements of which were applied
in CFView. I also thank Prof. Herman Deconinck for offering me the opportunity to give a lecture on objectoriented programming in computer graphics at von Karman Institute in 1991, which provided momentum to my
work to this exciting research area. For our interactions on unstructured modeling, I thank Jean-Marie Marchal
and gratefully remember the late Yves Rubin, whose suggestions prompted the development of CFView in the
finite-element-method domain. Special thanks go to Patrick Vankeirsbilck for many enlightening discussions on
object-oriented programming practice.
I would like to thank Koen Grijspeerdt for our teamwork in the IWT-funded LCLMS project, and for his effort
in introducing CFD and visualization in food-production applications.
Special thanks go to Birinchi Kumar Hazarika for the work in the EC-funded ALICE project and to Cristian
Dinescu, John Favaro, Bernhard Snder, Ian Jenkinson, Giordano Tanzini, Renato Campo, Gilles Gruez,
Pasquale Schiano and Petar Brajak.
I take the opportunity to thank my collaborators Danny Deen, Emil Oanta and Zvonimir Batarilo for successful
development of the Java visualization components in the SERKET project. I would like to thank Prof. Jan
Cornelis and his collaborators Hichem Sahli, Rudi Deklerk and Peter Schelkens for their contributions when
extendeding scientific visualization to general-purpose information systems. Special thanks also go to Claude
Mayer, Franois-Xavier Josset, Tomasz Luniewski, Jef Vanbockryck, Claude Desmet, Karel Buijsse, Christophe

iv

Herreman, Luc Desimpelaere and Richard Aked for their help in shaping the visualization development in ITEA
projects.
Two EU-funded TEMPUS projects, made it possible for me to interact again with Croatia and Macedonia, and
my thanks go to Prof. Zdravko Terze and Prof. Milan Kosevski for encouraging me to complete this thesis.
Special thanks go to Bernard Delcourt for his thorough proof-reading and improving the written English, and
sharing with me the last moments before the publishing of this thesis.
The funding of the European Commission (EC) and the Flemish institute for Innovation and Technology (IWT)
is gratefully acknowledged; the LCLMS, ALICE, LASCOT, QNET-CFD and SERKET projects have been
instrumental in allowing me to carry out my research. I am grateful to Vrije Universiteit Brussel for providing
the necessary research and computer facilities, not only to accomplish this work, but also to complete the
engaged projects.
I would like thank my parents for their moral and financial support, which made it possible for me to come to
Belgium; I will always remember my mother for her unfailing enthusiasm for research work, an attitude she has
passed on to me and for which I will ever be grateful. I thank my father and brother who kept chasing me up and
pushing me to complete this work.
Finally and most importantly, I wish to thank my wife and children for their love, patience, support and
encouragement during these many years, months, weekends, days and evening hours that went into this longerthan-expected undertaking.

Dean Vuini
Brussels, July 2007

Nomenclature

for all
there exist
equals
summation
product

Expressions
gij

:
metric tensor components (i,j = 1,2,3)

i, j, k
I, J, K
J
G
n
n
p
r, x
S
A
u, v, w
U
x, y, z
X

grid indices in u, v and w directions

grid numbers in i, j and k coordinate index directions
Jacobian of transformation
metric tensor
normal distance
normal vector
grid point coordinates

Symbols

position vector
structured grid
transformation matrix
parametric curvilinear coordinates
unstructured grid
Cartesian coordinates
hybrid grid
:
Kronecker delta; search radius
partial differential operator; boundary
error of numerical solution

.
x

Gradient operator

Laplacian operator
forward difference operator
spacing in parametric coordinates
spacing in Cartesian coordinates

u, v, w
x, y, z

Divergence operator
curl of vector

Subscripts and superscripts:

i, j, k
position indices for structured meshes
u, v, w
direction of differentiation or interpolation in parametric space
1,2,3
space coordinate axis
Abbreviations
ADT
AFX
AI
ANC
ATDC
AVS
BC
B-rep

:
Abstract Data Type
Animation Framework Extension
Artificial Intelligence
Automatic Naming Convention
After Top Dead Centre
Advanced Visual Systems
Boundary Condition
Boundary Representation
vi

CA
CAD
CAE
CFD
CG
CFView
CON
CPU
DFD
DNS
DPIV
EFD
ERM
EXT
FD
FE
FEA
FO
FV
GPU
GUI
HOOPS
HWA
IME
INL
J2EE
JOnAS
KEE
LAN
LG
LDV
LSE
LSV
MB
MFLOPS
MIPS
MIMD
MPEG
MVC
MVE
OO
OOM
OOP
OOPL
OUT
PC
PDE
PER
PIV
PHIGS
PVM
QFView

Crank Angle
Computer Aided Design
Computer Aided Engineering
Computational Fluid Dynamics
Computer Graphics
Computational Flow Field Visualization
Connection BC
Central Processing Unit
Data Flow Diagram
Direct Numerical Simulations
Digital Particle Image Velocimetry
Experimental Fluid Dynamics
Entity Relationships Model
External BC
Finite Difference
Finite Element
Finite Element Analysis
Function Oriented
Finite Volume
Graphics Processing Unit
Graphical User Interface
Hierarchical Object Oriented Picture System
Hot wire Anemometry
Integrated Modeling Environment
Inlet BC
Java 2 Platform, Enterprise Edition
Java Open Source J2EE Application Server
Knowledge Engineering Environment
Local Area Network
Local to Global index mapping
Laser Doppler Velocimetry
Large Eddies Simulation
Light sheet visualization
Mega Bytes
Millions of Floating Point Operations per Second ("MegaFlops")
Millions of Instructions per Second
Multiple instruction, multiple data
Moving Picture Expert Group
Model View Controller
Modular Visualization Environments
Object Oriented
Object Oriented Methodology
Object Oriented Programming
Object-Oriented Programming Language
Outlet boundary condition
Personal Computer
Partial Differential Equations
Periodic BC
Particle Image Velocimetry
Programmers Hierarchical Interactive Graphics System
Parallel Virtual Machine
Quantitative Flow Field Visualization

vii

QoS
RAM
RANS
RG
RMI
RMS
ROI
SDK
SGS
SIMD
SISD
SNG
SOL
SOAP
STD
SYM
SV
SVS
TKE
VG
VisAD
VTK
VUB
WWW
WS

Quality of Service
Random Access Memory
Reynolds Average Navier Stokes
Raster Graphics
Remote Method Invocation
Root Mean Square
Region of Interest
Software Development Kit
Sub Grid Scale
Single instruction, multiple data
Single instruction, single data
Singularity BC
Solid wall BC
Simple Object Access Protocol
State Transition Diagram
Symmetry BC
Scientific Visualization
Scientific Visualization System
Turbulent Kinetic Energy
Vector Graphics
VISualization for Algorithm Development
Visualization ToolKit
Vrije Universiteit Brussel
World Wide Web
Web Services

List of Figures
Figure 1: The scientific visualization role _______________________________________________________2
Figure 2: The Scientific Visualization Model _____________________________________________________4
Figure 3: The Visualization Data Sets __________________________________________________________5
Figure 4: Integrated Computational Environment _________________________________________________6
Figure 5: CFView the scientific visualization system_______________________________________________9
Figure 6: The OpenDX Application Builder_____________________________________________________10
Figure 7: VisAD application example _________________________________________________________11
Figure 8: The integrated modeling environment from Dassault Systmes and ANSYS, Inc _________________12
Figure 9: The comparison of Hardware/Software productivity ______________________________________13
Figure 10: Graphics Engine as combine software/hardware solution _________________________________15
Figure 11: Software Components Distribution___________________________________________________15
Figure 12: QFView Web Interface ____________________________________________________________17
Figure 13: Use of EFD and CFD tools ________________________________________________________18
Figure 14: Laminar flow at Ret = 60: DPIV of nearly axisymmetric flow, LSV of vortex shedding and CFD at
various degrees of non-axisymmetric flow ______________________________________________________18
Figure 15: PIV system at VUB _______________________________________________________________19
Figure 16: Flow pattern for 90oATDC: (a) visualization at 20 rev/min, valve lift = 10 mm (b) Average velocity
field at 5 rev/min, valve lift = 10 mm __________________________________________________________19
Figure 17: Turbulent kinetic energy field at 90oATDC at 5 rev/min, valve lift=10 mm ____________________20
Figure 18: Average vorticity field at 90oATDC at 5 rev/min, valve lift=10 mm__________________________20
Figure 19: Experimental and CFD model for flow analysis between corrugated plates ___________________21
Figure 20: Workflow for the integration of EFD and CFD simulations _______________________________21
Figure 21: Example of an Integrated Modeling Environment [57] ___________________________________22
Figure 22: Example of the 3D virtual car model testing ___________________________________________23
Figure 23: Software model as communication media in the software development process ________________25
Figure 24: Entity-Relationship Model _________________________________________________________26
Figure 25:.Data model decomposition _________________________________________________________28
Figure 26:.Cell classification ________________________________________________________________30
Figure 27: Developed cell ERM ______________________________________________________________31
Figure 28: 1D & 2D Cell topologies __________________________________________________________32
viii

Figure 29: 3D Cell topologies_______________________________________________________________ 33

Figure 30: Cell ERM______________________________________________________________________ 35
Figure 31: The Zone Composition and Topology ERM ___________________________________________ 37
Figure 32: Bridge orientation _______________________________________________________________ 38
Figure 33: Inner and Outer Frames of 2D and 3D zones __________________________________________ 39
Figure 34: Zone-Cell ERM _________________________________________________________________ 39
Figure 35: Developed Cell-Zone ERM ________________________________________________________ 40
Figure 36: Coherent and non-coherent cells____________________________________________________ 41
Figure 37: Surface types: (a) open, (b) closed __________________________________________________ 42
Figure 38: Simply and multiply connected and disconnected surface regions.__________________________ 43
Figure 39: Zone and cells point classification__________________________________________________ 44
Figure 40: Curve, surface and body neighborhoods______________________________________________ 45
Figure 41: Cell point classification___________________________________________________________ 45
Figure 42: Manifold and non-manifold neighborhoods ___________________________________________ 46
Figure 43: Zone topology concepts ___________________________________________________________ 46
Figure 44: Boundary LG map _______________________________________________________________ 47
Figure 45: The sup-sub zone relationship ______________________________________________________ 48
Figure 46: Node indexing in structured grids ___________________________________________________ 48
Figure 47: Cell topology and cell connectivity data structure ______________________________________ 49
Figure 48: Minimum node grouping rule ______________________________________________________ 50
Figure 49: Cell connectivity for different parametric zones ________________________________________ 51
Figure 50: Closed boundaries_______________________________________________________________ 52
Figure 51: Collapsing algorithm for multiple-connected and disconnected regions _____________________ 53
Figure 52: Surface parts (a) multiple-connected, (b) disconnected and (c) mixed _______________________ 53
Figure 53: Dual topology concept between nodes and cells ________________________________________ 54
Figure 54: Set of cells with duplicate nodes.____________________________________________________ 55
Figure 55: Navigation around the node: (a) one node (b) two nodes _________________________________ 56
Figure 56: Domain with boundary indexing in 2D & 3D and boundary identification ___________________ 57
Figure 57: Domain connectivity in 2D ________________________________________________________ 58
Figure 58: Segment orientation cases _________________________________________________________ 58
Figure 59: Multi-domain (multi-block) connectivity in 3D _________________________________________ 60
Figure 60: Cell topology ERM ______________________________________________________________ 61
Figure 61: WEB model viewed from the cell interior _____________________________________________ 62
Figure 62: WEB model edge analysis _______________________________________________________ 62
Figure 63: The edge orientation a) local to face b) global to cell ___________________________________ 62
Figure 64: Intersection pattern and polygon orientation __________________________________________ 64
Figure 65: Dual Cells _____________________________________________________________________ 65
Figure 66: Edge traversal direction for nodes labeled as: (a) FALSE (b) TRUE________________________ 65
Figure 67: The navigation graph for tetrahedron ________________________________________________ 66
Figure 68: The navigation graph for pyramid __________________________________________________ 66
Figure 69: The navigation graph for pentahedron _______________________________________________ 67
Figure 70: The navigation graph for hexahedron ________________________________________________ 67
Figure 71: Some intersection patterns for the hexahedron cell with the polygon partitions________________ 68
Figure 72: Polygon with maximum number of nodes, hexahedral cell. _______________________________ 69
Figure 73: Pathological intersection cases results in non-manifold 2D & 3D geometries_________________ 70
Figure 74: Pathological case of the star-shaped polygon, and its polygon partition _____________________ 70
Figure 75: Coordinates transformations characteristics __________________________________________ 73
Figure 76: Coordinates transformation _______________________________________________________ 80
Figure 77: Surface normal and quadrilateral cell _______________________________________________ 84
Figure 78: Possible singular cases of intersections with a node, an edge and a face_____________________ 91
Figure 79: Seed cells from boundary _________________________________________________________ 92
Figure 80: Hexahedron cell intersected with the plane ___________________________________________ 93
Figure 81: Cutting plane example of multiple connected topology___________________________________ 94
Figure 82: Disconnected components of an isosurface. ___________________________________________ 95
Figure 83: Ten different thresholds in a triangle ________________________________________________ 96
Figure 84: Fifteen different thresholds of quadrilateral ___________________________________________ 97
Figure 85: Ambiguity cases for quadrilateral thresholds __________________________________________ 97
Figure 86: Node traversal in the isoline algorithm for unstructured and structured surfaces ______________ 98
Figure 87: Close and Open curves for Isolines Representation _____________________________________ 99
Figure 88: Line-Surface intersection concepts _________________________________________________ 100
Figure 89: Line Surface multiple intersections and possible traversals ______________________________ 100

ix

Figure 90: Cell ray intersection _____________________________________________________________101

Figure 91: Intersection parameter description__________________________________________________102
Figure 92: Plane line intersection. ___________________________________________________________102
Figure 93: Rays and intersections with extended cell boundary ____________________________________103
Figure 94: Boundary normals for different parametric dimensions__________________________________103
Figure 95: Point location on cell boundaries___________________________________________________104
Figure 96: Possible positions of the point leaving the cell in the neighborhood of the triangle cell _________105
Figure 97: Possible positions of the point leaving the cell in the neighborhood of the quadrilateral cell_____107
Figure 98: Possible positions of the point leaving the cell in the neighborhood of the tetrahedron cell ______108
Figure 99: Possible positions of the point leaving the cell in the neighborhood of the prism ______________108
Figure 100: Possible positions of the point leaving the cell in the neighborhood of the pyramid ___________109
Figure 101: Possible positions of the point leaving the cell in the neighborhood of the hexahedron ________110
Figure 102: Tangency condition for the vector line analytical and numerical treatment _________________113
Figure 103: The cell boundaries parametric space in 2D _________________________________________116
Figure 104: The cell boundaries parametric space in 3D _________________________________________117
Figure 105: The map of a cell boundary point between connected cells in 3D _________________________118
Figure 106: The map of a cell boundary point between connected cells in 2D _________________________118
Figure 107: Local profile and Particle paths for a 2D airfoil ______________________________________123
Figure 108: Local values, Isolines and Particle paths representations in 3D __________________________124
Figure 109: Main Quantity menu ____________________________________________________________125
Figure 110: Field and Solid Quantity menu ____________________________________________________126
Figure 111: Validation Data menus __________________________________________________________126
Figure 112: Plot Data menu________________________________________________________________126
Figure 113: Structured topology for 2D and 3D geometry ________________________________________127
Figure 114: Unstructured topology for 2D and 3D geometry ______________________________________127
Figure 115: Domains connectivity for structured 3D multiblock grids _______________________________128
Figure 116: Initial representations: boundaries in 2D and solid surface boundaries in 3D. ______________129
Figure 117: Geometry/Surface menu _________________________________________________________129
Figure 118: Create Surface dialog-box _______________________________________________________130
Figure 119: Structured multiblock surface patch manipulation_____________________________________130
Figure 120: Cutting plane and isosurface examples for structured and unstructured meshes______________131
Figure 121: Surface dialog-box showing the cutting plane and isosurface instances ____________________132
Figure 122. Geometry menu ________________________________________________________________132
Figure 123: Surface geometry with boundaries outlines __________________________________________132
Figure 124: Geometry repetitions types: mirror, translation and rotation ____________________________133
Figure 125: Render menu__________________________________________________________________133
Figure 126: Rendering of the space shuttle ____________________________________________________133
Figure 127: Scalar and vector representation menus ____________________________________________134
Figure 128.: Isolines representations for 2D and 3D geometries ___________________________________135
Figure 129: Isolines menu item and dialog-box _________________________________________________135
Figure 130: Color Contours menu ___________________________________________________________136
Figure 131: Color contours based on different rendering algorithms ________________________________136
Figure 132.: Threshold color contours________________________________________________________136
Figure 133: Vector Field dialog-box _________________________________________________________137
Figure 134: Structured and unstructured vector fields ___________________________________________137
Figure 135: Vector Thresholds on cutting planes _______________________________________________137
Figure 136: Local isolines, scalars and vectors assisted with coordinate axis tool______________________138
Figure 137 Range menu ___________________________________________________________________138
Figure 138: Cartesian Plot menu ____________________________________________________________139
Figure 139: Scalar distribution along solid boundaries and sections ________________________________139
Figure 140: Cartesian plot of the shock wave location in 3D ______________________________________140
Figure 141: Scalar and Vector curve based extractions in 2D _____________________________________141
Figure 142: Isolines and Surface Particle Paths in 3D ___________________________________________141
Figure 143: Local profile representation for boundary layer analysis _______________________________142
Figure 144: Vector Local Profiles ___________________________________________________________143
Figure 145: Cutting Plane dialog-box ________________________________________________________143
Figure 146: Cutting planes representations in 3D _______________________________________________144
Figure 147: Cutting plane with Vector representations ___________________________________________144
Figure 148: Several isosurface representations of the Temperature field around the airplane_____________145
Figure 149: Combined use of Particle trace, Cutting plane and Isosurface probes _____________________145
Figure 150: Vector Line menus and representations _____________________________________________146

Figure 151 Surface and 3D Streamlines generation from a cutting plane surface ______________________
Figure 152: Vector lines representations from structured surface points, with the required toolbox in action
Figure 153: ERM of the interaction process ___________________________________________________
Figure 154: The menu structure ____________________________________________________________
Figure 155: CFView GUI layout____________________________________________________________
Figure 156: Different view types ____________________________________________________________
Figure 157: Evolution of GUI ______________________________________________________________
Figure 158: Reminders for different interactive components ______________________________________
Figure 159: Cube model for sizing the viewing space____________________________________________
Figure 160: Clipping planes in viewing space _________________________________________________
Figure 161: Coordinates system and 3D mouse-cursor input______________________________________
Figure 162: View projection types __________________________________________________________
Figure 163: Viewing buttons _______________________________________________________________
Figure 164: Camera model and its viewing space ______________________________________________
Figure 165: Camera parameters and virtual sphere used for camera rotation ________________________
Figure 166: Symbolic calculator for the definition of new field quantities ____________________________
Figure 167: EUROVAL visualization scenario for the airfoil test case ______________________________
Figure 168: EUROVAL visualization scenario for the Delery and ONERA bump ______________________
Figure 169Setting of graphical primitives_____________________________________________________
Figure 170: Superposing different views______________________________________________________
Figure 171: Different graphical primitives showing the same scalar field ____________________________
Figure 172: Different colormap of the same scalar field _________________________________________
Figure 173: Analytical surfaces generation for comparison purposes _______________________________
Figure 174: Comparison of the traditional and object-oriented software development life-cycle __________
Figure 175: Object concept________________________________________________________________
Figure 176: Abstract data type structure _____________________________________________________
Figure 177: Point object __________________________________________________________________
Figure 178: Single and multiple inheritance___________________________________________________
Figure 179: Polymorphism ________________________________________________________________
Figure 180: DFD of the streamline example___________________________________________________
Figure 181: The partial ERD of the streamline example. _________________________________________
Figure 182: STD of the streamline example ___________________________________________________
Figure 183: Class hierarchy diagram ________________________________________________________
Figure 184: Class attribute diagram_________________________________________________________
Figure 185: The MVC model with six basic relationships ________________________________________
Figure 186: MVC framework for Surface manipulation __________________________________________
Figure 187: Visualization system architecture _________________________________________________
Figure 188: Hierarchy of Geometry classes ___________________________________________________
Figure 189: 3D View Layer________________________________________________________________
Figure 190: Class hierarchy of the controller classes ___________________________________________
Figure 191: Event/Action coupling __________________________________________________________
Figure 192: Eclipse Integrated Development Environment _______________________________________
Figure 193: Knowledge domains involved in interactive visualization_______________________________
Figure 194: Conceptual overview of the SIMD/MIMD Parallel CFView system _______________________
Figure 195: QFView an Internet based archiving and visualization environment ____________________
Figure 196: The QFView framework ________________________________________________________
Figure 197: VUB Burner Experiment ________________________________________________________
Figure 198: The eight QNET-CFD newsletters_________________________________________________
Figure 200: The LASCOT scenario__________________________________________________________
Figure 201: The security SERKET scenario ___________________________________________________
Figure 202: The SERKET application________________________________________________________
Figure 203: Visualization of 3D Model_______________________________________________________
Figure 204: Components of a 3D Model______________________________________________________
Figure 205: Graphical and Textual Annotations _______________________________________________
Figure 206: Representation of a Measurement ________________________________________________
Figure 207: Cone Trees __________________________________________________________________
Figure 208: Reconfigurable Disc Trees ______________________________________________________
Figure 209:- Mobile Device Controlling Virtual Worlds _________________________________________
Figure 210: Mobile Application Over Internet _________________________________________________
Figure 211: Alternative User Interaction Devices ______________________________________________
Figure 212: Handheld Devices _____________________________________________________________

xi

147
147
151
152
152
153
154
155
156
156
157
158
158
158
159
160
161
162
163
163
164
165
166
169
172
173
174
175
176
180
181
182
183
184
193
195
197
199
200
203
204
207
208
210
212
213
214
215
217
218
219
221
221
221
221
222
222
223
223
223
223

Figure 213: New generation of miniature computers and multi touch-screen inputs ____________________223
Figure 214: 3D Model of Machine on Display Wall _____________________________________________224
Figure 215: Scientific Visualization with Chromium ____________________________________________224
Figure 216: Example of Augmented Reality ____________________________________________________224
Figure 217 :NASA Space Station on Display Wall _______________________________________________224
Figure 218: Collaborative Visualization ______________________________________________________224
Figure 219: 6xLCD Based Display Unit ______________________________________________________224
Figure 220: Parallel Rendering _____________________________________________________________224
Figure 221: 3D Model of Visualization Lab____________________________________________________224
Figure 222: Overview of the heterogeneous and distributed environment used for the theoretical benchmarks 248
Figure 223: The theoretical random-base meshes (a) 20x20x20 (b) 200x200x250 ______________________249
Figure 224: Mesh size 200x200x250 (a) Cutting plane and Particle traces (b) Isosurface ________________249
Figure 225: Average execution times in seconds for the algorithms on the different machines (with caching
mechanism enabled for the parallel implementations). ___________________________________________252
Figure 226: Average execution times in seconds for the SIMD and MIMD implementations of the isosurface
algorithm, with respect to the number of triangles generated (caching mechanism on) __________________253
Figure 227: Execution times in seconds for particle tracing with respect to the number of particles ________254

List of Tables
Table 1: Layered Software Architecture________________________________________________________14
Table 2: SEGMENT skeleton table____________________________________________________________32
Table 3: TRIANGLE skeleton table ___________________________________________________________32
Table 4: QUADRILATERAL skeleton table _____________________________________________________32
Table 5: TETRAHEDRON skeleton table _______________________________________________________34
Table 6: PYRAMID skeleton table ____________________________________________________________34
Table 7: PENTAHEDRON skeleton table_______________________________________________________34
Table 8: HEXAHEDRON skeleton table________________________________________________________35
Table 9: Structured zone parameterization _____________________________________________________47
Table 10: C++ implementation of the hashing value ______________________________________________53
Table 11: Boundary indexing for 2D and 3D structured grids_______________________________________57
Table 12: Domain connectivity specification in 2D _______________________________________________58
Table 13: Domain connectivity specification in 3D _______________________________________________59
Table 14: The WEB model record ____________________________________________________________61
Table 15: WEB model of the tetrahedron _______________________________________________________63
Table 16: The lookup table for the tetrahedron __________________________________________________64
Table 17: Polygon Subdivision _______________________________________________________________68
Table 18: Records from hexahedron lookup table with polygon partitions and multiple connected regions____69
Table 19: Lookup table for the triangle ________________________________________________________70
Table 20: Lookup table for the quadrilateral ____________________________________________________71
Table 21: Shape function for 3D isoparametric mapping __________________________________________78
Table 22: Reduction of multiplication operations ________________________________________________80
Table 23: Triangle truth table ______________________________________________________________104
Table 24: Triangle constrains path __________________________________________________________106
Table 25: Quadrilateral truth table __________________________________________________________106
Table 26: Quadrilateral constrains path ______________________________________________________107
Table 27: The mapping procedure of a cell boundary point between connected cells in 3D _______________119
Table 29: Standard notation for boundary conditions ____________________________________________128
Table 30: Comparison of classical and user- centered approach ___________________________________148
Table 31: Software quality factors ___________________________________________________________169
Table 32: Graphics primitives for different geometries and text ____________________________________201
Table 33: Average times (s) for Sequential, SIMD and MIMD implementations of Cutting Plane and Isosurface
algorithms (wall-clock time)________________________________________________________________211
Table 34: The lookup table for the pentahedron_________________________________________________234
Table 35: Research Projects Timeline ________________________________________________________239
Table 36: Average times for Cutting Plane (wall-clock time in seconds)______________________________251
Table 37: Average times for Isosurface (wall-clock time in seconds) ________________________________251
Table 38: Average times for Particle Trace (wall-clock time in seconds) _____________________________251
Table 39: Evolution of the execution times in seconds with the number of particles used _________________252
Table 40: Execution times in seconds for Isosurface on MIMD for different machine configurations (wall-clock
time) with varying number of processors ______________________________________________________252
xii

Introduction
Fluid motion is studied in Fluid Dynamics [1, 2] by performing experimental and computational simulations that
researchers analyze in order to understand and predict fluid flow behaviors. This scientific process yields large
data sets, resulting from measurements or numerical computations. Scientific visualization comes naturally in
this process as the methodology that enhances comprehension and deepens insight in such large data sets. The
term Scientific Visualization was officially introduced and defined as a scientific discipline in 1987 at
SIGGRAPH[3] and contributes to the role of computing, as quoted by Richard Hamming:

The purpose of computing is insight, not numbers.

Scientific visualization [4, 5] is usually performed through specialized software, which combines visualization
techniques to display and analyze scientific data. The scientific visualization methodology defines methods to
manipulate and convert data into comprehensible images. The scientific visualization process starts with the
transformation of data sets into geometric abstractions, which are further processed in displayable images,
created by computer graphics algorithms [6, 7]. Finally the human vision, possessing the highest-bandwidth of
humans information input, is exploited to understand the computer generated images.
Computational Fluid Dynamics (CFD) simulations are important source of scientific and engineering data, for
example, computations of complex three-dimensional (3D) internal flows in Turbomachinery and external flows
around complete aircraft or space shuttle configurations. The numerical simulation software used in the industry,
typically based on 3D Navier-Stokes solvers, requires easy and efficient visualization and presentation tools to
process the flow fields. The numerical simulation environment is expected to be ergonomic, so as to incite the
CFD specialist to investigate complex flow topologies and explore the physics of fluid flows in a user-control
manner. A key objective of the visualization process is to enable the creation of meaningful images from large
amounts of data.
In order to develop a scientific Visualization Software (VS) it is necessary to combine Computer Graphics (CG)
and User Interface (UI) design know-how with engineering knowledge. Thus, we need to consider and integrate
the mentioned methodical domains, when addressing the software development of scientific visualization
software. The well adapted approach to address scientific visualization issues is through the introduction of
interactive visualization tools for the analysis and interpretations of numerically generated data, which need to be
developed by taking into account and combining computer hardware (processor, disks, graphics boards) and
computer software (graphics application programming interface (API) and scientific visualization methods).
Scientific visualization methods give the possibility to the researcher to analyze the same problem in a variety of
ways. They promote the concept of virtual laboratory where the user performs virtual measurements. These
virtual measurements are the product of the interactive visualization process that the user applies on the data
he/she analyzes. For example, the interaction may lead to the researcher locating a flow re-circulation area, a
region of high turbulence, or the points where the value of enthalpy reaches its minimum and maximum values.
Usually, the goal is not only to see the flow pattern, but also to understand why and how it develops. The
possibility to apply several visualization tools in an interactive way makes it possible to maybe uncover the
problem under investigation. The exploration of numerically generated data leads to the concept of virtual
instrument, namely a software probe that can be interactively manipulated. Two important questions are to be
answered:
How does software probe (virtual instrument) measure a value?
Filtering and searching through numerical data.
How does it graphically display a measured value?
Applying computer graphics techniques.
1

Physical phenomena

Modeling

Computational
model

Experimental
model

Simulation

Simulated data

Scientific
visualization

Figure 1: The scientific visualization role

As shown in Figure 1, numerically generated data are the main input to the visualization system. The data
sources are experimental tests and computational models which yield high-resolution, multi-dimensional data
sets. Such large and complex data sets may consist of several scalar, vector and/or tensor quantities defined on
2D or 3D geometries; they become even larger when time-dependency or other specialized parameters are added
to the solution. Fast and selective extraction of qualitative and quantitative information is important for
interpreting the visualized phenomena. The performance (response time) of the visualization system becomes
critical when applying iterative optimization procedure to the model under analysis. An adequate computer
system performance (response loop) must be in place, in order to match the activity of the human visual system
and the computer display of the extracted flow features to enable the user to enjoy the benefits of a truly
interactive visualization experience. The visualization system must also provide extensive interactive tools for
manipulating and comparing the computational and experimental models. For example, to lower the costs of a
new product development, we can reduce the range of experimental testing and increase the number of numerical
simulations, provided we are able to effectively exploit the existing experimental database.
The main role of SV is to present the data in a meaningful and easily understandable digital format. Visualization
can be defined as a set of transformations that convert raw data into a displayable image; the goal is to convert
the raw information into a format understandable by the human visual system, while maintaining the
completeness of the presented information expected by the end user. In our work, we used Vector Graphics (VG)
for applying colors and plotting functionality on geometrical primitives such as points, lines, curves, and
polygons. This is in contrast to the Raster Graphics (RG) approach, which represents images as a collection of
pixels. Other scientific visualization techniques rely on comparison and verification methods based on RG highresolution images. Such visualization systems to analyze RG images, for example coming from satellites and
scanners are not developed in this work, but their results are integrated, as part of the performed physical
experiments.
The present work contributed to the development of Scientific Visualization by addressing the following
questions:

Why scientific visualization?

to help scientists and engineers visualize and explore their results in the simplest possible way,
to improve scientific visualization techniques,
to provide new and better ways of displaying data,
to discover new visual interpretations through image synthesis,
to standardize knowledge of scientific visualization,
to promote and improve software development platform for industrial applications,
to discover new phenomena in the regions where measurements cannot be taken,
to validate computational and experimental results,
to improve the learning process, as a pedagogical tool for aiding students, engineers in the fluid flows
modeling.
How to do scientific visualization?
Utilizing interactive scientific visualization software.

How to develop interactive scientific visualization software?

This is the fundamental question which was addressed in the present work:
multidisciplinary research, CFD and Computer Science
scientific visualization tools
object oriented methodology and software engineering,
new ways of interacting with numerical codes and computer environment,
new algorithms improving user perception (color, shape),
graphical user interface design
application of windowing (X11, Windows) and graphics standards (PHIGS, PEX, OpenGL)
implementations based on object oriented programming languages in C++ and Java
interactive tools for large data sets applying network computing with parallel machines
A blend of many developments in computational and computer graphics hardware and software has made it
possible to design interactive SV software. Today computer hardware and software give enough performance and
throughput for graphical and computational tasks to achieve the expected level of functionality, namely:
Computational task:

search and extraction algorithms that can identify, compute and extract the
geometrical and quantitative data from selected data sets.

Visualization task:

3D computer graphics display and user-interaction with the graphical objects.

Scientific Visualization Model

The scientific visualization model defines a series of transformations which convert data into images displayed
on the computer screen, as shown in Figure 2. The model encompasses four visualization data types: simulation,
derived, graphics and displayable data. The visualization process is defined as a set of mappings, which
transform the visualization data types from one data type to another.
SCIENTIFIC VISUALIZATION MODEL

DATA

TRANSFORMATION

Simulated
extraction
refinement
Derived
enhancement
enrichment
Graphical
rendering
Displayable

Figure 2: The Scientific Visualization Model

The first extraction/refinement transformation is applied to reduce the input simulation data to a sizeable and
formatted data set, in order to make it acceptable for subsequent processing steps. Typically, the simulated data
are the values of some physical variable at discrete locations of the solution domain. It is often necessary to use
some form of interpolation of the given values to obtain the new data at other specified points. For example, it
may be necessary to calculate the values of the variable on an arbitrary curve which is itself defined by a set of
discrete points. Such process requires the creation of additional data sets in order to approximate parts of the
continuum domain. For example, an additional data set would be needed to define a set of surface normals,
which must be available to the rendering transformation for it to display the surface as a smooth surface, i.e.
without visible edge discontinuities.
The second step is an enhancement/enrichment transformation, which relates data values to graphical attributes
such as color, transparency, luminosity, texture and reflectance. Usually, the mapping functions between values
and graphical attributes are interactively manipulated by the user. Nonlinear mappings tend to be more useful
than linear ones, as they can better reveal details of the visualized phenomena. As there are no methods capable
of identifying the best transfer functions, the user-driven interactive approach can be quite effective in
exploring different transfer functions.
The third and last step consists of rendering transformations which produce displayable images. Typical
rendering operations include view mappings such as rotation, translation and scaling, followed by perspective
mapping and clipping. The rendering of 3D volumetric effects include hidden line/surface removal, shading,
anti-aliasing and so forth. More elaborate algorithms deal with transparency and shading (ray tracing).

Figure 3: The Visualization Data Sets

The SV model needs to be well understood by the user -- the investigator -- in order for him/her to correctly
interpret the displayed data. The user must be able to fully control the amount of visualized information in order
to extract meaningful information from the data without being overloaded with graphical content.
Figure 3 shows examples of results of the visualization process. The solid model, mesh grid and computed
physical quantities are initial simulation data. Applying the next extracting transformation permits the interactive
generation of user-defined surface or curves. The mirror surface extracted from the double ellipsoid test case
represents the reduced data set. In the third step, the image is enriched and enhanced by a color mapping.
Computer-graphics primitives, such as polygonal meshes, store the data in the format treatable by the underlying
computer graphics engine. The final rendering transformation treats the 3D graphical objects with viewing and
lighting transformations.

Scientific Visualization Environment

The scientific visualization system must be an integrated part of the computational environment (Figure 4) if it is
to efficiently assist the scientist/engineer in his/her numerical simulations data analysis. Such software systems
are used today in research laboratories and in the industry. In the industry, visualization is used to gain a more
quantitative understanding of the simulated phenomena (ex. aerospace product design); the results of
visualization are also used in management and commercial presentations. In contrast, in the research laboratory,
scientists develop codes and try to understand qualitatively how the simulation algorithms behave; in this
context, they tend to use SV as a debugging tool. In both cases, the computational environment includes software
that supports geometrical definition (as in CAD systems), mesh generation (pre-processing), supervision of the
simulation (co-processing) and display and/or analysis of results (post-processing).
It is understood that in order to be effective, CFD software must be integrated into the (usually large) software
platforms used by the research institutes or industries. Clearly, the success of such software integration depends
on the sound application of software engineering methods, not on advances in CFD models or algorithms.
Building such systems requires the coordinated, multi-disciplinary effort of CFD and computer experts.
Developing integrated computation/visualization platforms has an impact on the engineering of simulation
software, which must interoperate with the visualization software; this, in turn, places new constraints on the
computer hardware which must meet the computational and graphical demands of the integrated software
solution.

Figure 4: Integrated Computational Environment

6

The CFD simulation process consists of:

geometry definition
grid generation
numerical simulation
solution analysis and comparison
In each of these phases, applying scientific visualization is of importance.
Geometry definition involves the creation of a spatial model of the object of interest (for example, a car or a part
of a plane) which includes defining the shape of the object. Visualization is applied here to explore (variants of)
the resulting model and to validate the input data received from a CAD like system.
Grid generation involves the discretization of the problem domain (on a structured or unstructured grid).
Visualization and graphical input techniques are applied to construct geometrical models, to evaluate the results
of the modeling, to detect errors and to control and monitor the grid generation process.
Numerical simulation generally involves discrete representations of the field variables and approximate
statements of the boundary/initial value problem. The analytical mathematical problem is transformed into an
approximate mathematical problem that can be solved using numerical methods. In nonlinear and transient
solutions, visualization is used to monitor the behavior and evolution of the simulation. Here, visualization can
rapidly reveal unstable and divergent (erroneous) solutions. Visualization can also help the investigator to tune
the solution parameters interactively so as to maintain the accuracy and stability of the computation. Clearly,
visualization modifies the very nature of the investigation process by making it highly interactive, providing
more control to the engineer/scientist who can better tune his/her computational experiment, optimize the use of
computer resources and dramatically reduce the time to results.
The result of the numerical simulation is a solution, i.e. the field values of some primary quantities. In the
solution analysis phase, secondary (or derived) quantities are computed using data extraction and postprocessing calculations; the secondary data are essential for validating and presenting the results of the
simulation.
Current best practice suggests four visualization phases, involving different performance and memory
requirements for computation, graphics, mass storage and communication:
pre-processing
co-processing
post-processing
interactive steering
Pre-processing verifies input settings before the simulation is computed. The post-processing mode treats the
simulation output data stored for visualization. This mode is often appropriate since the cost, time or effort of
repeating the simulation is usually larger than the cost of storing the results. It supports interactive visualization
and real-time animation (useful when the simulation itself cannot be performed at animation rates). During the
co-processing, the visualization system monitors in real-time the evolution of the simulation in progress by
displaying the computed data at each iteration. Data histories can be accumulated to permit animation of partial
results. This allows the user to monitor the behavior of the solution, to identify convergence or other problems
and to abort the simulation if errors occur. The fourth mode -- the most demanding one as far as computerresources are concerned --, is called interactive steering[8]; it combines simulation and visualization in a
system-user closed loop where the user can interactively modify the simulation parameters with immediate
visual feedback. A data accumulator can be added to the pipeline at the end of the visualization mapping to

store intermediate results. If the systems rendering module is capable of real-time animation, the accumulated
data can be visualized while the main numerical computation continues in parallel.
Interactive visualization accelerates the CFD design cycle by allowing the user to jump at will between the
various phases so as to optimize his/her CFD analysis. The user conducts the investigation in a highly interactive
manner, can easily compare variants of a simulation/analysis and may intuitively develop a deep understanding
of the simulation and of the calculation details. An example of an integrated environment application is the
Virtual Wind Tunnel [9], which reproduces a laboratory experiment in a virtual reality environment, where a
virtual model can be created and put to test with dramatic cost and time savings compared to what is done in the
real laboratory.

Scientific Visualization Software - state of the art

SV software has progressed enormously during the past two decades. One reason is the exponential increase in
the power of the computer (central and graphical processors), which has led to todays low-cost PCs providing
as much power as the high-end mainframes of some years ago. Development of advanced SV tools is no longer
the prerogative of specialized labs with costly computer equipment. Yet, there is an un-diminished demand for
new visualization-enabled software, driven by continuous hardware changes and the emergence of new software
platforms. Interactive visualization remains a key element of advanced engineering/scientific software, and their
design must account for this fact. There are presently many commercial interactive visualization products on the
market which provide SV functionality with increasing success. Such visualization systems are widely used in
application areas as diverse as nuclear energy exploration and atmospheric research. In the field of engineering,
such products are commonly used to visualize flow patterns and stress fields, and generally to study large multidimensional data sets. SV applications are used in many industries including aerospace, medicine, power
production, shipbuilding, geophysics, automotive, electronics, oil, agriculture, food production, etc. SV
applications are now ubiquitous in engineering and science, be it in:
Fluid Mechanics,
Structural Analysis,
Electromagnetic,
Thermodynamics,
Nuclear Physic, etc.
For the sake of completeness, let us mention that SV has been (and is) instrumental in advancing the state of the
art in industrial applications involving fluid flow modeling, such as:
Aerodynamics of trains, cars and airplanes.
Hydrodynamics of ships and floating structures.
Flow in turbo-machinery and torque converters.
Cryogenic rockets, combustion chambers simulations.
Flow in manifold, pipes and machinery.
Medical researches, circulation of blood in veins.
It is evident that advances in CFD software are driven by demands from many application areas, which in turn
places requirements on the associated visualization software. Today, visualization software solutions with
interactive 3D graphics capabilities can be categorized into four groups:

Figure 5: CFView the scientific visualization system

1. Visualization Applications
2. Modular Visualization Environments
3. Visualization Toolkits
4. Integrated Modeling Environments
1. Stand alone visualization applications are software solutions which offer direct functionality to the user, who
is responsible for defining the data set to be loaded for visualization. Known visualization applications in
CFD and Finite Elements Analysis (FEA) engineering are as follows:
EnSight from CEI [10],
FieldView from Intelligent Lights[11, 12]
TecPlot from Amtec Engineering Inc.[13]
CFView from NUMECA [14]
PLOT 3D, NASA [15]
FAST (Flow Analysis Software Toolkit),NASA [16]
VISUAL2-3 from MIT [17],
FLOVIS from CIRA
HighEnd from DLR, [18].
ParaView from VTK [19]
VisIt from Lawrence Livermore National Laboratory [20]

Such programs are appropriate for users who need off-the-shelf visualization functionality. Such software
implements the event-driven programming paradigm which is suitable where all functions are launched by
the user interacting at with the Graphical User Interface (GUI). This is the case for CFView [21], see Figure
5, a scientific visualization application developed by the author over the 1988-98 period. CFView started as
an academic application in 1988 and was continuously upgraded in the following years. In the mid 90s,
CFView was taken over by the VUB spin-off company NUMECA and integrated in FINE, NUMECAs
engineering environment. FINE is an environment that nicely illustrates the variety of visualization tasks that
need to be performed to solve an engineering problem, especially addressed to turbomachinery applications.
2. Modular Visualization Environments (MVE) are programs often known as visualization programming
environments; examples are[22]:
Advanced Visual Systems AVS [23],
Iris Data Explorer from Silicon Graphics[22, 24],
OpenDX the IBMs Data Explorer[25],
PV Wave from Visual Numeric [26].
Their most significant characteristic is the visual programming paradigm. Visual programming intends to give
users an intuitive GUI for them to build customized visualization applications. The user graphically
manipulates programming modules displayed as boxes, which encapsulate the available functionality. By
inter-connecting boxes, the user defines the data stream from one module to another, thereby creating the
application. The MVE can be viewed as a visualization network with predefined building blocks, and which
often needs to be quite elaborate in order to be useful to the user. The freedom given to the users to design
their own visualization applications is the strength of so-called application builders. This class of software
implements the data flow paradigm, with the drawback that iterative and conditional constructs are difficult
to implement. For example, PV Wave uses an interactive fourth-generation programming language (4GL) for
application development, which supports conditional logic, data sub-setting and advanced numerical
functionality in an attempt to simplify the use of such constructs in a visual programming environment. The
interactive approach is usually combined with a script-oriented interface, and such products are not easy to
use right out of the box and have a longer learning curve than stand-alone applications.

Figure 6: The OpenDX Application Builder

10

There is an ongoing debate on whether the best way to procure visualization software is to use stand-alone
applications or to build applications using MVEs. Time has shown that both approaches are equally accepted
as there is no alternative. The approach that we chose to follow in our work is a compromise between the two
options. The GUI of our CFView software looks very much like that of a stand-alone visualization
application; internally though, CFView is an object-oriented system which has the flexible, modular
architecture of an application builder. This is to say that a new component can be integrated in the core
application structure with a minimum coding effort; also, that the propagation effects resulting from the
modification are kept limited.
3. Visualization Toolkits are general-purpose object-oriented visualization libraries, usually present as
background components of SV applications. They emerged in the mid 1990s, and the two representative
examples are VTK[27] and VisAD[28]:
The Visualization ToolKit (VTK) is an open-source software system for 3D computer graphics,
image processing and visualization, now used by thousands of researchers and developers in the
world. VTK consists of a C++ class library and several interpreted interface layers including Tcl/Tk,
Java, and Python. VTK supports a wide variety of visualization algorithms (including scalar, vector,
tensor, texture and volumetric methods), advanced modeling techniques (such as implicit modeling,
polygon reduction, and mesh smoothing, cutting, contouring and Delaunay triangulation). In
addition, dozens of imaging algorithms have been directly integrated to allow the user to mix 2D
imaging / 3D graphics algorithms and data.
The VISualization for Algorithm Development (VisAD) is a Java component library for interactive
and collaborative visualization and analysis of numerical data. VisAD is implemented in Java and
supports distributed computing at the lowest system levels using Java RMI distributed objects.
VisADs general mathematical data model can be adapted to virtually any numerical data that
supports data sharing among different users, different data sources and different scientific
disciplines, and that provides transparent access to data independent of storage format and location
(i.e., memory, disk or remote). The general display model supports interactive 3D (Figure 7), data
fusion, multiple data views, direct manipulation, collaboration, and virtual reality. The display
model has been adapted to Java3D and Java2D, and virtual reality displays.

Figure 7: VisAD application example

11

4. Integrated Modeling Environments (IME) is software that combines two or more engineering applications
and visualization systems to solve a multi-disciplinary problem. For example, the naval architect shapes the
ship hull in order to reduce the ships hydrodynamic drag, while the stress engineer calculates the ships steel
structure. Both use visualization to analyze the data generated by the hydrodynamics and stress calculation
solvers. The visualization software may be able to process the CFD flow-field solver data and the FEA stressfield solver data in a unified manner, giving the two engineers the possibility to work on compatible,
interfacing 3D representations of the hydrodynamic and structural problems. An example of such integration
is the Product Life-cycle Modeling (PLM) developed by Dassault Systmes and the CFD solver technology
developed by ANSYS, Inc, where the FLUENT CFD flow modeling approach is integrated in CATIA CAD
tools throughout the whole product lifecycle [29].

Figure 8: The integrated modeling environment from Dassault Systmes and ANSYS, Inc

12

Object Oriented Methodology

Computer hardware has improved drastically in quality and performance in the last 30 years, much faster than
software quality and complexity. The trend is drawn qualitatively in Figure 9. The main reason for this situation
is to be found in the reusability of hardware components (chips), which are the cheap and reliable building
blocks of hardware systems, small and large. To date, software components with similar properties simply do not
exist, and reusable software chips are not commercially available. The effort to design and produce such
software would be too large, and standardization is not pursued by software makers who keep customers captive
with proprietary software and computer platforms. As a result, software production cannot keep pace with the
hardware technology, a situation often recognized as symptomatic of a software crisis.
Software/Hardware productivity
4.0

3.5
Hardware
Software

3.0

Complexity

2.5

GAP

2.0

1.5

1.0

0.5

1980

1990

2000

2010

2020

Year

Figure 9: The comparison of Hardware/Software productivity

A concern in this work was to try and produce visualization software that could, intrinsically, evolve as fast and
as cheaply as hardware. This led us to select an Object Oriented Methodology (OOM) for constructing software
components, specifically the component parts of our SV and CFD software.
OOM is a fairly universal approach that can be applied to solve many types of complex problems. The goal of
OOM is to reduce the system complexity by decomposing it in manageable components called objects.
Experience has shown that solving problems in a piece-wise manner leads to better quality and easily scalable
solutions. The system is cut into component pieces represented by objects that interact, through well-defined
interfaces, by exchanging information through messages.
The interesting feature of OOM is that objects can be created and developed independently, even with no a priori
knowledge of the application in which the objects will be used. The existence of an object is independent of any
specific application. The concept can be illustrated considering the Partial Differential Equation (PDE). The
PDE exists as a mathematical concept which can be applied, for example, in ship propulsion design or turbomachinery flow analyses. The key aspect here is that the mathematicians do not need to know anything about the
application of the PDE to develop solutions to the PDE. Hence, by encapsulating the knowledge on the PDE in
an object (software component), one makes it available to and reusable by any application or software that may
require it. The benefits are clearly a reduction in the application development effort, shorter production times and
higher application maintainability. The benefits stem from the fundamental property of reusability of the PDE
object. Any improvement to the PDE object would immediately benefit all the applications that use it. An
interesting consequence of having many reusable software objects would be that it would then make sense to get
hardware designed to fit the available software (and not the reverse as is the case today). The principles of
software reusability and portability have been fundamental to all the work described in this thesis.

13

Reusability is an intrinsic feature of all OO software and their efficient exploitation promotes the computer
network to become a commercial market place, as the Internet, in which such general-purpose and specialized
software components needs to be available, validated and marketed [30].
The OO approach has led to the emergence of Object Oriented Programming (OOP) with specialized OO
programming languages -- such as Smalltalk[31], CLOS, Eiffel[32-34], Objective C [35], C++ [36], Java, C#
and other derivatives -- which apply encapsulation and inheritance mechanisms to enhance software modularity
and improve component reusability. It is important to stress that the highest benefit of OOM is obtained when
OOM covers the full software life-cycle, from the requirements specification phase to the software delivery
phase. When an application is created applying OOM, reusability in different development phases can be
expected. First, the OOP brings in object-oriented libraries, which provide components validated in previously
developed applications. Second, the software design of the previously modeled software could be reused through
the established design patterns. Previously developed components may be reused for the new application which
does not need to be designed from scratch, which is an obvious advantage. Improvements that could be brought
to the existing, re-used objects would also improve the older applications that use the same objects.
It is interesting to note that the OOP paradigm has shifted the emphasis of software design from algorithms to
data (object, class) definitions [37-39]. The object-oriented approach can be summarized in three steps. The
system is first decomposed into a number of objects that characterize the problem space. The properties of each
object are then defined by a set of methods. Possibly, the commonality between objects is established through
inheritance. Actions on these objects and access to encapsulated data can be done randomly rather than in a
sequential order. Moreover, reusable and extensible class libraries can be created for general use. These are the
features which make OOP very attractive for the development of software, in particular for interactive software.
It should be mentioned that OOM does not directly reduce the cost of software development; however, it
markedly improves the quality of the code by assuring consistent object interfaces across different applications.
Estimated software construction times are often incorrect. Time and resource allocation tend to be largely
underestimated in software projects, not uncommonly by factors of 2 to 5, especially where innovative features
are to be developed. Unfortunately, for the software construction planning we do not have an underlying
engineering, scientific or mathematical model to calculate the software development time required, when starting
the new software development process. The theoretical basis of how to best construct software does not exist.
The ability to plan the project costs, schedule millstones, and diagnose risk is ultimately based on experience,
and could be only valid for a very similar application done in the past and applying the same development
environment.

RESPONSIBILITY

LAYER

TASK

DEVELOPER

Application - Database

simulation data

Visualization System

graphics data

Graphics & Windowing System

displayable data

Device Drivers

low-level graphics instructions

Operating System

select hardware unit

Hardware Platform

I/O devices, CPU and memory

VENDOR

Table 1: Layered Software Architecture

14

Figure 10: Graphics Engine as combine software/hardware solution

It is also important to ensure the production of portable code, i.e. code that can run without need of adaptation
on computing platforms other than its native platform. Porting -- adapting software to a computer system other
than the one for which it was originally designed -- can be a tedious and costly process. Portability can be
improved by adopting standards which various hardware/system platforms support. For example, one may adopt
the OpenGL standard which is supported by graphic boards. This ensures that only a small kernel of code must
be modified before recompilation for another hardware platform. This is why it is necessary to breakdown the
software architecture in carefully-defined layers as seen in Table 1. Understanding the lower layers of software is
of little or no interest to most scientists; these layers involve graphics languages, windowing systems,
communication mechanisms and device-drivers. The system vendors have understood this and provide graphic
platform systems which implement the four layers below the visualization system layer. As depicted in Table 1,
the upper two system layers are under the responsibility of the software developer; they include data extraction
and enrichment. The visualization system maps numerical data onto graphical data. The rendering of graphical
data is delegated to the graphics and windowing layer. In this way, porting of the software system to any
hardware platform is possible without modifying the visualization application.
There is often a need for graphics hardware (graphics boards) that can enhance the performance of the most
demanding 3D-rendering tasks requested from the graphics and windowing layer. The vendors layers deliver a
well-defined set of callable graphics routines -- examples: OpenGL, DirectX -- collectively called the graphic
engine. Todays graphics engines become increasingly powerful as software implementations are replaced by
hardware implementations (see Figure 9). This is the case for a variety of UNIX platforms, with graphics
engines implemented in specialized hardware such as: Silicon Graphics GL and OpenGL, HP Starbase, HP-GL,
SUN XGL, Xlib and PEXlib, Microsoft GDI, Quick Draw and Postscript.
In the beginning of the 1990s, CFView was ported onto most of these graphical engines.
A graphics engine typically processes floating-point input data to generate graphics. Example of graphics data
models are lines and polygons. Hence one assumes that line drawing and polygon filling are functions provided
by the graphics engine, and one needs not be concerned with developing low-level graphics routines. One can
therefore focus on generating the data sets that are needed to feed the graphics engine.

Figure 11: Software Components Distribution

15

To develop the visualization software, our approach must be multi-disciplinary in the sense that it puts together
an application engineer and a computer specialist in order to develop different application layers, as shown in
Figure 11. The software development environment needs to enable the evolution of the software under
development and has to provide a framework for porting applications across different hardware/operating
systems/windowing systems. Also, it has to simplify the process of the creation of interactive graphical
applications, with enabling the application engineer to have under control the application software layer and hide
the lower software layers of the system, as depicted in Figure 11. Thus, the object-oriented approach was
selected to introduce the necessary abstraction levels, necessary for organizing the inherent complexity present in
the development of the scientific visualization software.

R&D projects history

In 1988, we started developing the SV system that was to be named CFView [40]. At the time, the basis of the
Object Oriented approach [41] for developing SV systems was being established. The object-oriented
programming language (OOPL) C++ was chosen, because of its high performance in number crunching and its
capability to support OOP constructs.
CFView [21] was designed to work with structured and unstructured mesh types. CFView was found to be
particularly valuable when treating the CFD simulations of the European Hypersonic Database [42] and
validating the simulations against experimental data. CFView allows the simultaneous inspection of fluid flow
simulations with structured and unstructured grids; as it uses data extraction algorithms such as plane-cutting,
iso-surface and particle-tracing for both grid models, to uncover interesting features hidden in such large data
sets. These routines require large computational resources and tend to limit the interactive feedback of CFView.
In order to overcome this problem, a heterogeneous system named Parallel CFView was constructed which
distributes the computing load over several processors and permits real-time, interactive visualization of CFD
data. The development of Parallel CFView [43] was part of the EU-funded projects PAGEIN [44] and
EEC/ESPRIT PASHA [45].
In the PASHA project, CFView visualization system was ported to MIMD and SIMD platforms for comparative
benchmarking. The performance of the MIMD implementation was evaluated on two industrial test-cases
provided by the EUROPORT-1 project. Both SIMD and MIMD parallel machines were used to provide massive
back-end processing power. A key achievement of the Parallel CFView project turned out to be the interface for
communication between machines with heterogeneous architectures. An analysis was carried out to compare the
performance of such distributed computing environment consisting of both: the SIMD and MIMD
implementations and the sequential CFView system. The results showed that the parallel implementation of the
extraction algorithms speedup the performance of the visualization process by a factor of 10, which proved the
viability of such visualization system configurations.
PAGEIN demonstrated more effective resource utilization through the support of distributed computing and
collaborative engineering. The goal was to seamlessly integrate scarce human expertise in domains like,
supercomputers, massively parallel computers and data servers. PAGEIN enabled the engineers to exploit the
computing resources and reach the critical expertise mass without need for co-location. For the users PAGEIN
has integrated on top of wide area networks, applications and tools for supercomputing, visualization and
multimedia. It has demonstrated the viability of such combined technologies in CFD to be exploited on a
European scale by aerospace actors.
The Live Code Learning Multimedia System (LCLMS) [46] was the IWT founded project, in which the database
and network aspects for multimedia data sets were researched. This project strongly influenced the authors own
research towards distributed and collaborative environments for engineering applications based on low-cost PCplatforms. LCLMS provided the basic technology upon which QFView was designed.

16

Figure 12: QFView Web Interface

The development of the Quantitative Flow Field Visualization (QFView) system [47] was made in the ESPRIT
Project 28168 ALICE. QFView is a distributed software environment that integrates EFD and CFD data
processing, including flow field mappings with flow field visualization. QFView was devised to support a
unified treatment of data while providing for:

the validation of results from experimental and numerical systems, and

the archival and retrieval of data from unified (experimental and numerical) flow field database.

Based on proven Internet and World Wide Web (WWW) standard technologies, QFView provides an integrated
information system for fluid dynamics researchers (see Figure 12). QFView is a web-based archival, analysis and
visualization system, which enables the manipulation and extraction of data resulting from laboratory
measurements or computational simulations. The system is suited for combining experimental and computational
activities in a single operational context. This results in an increase of productivity, since the system facilitates
for the exchange of information between investigators who conduct the same or similar simulations/experiments
in different geographical locations, can be conducted in collaboration or independently.
The rapid progress in all facets of fluid dynamics research has made it essential that the research activities are
conducted in close cooperation between experimentalists, numerical analysts and theoreticians. In the early
stages of the CFD development the improvements of the process was so rapid that it was expected to eliminate
the role of EFD all together. However, experience showed that in order to study new phenomena experiments are
still the most economical approach. The CFD codes, ones validated against experimental data, are the most
effective tool for the production of data to build a comprehensive flow database [48]. The strengths of various
tools in EFD and CFD should be used judiciously to extract significant quantities required for problem solving as
shown in Figure 13.

17

Validation data
Statistical data
+Reynolds stresses
+Empirical parameters

PIV

Constants for Model tuning

Input
+Validation
+Truncation

Velocity field
+Vortricity field
+Length scale

DNS
Lower time scale

EFD

RANS

CFD

Resolution
requirement
Eddies
+SGS model

Fluctuation

HWA
LDV

Input
+Validation
+Eddies

LES

Figure 13: Use of EFD and CFD tools

Three laboratory experiments were performed at the VUB in order to test and demonstrate the capability of
QFView regarding web-based collaborative data access, namely:
1.

The flow downstream of a bluff body in a double annular confined jet [49-51]

2.

In-cylinder axi-symetric flows [52]

3.

Flow pattern of milk between 2 corrugated plates[53, 54]

The experiments generated large data sets -- measured and calculated values of physical quantities-- which were
visualized and compared in order to support the validation procedure of the obtained results.

The flow downstream of a bluff body in a double annular confined jet

The cold flow in the cylindrical combustion chamber of a prototype industrial burner was investigated. Two
concentric annular axial jets were used to approximate this complex flow field; the flow downstream of the bluff
body was investigated over a range of Reynolds numbers using flow visualization, particle image velocimetry
and laser Doppler velocimetry, see Figure 14.

Figure 14: Laminar flow at Ret = 60: DPIV of nearly axisymmetric flow, LSV of vortex shedding and CFD at
various degrees of non-axisymmetric flow

18

Figure 15: PIV system at VUB

In-cylinder axisymmetric flows

The main objective was to get an insight into the structure of in-cylinder flow during gas exchange processes due
to its capital influence on the combustion process and consequently on the efficiency of internal combustion
engines. A simplified in-cylinder flow in axisymmetric geometries and the flow through the inlet valve for
steady and unsteady conditions were investigated and experimental results represented with the QFView system.
Two test rigs were used for flow visualization studies and PIV measurements of the cylinder equipped with a
central valve:
1.

for the steady flow through the inlet valve and

2.

for in-cylinder transient flow (of water) in a mono-cylinder.

(a)

(b)

Figure 16: Flow pattern for 90oATDC: (a) visualization at 20 rev/min, valve lift = 10 mm (b) Average velocity
field at 5 rev/min, valve lift = 10 mm

19

The influence of the speed and valve lift on the in-cylinder flow in the motored mono-cylinder was investigated
while in the steady-state flow test rig; only the effect of the valve lift on the intake jet and stability of the tumble
motion was retained. The similarities and differences in the two types of flows were analyzed in order to
understand the relevance of the steady-state experimental results to model the unsteady flow in the motored
cylinder. Figure 16 (a) shows an example of visualization of the flow pattern in the mono-cylinder at 90o CA
after top dead centre (ATDC). The details of the flow structure were presented in terms of maps for average and
instantaneous variables such as velocity (Figure 16 (b)), turbulent kinetic energy (Figure 17), vorticity (Figure
18), normal and shear strain rates. The main findings were:

the integral length scales are the same for the investigated stationary and unsteady flows, a finding
which supports the hypothesis of the isotropy of the small-scale structures
the footprint of the flow is given by the root-mean square velocity and turbulent kinetic energy fields
the flow is a complex collection of local jet and recirculation flows, shear and boundary layers

Figure 17: Turbulent kinetic energy field at 90oATDC at 5 rev/min, valve lift=10 mm

Figure 18: Average vorticity field at 90oATDC at 5 rev/min, valve lift=10 mm

Flow pattern of milk between 2 corrugated plates

A detailed calculation of the flow pattern of milk between 2 corrugated plates was carried out using 2D and 3D
CFD calculations. The 2D calculation shows the influence of the corrugations shape, but the 3D calculations are
necessary to assess the importance of the corrugation orientation. A model was constructed that allowed for a
positive qualitative validation of the simulation results. The calculations allow identifying those regions where
turbulent backflows and thus higher temperature regions near the wall can occur. These regions are the most
sensitive to fouling and should be avoided as much as possible through better design. In this respect, CFD can be
regarded as a valuable assistant for the design optimization of plate heat exchangers.
20

Figure 19: Experimental and CFD model for flow analysis between corrugated plates

The development of the QFView environment highlighted the need for computer specialists and engineers to
collaborate. Our work demonstrated the indispensable role of multi-disciplinary teamwork for progressing
scientific visualization systems and developing next-generation engineering environments, which need to
combine CFD codes and EFD experiments facilities in an integrated information framework, applicable to
different sectors of industry, research and academia.

LASER

COMPUTATION

EXPERIMENT

VIDEO CAMERA
Computer for recording
&
Post-processing images

Database administration
&
post analysis

Database

SEEDING
S
FLOW
VISUALIZATIO
FIELD
N

VIDEO
RECORDING

Quantifying
moving images

ILLUMINATION

Figure 20: Workflow for the integration of EFD and CFD simulations

21

VELOCITY
FIELD
manipulation

After the September 11-2001 event, when the ALICE project ended, the authors research shifted towards
general-purpose (as opposed to scientific) visualization systems. This work was carried out in the LASCOT
project [55] funded by IWT, being part of the European Information Technology European Advancement
(ITEA) program for R&D of software middleware. The LASCOT visualization system demonstrated the
possibilities of 3D graphics to support a collaborative distributed knowledge-management and decision-making
applications. The research challenge was to deliver a visualization system capable of enhancing situation
awareness (i.e. information that the actor-user has to manipulate to solve a crisis situation), which was done by
providing a 3D interface that gave the user the possibility to navigate in and interrogate an information data-base
in a highly intuitive manner. It was a main requirement that the visualization technology should be capable of
assisting the user in performing decision-making and knowledge management tasks. The LASCOTs
visualization system was built upon the JOnAS [56] application server (developed on the European J2EE
architecture). These technology and architecture could be easily used for developing new engineering integrated
environments.
The work presently carried out by the author in the ITEA SERKET project is about security issues and protection
against threats, which relates to the development of 3D graphical models capable of integrating and rendering
data acquired from heterogeneous sensors like cameras, radars, etc.
Over the last 20 years, the author has initiated and worked on many research projects with a clear focus on
developing scientific visualization software and advancing the state of the art in this area. There are potential
avenues of future research, as discussed in the next chapter.

Towards an Integrated Modeling Environment

Todays trend in software development is towards more intelligent, multi-disciplinary systems. Such systems are
expected to capture engineering intelligence and to put in the hands of the engineer advanced tools for designing
new products or performing investigations.

Figure 21: Example of an Integrated Modeling Environment [57]

22

The Integrated Modeling Environment (IME) [58] concept is quite recent, yet its roots can be found in 1stgeneration CAD-CAM tools. An IME system attempts to offer to the engineer a homogeneous working
environment with a single interface from which various simulation codes and data sets can be accessed and used.
In the fluid mechanics application area, an IME system needs to integrate the latest CFD and EFD good
working practice; the system must be constantly updated so that at any time, it runs on the most-recent
software/hardware platform (see Figure 21).
An IME system consists of an Internet portal from which the investigator is able to access
information/knowledge/databases and processing functions, at any time and wherever they are located/stored.
He/she has access to accurate and efficient simulation services, for example to several CFD solvers. Calculations
can be performed extremely fast and cheaply where solvers are implemented as parallel code, and grid
computing resources are available. Results obtained can be compared with separate experimental results and
other computations; this can be done efficiently by accessing databases that manage large collections of archived
results. The possibilities for benchmarking and for exchanging knowledge and opinions between investigators
are virtually infinite in an IME environment. Clearly though, a pre-requisite for an IME environment to work is
its adoption by its user community, which agrees on a specific codex that enables and guarantees openness and
collaboration. Typically, an IME system will open Web-access to:

Computational Services: selection of simulation software and access to processing and storage
resources

Experimental Services: access to experimental databases with possibility to request new measurements

Collaborative Services: chat and video-conferencing, with usage of shared viewers (3D interactive
collaboration)

Visualization is required to support many tasks in an IME software. This poses the problem of building/selecting
data models that can be used by the visualization components to present the information correctly to the users,
whilst offering to them tools for real-time interaction in a natural, intuitive manner. The IME can include walldisplays connected to high-performance, networked computing resources. Such systems and architectures are no
longer a mere vision: they are becoming reality, which opens new challenges for scientific visualization software
researchers and developers.

Figure 22: Example of the 3D virtual car model testing

23

The usefulness of IME can be illustrated by considering how it would help the car designer. Assume that the
present goal is to find out how a new car performs under various weather conditions and to assess its safety level
before manufacturing begins. Assume that test-drive simulations are performed digitally, and that they are run in
parallel. Clearly, car behavior problems can be readily identified, and design can be modified rapidly and at little
cost; the expected savings in time and money are clearly substantial.
The QFView system that we have developed provides certain features of IME systems like the web-access and
data sharing but does not integrate distributed computational resources, which might be addressed in the coming
future.

Thesis Organization
The motivation for the research work presented in this thesis, and the objectives that were pursued are described
in the Introduction Chapter, which also covers the state-of-the-art and an outline of the authors research and
development work in relation to object-oriented visualization software.
The main body of the thesis is then subdivided in 3 Chapters. The first Chapter is dedicated to discussing
modeling concepts and fundamental algorithms, including discretization, i.e. the modeling of the continuum in
cells and zones. This Chapter also presents the theoretical foundations of the basic algorithms applied to
geometries, scalar and vector fields.
In Chapter 2, the visualization tools are explored to show different possibilities of extracting and displaying
analyzed quantities. Improved and adapted visualization techniques are illustrated in several examples.
Interactivity is then considered by explaining the Graphical User Interface design model. The Chapter ends with
a discussion of Visualization Scenarios which allow the standardization of the visualization process.
Chapter 3 is devoted to discussing the relevance and appropriateness of the object-oriented methodology (OOM)
for visualization systems, and the associated implementation issues. Object-oriented programming concepts and
object-oriented data model are reviewed. The design and implementation of the visualization system is presented
-- architecture, description of important system classes, management of input/output files, and development of
portable and reusable GUI code.
The last Chapter explains how OOM permitted the development of Parallel CFView, an extension of the basic
CFView visualization system that takes advantage of distributed and parallel processing. The Chapter covers the
upgrading of the QFView system to distributed data storage/archiving for scientific applications, then its
application for visualization in general-purpose information systems as prototyped in the LASCOT and
SERKET ITEA projects. We conclude this Chapter with suggestions for future research.
During the course of the work presented in this thesis, our CFView visualization system has evolved from an
academic laboratory prototype to a versatile and reliable visualization product providing a highly-interactive
environment capable of supporting the most demanding fluid flow investigations. Researchers can reliably use
and control CFView to extract, examine, probe and analyze the physics of flow fields, by simple commands
through an intuitive graphical user interface.
One of the objectives of this work was to explore object-oriented programming and design a new development
methodology for scientific visualization systems. This objective was achieved with the production of CFView, a
visualization system fully implemented in C++ (one of the OOP languages) [36].

24

1 Modeling Concepts and Fundamental Algorithms

The Object Oriented Methodology (OOM) software modeling concepts were applied and further developed in
order to establish the software model of the SV process. The fundamental concept in OOM is the object; it is
the elementary building block for mapping scientific and engineering concepts to their software equivalents.
The object is an abstract construct, which approximates (in a simplified manner) the understanding of real
concept in consideration, which is often quite complex. Consider, for example, how the physics of fluid flows
are described in terms of numerical equations and how these equations are modeled by software objects. These
objects are useful because they are identifiable elements with a well-defined purpose: each object performs a
given function by encompassing a certain mathematical or physical intelligence For example, an object
modeling a second-order differential equation; or a object modeling the viscosity of a liquid at a given
temperature; etc., in such a way that it can be re-used by the software engineer with no need to understand the
internal working details of the object. The obvious reused object in the real life is a car. We need it to go from
one place to another, but we do not need to know how it is build. We use it, and this is a way the software
engineer is supposed to reuse objects.

Figure 23: Software model as communication media in the software development process
The software model is a fundamental element in OOM software development. The model describes the
knowledge mapped in the software in a formal, unambiguously defined manner. Such a precise specification is
both the documentation and the communication tool between the developers and the users; recall that the term
developer includes application analysts, software designers and coding programmers (see Figure 23).
In the software development process, the analyst creates an abstract model that will be partially or fully
implemented. The designer uses that model as a basis to add specific classes and attributes to be mapped onto
one or more OOP languages. The designer specifies the detailed data structure and functional
operations/processes, which are required by the application specification. Finally, the programmer receives the
analysts and the designers models for implementation into source code. The source code is compiled to
produce the executable software. Software modeling is then the iterative and incremental process which maps
abstract concepts into formal constructs that eventually become reusable software entities. In OOM, the object
model comprises a data model and a functional model; see section 3.2 Object Oriented Concepts. The
specification of an object includes a description of its behavior and of the data necessary and sufficient to
support its expected functionality. The data model describes the pertinent data structures, the relations between
the objects and the constraints imposed on the objects. The functional model describes the objects behavior in
terms of operations. From the data model point of view, the primary concern is to represent the structures of the
data items that are important to the scientific visualization process and the associated relationships. The

25

modeling tool that was applied is known as the Entity-Relationship Model (ERM) [59]; it is well-adapted for
constructing static data aspects and ensures the completeness of and the consistency between the specified data
types. Figure 24 depicts an ERM with Surface, Section and Vertex entities (shown together with their
relationships). Modeling entities define the direct association between problem entities (visualization data) and
software objects (manipulative data), which establishes the reference basis for the incremental software
development. This is important, because software modeling is not a one-shot process but a continuous or
repetitive one, since software must be changed to account for the evolution of the user requirements and/or of the
technology platforms.
It is important to mention that the terminology introduced by OOM standardizes the names of the objects which
will constitute the system in all development phases, so that the Naming Convention must be strictly preserved
and complied with in the software model. ERM consists of three major components:
1.

Entities that represent a collection, a set or an object and which are shown as rectangular boxes
(Surface, Vertex). They are uniquely identified, and may be described by one or more attributes.

2.

Relationships that represent a set of connections or associations between entities; they are shown as
diamond-shaped boxes.

3.

Attributes that are properties attached to entities and relationships and which are shown in oval
call-outs.

In Figure 24, the integer numbers on each side of a relationship-type denote the number of entities linked by the
relationship. Two numbers specified on one side of a relationship indicate min-max values. For the consist of
relationship, the Vertex can be part of one and only one Surface and a Surface consists of M Vertices. A
dependency constraint on a relationship is shown as an arrow; for example, a Section can exist only if there is a
Surface for which an Intersection exists. The attribute can be of atomic or composite type. An example of
relationship attributes is the Intersection relationship, which has additional data attribute to perform the
intersection operation by interfacing Surface and Section entities. Composite attributes are shown in doublelined ovals. Underlined attributes are key attributes whose values are unique over all entities of that given type.
If an entity type does not have a complete key attribute -- like the Vertex because two vertices constructing two
different surfaces can have the same index --, it is called a weak entity type and shown in a double-lined box.
The attribute index of a Vertex is a partial key only and is denoted by a doted underline. Entity types can be
related by a is-a relationship, which define specialization or generalization of related entities. The entity types
connected by a small diamond
determine a hierarchy (or lattice). For example, possible surfaces can be
structured or unstructured. All the attributes, relationships and constraints on an entity type are inherited by its
subclass entity types.
index

type

parameters

name

description
coordinates

Section

Intersection

Surface

consist of

Vertex
index

data

Structured

Unstructured

Figure 24: Entity-Relationship Model

26

The entities in the ERM diagram are the basis for the Class decomposition because of their one-to-one
correspondence with the classes in the OOM Class Diagram. OOM preserves the same class decomposition
throughout the software development process; see Section 3.7, where detailed ERM diagram of the scientific
visualization model is shown, applying the ERM modeling elements showing objects, relations and constraints.
For example, Mesh or Surface objects are entities, qualified by specific attributes (for example color or
coordinates). Their relation includes functional dependencies between them, while constraints are applied on
attribute values.
ERM naturally fits in OOM, because it improves the specification of the object data model. ERM is one of the
semantic data models that support a rich set of modeling constructs for representing the semantics of entities,
their relationships, and constraints. The entity types are mapped to classes and their operations. Constraints that
cannot be specified declaratively in the model are coded in class methods. Additional methods are identified in
the classes to support queries and application functionality. ERM makes the software model more tangible to
users who just have CFD application knowledge. It is advisable to use ERM as a communication tool between
users and developers because ERM gets better the specification of the software model and its analysis.
The modeling concepts are not simply data-oriented: the object data model comes together with the fundamental
algorithms which will generate the inputs to the computer graphics algorithms. Algorithms represent a suitable
abstraction of the systems behavioral characteristics; their modeling is done in conjunction with the data
structures, and puts together all information required to make them work. The algorithmic models define the
computational aspects of the system; these models can be analyzed, tested and validated independently of the full
systems implementation. The algorithmic solutions influence directly the performance of the implementation;
hence, they should be modeled as simple, efficient and traceable components. To achieve clarity, effectiveness of
expression, and conciseness in describing algorithms, we will use a pseudo-language which can be routinely
mapped to any formal high-level programming language. An algorithm is written as a sequential order of
statements; a statement can be expressed in a natural-language syntax or in as mathematical notation with formal
descriptors and list of variables. The important statements are:

assignment:

a = b;

conditional:

if condition then statement else statement;

loop:

for variable = value until value do statement;

while condition do statement

In OOM, the data structures are encapsulated in the objects together with the algorithms. Normally, the
algorithms operate just on the internal data structure of the object they are associated with. More complex
algorithms could involve several objects; in this case, the data structure supporting their interaction needs to
relate all of them. For example, the cutting-plane algorithm involves the geometrical data structure of the domain
and of the surface which is created as a result of applying the algorithm. In what follows, we have grouped the
algorithms according to the data structures involved in the visualization processes, i.e. in terms of: combinatorial
topology (cell connectivity, node normal), computational geometry (cutting plane, section, local value) and
quantity representations (iso-lines, iso-surfaces, thresholds and particle traces).
Computer graphics algorithms which perform operations such as coloring, rendering and shading are commonly
implemented in hardware (and are of no interest here). What is important is the way in which we configure their
set-up and inputs in order to create the desired representations (this is partially discussed in Chapter 2 Adaptation
of Visualization Tools dealing with Representations).
This Chapter describes the algorithms in the following order: starting with topology, extending them to geometry
and ends with quantity related algorithms. This order was chosen because it incrementally introduces new data
structures which are sufficient and necessary for the algorithms to be operational. For example, the cutting-plane
algorithm cannot be efficient without the topology information.
27

1.1 Data Modeling

Data modeling is an important step in the visualization process since the data to be visualized needs to be
appropriately organized and structured. The Data Model has a direct impact on the performance of the
visualization system, and this is especially important when one must handle and store vast amounts of persistent
data. The data modeling definition is essential in supporting different visualization techniques (as explained in
Chapter 2 Adaptation of Visualization Tools).
CFD is concerned with the finite set of objects defined using discrete mathematics concepts [60]. Such objects
might be nodes, cells and similar types of objects describing the numerical simulation problem. For example, a
grid line is defined by its starting and ending nodes and all the intermediate nodes; it is a countable object since
it is composed of a finite number of nodes. Other objects -- such as planes or lines -- are composed of infinite
set of nodes as depicted in continuum mathematics. Discrete methods and combinatorial reasoning provide the
theoretical basis for defining data structures and for analyzing algorithms. There is a tight relationship between
mathematical objects and software objects; such identified relationships form the stable scientific background for
software development and are clearly at the core of visualization system software design.
The geometric model in the Euclidian space [61], and its manifold model in the topological space [62, 63], are
well covered by numerical models; we extended their application to provide the basis for creating graphical
models for scientific visualization purposes.
Mathematically, a set is a collection of distinguishable objects that share some common property (or properties)
that determine(s) their membership in the set. The objects in a set are called members. The set of nodes and the
set of cells are the elementary sets required to discretize a continuum. The discretization process consists in
dividing a spatial region into a finite number of non-overlapping cells, each of which is defined by its set of
nodes. To correctly approximate space, we use a variety of cell types, each type being characterized by the
interpolation function that is used to compute a physical quantity value at any arbitrary point within the cell. The
approximation must be such that (physical) continuity is satisfied between adjacent cells. The interpolation
algorithms work using the (physical) variable values (known) at the node coordinates. The set of nodes and the
set of cells are modeled with the so-called Zone object.
Cell

grouping

Zone

< sub - sup >

vertical decomposition

< part of >

horizontal decomposition

Figure 25:.Data model decomposition

28

The visualization data model covers the objects needed for storing and manipulating data; it includes two
decomposition lines:
vertical decomposition, based on sup-sub relationship, and
horizontal decomposition, based on part of relationship.
Both decomposition approaches describe the Cell and/or Zone geometry and topology with the grouping
principle (see Figure 25). The vertical decomposition describes the sup-sub relationship as the principal concept
in the boundary representation (B-rep) model [64-66], which defines geometrical shapes using their limits. The
sup-objects are here described by the group of sub-objects which are defined in the parametric space of lower
dimension. The horizontal decomposition implies that the grouping principle puts together objects from the same
parametric dimension, so that the resulting object remains in the same parametric space. The geometric model is
fully described with the parametric and modeling coordinates, for example a point in a curve or surface is
completely defined by its coordinates. The geometric model is enriched with the topology where boundary
relationships and connectivity can define multiple regions. The multiple region objects with connected parts are
defined by maintaining information about the common boundaries between them.
The Cell and Zone models are fundamental to the discretized geometry and topology models; according to the
vertical and horizontal decomposition principle:
the cell is explained in the light of the vertex-based boundary model
the zone is terms of its horizontal decomposition of as a collection of cells.

1.1.1 Cell class

A cell is the smallest identifiable element in/of a zone: it can be seen as the minimal zone, i.e. as a zone that
consists of just one cell. Hence, the 3D-cell represents the smallest element of the 3D Euclidian space. In the
same way as a 3D-cell defines the smallest 3-dimensional volume, the 2D-cell defines the smallest surface area
in two dimensions, and 1D cell defines the smallest curve in the plane. Cells of parametric dimension equivalent
to the zone cannot be decomposed further without introducing additional nodes. Hence the cell is the simpleconnected manifold of a modeled geometry [67] compose the curve, surface or space Zone. The cell is defined
in the modeling and in the parametric space where the mapping functions are known, as described in section
1.3.1 Isoparametric mapping and shape functions. These mapping functions are used to calculate the cell
properties at any arbitrary cell point.
The cell concept is used in the Finite Element Method (FEM) (Zienkiewicz [68, 69]) and in Computational Fluid
Dynamics (CFD) (Hirsch [60, 70]). The special types of elements, called iso-parametric elements, are applied
in this thesis to support the Cell modeling. The name iso-parametric derives from the fact that the same
parametric functions describe the geometry and the quantity variations within the cell. With iso-parametric
elements the variation of a quantity is interpolated using non-dimensional iso-parametric coordinates. The
cells parametric space is the region that fully describes the cell interior and boundary; it corresponds to the
number of parameters required to define a point in it. For example, there is only one parameter needed to specify
a point in a 1D-cell (as is the case for a point on a curve). The modeling dimension is the number of coordinates
required to specify a point in the Euclidean space. If the cell description is dimension-independent, it allows
geometry manipulation in the parametric space. An iso-parametric mapping is applied only when precise
coordinates inside the cell interior are required.
Parametric forms are modeled with Map classes that support the mapping between the parametric and
modeling spaces. The iso-parametric mapping is defined for each cell type (see Section 1.3.1). Maps are
mathematical objects which satisfy smooth, continuous conditions, as a functional mapping of the form:
M(p): D I,

1- 1

29

Solid

Face

Edge

Node

a) topology

Point cell

Curve cell

Surface cell

Body cell

b) parametric dimensions

Figure 26:.Cell classification

where D is the domain of map M in the parametric space Rp, and I is the image of M in modeling space Rm
(where all M(p), pD belong). The parametric dimension is defined by the cell itself as 1D for curves, 2D for
surfaces and 3D for solids. Every object is embedded in a Euclidian space of some dimension called the
modeling space. The cell map defines the underlying extent of the geometric object through a fixed number of
parameters. Every underlying extent is represented as a parametric function, for example, a 3D surface:
x = X(u,v)
1- 2
y = Y(u,v)
z = Z(u,v)
x, y and z in this map define the modeling coordinates and u and v define the coordinates in parametric space.
The parametric dimension defines the extent of the cell and the modeling dimension is the dimension of the
Euclidian space (as seen in the real world). Since the cell is a finite part of an infinite region, it does not consist
of the entire curve or surface, but only of a trimmed region (see Figure 26).
The cell types may be distinguished by:
TOPOLOGY:
- number of nodes:

1, 2, . . . N

- skeleton

recursive structure of boundary cells

GEOMETRY:
- modeling space :

node coordinates: 0D:(0), 1D:(x), 2D:(x,y) or 3D:(x,y,z)

- parametric space:

node coordinates: 0D:(0), 1D:(u), 2D:(u,v) or 3D:(u,v,w)

- mapping function:

transformations between modeling and parametric spaces

- interpolation function:

inside modeling or parametric space

The cell is called 0D, 1D, 2D or 3D in accordance with the cell parametric dimension which describes the region
of a point, curve, surface or body (see Figure 26 (b)). The cell boundaries are defined as cells of lower
dimension in respect to the cell they bound (see Figure 26(a)): they delimit the cell region from the underlying
Map. For example, three edges of parametric dimension 1D bound the face of topological dimension D2. Thus,
1D-curve cells bound 2D surface cells. The edges are defined as regions with infinite line maps bounded by
respective nodes.
30

composed of
part of

Cell

Skeleton

M
S

Cell 3D
Solid

Cell 2D
Face

composed of
part of

S
Hexahedron

composed of
part of

Pentahedron

Piramid

Tetrahedron

Quadrilateral

Triangle

Cell 1D
Edge

composed of
part of

Cell 0D
Node

Segment

Vertex

Figure 27: Developed cell ERM

A cell is defined by a set of nodes: it represents a bounded region of a geometrical space described by a
boundary representation model (B-rep) model, defined in the indexing topology space. The same cell can be
embedded in Euclidean spaces of different dimensions. A triangle is part of an infinite plane bounded by three
lines; a tetrahedron is part of a 3D-space bounded by 4 planes. The same triangle could be embedded in the 3D
space as a part of the tetrahedron B-rep. The definition of the lower-dimensional boundary cells (the sub-cells)
provides an elegant recursive structure to describe a cell topology (see Figure 26), since boundary cells can in
turn be defined by lower-dimensional boundary cells.
The cell classification identifies possible cell types (see Figure 26). A higher-dimension cell can always be
described with cells one-dimension lower. This structure ends at 0-cell node. The sub-cell is a part of the
boundary which defines higher-dimensional cells (named sup-cells). The sup-cell knows which cells are its
boundaries. For example, a tetrahedron is a 3D-cell bounded by four 2D-cells called faces. Each face is a 2Dcell bounded by tree edges. Each edge is a 1D-cell bounded by two nodes. Each node is a 0D-cell with no
boundary.
The composed of relationship is the main relationship used to decompose a cell in its vertex-based B-rep: it
shows the cell identification path that goes from a cell of higher dimension (3D-cell) to a cell of lower dimension
(OD-cell). The part of relationship is the inverse relationship; it identifies the sub-cells which can define more
than one sup-cell. For example, a face is composed of edges and an edge can define 2 faces as part of the
relationship. Both relationships define a collection of cells. In our case the composed of relationship is chosen for
the implementation and the part of relationship is calculated from the composed of relationship, when necessary.
It is convenient, because of the variety of cell types, to group cells according to topology (number of cell nodes),
geometry (parametric space) and interpolation functions. Topology permits to describe uniquely, for example, a
3D-grid as collection of nodes, edges, faces or 3D-cells. The thesis describes in details the 3D cells:
hexahedrons, pentahedrons, pyramids and tetrahedrons and 2D cells (quadrilaterals and triangles). The definition
of a cell is dimension-independent and is completely analogous for cells of higher or lower dimensions. For each
cell type, the cell skeleton is defined and shown in appropriate cell tables (see Figure 27). The skeleton provides
an easy way to traverse the topology of an object in a dimension-independent manner. A skeleton of specific
dimension defines a set of cells. For example, consider the pyramid:
the 0-skeleton are the five nodes,
the 1-skeleton are the eight edges,
the 2-skeleton are the five faces.

31

n0

n1
u

e0

n2

e2

e2

n3

e1

e3

e0
n0

n2

n1

e1

e0
n1

n0

Figure 28: 1D & 2D Cell topologies

The applied indexing for uniquely identifying the cells topological properties is described in the following
skeleton tables and related Figure 28 and Figure 29:
1D: segment

Mesh type
: unstructured & structured
Cell topology : SEGMENT 1D, 2D & 3D

Coordinate System

Cell
1

Nodes
3

Axis
1

Nodes
3

T1N2

0-1

0-1

Table 2: SEGMENT skeleton table

2D: triangle
Mesh type
: unstructured
Cell topology : TRIANGLE 2D & 3D

Coordinate
System

Edges
3

Nodes
3

Cell
1

Nodes
3

Axis
2

Nodes
3

0-1

T2N3

0-1-2-3

0-1

1-2

0-3

2-0
Table 3: TRIANGLE skeleton table

2D: quadrilateral

Mesh type
: unstructured & structured
Cell topology : QUADRILATERAL 2D & 3D

Coordinate
System

Edges
4

Nodes
4

Cell
1

Nodes
4

Axis
2

Nodes
4

0-1

T2N4

0-1-2-3

0-1

1-2

0-3

2-3

3-0
Table 4: QUADRILATERAL skeleton table
32

tetrahedron

w
v

n3

e5
e3

e5

e4 e2

e4
e3

e5

f1

n2

e2

e3

e1

f3

e4

e1

e2

f0
f2
e0

e0
e0

n0

e1

n1

pyramid

n4

v e7

e7

e6

f1
e4
n3

e5

e2

e4

n2

e1

e3

e2
e2

f3

e1
e1

f0

f2

e0

e0

e0

f4

e5

e5
e3
e3

e4

e6
e6

e7

n1

n0

pentahedron
w

n5

e8
e8

e8

n3

e6
e6

e7
e3

n4
e2

e5

e3

n2

f2
e2

e1

e0

e4

e5

f3
e2

n0

e0

e7

f1

e3

e6

e7

f4

e5

e4

e4
e0

f0

e1 e1

n1
u

e10
e10

hexahedron
n7

e11

e11

n2

e2

e3

n0

e2

f0
e5

f2

e1

e0
u

e0

e0

n1

Figure 29: 3D Cell topologies

33

e9 e6
e6

e3
e3

e4

e9

e2

e8
e4

e5
n3

f4

e8

f1

e6

n5

f5

e7

e7

v
e4

e11

e9

e7
e8

n4

n6

e10

f3
e5 e1
e1

Mesh type
: unstructured
Cell topology : TETRAHEDRON 3D

Coordinate
System

Edges
6

Nodes
4

Faces
4

Nodes
4

Cell
1

Nodes
4

Axis
3

Nodes
4

0-1

0-2-1

T3N4

0-1-2-3

0-1

1-2

2-0-3

0-2

2-0

0-1-3

0-3

0-3

1-2-3

1-3

2-3
Table 5: TETRAHEDRON skeleton table

Mesh type
: unstructured
Cell topology : PYRAMID 3D

Coordinate
System

Edges
7

Nodes
5

Faces
5

Nodes
5

Cell
1

Nodes
5

Axis
3

Nodes
5

0-1

0-1-2-3 T3N5

0-1-2-3-4

0-1

1-2

0-4-1

0-3

2-3

1-4-2

0-4

3-0

2-4-3

0-4

3-4-0

1-4

2-4

3-4
Table 6: PYRAMID skeleton table

Mesh type
: unstructured
Cell topology : PENTAHEDRON 3D

Coordinate
System

Edges
9

Nodes
6

Faces
5

Nodes
6

Cell
1

Nodes
6

Axis
3

Nodes
6

0-1

0-2-1

T3N6

0-1-2-3-4-5

0-1

1-2

2-0-3-5

0-2

2-0

0-1-4-3

0-3

0-3

1-2-5-4

1-4

3-4-5

2-5

3-4

4-5

5-3
Table 7: PENTAHEDRON skeleton table

34

Mesh type
: unstructured & structured
Cell topology : HEXAHEDRON 3D
Edges
12

Nodes
8

Faces
6

Nodes
8

0-1

0-3-2-1 T3N8

1-2

2-3

Coordinate
System

Cell
1

Nodes
8

Axis
3

Nodes
8

0-1-2-3-4-5

0-1

3-0-4-7

0-3

0-1-5-4

0-4

3-0

1-2-6-5

0-4

2-3-7-6

1-5

4-5-6-7

2-6

3-5

4-5

5-6

10

6-7

11

7-4
Table 8: HEXAHEDRON skeleton table
Space

Dimension

Step
order 0
Modeling

Map
Cell

Parametric
Linear
order 1
S
Quadratic
order 2
Cubic
order 3

M
compose of
part of

Cell

cell node
indexing

Node
Index

M
S

Boundary
(sub-cell)

Bridge
Cell

Star
(sup-cell)

1
Frame

Figure 30: Cell ERM

35

0D

1D

2D

3D

By convention, the local coordinate system of a cell is in parametric space: its origin is at the first cell node, and
the coordinate axes are defined based following the local node indexing within the cell, as described under the
Coordinate System header in each skeleton table. The orientation of the cell boundary is considered positive if its
face normal points from the interior of the cell to the exterior. A cell is always embedded in a geometrical space
of equal or higher dimension than the cells intrinsic dimension, and can be composed of cells of intrinsic
dimension lesser than its own. The cell with lowest 0D dimension is a node. Any cell can be described by an
ordered collection of nodes. The topology for nodes, edges, faces and cells is predefined for each cell type so that
some properties are implicitly defined. For example, the order of the nodes of a triangular cell enables us to
compute the line that is normal to that triangle in a 3D space. Another example is the extraction of faces from a
3D-cell applied in a marching-cell algorithm, see Section 1.4.1.
B-rep defines the lower dimensional cells as sub-cells, while the cell itself is defined (bounded) by the
mentioned sub-cells. As shown in Figure 30, each cell can have more than one sup-cell (named Stars), and a
cells boundary can be composed of more than one Boundary sub-cell. The sup-sub relationship is defined by the
Bridge classes. The Bridge class defines the relationship between a cell and one of its boundaries; it specifies the
direction, when moving from the boundary in the cell interior. Each boundary cell is a bridge; for example, a
triangle has 3 bounding bridges (edges) which determine the boundary sub-cells. Boundary cells can be
connected in a prescribed way by showing the boundary topology, named Frame. The group of Bridges (edges)
forms the Frame, which defines their sup-cell (face).

1.1.2 Zone class

The basic class that organizes and allows the manipulation of geometrical data is called the Zone class. A Zone
is a geometric concept that defines a portion of the Euclidian space; it is defined as an ordered collection of
nodes (also known in mathematics as a point set). A geometry described with the nodes topology defines curves,
surfaces and bodies. Each node represents a unique set of real numbers, called coordinates, in the right-handed
rectangular Cartesian coordinate system.
P identical (x,y,z) as P(x) or P(xi), i=1,2,3.

1- 3

The geometry of a zone, described by the nodes coordinates and the topology is a composition of cells. The zone
topology is defined as the union of cells according to the B-rep cell model extended to support:
the intersection of every pair of the zone cells,
the union of any number of the zone cells.
The finite space z is discretized into a finite number of regions. The zone Z is expressed as the union of the
finite space z and its boundaries z.
Z = z z

1- 4

Cell C is expressed as the union of the finite region c and its boundary c
C= c c

1- 5

The boundaries of the cells represent their connections with the neighborhood cells:

c1 c2 = c1 c2.
The common boundary c1

c2

1- 6

is also the zone cell. The zone is defined as a union of all its cells, whatever

parametric space they belong to:

N

Z = z z =

, where N is the total number of cells.

1- 7

c =1

The Zone model is defined with a boundary representation model, B-rep, because we always treat a finite region
in the generalized modeling space. A zone encapsulates the relationship with its boundary. As the cell is the
36

atomic entity in the modeling and parametric spaces, it represents a Zone unit space. The only difference is that
the Zone B-rep has in addition the cell connectivity to depict the zone topology. The zone topology incorporates
schemes that specify the geometry of the zone boundaries and their inter-connectivity. It explicitly maintains cell
connectivity by ensuring that adjacent cells share their common boundaries. The zone topology depends solely
on the cells structure, not on their geometry. If the geometry of a zone is modified, its topological structure
remains intact. The invariant nature of the zone topology allows its separate treatment (see Section 1.2). The
combined use of geometry and topology is present in the algorithms described in Section 1.3 and Section 1.4.
Grids can be discretized using different types of zones, so one can establish sup-sub zone relationships and
define a zone as a region bounded by zones of lower dimension -- the sub-zones. A sup-zone is a zone bounded
by sub-zones. The sup-sub relationship yields an elegant recursive structure called the zone topology. The B-rep
modeling naturally supports aggregation, thus it can be extended to model the set of zones through composition
based on the recursive sub-zone relationship. B-rep describes the topology of a zone modeled with all its
boundary zones and their connectivity.
The presented topological structure and its general functionality is defined in the Geom class, thus applicable to
Zone and ZoneGroup sub-classes, which are responsible for modeling the collection of zones as a Set. The
additional topological classes are: Bridge, Frame and Skeleton.

M
Geom
1
S

Zone Group
1

S
1

Skeleton

Zone

Node

Boundary
(sub-zone)

Star
(sup-zone)

Bridge
M

1
Frame
S

Inner
Frame

Outer
Frame

Figure 31: The Zone Composition and Topology ERM

37

Cell

n1
b1
S1

t1

t2

n2

D1-bridge
S2

b2

Figure 32: Bridge orientation

The Bridge class describes the relationship between a zone and one of its boundary zones. It specifies the
orientation of the boundary so one can identify on which side of the boundary is the interior of the zone. It
specifies not only that the sub-zone bounds the sup-zone, but also the pre-image of the sub-zone on the sup-zone
domain. The pre-image map is defined as the map Mp which satisfies:
Msup (Mp (p)) = Msub (p)
for all the parameters p in the interior of the sub-zone, where Msup is the bridge sup-zone map and Msub the subzone map.
Pre-images relate points in sub-zones to parameter-values of the zone they bound. This functionality is notably
used when moving to the interior of a surface from a point on its boundary. The inward direction from a
boundary can be approximated by the planes tangent to the sub-zone and the sup-zone. This is not sufficient,
however, since several sup-zones can use the same sub-zone as their boundary, and an imposed orientation is
required to uniquely solve the problem. The sub-sup relationship imposes the bridge orientation, a Boolean
variable which specifies the orientation of the sub-zone as natural or inverted. The bridge normal is defined
as the vector perpendicular to the sub-zone at the point considered and which lies in the plane tangent to the supzone at that point. There are 2 kinds of bridge normals:
1.

The algebraic normal is computed from the natural orientation of the sub-zone and sup-zone.
D0-bridge: point bounds a curve, the normal is simply the oriented tangent of the curve.
D1-bridge: curve bounds a surface, the algebraic normal is defined as n t where t is the oriented
tangent to the curve and n is oriented normal to the surface.

2.

The topological normal is the algebraic normal, but its orientation can be inverted if needed. It
represents the inward direction from the sub-zone to sup-zone. The inward normal is closely related
to the relative orientation of the bridge.

The interior of a zone is a region of the zone bounded by lower dimensional zones. A node n of Z is said to be an
interior point of Z, if there exist at least one sub-zone about n whose all points belong to Z. A node n is said to be
exterior to a zone Z, if none of the sub-zone points belongs to Z, in which the node is located.
The boundary of a zone is the zone of topological dimension-1 composed of the nodes that are exterior to Z. A
Frame combines bridges into a connected enclosure. Enclosures can have one Outer frame and several Inner
ones (see Figure 33). The dimension of the Frame is that of the sub-zone whose boundaries it represents. Frames
can be active or inactive depending on whether they have been assigned to the sup-zone or not. The Geoms
can be constructed from Frames and Bridges.

38

holes

Figure 33: Inner and Outer Frames of 2D and 3D zones

The Frame object represents the boundary as the set of Zones, which are connected and form a closed loop (see
Figure 33 showing the case of a surface and a body with internal holes). Bridge and Frame are useful when the
topological structure is manipulated in point location based algorithms. The Skeleton provides an easy way to
traverse the topology of object with different parametric dimensions.
The Range object represents a rectangular box aligned with the coordinate axes and representing the minimum
space holding the zone. The Geom hierarchy allows working with a single zone or with a set of zones. The Zoneoperations apply the boundary topology for extraction algorithm, while the ZoneGroup operations are related to
zone composition (assembly) and distributions of the underlying maps and topologies.
A set of cells defines a Zone when all the cells have disjoint interiors. A Zone is a grid of polygonal cells with
tangent discontinuities at the cells boundaries. A zone is a topological entity which defines connectivity between
cells. Each zone has a set of cells associated with it, together with a set of nodes from which cells are created. A
Cell provides a piece-wise linear approximation and is always associated with a unique Map. Each cell is part of
only one zone. The cell provides support for query/modification operations and can access its nodes. A Zone is
defined only in the modeling space and it uses the cell maps for preserving the continuity of the geometry and
quantity fields it is defining.
Space

defines

Dimension

limits

S
S

Range

0D

1D

2D

3D

1
Modeling

Map
Cell

Step
order 0

Parametric

1
1

Linear
order 1

Geometry

Topology

S
Quadratic
order 2
Cubic
order 3

Point
Coordinates

Zone

Cell

local node
indexing

Figure 34: Zone-Cell ERM

39

cell node
indexing

Node
Index

PointSet

Curve

Zone

Space

Surface

Body

Hexahedron
T3N8

Unstructured

Structured

3D

2D

1D

composed of
part of

composed of
part of

embeded

embeded

embeded

Pentahedron
T3N6

Piramid
T3N5

Tetrahedron
T3N4

Quadrilateral
T2N3

Triangle
T2N3

Segment
T1N2

Vertex
T0n1

S
composed of
part of

Cell 3D
Solid

Cell 2D
Face

composed of
part of

composed of
part of

Cell 1D
Edge

Cell 0D
Node

S
M
compose
is part

Cell

Skeleton
M

Figure 35: Developed Cell-Zone ERM

The Zone class models geometry reusing the cell definition in the modeling and parametric spaces. The point
location algorithm, applies the marching-cell algorithm, which uses the zone topology, and ones inside the cell,
the cell Map is applied to control the underlying cell extent in the parametric space. The cell boundaries split the
parametric space in two parts, first the cell itself and second the rest of the Map. The rest of the cell Map is of no
interest since it is not a part of the Zone. A Zone Map is a collection of independent Cell Maps.
The Node is the geometric element used to define the Zone points. A node defines a point in the Zone Map and
also in the Cell Maps. The Cell Maps provide mechanisms to query the node coordinates and the derivatives of
defined variables.
The cells connectivity is only defined for unstructured meshes. By definition, a zone is said to be connected if
one can move between any two points in the zone along a curve that fully lies in the zone. As the interactively
created Zones are unstructured, the zone connectivity needs to be generated dynamically during the interactive
session.
Each zone subclass is specialized according to its intrinsic parametric cell dimension as 0D, 1D, 2D or 3D zone
as the nodes are composing cells (see Figure 35, point, segment, polygon, polyhedral). The Zone class, which is

40

the base geometry class, is defined with different subclasses, which are each representing collections of
homogeneous cells. The PointSet, Curve, Surface and Body are defined with the parametric dimension,
respectively 0D, 1D, 2D and 3D. Recall that a zone is both a collection of nodes and a collection of cells. The
highest parametric dimension of the cells in a zone determines the dimension of the zone and its parametric
space. The parametric dimension of the zone (topology) should not be confused with the modeling dimension of
the zone (geometry); the same holds for cells. For example, a surface with a 2D topology could have a 2D or 3D
geometry definition, and its node coordinates would be (x, y) or (x, y, z) in the respective geometrical space.
As explained, different topologies can exist for the same geometry, and different geometries can exist for the
same topology. Geometry and topology are defined in the Zone class.

1.1.3 Coherence and Tolerance Concept

A Zone can be either in a coherent or in a non-coherent state. The coherence status of a zone can be queried, and
a non-coherent zone can be processed to become coherent. In this process, the map of the zone is modified so
that its B-rep structure is coherent with the cell boundaries.
4

Coherent

Non-coherent

2
3

6
n7

n7
n9
n4
n4

n5

n6
n6
n5
n8
n1
n1

n3

n3

n2
n2

Figure 36: Coherent and non-coherent cells

Zone coherence requires that the connected zones match along their all lower-dimensional boundary zones or not
at all. For example, an example of not coherent curve is when it is self-intersecting (see Figure 36).
Coherence of 2D zones is achieved by combining triangular and rectangular cells. As cells do not have holes,
they are simply-connected regions. Cells coherence guarantees that the Zone corresponds exactly to the trimmed
map region, defined as the zone interior. This property assures that the connected surfaces meet across their
common edges. As the result of the imposed coherence is that they must have the same nodes defining their
common boundaries. The boundary representation imposes several geometrical constraints, for example, when
an edge is constrained to lie completely on a surface. Such geometrical constraint is impossible to obey exactly
since all computations are subject to round-off errors. Geometrical constraints can be met, only within certain
tolerances, because the geometrical data are numerically approximated.
The tolerance-model has two essential parameters:

the problem size that defines the coordinate units and

the unit tolerance that specifies the distance between two points. Two points are considered
one point if they are centered at the same origin when scaled down to the unit box of size 1.

41

Obviously, in practice, computation must be carried out at a given precision level, for example using doubleprecision floating point computation that yields an accuracy of about 10-16. The unit tolerance values are in
accordance of a larger magnitude order than the precision limit given above, so that the inaccurate input and the
accumulation of errors in the computation remain insignificant.
The effective tolerance for the problem is defined by multiplying problem size by input unit-size. Points are
considered as coinciding when located within the effective tolerance distance. The cell has nodes that define
modeling and parametric coordinates for points on the map. The cell nodes coordinates are created within
specified tolerances.

1.1.4 Zone classification and Surface class

The data model must integrate all the objects necessary for storing and manipulating the data. This includes the
Zone and Field classes that combine geometry, topology and quantity data for the objects created at input and for
the objects created as the result of the user interaction. The zone classification follows:
0D - general points:.
- local values,
1D - general curves:
- constant mesh line (for structured mesh),
- arbitrary curve section,
- general curve resulting from the intersection of two surfaces (for example, solid boundary
and arbitrary section),
- particle paths.
2D - general surfaces:
- mesh boundaries,
- constant I, J or K surfaces (for structured meshes only),
- arbitrary cutting planes,
- isosurfaces,
- surface-tangent vector (for example, stream-surface).
3D - general volumes:
- complete mesh,
- single domain of the mesh,
- arbitrary sub-mesh interactively generated.
In the next section, the 2D surface concept is explained in detail so as to clarify various aspects of interactivelycreated surfaces.
no
boundary

two
boundaries

(b)

one
boundary
(a)

(a)

Figure 37: Surface types: (a) open, (b) closed

42

The analysis of a quantity field is usually performed on interactively-selected zones of different types which
represent discretized regions of space. The most important Zone class is the Surface class; it is defined by a set of
nodes and 2D cells. The surfaces finite geometry is in the 2D or 3D space. A surface lies in a bounded region of
space and has two sides. To pass from one side to the other side of the surface, one must either cross the surface,
or cross the curve that bounds the surface area. A surface without a boundary curve, (for example a sphere, or
any surface that can be transformed continuously into a sphere) is called a closed surface, in opposition to and
open surface which has at least one boundary curve (see Figure 37). The boundary curve is always a closed
curve; if a boundary curve can continuously be deformed and shrunk to a point without leaving the surface
space, then the surface is characterized as simply-connected.

3
2

1
3
1

multiple connected region

multiple disconnected region

Figure 38: Simply and multiply connected and disconnected surface regions.
The visualization system makes an extensive use of the surface as a main identification and manipulation object;
most visual representations are related to surfaces, as the surfaces are the most appropriate containers of the data
extracted from volume data sets.
The main surface types are:
mesh surfaces
mesh boundary surfaces
slicing or cutting planes
isosurfaces
The mesh surfaces are part of structured grids (consistent I, J, K surfaces or surfaces defined by faces), and other
surfaces can be created for structured and unstructured meshes. The mesh and boundary surfaces are given at
input, when cutting planes and isosurfaces are computed during an interactive session (surfaces and associated
visualization tools are described in the Chapter 2).
The cells connectivity defines the relationship between nodes and various cells. A Surface can be queried for:
which are the cells adjacent to a given cell,
which are the cells that share a given node,
which edges are defined with a given node,
which two nodes define the edge connecting faces.
Geometry algorithms rely on the mathematical form of the underlying maps and on the topology that defines
boundaries. They are:

the bounding cube that computes the range (cube) which contains the geometric objects. It is used as
rough bound of objects before proceeding to more precise computations.

the closest point that computes the minimum (geometrical) distance between objects,

the extent that computes the quantities as length, area, volume of objects,

the center of extent that defines the center of object area, volume, etc.

43

1.2 Computational Topology

Computational (or algorithmic) topology is concerned with the development of efficient algorithms for solving
topological problems. Computational topology is applied here to geometrical objects defined by point sets, [61,
63]; the resulting algorithms combine topology and algebra. The topological relationships are deduced simply by
counting and indexing; these elementary techniques are sufficient to create the topologies of the constructed
geometries, for example the polygons generation method applied in the cutting-plane algorithm uses the cell
intersection pattern, which defines the cell topology, not requiring taking into account the coordinates of the cell
nodes.
In this work, the topology algorithms are important for the unstructured meshes generation. For example, when a
cutting-plane or iso-surface is generated, but not keep for further use, in that case it is sufficient creating a set of
polygons and pass them to the graphics algorithm to produce the desired surface visualization (see Section
2.4.5). Thus, this process does not require the definition of the surface topology, but if such surface has to be
kept for further analysis, then the polygon set representing the initial surface, becomes an input for the creation
of the fully defined unstructured surface. The creation process of the unstructured surface requires the
computation of the Zone topology consisting of:

unique node and cell indexing

cell topology

cell connectivity, and

unique local to global indexing map

This topology information enables different visualization tools of the SV system to operate on interactivelycreated surfaces. The first algorithm is the cell connectivity algorithm, which pre-processes cell connectivity
information as part of the input to the visualization system. The second algorithm is the node topology algorithm,
including the node reduction algorithm. This algorithm is used to determine surface normals which are required
to identify the cells around a node and to calculate the unique node normal. This algorithm provides the
necessary input to the surface-shading algorithm. Both algorithms are performing the nodes and cells topological
mapping from local to global indexing space.

1.2.1 Topological space

The zone Z is a node-set describing a continuum concept. The topology C is a cell-set defined for a node-set Z
having the following properties:
1. n Z and c C
2. the union of any number of cells of C is in C
3. the intersection of any two cells of C is in C.

(a) Zone and Cell boundary points

(b) Zone internal and Cell

boundary point

(c) Zone and Cell internal point

Figure 39: Zone and cells point classification

44

The pair (Z, C) is called a topological space. In addition, each cell in C defines its topological space because it is
defined as a node-set and as a cell-set containing its boundary cells. If the cell boundaries are excluded the open
cell represents the open set of points which contains only interior points. The closed cell refers to a closed set of
points and contains in addition its boundary points. A set is closed if the complement is open. All cell points that
are bounded by its boundary are called cell interior. A point p is an interior point of c, where c C; if there exists
an open set contained in c and at the same time contains p. The set of all interior points of c is called the interior
of c. If the point is classified as interior point of all c, it is in addition the interior point of cell-set C as c C.

curve cell

surface cell

body cell

Figure 40: Curve, surface and body neighborhoods

The cell interior is always a manifold because every interior point of the cell has its infinitesimal neighborhood which
reflects the similar cell shape. The shapes of the curve, surface and body cell topology are shown in Figure 40, where
the neighborhood of the interior point looks like the primary cell. Boundary points are not interior points or part of the
manifold, because its neighborhood is not complete, see Figure 41. The concept of neighborhood is important in order
to understand the definition of the boundary cell. The neighborhood is defined as infinity small region around the
arbitrary point. It is a set of points inside the cell interior having the same parametric dimension as the analyzed cell.
The neighborhood which completely belongs to the cell interior is defined as -full, and can topologically deform up
to the analyzed point. The -full neighborhood classifies the point as the cell interior point.

zero
exsterior point

half
boundary point

full
interior point

Figure 41: Cell point classification

45

non-manifold

T
x

manifold

Figure 42: Manifold and non-manifold neighborhoods

When the point is on the boundary which is not connected to another cell, the neighborhood is called -half. The
boundary sub-cell represents -half connection to the sup-cell. The sup-cells interiors that are immediately
adjacent to that point form the -full neighborhood. For example, the boundary cell in such case represents an
internal edge. The boundary cell retains non-manifold configuration. The non-manifold cells are cells in whose
interior the shape of the original cell is perturbed, like shown in the Figure 42.

1.2.2 Structured and Unstructured Topology

The structured and unstructured topologies are the two forms through which the zone topological information is
accessed. In this approach the zone topology is defined by unified concepts of:
cell topology
defines cell nodes
cell connectivity
defines adjoin cells

cell topology

cell connectivity

Figure 43: Zone topology concepts

46

Global indexing

Local indexing

2
3

5
15
0
14

13

7
6

16
11

12

10
6

10

Figure 44: Boundary LG map

The topological entities found in the zone topology are indexed cells and nodes. The cell topology contains the
node indices defining topologically the zone as a set of cells. The cell connectivity contains the indices of cells
connected to their cell sides, see Figure 43. The cell side represents the intersection between two cells. The cell
which contains a non-connected side represents the cell on the zone boundary. The node and cell are uniquely
defined inside a zone with three types of indices:
global,
local,
cell.
The local indexing establishes the correspondence between the zone and cell nodes. The global indexing
supports the node and cell identification between sup-sub zone topologies, which are usually applied in the cell
based algorithms. The cell indexing enables the identification of the local node index inside the zone when the
local cell index and the node cell index are known. The sub-sup relationship is the connectivity with external
topological space. It is defined through local to global indexing, named LG map, in the case when the sup-zone is
known. The LG map for nodes and cells defines global indices in the sup-zone, for nodes and cells respectively,
see Figure 45.For example, if the zone is the surface partitions of its sup-zone boundary, the LG map with the
internal bounded volume is defined, see Figure 44. Each zone topology is independent of its sup-zone, but
occasionally, it is important to navigate through zone with different topological dimension in order to control the
specific surface/volume region. This relationship is particularly important in the marching cell algorithms
applied for extraction of section and cutting plane.
For structured grids the zone topology is given implicitly with the parameterized nodes order and all other subtopologies can be deduced from such imposed nodes order, see Figure 45. This is not the case for unstructured
grids where each node and cell index is explicitly numerically prescribed. In structured, topologies the nodes and
the cells are identified with imposed number of parameters based on the parametric zone dimension. For
structured zones the parameters are classified in Table 9.

dimension

zone

parameters

number of nodes

number of cells

1D

curve

(i)

I-1

2D

surface

( i, j )

IJ

(I-1) (J-1)

3D

body

( i, j, k)

IJK

(I-1) (J-1) (K-1)

Table 9: Structured zone parameterization

47

(i,j)

(i,j,k)

(i)

i
i

i
(a) structured

3
1

(b) unstructured

Figure 45: The sup-sub zone relationship

(1) surface of the body, (2) curve of the surface, (3) node of the curve
In the Figure 45(a) the structured decomposition shows the possibility for the extraction of family of surface by
making constant one of the ijk indices. The procedure can be repeated for surface where the i- or j- family of
curves can be extracted. Finally the node in the curve is identified with one index. The ordered node indexing
makes possible the parametric manipulation of structured zones, see Figure 46.
For the unstructured grid the zone topology can be defined as homogeneous or heterogeneous set of cells. The
simpler model is homogeneous zone topology which is constrained with only one cell type. For example the
homogeneous zone topologies are surface of triangles or body of hexahedrons cell types. In the most general
case the unstructured grid is composed of heterogeneous cells, which have varying number of nodes, edges and
faces.
surface

curve

body
i,j,k+1

i,j+1

i,j+1,k
i-1,j,k

i-1

i+1

i-1,j

i+1,j

i+1,j,k
i,j-1,k

i,j-1

Figure 46: Node indexing in structured grids

48

i,j,k-1

cell

cell starting index

0
1
2
3
4
5
6
7
8
9

...

nodes

cell index vector CT

cell

last node cell #0

first node cell #1

...

0
1
2
3
4
5
6
7
8
9

last node cell #1

first node cell #2

cell topology

cells

cell starting index

last cell<>cell #0
first cell<>cell #1

last cell<> cell #1

first cell<> cell #2

cell index vector CC

cell connectivity

Figure 47: Cell topology and cell connectivity data structure

Thus, the rule is to apply the vectors of integers for indexing cell nodes and connected cells controlled with
additional vectors for cell indexing, see Figure 47. The internal structure of a cell is fully defined with the cell
topology and the cell index vector. The same structure is applied to the cell connectivity with associated cell
index vector. The cell index vectors vary in length. For the cell topology the cell index vector depends on the
number of all the cell nodes while for the cell connectivity it depends on the number of all the cell sides. The cell
index vectors identify the first cell node or connected cell, to the first cell side. In addition, the number of the cell
nodes and its connected cells are defined from cell index vector. For heterogeneous unstructured zones the
starting cell index for the cell topology CT and cell connectivity CC, are defined as:
cell topology

cell connectivity

CN

i = ni

( starting cell index ) CT

( starting cell index )CC

i =0

CN

= sj

1- 8

j =0

where n is a number of nodes per cell, s is the number of sides per cell and CN is the total number of zone cells
which precede the one for which the index is calculated. From the cell index vector the number of cell nodes are
defined as the difference between the starting cell index CT of the next and current cell. The equivalent
mechanism is applied to the cell index vector CC associated with the cell connectivity to define the number of
cell sides, see Figure 47. Thus, the number of cell nodes and cell sides are calculated as:
(number of cell nodes)i = CTi+1 - CTi

1- 9

(number of cell sides)i = CCi+1 - CCi

1- 10

1.2.3 Cell Connectivity

The Cell Connectivity model defines the Zone topology consisting of:
Input model parameters:

Additional output parameters:

parametric dimension

cell connectivity

number of nodes

LG map for cells

number of cells

LG map for nodes

cell topology

The dynamic memory storage is appropriate for such heterogeneous data model as the size of the complete zone
topology can be only defined if all the input topology is read in. The input data set are the first four parameters of

49

the Cell Connectivity model. The cell connectivity algorithm calculates the additional output parameters: cell
connectivity and LG maps for the following tasks:
to define zone topology,
to define boundary topology,
to define patches (segments) topology given as boundary conditions.
The cell connectivity information is required to make efficient the marching inside a zone when passing from
one cell to another. This concept is applied in the cutting plane and vector line algorithms. The cell connectivity
is calculated from the cell topology (nodes defining each cell) in order to keep the input data smaller and in
addition, to avoid the unnecessary errors, which could be detected if the input data would contain inconsistent
topology information. There are two types of cell connectivity calculations which are performed during
visualization process:
The first one is done during the preparation phase when the input files are generated. Such cell
connectivity information is of the static nature for the zones input and doesnt change during the
visualization process.
The second type of cell connectivity calculation is done when the surface is interactively created,
and more precisely, when saved for additional investigation. These are cases when the cutting plane
or the iso-surface is created, as the cell connectivity is generated for each newly created surface.
This calculation is computationally intensive and affects real-time interaction. Such aspects impose
further optimization and improvements to the cell connectivity calculation.
6

(a) grid

(b) grouping rule

Figure 48: Minimum node grouping rule

Initialization of empty sides set

for each cell:
for each side in cell:
if(side in sides)
connect and remove side from sides
else
insert side in sides
Metacode 1: The kernel of the connectivity algorithm
The kernel of the cell connectivity algorithm is based on the grouping of cell boundaries, being they faces or
edges, respectively for 2D or 3D topology. The cell connectivity algorithm receives the cell topology, which
consists of nodes defining each cell. The grouping occurs around the minimum node index composing the side,
see Figure 48, which is used as grouping key. This principle, which groups the sides around their minimum node
index, improves the traversal of the cells. The cell connectivity algorithm gets better performance because the
search is done just among the sides included in the node-side set. The supported side-types in 3D parametric
space are predefined faces: triangles or quadrilateral, while in 2D they are edges as segments, see Figure 49.

50

N2
N0

C2

C1

N1

C0

N3

1D cells

C1

C0
e0
n0

no

e0

C2

n0

e0

n1

n1

n1

N4

N3

C1

C0

C2
N5

N2

N1

2D cells

N0
n2

n2

n3

n2

e2

e1
e1

e2

e3

e1

C1

C0

e2

C2
n1

e0
e0

e0

n1

n1

n0

n0

n0
N9

N10
C3
N7

N5

N8

C1

N6

N2
N4

N3
C0
C2
N0

N1

3D cells

n3

n7
n1

C1

n2

n6

f5

n4

n2

n0

f0

n0
f2

f4

n4 n5

n4

n1

n3

n2
f3

n1

n3

n1

f2

f4

n0

n3

f0

C2

C0
n5

n2

n0

Figure 49: Cell connectivity for different parametric zones

51

At the beginning each node-side set is empty. The algorithm iterates over all the zone cells and constructs the
sides for each cell which are grouped according the minimum node index. When a side is created its minimum
node index is found. The side nodes are ordered from the minimum to the maximum node index. By definition,
the minimum node index is the first one in the array defining the side. This nodes order simplifies the
comparison between two sides. As the node set is of fixed size and contains all the zone nodes, the appropriate
data type to treat such set is the indexed array with the node index as entry. The node index makes possible to
find the set of all processed sides around the identified node.
The newly created side is compared with all the sides in the node-side set. If such a side is not found, the created
side is added to the node-side set. If there is a side in the node-side set, which matches the newly created side,
the connected cells for the newly created side and the matching side are set respectively inside the cell
connectivity table.
When discarding empty node-sides sets the resulting node set contains the node-sides sets, which contain non
connected sides. Such sides represent the cells forming the boundary sub-zone. The sub-zone of parametric
dimension-1 requires the complete definition of the zone topology. After the first traversal of all sides, the side
nodes are grouped in the unique node set, which by definition doesnt allow duplicate nodes. The sub-zone
nodes, when sorted in the monotonous increasing order, represent the LG map of the boundary nodes.
Simultaneously, the cell topology is created with the global node indexing of the sup-zone. Such sub-zone
topology is completely constructed if a node and a cell indexing are done in the local index space. The related
LG maps define the local indexing, which is applied to the cell topology and cell connectivity.
The result of the first iteration over boundary cells is:
the LG map of nodes,
the cell topology in global index space,
the LG map of cells.
The following step is the calculation of:
the cell topology in local index space,
the cell connectivity.
The cell topology in local index space is created from the cell topology in global index space where for each
global index the corresponding local index is found from the LG map of nodes. This algorithm is tuned with
appropriated hashing value to improve the searching performance.

(a) surface bounds volume

(b) curve bounds surface

Figure 50: Closed boundaries

52

unsigned int hash() const {

unsigned int i=iv.length(), j=0;
while(i--) j ^= iv(i);
return j;
}

Table 10: C++ implementation of the hashing value

The cell topology with the associated cell index array and parametric dimension of the cell types are sufficient
parameters to find the cell connectivity of the investigated zone. When all the cells are connected, the remaining
nodes which have additional sides represent the entry point for the definition of the zone boundary. The zone
boundary can have one or more parts. Thus, connected or disconnected regions can be found. Each such region is
defined as a closed zone. A curve bounding the surface or a surface bounding the body are examples of closed
zone types, see Figure 50.
outer Frame
collapsing front
inner Frame

boundary: collapsing multiple curves

boundary: collapsing curve

no boundary: collapsing point

Figure 51: Collapsing algorithm for multiple-connected and disconnected regions

The cell connectivity algorithm is repeated on the boundary because different disconnected or multipleconnected regions could be found. A separate calculation is made imposing the test based on the front marching
technique. In the collapsing algorithm, the necessary condition for the single-connected closed region is
accomplished when the boundary results collapse back to the point, thus these are no boundaries of examined
zone. If the collapsing is not accomplished the zone is an open zone with the associated boundary.

2
3

2
3

1
(a)

(b)

(c)

Figure 52: Surface parts (a) multiple-connected, (b) disconnected and (c) mixed
The algorithm includes the traversal of all the sides on the front allowing the cell to be considered only once.
Practically, the cell is added only once and removed only once from the front set of cells. The cell inclusion is
controlled with the cell-done vector. When the front reaches the surface/curve boundary the remaining sides are
not connected to any cell. The traversal of cells continues with the remaining ones which are not searched. The
next cell becomes the starting cell for the creation of the new curve/surface zone. Such treatment splits the zone
in multiple connected or disconnected regions, named ZoneParts, see Figure 51.
If the surface boundary exists it is always closed. For example, when surface has three boundaries it can form
different surface parts arrangements, see Figure 52. It is obvious that it is not sufficient to prove that for the
example shown in Figure 52 the surface boundary is the group of three curves. The question, if the zone is
53

multiple-connected or disconnected, still remains. The traversal of the cells starts with the arbitrary chosen cell
and follows the cell connectivity information. The surface is recognized to be a connected region when all the
cells are reached. In addition, if there are no boundaries the surface is closed. The algorithm is performed on
surface/curve level as the created and visualized objects are in 2D/3D space.
When all the sides are traversed, the remained entries into the node-side set represent the boundary cells of the
zone. These are, for a body the surfaces, and for a surface the curves. The boundaries are later used with the
definition of boundary condition patches. For the internal and boundary surfaces the topology has to be defined
from the global node index space. Thus, it is possible to reach sup-zones without leaving the zone itself from
every cell inside the zone.

1.2.4 Node Topology

The node topology algorithm defines for each zones node the surrounding cells, which are sharing it. This
information is very important for the conversion from the cell-centered numerical model to the vertex based one.
Such conversion is performed for improving the surface rendering (cell normals to define a node normal) or
when some derived quantities require cell values to be used to define the value at zones node. In order to fulfill
the continuity condition for the calculated quantities, the node value has to be unique for all the involved cells, so
an averaging mechanism, based on geometrical or physical properties, can be defined in order to calculate the
unique node value. For example, the surface rendering requires a unique normal definition at each node. This is
obviously not the case when each cell has its own normal and local mapping based on the vectors which define
the cell edges. The only case where the normal is unique for all the cells is the case when all the cells belong to a
unique tangential plane.
dual

original

node topology
node connectivity

cell topology
cell connectivity

Figure 53: Dual topology concept between nodes and cells

The node-cell relationship has the M<>M cardinality which implies that one node defines multiple cells, named
node topology, and one cell is defined with multiple nodes, named cell topology. The dual principle of node and
cell topology is shown in Figure 53. If we focus on the cell connectivity graph and we replace the cells with
nodes at their gravity centers the result is the node connectivity graph. For both the concepts, the topological
structure remains the same. Thus, it is generic to nodes or cells. In order to keep one data structure we recreate
the dual structure when necessary. The topology and connectivity for cells are defined and the node dual
structures are calculated when needed. The dual node topology structure is calculated in the case of the before
mentioned averaging mechanism when the quantity is calculated from the values associated with surrounding
cells. The node topology algorithm makes consistent the topological information in a zone. The achieved benefit
is the reduction of the memory allocation. For structured zones, this information is implicitly given through
structured indexing convention follows as trivial solution. For the unstructured case the algorithm is based on the

54

traversal of all the zone cells, where for each node the connected edges are found and surrounding cells are
identified. The algorithm manipulates dynamic information which implies that the exact length of arrays storing
topological information can be defined when complete cell/node traversal is performed. The required
intermediate step is the creation of temporary sets for the processed state of nodes and the two disjoint sets
storing surrounding cells. The two disjoint sets avoid recursive invocation of the cells which where previously
processed.
Initialization of the node_processed state
Initialization of the node_topology set.
for each cell:
for each node in cell:
if(node_processed(node)) continue
else
node_processed(node) processed
initialization of topology set
initialization of cell_not_processed set
insert cell in cell_not_processed set
while(cell from cell_not_processed)
insert cell in topology set
for node find local_node in cell
for cell find sides set containing local_node,
for sides set find connected_cells set
if(connected_cells not in topology set)
insert connected_cells in cell_not_processed
insert topology set in node_topology set.

Metacode 2: The node topology algorithm

The logic of the algorithm is outlined in the Metacode 2. The node topology set has all the necessary information
to recreate the node-centered topological information for a zone, see Figure 47 in Structured and Unstructured
Topology section.
The node reduction algorithm is a minor modification of the node topology algorithm where the cell connectivity
information is available and the unique node indexing has to be calculated for the given zone. The problem is
illustrated in Figure 54. The elimination of duplicate nodes is based on coherence and tolerance model, see
section 1.1.3. The nodes are assumed to be the same if they are within the range defined in the tolerance model.
The local node identification is not based on the local index, as in the case of node topology algorithm, but on
tolerance model. In addition, the new node index space is created.

9 1
7

2
3

Figure 54: Set of cells with duplicate nodes.

55

We assume that the connectivity information allows navigation around the node. The treatment of a particular
case, when the node is building two cells, which are not connected through one of the cell sides, is resulting in
having two nodes at the same place, see Figure 55. In such a case the boundary curve would intersect itself. This
is avoided by applying node reduction algorithm, which eliminates the creation of non-manifold geometry.

Figure 55: Navigation around the node: (a) one node (b) two nodes
The second variation of node topology algorithm is the calculation of node value based on the averaging
mechanism. The problem is to compute the cells surrounding the node and to calculate the node value from the
cell values. The calculation of the node value is reusing the navigation part of the node topology algorithm in the
simplified form for requested node. In addition, the cell can be customized to contain the information necessary
to apply the desired averaging mechanism.

56

1.2.5 Domains Connectivity

In this section the connectivity information for structured non-overlapping grids is described. The domain
connectivity is based on matching boundaries. It is shown how the orientation of the boundaries and segments is
predefined together with rules which establish the matching between the two neighboring segments. In addition,
the input description for the domain boundary conditions is given.
j

B4

B5

B3
B1
B1

B6

B4
B2

B3

k
B2

Figure 56: Domain with boundary indexing in 2D & 3D and boundary identification
The domain can be analyzed as a hexahedron cell type, see section 1.1.1, with the difference that the cells faces
are domain boundaries, see Figure 56. The domain itself can be arbitrary oriented, but it must be right handed.
Because of the structured nature of the domain node indexing (i,j,k), the boundary indexing logic is different than
face indexing of hexahedron cell. The rule defining the indexing is based on the permutation of the i,j,k planes in
increased order for minimum index value and in decreased order for maximum index value. Applying such a rule
the following Table 11 is defined:

Boundary

Index

Value

Curve (i)

B1

(i)

B2

(j)

B3

max

(j)

B4

max

(i)

Boundary

Index

Value

Surface indices(i,j)

B1

(j,k)

B2

(i,k)

B3

(i,j)

B4

max

(i,j)

B5

max

(i,k)

B6

max

(j,k)

Table 11: Boundary indexing for 2D and 3D structured grids

This indexing logic has and interesting feature because it simulates the playing dice where the sum of indices for
opposite boundaries is constant and equal 7. Note that it doesnt meter weather the boundary is viewed from the
outside or inside of the domain. The rule is that the lower index varies first and thereby the orientation is
uniquely determined. As the boundary can be constructed with different boundary condition, additional
decomposition of boundaries in segments (patches) is given. The segment is a part of the boundary with
minimum and maximum extension.

57

S2
i

D3

S1

j
i

i
II

D3

III

D2

S1

S1
j

S1
D2

S2

I
S2

D1
D1

S2

j
S1

j
i
i

Figure 57: Domain connectivity in 2D

In the 2D example, see Figure 57, there are 3 domains, each having 2 connected segments. For the domain D1,
the involved boundaries are B4 and B3. On the boundary B4 there are two segments and the second one is
connected to domain B2. Such identification path though domain boundary segment is the origin for the
following adopted notation. The unified segment identification as Dd.Bb.Ss, where lower case characters d, b, s
are replaced with numbers for specific case. Coming back to the example, there are three connected regions I, II
and III. The connectivity between them is specified in the following Table 12:

Connected region

Segment

Segment

Orientation

D1.B4.S2

D2.B4.S2

R (reverse)

II

D1.B3.S2

D3.B2.S1

E (equal)

III

D2.B4.S1

D3.B4.S2

R (reverse)

Table 12: Domain connectivity specification in 2D

There are eight possible orientations of the neighboring segments in 3D defined with reference node and
orientation, see Figure 58. The first four possible cases of neighboring segment can exist if the origin of the
segment is located in one of the four nodes as indicated in Figure 58, where the orientation between the
connected segments is equal. The next four cases define the inverse orientation.
0
n0

n3

+
i

+
n2

n1

3
j

j
j

i
4

Figure 58: Segment orientation cases

58

In the following 3D example, see Figure 59, the 3 connected surfaces areas are identified. The segment
orientation and the node reference are added to the 3D input. Applying the same symbolic path for the segment
identification as the one for 2D, the following Table 13 is constructed:

Connected region

Segment

Segment

Node

Orientation

D1.B6.S4

D2.B6.S3

R (reverse)

II

D2.B3.S2

D3.B5.S2

R (reverse)

III

D1.B6.S3

D3.B3.S2

E (equal)

Table 13: Domain connectivity specification in 3D

The connectivity for the non-matching structured boundaries is removing one-to-one comparison of their nodes.
and each passing from one boundary to the connected one is implying additional algebraic treatment of the point
location [71].

59

k
j

Domain 3

Domain 1

j
k

Domain 2

D3.B3
D1.B6
i
S1

j
D1.B6.S3

S1

D3.B3.S2

S2

D3.B5
j

i
i

S1

D3.B5.S2

D1.B6.S4
D2.B3.S2
S3
j

j
S1

D2.B3

i
S4
j

D2.B6.S3

S1
S2

D2.B6

Figure 59: Multi-domain (multi-block) connectivity in 3D

60

1.2.6 Cell Intersection Pattern

The cell intersection pattern is based on the cell model, see section Cell class. All components of cell topology
are simply-connected and convex regions. The 3D cells are bounded with oriented faces which are themselves
bounded with oriented edges. In order to define the cell intersection pattern the cell topology is modeled with the
Winged Edge Boundary (WEB) model [72-74]. The WEB model supports explicit identification of nodes and
faces starting from edges. ERM for 3D cell topology, see Figure 60, shows that an edge has the sup/sub
topological relationships with other cells. The edge is composing a face (sup-relationship) and in addition it is
constructed from two nodes (sub-relationship).
M

Edge
2
M

sub

sup

2
M
M

Face

Vertex

1
M

1
M
1

Solid

Figure 60: Cell topology ERM

The important difference between the B-rep implementations are the choices which define the relationships
stored explicitly in the WEB structure and the relationships left implicit (to be computed as needed). As
mentioned previously in this chapter, the main B-rep topological items are: faces, edges and vertices. A face is a
bounded portion of a surface; an edge is a bounded piece of a curve and a vertex lies at a point. Other elements
are a set of connected faces, a loop of edges bounding a face and a loop-edge relationship (WEB) which are used
to create the edge loops. In the full developed form the node model has three main relationships: edge <> node,
face <> node and solid <> node, see the skeleton tables 1-4 to 1-7. Such model is mapped to the WEB model in
order to have a form in which an edge identifies its sub/sup cell, respectively node/face, see Figure 60.

Node

Face

Edge A

Edge B

Edge

Start

End

Next

Prev

Next

Prev

nstart

nend

fA

fB

enA

e pA

enB

eBp

Table 14: The WEB model record

Each WEB relationship requires a labeled record, see Table 14. In the cell topology section different B-reps are
presented within the cells skeleton tables. These tables are defined with the vertex B-rep model, where edges and
faces are defined with ordered set of nodes. Nodes order preserves the edge/face orientation, namely counter
clockwise as seen from the outside of the cell, see Figure 61.

61

Figure 61: WEB model viewed from the cell interior

In the ER model of cell topology both node and edge models are present, see Figure 60. The WEB model extends
the edge-based boundary model by the connectivity information between edges composing the face. Each edge e
appears in exactly two faces. The other two edges eA and eB appear for each edge e in the faces description, see
Table 14. Moreover, each edge has the consistent orientation of faces, where e occurs exactly once in its positive
orientation and exactly once in the opposite orientation. The WEB data structure takes into account the
mentioned properties and identifies edges and faces based on the looping forced by the face orientation which
can be clockwise or counterclockwise, see Figure 61. The face is defined with the starting edge and the face
orientation. In addition, WEB includes the neighboring faces and the previous and next edge of the faces. As the
full WEB data structure includes the neighboring edges, the edges for each node can be extracted.
e9

e8

f2

f3

e5

f2

f3

e5 e5

e1

e0

Figure 62: WEB model edge analysis

The Figure 62 shows the WEB representation of the edge e5. The face f2 and f3 are considered as faces fA and fB
respectively. For example, the complete edge-based WEB model is constructed for the hexahedron; see Table 8,
where the first node and face for each edge are also given.

n4

e4

n0

e8

n4

n5

e8

n5

f2
e4

e5

e0

n0

n1

a) local

e5

e0

n1

b) global

Figure 63: The edge orientation a) local to face b) global to cell

62

The global edge orientation is given as direction from its start to its end node. This information is available from
node-based B-rep, where edges are defined in terms of nodes. In Table 15, the face f3 of the tetrahedron is
defined with the edges e1, e5 and e4. The face orientation determines the local orientation of edges, when defining
the face itself. The WEB representation of the face fA is based on the global edge orientation, which is aligned
with counter-clockwise orientation.

First Node-

First Edge-Face

Node

Edge

Face

Edge

n0

+e0

f0

-e0

n1

+e1

f1

+e2

n2

+e2

f2

-e3

n3

-e3

f3

+e1

Node

Face

Edge A

Edge B

Edge

Start

End

Next

Prev

Next

Prev

e0

n0

n1

f2

f0

e4

e3

e2

e1

e1

n1

n2

f3

f0

e5

e4

e0

e2

e2

n2

n0

f1

f0

e3

e5

e1

e0

e3

n0

n3

f1

f2

e5

e2

e0

e4

e4

n1

n3

f2

f3

e3

e0

e1

e5

e5

n2

n3

f3

f1

e4

e1

e2

e3

Table 15: WEB model of the tetrahedron

Such data structure treats faces and vertices in a completely symmetric manner. If the start node ns, end node ne,
and the connected faces fA and fB are exchanged, we end up with the dual cell of the original one, see Figure 65.
The intersection cell pattern is based on cell nodes states which for each node can have a label set to TRUE(+) or
FALSE(-). For each type of intersection pattern the node states are grouped together forming the node-mask. The
node-mask captures the intersection pattern and it is mapped to a binary representation of an integer. The number
of intersection patterns depends on the number of cell nodes describe with the following equation:
number of patterns = 2 number of cell nodes
The lookup table, contains all the possible intersection patterns for a given cell type. The unique integer label is
created from the node mask, where each bit stores the node threshold state. The threshold is calculated for
extraction algorithms like cutting plane and iso-surface are. For example, in the cutting plane algorithm the
threshold is its distance sign from the plane and for the isosurface the threshold is difference between the scalar
value and the given iso-value. Except for the computed node mask, based on geometry or quantity, and for the
calculation of the intersection point or iso-value position, the marching cube algorithm is used unmodified for
both: cutting plane and iso-surface algorithms [75].
63

Label
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15

Node Mask
0
0
0
0
0
1
0
1
1
0
1
0
1
1
1
1
0
0
0
0
0
1
0
1
1
0
1
0
1
1
1
1

0
0
0
0
0
0
0
0
1
1
1
1
1
1
1
1

0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1

n
0
3
3
4
3
4
4
3
3
4
4
3
4
3
3
0

Edges Intersected
0-3-2
0-1-4
1-4-3-2
1-2-5
0-3-5-1
0-2-5-4
3-5-4
3-4-5
0-4-5-2
0-1-5-3
1-5-2
1-2-3-4
0-4-1
0-2-3
-

Faces Connected
1-0-2
2-3-1
3-1-0-2
2-0-3
1-0-3-2
2-0-3-1
0-3-1
1-3-0
1-3-0-2
2-3-0-1
3-0-2
2-0-1-3
1-3-2
2-0-1
-

Table 16: The lookup table for the tetrahedron

+
P2

P1

Intersection pattern
2

P3

P0

3
3

P2
P3

Intersection pattern
2

P1

P0

Figure 64: Intersection pattern and polygon orientation

The header of the lookup table, see Table 16, consists of the intersection pattern label, the node mask and the
ordered sets of intersected edges and faces. The implied order of intersected edges guarantees that the scattered
points of the intersection pattern define a polygon and that its orientation is in accordance with the global
orientation of a created surface. Different orientations for the same intersection geometry have different node
masks, see Figure 64.

64

n4

n3

f1

f1

v
f4

n2

f3
f2
n0

n0

f2

n3

f3

f0

f0
n2

n1

n1
u

u
w

n5

n7

n4

n3

f5

f4

n6

f1
f1
f3

n4

f4

n5

v
f2

n2

n3

f2
f3

n0

n0

f0

f0

n2
n1
n1

u
u

Figure 65: Dual Cells

The lookup table for each intersection pattern identifies which edges are intersected together with the intersected
edge sequence, which preserves the polygon orientation. The edge is intersected, if its nodes have different
labels. For each node labeled FALSE, the traversal of the joint edges is done in the counter clockwise direction
and for the node labeled TRUE in the clockwise direction, as shown in Figure 66. This orientation is valid if the
node is viewed from the cell interior.

(a) FALSE

(b) TRUE

Figure 66: Edge traversal direction for nodes labeled as: (a) FALSE (b) TRUE

65

n3

n3
e5

n2

e5

e3
e4

e2
e0

n0

e4

n2

e1

e2

f3

f2

e3

f0
e0

n0

n1

e1

n1

f1

Figure 67: The navigation graph for tetrahedron

When the navigation graph is coupled with node mask, see Figure 67, the unique traversal path is defined with
the condition, where the looping is done around FALSE nodes in clockwise direction. The intersection pattern
identifies which edges are intersected. The edge ordering preserves the pattern orientation. The polygonal nature
of the intersection pattern imposes looping over intersected edges in the predefined order imposed by the node
mask. The algorithm takes advantage of the consistent edge/face orientation, see Metacode 3.

identification of node mask

traversal of the mask
node identification and its first intersected edge
the edge becomes starting edge of the traversal.
looping over intersected edges following the false node and face orientation
the looping ends on starting edge

Metacode 3: The cell intersection pattern algorithm

If we consider the other cell types the number of intersection patterns increase as shown:
24=16

25=32

26=64

28=256

n4

n4
e7
e4

e7

n3
e3

e4

e6

e6
f4

f3

e5

f1

e5

e2

n0

e3

n3

n0
e0

e2
e0
f2

n2
e1
n1

Figure 68: The navigation graph for pyramid

66

n2
f0

e1
n1

n5
e8
n5
n3
e7

e5

e5

e8

e6
e3

f1

n3

n4

e3

n
2

e2

n0
e0

f2

n0
e4

e1

n2
e2
f0 e1
f3

e7

n1

e6

e0
e4
n1

f4
n4

Figure 69: The navigation graph for pentahedron

For the reason of readability the example of the pentahedron lookup tables is given in appendix on page 231
The navigation graph for other cell types follows; see Figure 68, Figure 69 and Figure 70.

n7
e1o

n6
n7

e11
e7
n4

n3

e6
e8

n3

e2

e2

e11

f1

e3

f0

e1

n2

f3

e9

e5
e1

n0

e6
n2

n5

e4
e3

f4

e7

e9

n6

e10

n0
e4

e0

e0
f2
e8

n1

n4

n1
e5
n5

f5

Figure 70: The navigation graph for hexahedron

The cell intersection pattern doesnt imply that the created region is simply-connected. It can happen that we
have up to four disconnected regions. The next complication is that polygons with greater number of nodes than
triangles or quadrilaterals can compose the intersection pattern see Figure 71. In such case it is important to
preserve the internal cell connectivity necessary for the creation of the zone topology.

67

Figure 71: Some intersection patterns for the hexahedron cell with the polygon partitions
The underlying cell map for the general n-polygon cell type is not defined. An unstructured surface can be
composed only of triangles or quadrilaterals because they are the only cell types defined for 2D topologies. In
order to subdivide the n-polygon cell type a very simple logic of subdivision is applied. It minimizes the number
of created cells by generating quadrilaterals whenever possible. To eliminate random approach to the local
subdivision (geometry orientation and connectivity) the Table 17 was defined as one possibility to uniquely
define the polygon partitions.
N-polygon

Subdivision

triangle

quadrilateral

quadrilateral + triangle

2 x quadrilateral

2 x quadrilateral + triangle

Table 17: Polygon Subdivision

The number of created internal cells depends on the number of polygon nodes. The unique triangles and
quadrilaterals and their orientation (intersected edges order) are defined in the cases with more than four
intersected edges. The practice shows that the maximum polygon has 7-nodes. This happens in the case of
hexahedron cell. The subdivision generates one triangle and two quadrilaterals as shown in Figure 72. The
advantage is that such decomposition produces less number of cells than the triangulated surface. The
triangulation algorithm was made for the first version of marching cell algorithm.

68

Figure 72: Polygon with maximum number of nodes, hexahedral cell.

In the intersection pattern algorithm the orientation of each internal polygon cell has to be preserved within all
created cells. The connected faces are used in the marching cell algorithm, which can be combined with the cell
connectivity algorithm, when the proper topological surface has to be created.

Label nT
0
1
65
173
165

0
1
2
3
4

2
170
188
225
110
77
52

0
0
1
2
1
0
1

Triangles

Triangles

Edges
0-4-3
0-4-3 6-10-9
0-5-1 4-11-8 6-9-10
0-5-1 4-11-8 3-2-7
6-9-10
6-9-10
0-7-3 4-11-8
0-10-8
1-4-5

Faces
2-1-0
2-1-0 4-5-3
2-3-0 1-5-2 3-5-4
2-3-0 1-5-2 0-4-1
3-5-4
3-5-4
6-1-0 1-5-2
6-5-2
7-2-3

nQ

Quadrilaterals

Quadrilaterals

0
0
0
0

Edges
-

Faces
-

1
2
1
1
1
2
2

1-4-3-2
0-3-11-8 1-9-10-2
1-3-4-5
0-5-6-7
0-3-7-10
0-4-7-10 0-10-9-1
1-2-6-9 1-9-11-4

3-2-1-0
0-1-5-2 3-5-4-0
0-1-2-3
2-3-4-6
0-1-4-6
2-1-4-6 6-5-3-0
0-4-3-6 6-5-1-7

Table 18: Records from hexahedron lookup table with polygon partitions and multiple connected regions
The new lookup table consists of the header indicating: intersection label, the created triangles/quadrilaterals and
the connected faces for the marching cube algorithm. Obviously, the internal subdivision of polygon into
possible quadrilaterals and triangles requires the cell connectivity indexing between internal cells. In the
hexahedral example, see Table 18, there are defined faces which doesnt exist for hexahedral cell. This faces are
used for the internal cell connectivity indexing. It is assumed that any face with greater index then maximum
index of the intersected cell face is an internal polygon cell.

69

Figure 73: Pathological intersection cases results in non-manifold 2D & 3D geometries

The intersection faces are never connected internally and externally. Such possibility would setup non-manifold
configurations, as shown in Figure 73, where an edge or a point would be connected to more than two segments
or faces.
Correct

Made
3

2
2

4
0

0
5

Figure 74: Pathological case of the star-shaped polygon, and its polygon partition
The intersection pattern algorithm treats only topologically the intersected cell. Such method only guaranties to
produce sensible results when convex polygons are created. The pathological case, see Figure 74, shows the starshaped polygon, which could happen if the algorithm would generate overlapping cells. However, such cases are
not expected because the good numerical simulation grids are made from convex cells.
Integer
0
1
2
3
4
5
6
7

F
F
F
F
T
T
T
T

Bit representation
F
F
T
T
F
F
T
T

Edges intersected
e1, e3
e1, e2
e2, e3
e2, e3
e1, e2
e1, e3
-

F
T
F
T
F
T
F
T

Table 19: Lookup table for the triangle

70

Polygon created
l1
l1
l1
l1
l1
l1

Integer
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15

F
F
F
F
F
F
F
F
T
T
T
T
T
T
T
T

Bit representation
F
F
F
F
F
T
F
T
T
F
T
F
T
T
T
T
F
F
F
F
F
T
F
T
T
F
T
F
T
T
T
T

F
T
F
T
F
T
F
T
F
T
F
T
F
T
F
T

Edges intersected
e1, e4
e1, e2
e2, e4
e2, e3
e1, e2, e3, e4
e1, e3
e3, e4
e3, e4
e1, e3
e1, e2, e3, e4
e3, e2
e2, e4
e1, e2
e1, e4
-

Polygon created
l1
l1
l1
l1
l1, l2
l1
l1
l1
l1
l1, l2
l1
l1
l1
l1
-

Table 20: Lookup table for the quadrilateral

The only ambiguity that arises with is that for the intersection points, Figure 73, there are different possible
connection patterns:
connect the point on the next edge, clockwise or counter wise around the cell;
surround regions of higher or lower-valued scalar quality;
connect the opposite edges;
calculate the average of all the four nodes, which will indicate the connection of the isolines.
It must be noted that the presented cell intersection pattern algorithm treats separately all the possible intersection
cases for the defined cell types, without implying any symmetry or rotational similarity, in order to reduce their
number. In case of ambiguity, when several connecting possibilities might exist, the one aligned with the
imposed edge traversal is applied, see Figure 66. The presented algorithm generates unambiguous zone topology,
only considering the cell intersection patterns, which differs from existing algorithms, which treat the influence
of the quantity value change to remove ambiguities [76].

71

1.3 Continuity Assumption

The primary purpose of the scientific visualization system in Fluid Dynamics is to visualize the computed
discrete properties, which describe the fluid flow as continuous matter. The fluid flow description is discretized
throughout the numerical data model. The adopted Eulerian standpoint, as opposite to the Lagrangian coordinate
system, which follows matter as it moves, describes the fluid quantities at specified points:
Q = f (x, y, z, t)
where Q may be, for example, stress, density or velocity.
The continuity assumption describes the fluid fields: scalars, vectors or tensors, in the sense that the average
quantity Q in a fluid surrounding the point are assigned to the point itself. This assumption ignores the molecular
nature of the matter and the discontinuities which arise from this assumption. In addition, the motion of the
molecules is neglected. It is assumed that the fluid is a continuous medium and its quantities change
continuously. It means that the value of a quantity Q at point x and its value in the surrounding x+x can be
related to its value at x by a Taylor series

Q(x + x) = Q(x) +

Q
2Q ( x ) 2
x + 2
+ ...
x
x 2!

The first two terms are only required to indicate the behavior when x approach zero. Thus, one has a fictitious
continuum supported by mathematical techniques, especially interpolation explained in the following section
map the numerically generated data as discrete model in the continuum model.
In this section, the focus is on the development of techniques necessary to support the continuity assumption
using numerically generated data as field quantities on defined geometry. These data represent the approximation
of the given continuum by the set of points defined in the discretized space, called computational grid. The
assumption of continuity has to be kept in mind despite that each grid point is storing the physical solution in
discretized form. An interpolation method that expands the values of the discretized solution in the continuous
solution, in a whole domain, models the concept of the continuum throughout the computational grid. Thus, the
base for the continuum field analysis is formed and we can extract quantities related to different geometries as
point, curve or surface fields.

72

1.3.1 Interpolation Method

The numerical treatment of the continuum model is based on interpolation methods, which are used in
constructing new data points from the discrete set of known quantity fields points. The continuum data function
is approximated with the interpolation formula, which is defined with the set of discrete numerical values. Some
numerical simulations compute the solution at the cell nodes and the scientific visualization algorithms use
interpolation formulas to define the solution, for example to find a scalar or vector data, at any desired location
inside the cell. In this section the interpolation formulas are defined for the coordinate transformations, Jacobian
matrices and Metric tensors, see Figure 75, for the selected cell types, see section 1.1.1:
1

1D segment

U1 in E ,E ,E

2D triangle, quadrilateral

U2 in E ,E

3D tetrahedron, pyramid, pentahedron, hexahedron

U3 in E

The interpolation is based on the parametric definition of the point location, including the mapping between the
modeling and parametric space. Each cell has the following two important parameters:

the dimension of the parametric space k.

the dimension of the modeling space n

The Euclidian space En with dimension n=1,2 or 3 is defined with variables x, y and z used as coordinates in the
modeling space. The parametric space Uk with dimension k=1,2 or 3 is defined with variables u, v and w.
Modeling SpaceE3

Parametric SpaceU3
w

P(x)

x = A (u)

J=

A
u

X (x,y,z)

G = JTJ
y
P(u)
u(u,v,w)

Figure 75: Coordinates transformations characteristics

The mapping from the parametric space U to the modeling space E is defined as:
n
n
A :U E
k
k

1.3.1 - 1
and in vector notation:
x (x,y,z) = A [u (u,v,w)]

1.3.1 - 2

There are several geometrical variables, see Figure 75, which has to be defined for the mapping A, which
represents the base for the coordinate transformation, and in addition, it is applied for the definition of the
Jacobian matrix J and the Metric tensor G.

73

It is assumed that the cell is simply connected, thus topologically, uniquely describing a single portion of the
Euclidian space [62]. It is important to note that the topological structure of the cell, named cell topology is
preserved by the mapping A. The cell is a closure of an open connected set and it is assumed to be a closed
including its boundary Uk. This means in particular that the cell nodes are included in the mapping. The cell
boundaries in the parametric cell space Uk are aligned with the constant coordinate axis or lines/planes with unit
normal n(1,1,1). Before the mapping A is derived, a cell in modeling space must be numerically specified with
all its constituting nodes. The cell boundary Uk must be given in order to define the boundary mapping as:

Akn : U k E n
1.3.1 - 3
before the extension to the cell interior is done. Each point in the parametric space U is mapped to a unique point
in the modeling space E. Thus, for every point u Uk, there exist a unique point x En and for every point x
En, there exist a unique point u Uk. Such mapping is smooth and non-singular within the cell and preserves in
addition the cell orientation.
As it could be noted, the superscript and subscript indexing notation was used in this section to indicate the
contravariant and covariant nature of respective coordinates transformations to keep it general [77]. From now
on, we assume that the applied coordinate systems are orthogonal, which yields that the respective contravariant
and covariant coordinates are equal, and we used only the subscript indexing.

Isoparametric mapping and shape functions

It is assumed that the isoparametric mapping A is a polynomial form defining a general scalar field f at the
parametric coordinates u, with a set of coefficients a. The general formulations, which take into account the cell
parametric dimension (exponents r, s and t are increased respectively for 1D, 2D and 3D) and the number of cell
nodes (indexed by i) are:
i =1

1D:

f=

a (u
i

) i = a0 u0 + a1 u1

i =0
r 1

f = a0+ a1u
i =3

2D:

f=

ai( u v

) = a0 u0 v0 + a1 u1 v0 + a2 u0 v1+ a3 u1 v1

r s i

i =0
r +s2

f = a0 + a1 u + a2 v + a3 u v
i =7

3D:

f =

a (u v w )
r s

t i

i =0
r + s +t 3

f = a0 u0 v0 w0 + a1 u1 v0 w0 + a2 u0 v1 w0 + a3 u0 v0 w1
+ a4 u1 v1 w0 + a5 u1 v0 w1 + a6 u0 v1 w1 + a7 u1 v1 w1
f = a0 + a1 u + a2 v + a3 w + a4 u v + a5 u w + a6 v w + a7 u v w
1.3.1 - 4
For example, in the case of a quadrilateral cell, the 2D equation 1.3.1 - 4 can be satisfied for each cell node by
forming a set of simultaneous equations, for which we assume that the function value fN is known at each cell
node, as follows:

74

 f 0 1
f 1
1 =
f 2 1

f 3 1

u0

v0

u1

v1

u2
u3

v2
v3

u 0 vo a 0
u1 v1 a1
u 2 v2 a 2

u3 v3 a3

or in the vector notation

fN = C a

1.3.1 - 5
The unknown coefficients a can be calculated by finding the inverse of C when fN is defined at cell node, as:
a = C -1 fN

1.3.1 - 6
Once coefficient a is computed, it can be substituted in equation 1.3.1-4, and written as
f=pa

1.3.1 - 7
where p(u) = [ 1, u, v,........, u v,........,u.2........]
resulting in
f(u) = p(u) C -1 f N.

1.3.1 - 8
The shape function h is defined as
h(u) = p(u) C-1

1.3.1 - 9
and when combined with the node solution fN gives the polynomial form
i = M 1

f(u) =

hi(u) (fN)i

i =0

1.3.1 - 10
where M indicates the number of cell nodes. This equation reveals important constraints of the shape function h
expressed as:
hi = 1, for node i and hj = 0, for node j i

1.3.1 - 11
The shape function h has a value of unity at an arbitrary node i and is equal to zero at all other nodes. The
variation of the solution along all boundaries is retained in order to satisfy condition for continuity. The outlined
formulation has two know disadvantages [69]:
1. The inverse of C may not exist, and
2. The evaluation of C in general terms for all cell types involves considerable algebraic
difficulties.
As the definition of the shape function h is necessary for all the elaborated cell types, it is appropriate to define C
-1
more explicitly for the code implementation. This is accomplished by applying the interpolation polynomial in
the Lagrange form [78, 79], which is well known numerical interpolation method for its systematic definition of
shape functions. Lagranges polynomials L satisfy the constraint given in the equation 1.3.1-11 and the cell-tocell continuity condition. For example, their definition is given in explicit form, in 1D coordinate space,

75

hi hu Lni ( u ) =

(u uj )

(u u
j =0
j i

1.3.1 - 12
In the term Lin, n stands for the number of subdivision in the cell along the parametric coordinate axis. As the
analyzed cell types are linear the number of subdivision is 1 and it is constant for all of them. In the following
equations such indices are removed and replaced with the node number associated with the parametric
coordinate (u,v). The explicit form of the 2D shape function is defined in accordance with the analogy to 1D as
follows:
hi huv Lu(u) Lv(v)

1.3.1 - 13
and 3D shape function as:
hi huvw = Lu Lv Lw

1.3.1 - 14
The following relations satisfy the shape function conditions 1.3.1-12 for different dimensions spaces
2D: hi hj = ij

1D: hi =1

3D: hi hj hk = ijk

1.3.1 - 15

Lagrange polynomials provide characteristics to define non-linear cell types. As the development of scientific
visualization system could be diversified into non-linear cells, the presented structure represents a good basis to
capture future needs to develop interpolation methods, which will better approximate the imported results from
numerical simulation in the visualization system.
As explained, the isoparametric mapping A is based on simple products of Lagrange polynomials in the
parametric cell space U supported by values defined at cell nodes:

h ( u )( f )i
M

f = A(u) =

i =1

1.3.1 - 16
where M is the number of cell nodes and (fN )i are the nodes coordinates or solution. The mapping is called
isoparametric because the coordinates and solution are treated with same the shape function. The simplest shape
function of order 1 is defined for 1D as:
Lo (u) =

u u1 u 1
=
= 1u
u0 u1 0 1

L1 (u) =

u u0 u 0
=
=u
u1 u 1 0

1.3.1 - 17
Substituting 1.3.1-17 in equations 1.3.1-13, 1.3.1-14 and 1.3.1-15 the following result is obtained, respectively
for 1D, 2D and 3D. The fully developed mapping for linear cell types follows:
1D: segment
A

i =0

i =0

hi f i N ai u i = a0 + a1u1 = a0 + a1 u

A = h0 f0 + h1 f1 where

h0 L0 (u) = 1 - u
h1 L1 (u) = u
76

and after simple algebraic manipulations, the a coefficients are

defined as:

a0 = f0

f1

f0

n0

n1

a1 = f1 - f0
A = a0 + a1u

1.3.1 - 18

This is the isoparametric mapping for segment cell. Follows the derivation of the isoparametric mapping for 2D
cell types:
3

A =

h
i =0

i ( u ) fi

a (u v
i

r s i
)

i =0
r + s2

2D: quadrilateral
h 0 h00 = L0 (u) L0 (v) = (1 - u ) (1-v )

h1 h10 = L1 (u) L0 (v) = u (1 - v)

n3

h2 h11 = L1 (u) L1 (v) = u v

n2

h3 h01 = L0 (u) L1 (v) = (1-u) v

and after simple algebraic manipulations the mapping coefficients are
a0 = f0
n0

a1 = f1 - f0

n1

a2 = f3 - f0
a3 = f0 - f1 + f2 - f3 = - a1 + f2 - f3
A = a0 + a1 u + a2 v + a3 u v = a0 + u (a1 + a3 v) + a2 v

1.3.1 - 19

The isoparametric mapping A for triangles is a degenerated case of the quadrilateral one, where the shapes
function of the dummy node hD h11= 0 annihilate the influence of the a3 coefficient. The coefficients for 2D
triangle follow:
v

a0 = f0
nD

n2

a1 = f1 - f0
a2 = f2 - f0
hD h11= 0 u v a3 = 0
A = a0 + a1 u + a2 v

n0

n1

1.3.1 - 20
Note that the nodes indices are shifted according to the triangle node connectivity see Table 3: TRIANGLE
skeleton table on page 32. The most complex isoparametric mapping in the context of this thesis is the 3D
mapping of the hexahedron.
77

i ( u ) fi

i =0

ai ( u
i =0
r + s +t 3

r s

v w )

This mapping is fully developed before the specific mappings for tetrahedron, pyramid and pentahedron, cells as
they are degenerated cases of the hexahedron one. The shape functions for each node are defined in Table 21,
and the mapping A is:
A = a0 + a1 u + a2 v + a3 w + a4 u w + a5 u w + a6 v w + a7 u v w

Shape function
h0

h000

(1 - u) (1 - v ) (1 - w )

h1

h100

u (1 - v) (1 - w)

h2

h110

u v (1 - w)

h3

h010

(1 - u) v (1 - w)

h4

h001

(1 - u) (1 - v) w

h5

h101

u (1 - v) w

h6

h111

uvw

h7

h011

(u-1)vw

uv

uw

wv

uvw

+
+

+
+

Table 21: Shape function for 3D isoparametric mapping

From the Table 21, the coefficients are calculated from the node values f and after simple algebraic manipulation
they are defined as:

a0 = f0

n7

n4

a1 = f1 - f0
a2 = f3 - f0

n6
v

n5

a3 = f4 - f0

n3

a4 = f0 - f1 + f2 - f3 = - a1 + f2 - f3

n0

a5 = f0 - f1 - f4 + f5 = - a1 - f4 + f5

n2

a6 = f0 - f3 - f4 + f7 = - a2 - f4 + f7

n1

a7 = - f0 + f1 - f2 + f3 + f4 - f5 + f6 - f7 = - a4 + f4 - f5 + f6 - f3

The coefficients are grouped to reduce the number of multiplication operations as:
A = a0 + u [a1 + v (a4 + a7 w) + a5 w] + v (a2 + a6 w) + a3 w

1.3.1 - 21

The 3D cells are always embedded in the hexahedron, see figure above. As mentioned, degenerated cases of the
hexahedron are tetrahedron, pyramid and pentahedron cell. The shape functions hi, containing the non existing
nodes are removed from the isoparametric mapping A by making them equal to zero, equation 1.3.1 - 21. For the

78

tetrahedron the existing nodes are 0, 1, 3 and 4 of the hexahedron, thus the shape functions h2, h5, h6, h7, are zero,
see Table 21:
h2=0

h2= u v u v w

from h6= u v w = 0

uv=0

h5=0

h5= u w u v w from h6= u v w = 0

uw=0

h6=0

h6= u v w

uvw=0

h7=0

h7=v w u v w

from h6= u v w = 0

vw=0

and the coefficients are:

a0 = f0

a1 = f1 - f0
n3

a2 = f3 - f0
v

a3 = f4 - f0
n2

In the tetrahedron node notation the coefficients are:

n0

a0 = f0
a1 = f1 - f0
a2 = f2 - f0

n1

a3 = f3 - f0

and the mapping is A = a0 + a1 u + a2 v + a3 u v

1.3.1 - 22

For a pyramid the complete base of the hexahedron is included and in this way the hexahedron node indexing is
equal to pyramid ones. The last three shape functions h5, h6 and h7 are zero, see tetrahedron derivation, and the
w
existing coefficients are:
n4

a0 = f0
a1 = f1 - f0

v
n3

a2 = f3 - f0
n0

a3 = f4 - f0
n2

a4 = f0 - f1 + f2 - f3 = f2 - f3 - a1
n1

forming the mapping:

A = a0 + u (a1 + v a4) + v a2 + w a3

1.3.1 - 23

The last 3D cell is a pentahedron, called also the prism. It is a half of the hexahedron as it does not include
hexahedron nodes 2 and 6. Thus, the coefficients h4 and h6 are zero; see tetrahedron derivation and the
coefficients are:
n5
w
a0 = f0

n3

a1 = f1 - f0

a2 = f2 - f0

n4

a3 = f3 - f0

n2

n0

a4 = - a1 - f3 + f4
n1

a5 = - a2 - f3 + f5

79

A = a0 + u (a1 + a4 w) + v (a2 + a5 w) + a3 w

The mapping is

1.3.1 - 24

The number of multiplication operations is reduced with adequate grouping of mapping coefficient, as shown in
the Table 22.
Operation

Cell

Simple

Grouped

T2N3

T2N4

T3N4

T3N5

T3N6

T3N8

12

Table 22: Reduction of multiplication operations

Jacobian matrix
The modeling space E is discretized with an arbitrary set of cells and it is global to all of them. The parametric
space U is local to every cell. The isoparametric mapping A, applied to a point P, transforms its parametric
coordinates u to modeling coordinates x.
The cell coordinates are used as mapping support, equation 1.3.1 - 16, and the mapping A becomes the
coordinates transformation x(u), see Figure 76.
x xi ( u, v, w )

i = 1, 2, 3

1.3.1 - 25
Such mapping hasnt much value if the inverse mapping A , denoted u(x) doesnt exist. The mapping u(x)
allows backward coordinates transformation to the parametric space, as shown in Figure 76.
-1

u ui ( x, y, z )

i =1, 2, 3
1.3.1 - 26

Modeling space E3

Parametric space U3
w

z
1
1

u(x)

P(u,v,w)

ew
0

ev

ez
v

eu

ex

y
eu

x(u)

ey

ew

ev
P(x,y,z)

Figure 76: Coordinates transformation

80

and can be found if x is single valued and continuously differentiable in the neighborhood of a point P. Thus,
provided that Jacobian matrix

J=

x
u
1.3.1 - 27

has the Jacobian J which exists and doesnt vanish

x
u
x
y
J J =
=
u
u
z
u

x
v
y
v
z
v

x
w
y
w
z
w
1.3.1 - 28

This implies the existence of the inverse J -1. J is calculated from A, see equation 1.3.1 - 16, as
J=

x A (u )
=
u
u
1.3.1 - 29

For the cell origin J is defined with cell base vectors [eu, ev, ew], see Figure 76., as
eu
x
J = [eu ,ev ,e w ] = eu y
e
uz

evx
ev y
evz

ewx

ewy
ewz

In E3 the triad [eu, ev, ew] serves as a basis for U3 provided that they are not coplanar.
eu(eu ew)0
The three unit vectors have each, only one non vanishing component in U3
eu e(1)= (1,0,0)
ev e(2)= (0,1,0)
ew e(3)= (0,0,1)
The suffixes to the e are enclosed in parenthesis to show they do not denote components. The j-th component of
e(i) is denoted by e(i)j and satisfies the following relation
e(i)j=ij I
In U3 any vector a can be expressed in the form
a= ai e(i)
and the summation convention is also applied to suffixes enclosed in parentheses.
Isoparametric mapping for quantities is important, as it allows combined treatment of quantities interpolation and
mapping between parametric and modeling coordinates spaces. When the modeling quantities have to be
manipulated and therefore be defined in both coordinate spaces, the knowledge of J is required to define the
quantity components in both spaces, while the quantity itself is invariant. These characteristics are used for
vector line algorithm; see section 1.4.5, when the integration of the particle path is performed through the
parametric vector field. Another application of the Jacobian occurs in the calculation of derived quantities, for

81

example vorticity, is another application of the Jacobian matrix. Follows the derivation of Jacobian matrices for
predefined cell types:

Jacobian 1D matrix:

J 1D =

A A

u u
1.3.1 - 30

1D segment A = a0 + a1 u

J T1N2 =
= a1
u

1.3.1 - 31
Jacobian 2D matrix:

J 2D =

A A A

u u v
1.3.1 - 32

2D triangle
A = a0 + a1 u + a2 v

A
= a1 ,
= a2
J T2N3 =
v
u

1.3.1 - 33
2D quadrilateral
A = a0 + a1 u + a2 v + a3 u v

A
J T2N4 =
= a1 + a3 v,
= a2 + a3 u
v
u

1.3.1 - 34
Jacobian 3D matrix:

J 3D =

A A A A

,
,

u u v w
1.3.1 - 35

3D hexahedral
A = a0 + u [a1 + a5 w + v ( a4 +a7 w)] + v ( a2 + a6 w ) + a3 w

A A A
JT3N8 =
,
,

v
w

A
= a1 + a5 w + v (a4 +a7 w)
u
A
=a2 + u ( a4 + a7 w) + a6 w
v
82

A
= a3 + a6 v + u (a5 + v a7)
w
1.3.1 - 36
3D tetrahedron
A = a0 + a1 u + a2 v + a3 w

A
A
J T3N4 =
= a1 ,
= a2 ,
= a3
v
w
u

1.3.1 - 37
3D pyramid
A = a0 + (a1 + a4v) u + a2v + a3w

A
A
J T3N5 =
= a1 + a4 v,
= a2 + a 4 u ,
= a3
v
w
u

1.3.1 - 38
3D prism
A = a0 + u (a1 + a4w) + v (a2 + a5w) + a3 w

A
A
J T3N6 =
= a1 + a4 w,
= a2 + a5 w,
= a3 + a4 u + a5 v
v
w
u

1.3.1 - 39

Metric tensor
The metric tensor is one of the basic objects in differential geometry [77, 80] and it is related to geometrical
properties such as length, area or volume, respectively in 1D, 2D or 3D coordinate space. It is the ratio between
the two coordinate systems for which the isoparametric mapping A and the Jacobian matrix J are defined. The
metric tensor G can be written as
G = JT J
1.3.1 - 40
This relation is important in the case when parametric and modeling spaces differ in dimension. For example,
when a quadrilateral cell is placed in the 3D modeling space the Jacobian matrix is not of the square type:

J=

x
u
y
u
z
u

x
v
y
v
z
v

1.3.1 - 41

This is a significant inconvenience for the calculation of the inverse Jacobian matrix J-1, as in general J is a
rectangular matrix. Let us consider its Moore-Penrose generalized inverse J, see [81] for equalities on
generalized inverses. The metric tensor G = JT J is square and regular, thus its generalized inverse is equal to its
ordinary inverse, so that

83

G -1 = (JT J)
G -1 = J (J)T
G -1J T= J (J)T J T
G -1J T = J (J J)T
G -1J T = J J J
G -1J T = J
1.3.1 - 42
The inverse of the metric tensor G -1 can be explicitly calculated being by definition the square type matrix:
G = {gij}

1.3.1 - 43

where,
gij =

xk xk
,1 i, j k
k =1
i uj
n

or in the vector notation: xu 1 xu

xu 2 xv

gij = xu i xu j
1.3.1 - 44
The definition in vector notation shows that gij is the dot product of the tangent vector of i-th coordinate axis
with the tangent vector of the j-th coordinate axis analyzed from the modeling coordinates space.

n3
n

n
n2
v

3
n0

xv
0
n1

xu
1

y
x

Figure 77: Surface normal and quadrilateral cell

The vector (cross) product of these two vectors xu, xv defines the normal n of the tangential plane at the cell node
n0.
n=

x u xv
x u xv

1.3.1 - 45

84

The four nodes of the quadrilateral are defining the surface, see Figure 77. The surface point for which the
normal exists is the regular surface point and has to satisfy the following condition:
x ux v 0
1.3.1 - 46
If the above condition is satisfied, the vectors xu and xv are not collinear and they define the tangential plane. If
the condition is not satisfied the surface point is singular and the Jacobian of such coordinates transformation is
zero. Following the equation 1.3.1-41 the transpose Jacobian matrix is:

x
u
T
J =
x
v

y
u
y
v

z
u

z
v
1.3.1 - 47

and applying it to the equation 1.3.1-40 , we obtain the following expanded form:

x 2 y 2 z 2 x x y y z z
+
+

+
+

u
u
u
u
v
u
v
u
v

G = J TJ =
2
2
2
x x y y z z x y z
u v + u v + u v v + v + v

1.3.1 - 48
From which the G
equation 1.3.1-42.

-1-

can be calculated, and thus we have defined all the elements for calculating Jin the

Some additional characteristics of the Jacobian matrix and Metric tensor determinants, defined as
G =G
1.3.1 - 49
and, if the Jacobian matrix is of the square type, the metric G is defined explicitly with the Jacobian J as
J =J,

G = J2

1.3.1 - 50
For example, the calculation of cell length L, area A and volume V is possible by knowing the Jacobian of
isoparametric mapping:
volume
J=

dV
dV0

area
J=

length

dA
dA0

J=

dL
.
dL0

1.3.2 Derived Quantity

The derived quantities are the ones, which we are able to compute from the available data provided by an Input
Model. As the inputted data are usually coordinates, scalar and vectors, by simple algebraic operations we can
add, subtract, multiply and divide the existing quantities, defined as Unary Operations, for which a constants can
be introduced, for example when scaling the coordinates. We can also consider Binary Field Operations when
we use multiple quantities in an algebraic expression. The possibility to derive quantities from a vector field in
terms of components and magnitude is also considered, and it is described in section 2.8, where a Symbolic
Calculator is presented. In this section we will describe the isoparametric numerical model of the cell to be

85

applied when Gradient, Divergence and Curl operators are computed for the related scalar or vector quantity
field.
The scalar function S of position s when gradient operator grad ( ) is applied to it produces a vector s.
v s grad

s=

s
s
s
ex + e y + ez
x
y
z
1.3.2 - 1

There is also a need to calculate partial derivates of the quantity, as divergence:

s .v div v =

vx v y vz
+
+
x y z
1.3.2 - 2

The important physical meaning for the divergence of the velocity field v is that it represents the relative rate of
the space dilatation when computed along the particle trace. Consider the cell around the point P. By the fluid
motion this cell is moved and distorted. As the fluid motion cannot break up by the continuity law its volume dV0
=J dV and hence:
J=

dV (t0 )
dV (t )
1.3.2 - 3

defines the ratio of the cell volume at beginning and time t, called the dilatation or expansion [77].
Suppose a velocity vector field v(x) defined in a 3D Euclidian (modeling) space. The vorticity field (x) is the
circulation of the velocity field v around a cell area taken perpendicularly to the direction of ., and is
obtained by computing at each point the curl of the velocity field v.

= curl v

vz v y

y
z
v v
y = x z
z
x
v y vx
z =

x y
x =

1.3.2 - 4
Analytical Model

If v is given as function in the parametric coordinate system u(u,v,w), with a well know link to modeling space
x(x,y,z) trough isoparametric mapping, see section 1.3.1 on page 74. The components of the velocity vector v can
be written as:
v( u ) = [v x ( u , v , w ), v y ( u , v , w ), v z ( u , v , w )]

1.3.2 - 5
and by the law of partial derivation, applied to Jacobian matrix, see section see section 1.3.1 on page 80:

d -1
d u
J = i
dt
dt x j

( )

d ui vi
=
=
,
x dt x
j
j

i,j = 1,2,3
1.3.2 - 6

86

and defining the partial derivative of the vector component along the modeling coordinate axis:

vi vi u1 vi u2 vi u3 vi uk
=
+
+
=
x j u1 x j u2 x j u3 x j uk x j
1.3.2 - 7
where u(u1, u2, u3 )
u(u, v, w)
and for the specific case

vy
z

vy vy u vy v vy w
=
+
+
z u z v z w z
1.3.2 - 8
or putting it into the form of matrix multiplication

v v -1
=
J
x u
1.3.2 - 9
and expand the notation

vx
x

vy
x
v
z
x
The components of the

v
x

vx
y
vy
y
vz
y

vx vx
z u

vy vy
=
z u
vz vz

z u

vx
v
vy
v
vz
v

vx u u u
w x y z

vy v v v
w x y z
vz w w w

w x y z

matrix are applied in the equations for gradient divergence and vorticity. For the

equation 1.3.2 - 8 the remaining part

v is calculated locally for each cell node when the inverse of Jacobian J
u

exists.

Numerical Model
The velocity field is given at cell nodes and for each node

1D segment

v v

components are calculated for all cell types.

u u

v
)1,0 = v1 v0
u

u
n0

2D triangle

n1

v i
= ( v i )1 ( v i )0
v
u
( i )k = ( vi )1 ( vi )0
v i
u j
= ( v i )2 ( v i )0
v

87

2D triangle
w

vu
u
vv
u
vu
v
vv
v

n2

= ( vu )1 ( vu )0
j

= ( vv )1 ( vv )0

= ( vu )2 ( vu )0
= ( vv )2 ( vv )0

vi
u

v i
v

1-0

2-0

1-0

2-0

1-0

2-0

n1

n0

2D quadrilateral
v

vi

= ( vi )1 ( vi )0
u 1,0

n2

n3

j
k

vi

= ( vi )2 ( vi )3
u 2 ,3
vi

= ( vi )3 ( vi )0
v 3 ,0
u

n1

n0

vi

= ( vi )2 ( vi )1
v 2 ,1

1-0

3-0

1-0

2-1

2-3

2-1

2-3

3-0

j
n3

0
1
2
3

v
n2
n0

n1
u

88

vi
v

3D tetrahedron

vi
u

vi
u

vi
v

v i
w

1-0
1-0
1-0
1-0

2-0
2-0
2-0
2-0

3-0
3-0
3-0
3-0

3D pyramid
w

n4

j
k

n3

0
1
2
3
4

n0

n2
n1

vi
u

vi
v

v i
w

1-0
1-0
2-3
2-3
1-0

3-0
2-1
2-1
3-0
3-0

4-0
4-0
4-0
4-0
4-0

vi
u

vi
v

v i
w

1-0
1-0
1-0
4-3
4-3
4-3

2-0
2-0
2-0
5-3
5-3
5-3

3-0
4-1
5-2
3-0
5-2
5-2

vi
u

vi
v

v i
w

1-0
1-0
2-3
2-3
5-4
5-4
6-7
6-7

3-0
2-1
2-1
3-0
7-4
6-5
6-5
7-4

4-0
5-1
6-2
7-3
4-0
5-1
6-2
7-3

3D pentahedron

j
k
0
1
2
3
4
5

3D hexahedron
w

n7

n4

k
n6
v

n5
n3
n0

n2
n1

0
1
2
3
4
5
6
7

The linear derivatives for the considered cell types are introduce in the node connectivity algorithm for the
calculation of the uniquely defined quantity at each cell node.

89

1.4 Extraction Algorithms

The numerical data model with geometry and quantity definitions requires effective extraction algorithms for
filtering and manipulating large data sets. Extracted data sets, to be visualized, are modeled as representations of
surface, curve or point type. The numerical probes are represented by points, lines or planes; they are used to
intersect grid geometry and quantity fields. In practice, the most general 4D data set is expressed as:
q = q(x, y, z, t)
and its form in nD space is extended and given by:
q = q(x1, x2... xn )
The quantity q can be associated with a scalar, vector or tensor field. Visualization techniques in order to make
extracted data set displayable require geometrical representations in 2D or 3D space. The mapping consists in
converting the data set into a geometrical entity, like a surface, and to associate color to quantity value in order
to provide added-value to the visualization. The possibilities of extraction algorithms are multiple since they
can create:
surfaces: cutting planes or iso-surfaces,
curves: iso-lines, sections and particle traces,
points: locations of local values.
In extraction algorithms, a threshold, which is the difference between the requested and the node value, is
applied to each mesh node. The cells, where at least one of the nodes thresholds has a different sign than the
others cells nodes are identified as intersected cells, to which the cell intersection pattern is going to be
applied. The cell connectivity model is essential for the extraction algorithms, because it allows the application
of the marching cell algorithm for cutting plane and iso-surface algorithms. The same observation can be made
when comparing the section and the iso-lines algorithms. The only difference is the lowering of the parametric
dimension of the zone from which the extraction is made.
The calculated threshold can be the difference between the given values in respect to geometry or quantity values
at the cell nodes. The threshold property defines:
geometry for cutting plane, section and local value
quantity for isosurface, isolines and particle traces.
This chapter makes difference between geometry/quantity extraction algorithms. It defines two types of
marching cell algorithms based on the above mentioned threshold property. In addition follows the section on
threshold regions where the sub zones are extracted from zones. The point location and particle trace algorithm
for volume and surface are closing this chapter. They are treated separately because they require the point cell
relationship with the objective to state that the point is in or out of the cell.
The following sections describes the marching cell algorithm which represents a common part of the cutting
plane and isosurface algorithms, which are the most important interactive visualization tool to investigate
volumetric data.

1.4.1 Marching Cell Algorithm

The marching cell algorithm is the core part of all extraction algorithms and represents the generalization of the
marching cube algorithm [82]; it performs the following five tasks:
1. Identification of seed cells, which are initial cells from which the cell traversal starts,
2. Identification of intersected cells,
90

3. Marching through cells applying connectivity,

4. Application of intersection pattern with creation of new cells,
5. Creation of the appropriate zone from the collection of created cells.
The identification of intersected cells is based on the point-to-plane distance for the cutting plane and section,
where as for the iso-surface and the iso-line it is the difference between the node value and the value provided by
the user input. The cutting plane threshold explanation follows in more details as the cutting plane input has
more parameters than iso-surface one, due to the interactive translation and rotation functionality. Each
interactive re-positioning of a cutting plane requires the recalculation of geometry and quantity data. To identify
an intersected cell, its nodes are labeled according to the distance sign of the nodes with respect to the cutting
plane. The equation of the cutting-plane S is:
S(x) = ( x - x0 ) n = 0,
where x0(x0,y0,z0) is an arbitrary point of the plane and n(nx,ny,nz) is the unit-vector normal to the plane. A point x
for which S(x)>0 is conventionally at a positive distance from the plane. The plane equation is written in
(right-handed) Cartesian coordinates in the form:
Ax+By+Cz+D=0

Ax+D=0

or

1.4.1 - 1
where the vector A (A,B,C) is along the normal to the plane. The distance d0 between the point x0(x0,y0,z0) and
the plane:
Ax + By0 + Cz0 + D Ar0 + D
d0 = 0
=
A
A2 + B 2 + C 2
d0 = n x0 - p

where n = A and
A

p=

D
A

1.4.1 - 2
where the sign of the square root is chosen to be opposite to that of D. The definition of the distance sign is:
sign (d0) = sign (A x0 + D)

1.4.1 - 3

for all cell nodes. sign(d0) defines on which side of the plane the node r0 is located.
Each node position vector is projected on the normal n of the cutting plane in order to determine its sign (d0)
(positive or negative). The set of the cell nodes sign (d0) describes the cell intersection pattern, as shown in the
lookup table where all the possible intersection patterns are labeled and defined, see Section 1.2.6.
0 sign false

Node

0 < sign true

Edge

Face

Figure 78: Possible singular cases of intersections with a node, an edge and a face
91

SP=0

LC=P

LL=0

S
SP=C

L
S

S
L
L
L

P
SC

VS

Figure 79: Seed cells from boundary

Cells which have nodes with positive and negative distance signs are intersected by the cutting-plane (cells
where distance signs are all positive or all negative are not intersected).
The cutting-plane algorithm is improved by traversing the domain boundary, in order to avoid scanning all the
domain cells, as shown in Figure 79. Every domain has one or more boundaries, which in 3D space consists of
one or more surface boundaries. The finding of seed cells is specific to the extraction algorithms. The zone
representation is based on B-rep model where each zone has its own boundary. From the seed cells the marching
cell algorithm starts. We apply the extraction algorithm to the boundary, which is obviously faster than
traversing the whole zone. We identify the intersected boundary cells. This is done in 2D, where surfaces are
bounded by curves or in 3D where volumes are bounded by closed surfaces (see Figure 79). If we intersect a 3D
surface, it is not sufficient to intersect its boundaries, as it can happen that the plane intersects the surface but not
its curve boundary. However, the existence of the intersection with a surface boundary implies the existence of
surface/plane intersection. It is obvious that a curve/line intersection in 2D (or a plane/surface intersection in 3D)
detects all the points (or curve cells) necessary to complete the marching-cell algorithm.
A difference between the cutting-plane/section and the iso-surface/iso-line algorithms is that the former includes
the generation of seeds from its boundaries. For the volume boundaries, the section algorithm is applied to find
the seed cells. A main advantage is the improved performance of the cutting- plane algorithm compared to the
iso-surface one, since the number of cells and nodes to be analyzed is one order of magnitude smaller. Surface
cells have 3 or 4 nodes (triangle or quadrilateral), where volume cells have 4 or 8 nodes (tetrahedral or
hexahedral). The number of cells that are visited on a 3D structured grid scales like O(N3) for the iso-surface
algorithm and like O(N2) for the cutting-plane algorithm (where N is the number of traversed cells).
The surface extraction procedure can be performed by the cutting-plane and the iso-surface algorithms. Both
algorithms produce unstructured surfaces, i.e. surfaces composed of the cells constructed by the cellintersection pattern. To reduce the time it takes to identify the intersected cells, a technique called the marchingcell algorithm is used. The marching cell algorithm exploits the fact that a plane, intersecting a mesh (for
example a structured mesh of cubic cells) cuts only a limited number of cubes (Figure 80). Inside the domain, it
is possible to start from any intersected cell and to march to the next intersected cell until all intersected cells
are visited. Connectivity is exploited to avoid scanning all the cells of the mesh. The marching-cell algorithm
improves the sequential algorithm, which is not the case for the parallel one [43].
The cutting-plane algorithm comprises the following steps:

find all intersected cell on the volume boundary and create seeds set

while(seeds not empty)

create node mask

92

for each intersected cell:

compute intersection polygon applying cell intersection pattern.

add connected cells to the seeds if not yet intersected

The algorithm selects the appropriate cell type based on the parametric cell dimension and the number of cell
nodes. The node mask defines the cell intersection pattern and the intersection can be calculated. To keep track
in the marching-cell algorithm the set of intersected cells seeds must be removed from the seed set of cells to
avoid possible duplication in processing of intersected cells. For each intersected cell, the edges intersections
with the cutting plane are computed. The nodes on the intersected edges define polygons, see Section 1.2.6,
which all together define the cutting plane. In principle, these polygons are used for defining the graphical output
primitive, see Section 3.7.2. The intersection of edge and plane is calculated in parametric form using a linear
interpolation between two edge nodes; the resulting ratio is applied to compute the quantity value at the point of
intersection. When the surface is kept for further processing, the created parametric field (nodes and the quantity
parameter) is saved in order to speed up the recalculation of other available quantities.
The orientation of all created cells must be the consistent, because of the graphics rendering techniques, which
require the normals definition as input to the shading and contouring algorithms. The orientation is implicitly
given in the lookup table, where the order of triangle and quadrilateral nodes defines the local coordinate system
and thus the normal to the generated cells (see Section 1.2.6). The same approach is used for pyramids, prisms
and hexahedrons. Situations with intersected patterns with several internal pieces may be more complex and
result in larger lookup tables, but processing remains extremely rapid with no need for conditional branching
code.

(u,v,w+1)
i, j, k+1

(u,v+1,w)
i, j+1, k

(u,v,w)
i, j, k

(u+1,v,w)
i+1, j, k

Normal mode

Plan

Save mode

Figure 80: Hexahedron cell intersected with the plane

This algorithm has 2 modes of operations: display and save. It is not always necessary to define the topological
relationships and save the surface topology. In the display mode, the sole issue is to display the colored contour
map or vector field. If other representations are needed -- like distribution of a scalar quantity in a x-y plot -- the
cutting plane must be saved, which implies the creation of an unstructured surface (made up of triangles and

93

quadrilaterals). The display mode manipulates polygons only so as to speed up interactive processing. The
saving process calculates the needed cell-connectivity information, surface boundary and quantity parameters
taking into account the original 3D domain from which the surface is extracted. The parametric field allows
recalculating the surface field for an arbitrarily chosen quantity, thus reducing the processing and memory
requirements to perform such operation. The save mode algorithm is decomposed in the following steps:

find all intersected cells on the boundary and create seeds set
while(seeds not empty)

initialize set of cells to be intersected xcells and add one seed cell to it

initialize collection of intersected cells

while(xcells not empty)

create node mask

for each intersected cell:

compute intersection polygon applying cell intersection pattern.

partition the polygon and compute connectivity

add polygon partition to the collection of intersected cells

add connected cells to the xcells if not yet intersected and remove from seeds if exist

create topology from the collection of intersected cells

After calculating edge intersections and polygon partitions, cell connectivity is examined. At the time one
computes the intersections in a given cell; its connectivity is not necessarily locally definable since it can be that
the adjacent cells are not yet constructed. Overall connectivity is preserved through face indexing, an indexing
convention that depicts (internal and external) cell connectivity. Face indexing consists of indexing the virtual
faces between polygonal cells with integers which are bigger than the number of faces of the cell. Each
additional cell in the polygon is given an incremental index which is bigger than the number of cells in the zone.
The resulting collection of all intersected cells must be further processed by topology algorithms that establish
the topological description of the cutting plane as an unstructured surface. The connectivity of the intersected
cells is defined applying the global SupZone indexing, to which internal indexing is added. The collection of
cells is sorted using the global indexing. This characteristic allows a fast transformation where for each cell with
greater index than the current one the cell connectivity is calculated with preserving local indexing.
The node-reduction algorithm takes into account the relation between node and cell, where each node can build
more than one cell (1.2.4 Node Topology). The alternative is to eliminate the multiple intersections of the same
edge. The detection of an intersected edge and the addition of new one in an ordered set imposes local node
indexing (see Section 1.2.3 Cell Connectivity). The intersected supZone edges are stored in the set based on
edge-nodes key. If an edge is intersected, the result of intersection is reused and the unique node indexing is
preserved. For a convex domain, there is one single connected set of cells, defining the arbitrary cutting-plane
surface. If the domain is not convex, multiple cutting-plane regions may exist, depending of the position and
orientation of the cutting plane (see Figure 81).

Figure 81: Cutting plane example of multiple connected topology

94

Figure 82: Disconnected components of an isosurface.

The iso-surface algorithm computes the unstructured surface for a given constant scalar value. An iso-surface is
not necessarily a plane and can have more general shape. The iso-surface algorithm differs from the cuttingplane in two respects: first, all the cells are checked; second, the difference between the scalar and threshold
values replaces the distance-sign value. The algorithm is decomposed in following steps:
for each node find threshold
for each cell
create node_mask
for each intersected cell:
compute intersection polygon applying cell intersection pattern.

The save mode algorithm is decomposed in the following steps:

for each node find threshold
for each cell
find intersected cell from not inspected cells
initialize xcells collection with intersected cell
while(xcells not empty)
create node mask
for each intersected cell:
compute intersection polygon applying cell intersection pattern.
partition the polygon and compute connectivity
add polygon partition to the collection of intersected cells
add connected cells to the xcells if not yet intersected and remove from
seeds if exist
create simply connected region from the collection of intersected cells

The 3D surface generated by this algorithm is unstructured; it is constructed in exactly the same manner as in
the case of the cutting-plane. The marching-cell algorithm cannot be readily applied if the iso-surface has
disconnected regions: as shown in Figure 82. The disconnected components may be open- or closed-type
surfaces; indeed we can group any number of disconnected iso-surface parts into a set, which the user still refers
to as a single surface. An iso-surface which has been saved has the functionality to recreate any other quantities
at its nodes; it is classified as a parametric field. The nodes of the original domain determine the intersection
edges, together with the linear interpolation coefficient.
Distinguishing between structured and unstructured meshes is important when it comes to considering memory
requirements. Unstructured meshes require more storage than structured ones, since one needs cell-connectivity
to be defined explicitly. There is no significant difference between structured and unstructured meshes from the
algorithmic point of view since intersection calculations are based on the cells lookup tables in both cases. The
output of the cutting-plane algorithm is necessarily an unstructured surface, whether the base mesh is structured
or not. An algorithmic difference exists, however, in how connectivity information is accessed. For structured

95

meshes, cell-connectivity is explicit to and deduced from the cell ordering sequence; this is in contrast to the case
of unstructured meshes where cell-connectivity is provided as input data.

1.4.2 Threshold Regions

The threshold algorithm is an efficient extraction technique used in visualization; it can be applied in the
parametric (indexing), modeling or quantity spaces. The idea is to create a SubZone (SubField) from the original
Zone (Field).
The threshold in the index space is commonly used for specifying surface parts on the structured surfaces, when
we define boundary conditions, for example a solid boundary. In that case, the imposed index threshold is the
span between the two I and J indices. Since the indices are integer numbers, the resulting extraction is the
structured 2D surface. A similar logic is applied for extracting edges and nodes. The difficult part is to re-create
the subsurface because the nodes and the edges can determine distinct surface topologies. Such index based
threshold can be applied to volume, surface or curve data sets (just not on point sets since one cannot construct
SubPoint as SubZone). SubSurfaces and SubCurves are examples of SubZones for (un)structured grids (see
Section 1.2.2).
In the geometrical space, one handles ranges of coordinates. A thresholded region can include part of a cell. The
new SubCells must be completely defined as polyline, polygon or polyhedron; the goal is to be able to handle
these poly-entities with the available graphic primitives.
The Quantity Threshold representation is a frequently applied representation which extracts SubSurfaces
between 2 limiting iso-lines. It permits considerable reduction of the generated graphics (compared to the
complete surface color-mapping) and enables the analysis of multiple surfaces. The threshold algorithm removes
the possible drawback of having a surface masking another one. The threshold algorithm for both scalar and
vectors quantities requires:
the variable threshold type,
identification of cells containing the threshold,
construction of the SubZone from extracted cells

S < Smin
Smin < S < Smax
Smax < S

The polygon filed by is

plotted

Figure 83: Ten different thresholds in a triangle

96

S < Smin

Smin < S < Smax

S < Smax

Figure 84: Fifteen different thresholds of quadrilateral

The generation of the SubCells is determined by the cell threshold pattern. The cell threshold pattern depends
on the node threshold label -- which can be in any one of the 3 states: included, excluded min or excluded
max. For triangular and quadrilateral cells, the threshold analysis proceeds in much the same way as the cell
intersection pattern defined with the binary node mask; here the processing is n-ary instead of binary, and
following equation applies:
label index = so 30 + s1 31 + ... + si 3i+ ... + sN-1 3N-1
The number of possible threshold patterns for a cell with n nodes is given by:
number of threshold patterns = 3n
so that for a triangle there are 27 threshold possibilities, and 81 for a quadrilateral cell; some of them are shown
in Figure 83 and Figure 84. By imposing that the internal area of a threshold pattern is always a single connected
region, one obtains patterns like those illustrated in Figure 85.
accepted

Figure 85: Ambiguity cases for quadrilateral thresholds

97

1.4.3 Curve Extraction: section and isoline algorithm

The logic of the cutting-plane and the iso-surface algorithms is readily applied to the section algorithm which
simply lowers the parametric dimension of an intersected zone from volume to surface. An optimization is
possible for 2D projects since a 2D domain is always an open surface and its boundaries are always intersected.
The connectivity of the extracted zone is solved by reusing the connectivity of the involved zone cells.
A unified approach for the cutting-plane and the section algorithms has been constructed [75]. The section
algorithm determines the surface/plane intersection and provides the quantity distribution in the created section;
the section algorithm is the mere 2D version of the cutting-plane algorithm.

Isolines
The isoline representation is the primary technique for displaying scalar field information. The isoline algorithm
calculates the curves that connect the surface points where the value of the scalar quantity is equal to a given
input value; in effect, it is the 2D version of the iso-surface algorithm.
The scalar field, known by its values at the nodes of the cell, is filtered against the calculated intersection mask
so as to detect a sign change; if this occurs, the cell is known to contain the isoline. The intersection is calculated
by applying a linear interpolation between the nodes of the edge:

qiso q1 x iso x1
=
q 2 q1
x 2 x1
When a cell is investigated, all its nodes are tested against the criterion qn - qiso > < 0, and the calculated sign is
applied in the definition of the cell intersection mask. The intersection mask is used to find the intersection
pattern in a cell lookup table. For surface cells, 2 cell tables are defined, the first one for triangles and the other
one for quadrilaterals. There can be 0, 2 or 4 intersections for a quadrilateral. The ambiguity case is when 4
intersections are detected; in such case, the 4 possibilities are drawn (see Figure 85).
There is no ambiguity case for the triangle. The marching algorithm generates open- and closed-curve types.
Topologically, open-curves are always connected to the surface boundary, whilst closed-curves can always be
transformed to a circle. A linear interpolation between edge node values is also used for defining the isoline
colors (in relation to the color map).
There are several ways of applying the isoline algorithm. The straight forward method consists of traversing all
surface cells for which the resulting isoline is created when connecting the intersected cells. When displayed, the
resulting line segments are perceived as a map of open- and closed-type isolines. In this approach, the isoline
connectivity table is not created which simplifies the algorithm but increases the calculation overhead, since
every isoline point is shared by two adjacent cells and needs to be calculated twice. Hence the isolines are drawn
with duplicated set of points, and this increases the display time, which can be problematic when interactive
viewing actions like rotation and zooming are performed.

4
1

(a) unstructured
5 times duplicated test

(b) structured
4 times duplicated test

Figure 86: Node traversal in the isoline algorithm for unstructured and structured surfaces
98

Figure 87: Close and Open curves for Isolines Representation

The threshold sign is checked at each cell node. The node calculations are performed repeatedly since one
node can be included in several cells (see Figure 86). On a structured surface, a node internal in the surface is
shared by four adjacent cells; on an unstructured surface, a node may be shared by more than 4 cells.
There are two types of isolines that are surface curves (see Figure 87):
1. open isolines whose starting and ending points are on the surface boundary,
2. closed isolines that are parts of the surface interior.
The isoline algorithm involves edge detection:

for each cell

for each edge

for each node

To generate curves for which there are no duplicated or disconnected points, the following algorithm is
applied:
1. generation of a collection which contains at least one point on each closed isoline;
2. marching trough the surface grid starting from the boundary (case of open isolines);
3. marching trough the surface grid starting from the points in the list of remained isoline points,
applied to define closed isolines.
1. Calculation of the threshold sign for all the nodes q - qiso + or -;
2. Identification of boundary and internal cells;
3. Traversal of internal cell-edges,

( # cell # edge # n 1 # n 2);

4. Forming of the boundary cell-edge,

( # cell # edge # n 1 # n 2).

Starting from the intersected boundary cells, one follows the isoline by the existing cell-edge combination
in the internal #cell-edge list, from which the combination is removed, when the isoline point becomes part of
the isoline curve. The marching algorithm continues until the boundary-cell edge is reached. Each isoline
point is added to the isoline curve. The next point defining the isoline is found from the cell intersection
pattern following the cells connectivity. The ambiguity cases (if any) are handled without special attention.
The ambiguity is locally checked if such cell pattern is met. When all the located boundary cells are removed
from the list, all open isolines of the traversed surface are found.
The remaining internal cell edges are used in the same way as the boundary ones that have been intersected
during the first cell-edge test. If that combination is found the isoline is closed. When the collection of
internal cell-edges is empty, it means that all the closed isolines have been found.

99

auxiliary area

local value point

view plane normal and
view plane point

n view
plane

surface/line line
intersection point
C

na
L

arbitrary point diferent

from local value point

nsnp

minimum
projected
area

ns

np

n s x nv =
na

ns

P
D

D
A

section plane point

section plane

Figure 88: Line-Surface intersection concepts

1.4.4 Point Extraction: local value algorithm

Point extraction is computed by certain visualization tools, which require the identification of a surface cell
containing the point indicated by the user. These visualization tools are: local value, local vector, local isoline,
two points of the section, starting point of the particle trace and picking of the surface.
The point extraction algorithm solves the well-known geometrical problem called the point location problem,
[83, 84] The algorithm applies transparently to 2D- and 3D-surfaces. The 2D problem can be simplified to the
problem of 2 lines intersecting; the extension to 3D requires calculating the line at the intersection of two planes
(see Figure 88). In all cases, the surface is intersected with the objective to find the intersected surface cell in
which the point is located and the points coordinates on the surface itself. For an open surface, the algorithm can
be optimized to avoid the full traversal of all the surface cells by finding the boundary cells which are
intersected. This is the same idea as the one used for the section algorithm (see Section 1.4.3). The boundary
cells that are found are used as the starting points for the marching-cell algorithm which then proceeds towards
the interior of the surface. Since a surface consists of connected cells, cell connectivity can be used to identify
the adjacent cells and the intersected edge uniquely identified cell to stat the marching cell algorithm. The
intersected cells are then ones more intersected with a second, auxiliary plane (see Figure 89). If there is no
intersection with the auxiliary plane, the inputted point is not located in the surface cell and the algorithm
continues to search in the direction prescribed by the section plane. This algorithm is an extension of the
plane/surface intersection algorithm, with the added point inclusion test which is performed for each cell
intersecting with the auxiliary plane.

Figure 89: Line Surface multiple intersections and possible traversals

100

The algorithm is decomposed in an outer part, which controls the traversal of the cells and an inner part
which deals with line/plane intersections and the point-inclusion test. The line/plane intersection algorithm is
applied for:
the boundary curve when curve cells are tested against possible section plane intersection
and during the surface cells traversal where the cell edges are intersected.
The intersection point at the cell boundary (set of edges) is identified through the set of constraints applied to the
point inclusion test. This part of the algorithm includes a combination of boundary constraints checks and
line/plane intersections with surrounding cell edges. If the point is found to be in the cell, it is added to the set of
intersections points, and traversal is pursued till all the seed cells are visited.
Inside the located cell, a Newton-Raphson method is applied to define more precisely the surface (see page 111).
The point-location algorithm is decomposed in:
line-plane intersections,
point inclusion test,
Newton Raphson method.

Line Plane Intersection

As mentioned line-plane intersection algorithm has several applications in the identification of region internal to
the specific discretized zone. The examples are the line-surface intersection, and clipping algorithms.
When the particle trace integration generates the point outside the cell, the intersection of the trace with the cell
boundary is found, and the integration continues in the adjacent cell. The last two points of the particle trace
represent the oriented line segments, called ray. The ending point of the ray exerts the point inclusion test, which
defines if the point is inside or outside of the cell. The side effect of the testing is the intersection point with the
cell boundary, where a ray exits the cell.
In the clipping algorithms the sections or surfaces are imposed by predefined limits. A parametric line clipping
algorithm, [6, 7, 85], is extended to different cell types in 2D and 3D parametric cell space, (intersection with a
cell edge or face). For 2D the algorithm is simplified with two 2D lines intersection between the cell edge and
the ray shown in Figure 90.

p1
p1
p0

p0
p1

p0

p1
p1

Figure 90: Cell ray intersection

The two points of the oriented segment p0 and p1 define the ray which intersects the cell boundary. The line or
the plane P is intersected with ray L, respectively in 2D and 3D cell space.
The ray L is defined parametrically as

101

L p= p0 + (p1 - p0 )
L p (t) = p0 + t dp,

where t = 0 at p0 and t = 1 at p1,

1.4.4 - 1

The plane is defined with normal n and the point rp lying inside the plane P. The plane equation in the pointnormal form is:

P (r - rp) n = 0
1.4.4 - 2
In Figure 91, the imaginary line/plane P divides the cell plane into the outside and the inside half plane. The
outside is indicated with positive direction of the normal n. If the point p of the ray L is introduced in the
equation 1.4.4 - 2 we define the half-space in which the line point is located:
P
L
inside half plane
px

p1(t=1)

dp
n

t>0

t<0

p0(t=0)

rp
outside half plane

n (rp - p) < 0

n (rp - p) > 0

Figure 91: Intersection parameter description

n(rp - p) > 0 PL < 90 or PL > 270
o

n(rp - p) < 0 PL > 90 or PL < 270

o

1.4.4 - 3
When the equation 1.4.4 - 1 is introduced in equation 1.4.4 - 2 the parameter t for the edge/face intersection point
px is:
t n(rp p0 ) ,
t>0
=
n( p1 p0 )
1
424
3
dp

1.4.4 - 4
The valid value of t can be computed if the denominator is non zero. The algorithm checks that:
n 0,

dp 0, thus (p1 p0)

n dp 0, thus the edge and the segment are not parallel

L
n
p1
dp
px
rp - px

rp
p0

Figure 92: Plane line intersection.

102

and the intersection point is at:

px = p0 +

n( r p - p0 )
n dp

dp

1.4.4 - 5

Point location algorithm

The point location algorithm defines if the ending point of the ray is inside or outside the cell. In addition, if the
point is outside the cell, it defines the intersection point of a ray with the cell boundary. The cell boundary is
edge for 2D zones and face for 3D zones. The boundary is applied for the identification of the neighboring cell,
see Figure 93. The intersection point can be transformed into the neighboring cell parametric coordinates,
without the need to apply the time consuming Newton-Raphson algorithm, as discussed in the next section. The
objective is to reduce the number of tests necessary to state if the point is in or out for the analyzed cell.
v
p1

p0
u

p1

Figure 93: Rays and intersections with extended cell boundary

The point inclusion algorithm is based on the ray which is checked against several constrains defined for each
cell type. The ray is defined with the starting point p0 which is always inside the cell and the ending point p1 for
which the point inclusion algorithm is performed. The boundaries are extended in order to apply the lineplane/line-line intersection algorithm. The parameter t, equation 1.4.4 - 4, is calculated for each boundary and the
solution is the intersection point with minimum t.
s
ns
e
ne
nf

Figure 94: Boundary normals for different parametric dimensions

The extended cell boundaries split the external cell neighborhood in several regions. In order to define the ray
exit point the extended boundaries define constrains which defines these regions. The number of regions R is:
R=2C,

where C is the number of constraints.

1.4.4 - 6

103

The ray ending point is checked against cells boundary which is further refined with the intersection of the ray
with the boundaries of the cells. As an edge is a part of the boundary of the 2D cell and the face is part of the
boundary of the 3D cell the plane line intersection algorithm, see page 101, is applied to find the point where the
ray leaves the cell, see Figure 95.

(a) line-plane intersection

(b) line-line intersection

Figure 95: Point location on cell boundaries

The algorithm sequence follows the right handed-orientation of the cell parametric coordinate system from left to
right.

u < 0 left cell

u > 1 right cell

1D segment:

left

right

A point can be located in only one of these regions. In case of a triangle cell T2N3, there are 6 regions, see Figure
96, with the following constrains:

Edge ei

e0

e1

e2

Constrain ci

c0 v<0

c1 1 - u -v <0

c2 u <0

Each constrain reduces a possible point location to a half plane. As shown in Figure 96, there are 8 possible
locations. The following proposition is a combination of constrains:
(c0 ^ c2 ) c1
1.4.4 - 7
and when applied, it reduces the number of possible point locations to 7. The complete number of locations is
shown in the truth Table 23, indicating true T or false F, when tested against boundary constrains of the
prescribed regions. The region x is not existing because the point would need to be located in R1 and R2 at the
same time, which is impossible because the two regions do not overlap, see Figure 96.
Regions

IN

c2

c0

c1

Constrain

Table 23: Triangle truth table

104

v
R5

R3,4,5

P5

n2
R1

R2

cell interior

cell interior

e1

e2
PIN
n0
0

P
n1

e0

1
P4
R4

R0

P3

R3

R5

R0,1,2

n2

R2

P2

R1

P1

cell interior

e1

e2

cell interior

PIN
0
n0

e0

u
n1

R3

R4
R0
P0

Figure 96: Possible positions of the point leaving the cell in the neighborhood of the triangle cell
Constrains in Table 23 can define a point location inside or outside of the investigated cell. This is not sufficient
for the evaluation of the ray exit edge/face and the calculation of intersection with exit edge/face. The orientation
of the ray helps to identify more rapidly the intersection by introduction of the ray intersection constrains as
follows:

105

Constrain ri

r0= u >1

r1= v <0

r2= v >1

IN

C
T r1(e2)

c2
c0

T e0

c1

F r2(e2)
T e1

T r0(e2)

F e2

T e1

F e0

T e1

F IN

Table 24: Triangle constrains path

From the Table 24 we can find out the number of tests and the number of intersections with boundaries necessary
to complete the identification of the ray exit and its intersection coordinates with the cell boundary. The most
complex computation is required when the exit point is located in R5 and R2. In that case, three tests and two
intersections must be performed.
For the quadrilateral cell T2N4, the following constrains are checked for the identification of an exit point
location in the cell external regions, see Figure 97:

Edge ei

e0

e1

e2

e3

Constrain ci

c0 = v < 0

c1 = u >1

c2 = v >1

c3 = u < 0

Each of these constrains reduce the possible point location to the half plane as indicated in Figure 91. There are
16 possible situations in the truth table, and the following propositions reduce this number to 9.
c0 c2
c1 c3

1,2

c0

c1

c2
c3

11

12

13

14

15

IN

10
2

Table 25: Quadrilateral truth table

In addition, to the boundary constrains the ray intersections constrains are introduced.
Constrain ri

r0= v <0

r1= v >1

Table 26 after the elimination of impossible cases and identification of external cell regions defined in the truth
Table 25.

106

Ri

IN

ci
T r0(e3)

c3

F r1(e3)

c1

Te0

Tr0(e1)

c0

Te0

c2

Te2

Fe3

Fr1(e1)
Te2

Fe1

Te0
-

F
Te2

FIN

Table 26: Quadrilateral constrains path

v
P
R7

P7

R2

R6

P6

cell
interior

n2

n3

cell
interior
e2

R3

R1

PIN
0
n0

R4,5,6,7

e0

n1

P5

P4
R4

R0

R5

P2
R2

R7

R6
R0,1,2,3

n2

n3

R1

cell
interior

P3

e2

0
n0

P1

PIN
R3

cell
interior

e0

n1

R4

R5
R0
P0

Figure 97: Possible positions of the point leaving the cell in the neighborhood of the quadrilateral cell

107

The next cell types are 3D cells. The simplest case is the tetrahedron which can be associated with quadrilateral
because they have the same number of constrains:

(a) node
(T3N4)

(c) face

(b) edge

Figure 98: Possible positions of the point leaving the cell in the neighborhood of the tetrahedron cell

Figure 99: Possible positions of the point leaving the cell in the neighborhood of the prism

108

Figure 100: Possible positions of the point leaving the cell in the neighborhood of the pyramid

109

(T3N4)

(a) node

(b) edge

(c) face

Figure 101: Possible positions of the point leaving the cell in the neighborhood of the hexahedron

110

Newton-Raphson method
The Newton-Raphson method is used in a completion point extraction phase where a candidate cell and point are
in the neighborhood of the point to be found. The point extraction algorithm in its first phase finds the candidate
points in modeling coordinates x. The Newton-Raphson method computes in iterative loop a sequence of
positions which lie increasingly close to the ultimate location. This can be presented as a multidimensional rootfinding problem on a distance function d that measures the error between the specified constant location x and
the guess solution A(u), see section 1.3.1.
d(u) = A(u) - x 0
1.4.4 - 8
The prerequisite to find the candidate point in the neighborhood of the proper minimum ensures that the
subsequent refinement phase doesnt inadvertently fall into an erroneous local minimum, as the field can have
several non-zero minimum. The iteration is initialized from an arbitrary location inside the cell u0 and is
repeatedly shifted through a sequence of new locations ui+1 which approach to the unknown location u. The
Newton-Raphson method is defined with the recursive equation
ui +1 =ui

d ( ui )
d' ( ui )

where

d' ( ui ) =

A( ui )
J ( ui )
ui
1.4.4 - 9

The denominator of the equation contains partial derivatives of the interpolating function, where Jacobians
matrices and their inverses are defined for each cell type; see section 1.3.1 on page80. The inverse of the
Jacobian matrix maps the modeling space error into parametric form.
ui +1 =ui

d ( ui )
J ( ui )

After introducing the term above into the equation 1.4.4 - 8,

ui + 1 =ui J -1 [A( ui ) x ] .

1.4.4 - 10
The iterative algorithm applying the equation 1.4.4 - 10 has to result in the quite small distance d between
successive points ui. If the distance is increasing the point parametric coordinates are outside the range [0...1]
and the analyzed point is not located in the investigated cell. For the calculation of the starting point (seed) in the
parametric space the applied Newton-Raphson method uses the cell connectivity and the point location tests,
which define in addition an exit cell side. Thus, the Newton-Raphson algorithm can be restarted with the new
cell guess.
The identification of the starting cell is done applying the local value algorithm; see section 1.4.4, for finding the
point on the surfaces and the cell containing the point. Because the surface is part of a 3D domain, the surface
knows about the volumetric cell and edges defining that surface cell. Ones the surface cell is identified the
volumetric cell is defined based on the surface cell index mapping, through which each surface cell has a link to
the volume cell origin.
This algorithm utilizes the knowledge of the normalized cell boundaries aligned with parametric coordinate axis
so the points are checked against the simple boundary conditions for example if the (u,v,w) value is between
(0,1).

111

An application of the Newton-Raphson point location algorithm is related to the usage of numerical probes tools
when for example the user provides input for one of the following Representations: Local Value, Local Isoline or
Vector Line, discussed later in the thesis, see section 2.4.3. All of them have in common that the point location in
the parametric cell space needs to be computed.
Given a point in a physical space (x,y,z), the problem is to find out if a investigated cell is containing or not this
point. The algorithm has to perform the in-out checks applying the cell normalized coordinates (u,v,w), and it
turns out that this computation is one of the most computationally intensive task when the before mentioned
representation are constructed. The objective is that such algorithm is efficient. Conceptually, the task involves
the application of the Jacobian matrix of the isoparametric mapping, which provides only local information for
each cell. The marching cell algorithm can be an applied for getting a correct cell. The optimization of the point
location algorithm to define the starting seed cell for the marching cell algorithm is done by the traversal of the
domain boundaries. After that, just the selected cells are taken into account. When these cells are analyzed the
Jacobian matrix for the analyze cell is calculated and a center of the cell represents the initial point location
guess.
u
x


-1
-1
v = J y u = J x
w
z


x
u


y = J v Jx = u( 1 )
z
w


Ju = x

1.4.4 - 11
Note that even within the current cell the metric terms vary, so that the algorithm is of iterative nature until
u=v=w=0. If (u,v,w) are outside the range [0,1] or depending of the cell type boundary constrain, the target
point is outside the current cell and another neighbor cell will be analyzed.
Given an arbitrary point x, and the good guess of the cell where the point is located, a Newton-Raphson
procedure is used to calculated the corresponding u. The initial guess u(0) is taken to be a center of the cell and
than we apply the previously mentioned equation, in an iterative algorithm as follows:
u(i+1)= u(i)+u(i)
u(i+1)= u(i)+ J (-1u)x (i )
1.4.4 - 12
If the connected cell is searched, the algorithm converges quadratically to the current point. If the cell being
searched doesnt contain the point, the new value of u will exceed the cell normalized space ([0,1] interval). The
search is than switched to the neighbor cell in for u(i+1). With the possibility to find the connected cell, the
algorithm is repeated for the next connected cell, till the solution or boundary of the domain is reached. In the
case that the point is not found, and there are no other connected cells because the boundary of the domain is
reached, the algorithm looks for other seeds (boundary) cells and the marching cell algorithm continuous on the
cells, which are still available for traversal. Here the problem is to identify the cell boundary through which the
point moved out of the cell. The whole algorithm is situated in the normalized cell space where the boundaries
are simplified, and nicely aligned with the coordinate axis planes. It is interesting to reuse the found point and
cell information for the next point inclusion calculation, because the algorithm is often repeated for the point in
the neighborhood of the previous one.

112

1.4.5 Particle Trace Algorithm

Scientific visualization tools for particle tracking apply algorithms, which difference comes from the calculations
space (modeling or parametric) in which the integration is performed. If the calculation is made in the parametric
space [86, 87] the particle trace points are found through the mapping coefficients, which relates parametric and
modeling space. As these coefficients are computed in the parametric space, they imply the calculation of
Jacobian matrices, which lower the algorithm performance, and in addition the particle trace points have to be
mapped back from the parametric to the modeling space, in order to be displayed. This method is constrained to
the irregularities of the modeling grid, which can introduce errors to the computed vector field transformations.
A more direct method for calculating particle traces [88], is to use the modeling coordinates, but then the
performance penalties are due to the difficulties in calculating the particle trace points location. The method
based on generalized stream function for 3D [89] describes a steady flow applying two coincident scalar fields
representing dual stream functions, where the particle trace is found as the intersection of these two iso-surfaces
[90]. The limitation of this method is the coarse resolution of the vector line inside the cell, which is represented
only as a straight line segment. More conventional approaches includes the adaptive step sizing and more
efficient integration formulas based on Runge-Kutta or Adams-Bashforth schemes [91].
This section describes the particle trace algorithm [92], named also Vector Line algorithm in this thesis, and
applied to the velocity fields defined on structured and unstructured mashes.
A vector line is an imaginary curve where a direction at any of its points is the direction of the vector field at
these points. Vector lines never intersect, because a point in the vector field can have but one direction, only one
line can pass through it. It is assumed that the vector field is stationary, continuous and single valued and that the
particles are mass less, definitions follows:
Stationary
Continuous
Single valued

means that the vector field is independent of time. The trajectories of the particles are called
streamlines.
means that streamlines do not break up.
means that the particle cannot split and occupy neither two places nor two distinct particles
occupy the same place. Streamlines never intersect.
analytical

xi+2

numerical

vector line
Pi+1(t+dt)

+
x(t

dt )

dx

t
vi(x)

v(x)
dt

Pi+1(t+t)

x i+1

x(t)

xi

Pi(t)

Pi(t)

xi-1

Figure 102: Tangency condition for the vector line analytical and numerical treatment

113

The mathematical concept of the particle motion is described by a point transformation of the particle position x
during time t, see Figure 102. Consider the point P in a vector field v. The vector line through the point is a
general 2D/3D curve. The position vector x gives the location of P as a function of some parameter t that varies
along the vector line. The tangent to the curve at P determines the vector line
dx
= v [x(t)]
dt

1.4.5 - 1
The numerical integration takes into account the discrete points of the computational grid which defines the
vector field v. Each vector line is a solution of the initial value problem governed by a vector field v and an
initial seed point x0. This curve is depicted by a ordered set of points (x0, x1, x2 ... xn) defined by
t i+1

xi + 1 = x +

v [ x(t)]dt
ti

1.4.5 - 2
Adjacent points are connected and define the curve geometry which is displayed. The Vector Line representation
can be applied to surface and volume vector fields defined in 3D space. Thus, there are two different vector line
algorithms: one for the treatment of volume vector fields, and another for the treatment of surface vector fields.
For example, the last one can be applied to cutting plane and isosurface vector fields. The flow in the surface is
taking into account the projection of the tangential component of the velocity field. The following two sections
treat these two aspects.

Volume Vector Line

The vector lines are the solutions x(t)=[x(t),y(t),z(t)] of the system of ordinary differential equations, see
equation 1.4.5 - 1. The value of the v(x) is defined by the interpolation algorithm local to a cell, which contains
the point x, where the vector field v is described with a finite number of vectors vs given at cell nodes xs. The
applied interpolation algorithm is the isoparametric mapping, as developed in section 1.3.1 for each cell type,
where the vector field at an arbitrary point inside a computational cell is obtained by the interpolation in the
parametric space, and transformed back to the modeling space as follows:
x(x,y,z) = A[u(u,v,w)])
1.4.5 - 3
The isoparametric mapping operator A defines the coordinates transformation of the vector line points from the
parametric space (u,v,w) to the modeling space (x,y,z). The applied Runge-Kutta integration method [93],
explained later on, on page 119, requires the parametric cell vector field g(u,v,w) at the specified point inside the
cell. The application of the equation 1.4.5 - 3 to the vector line's equation 1.4.5 - 1 in the parametric cell space
leads to the following result:
dA( u )
= v[A(u)]
dt

1.4.5 - 4
which is equivalent to:

du
= g[u(t)]
dt
1.4.5 - 5
With

g(u)=J -1(u) v[A(u)]

1.4.5 - 6

114

and the Jacobian J of the isoparametric mapping A:

J(u) =

A
u
1.4.5 - 7

The equation 1.4.5 - 1 is computed with interpolated values from g(u). The parametric vector field g(u) is
computed for each cell node and applied for the definition of the inverse mapping J-1. The isoparametric
algorithm is efficient when processing:

the vector value at the point inside the cell, requires J-1,

the point inclusion test that defines if a point is located inside the cell, requires A.

The point inclusion test is efficient because the cell boundaries are planes or lines aligned with main coordinate
axis of the cell parametric space, see section 1.1.1. The conditions that the mappings A and J-1 exists are:

the vector field v should be single-valued, i.e., when the mesh contains singular points (several
different grid points occupy the same location in space) a numerical solution has to ensure that these
points have identical values for v.

The Jacobian matrix J must be non-singular in order to be inverted.

The mapping for the right hand side of equation 1.4.5 - 6 calculating g(u) must guarantee enough
continuity for the solution of u(t) throughout the approximation of J.

Continuity should be ensured across two cells. This is satisfied with the piece wise (cell-by-cell)
isoparametric mapping. It must be noted that too distorted cells must be avoided.

Let xs be the cell nodes associated with vector field vs in modeling space, and let Ui be the cell parametric space

(u,v,w), oriented according to the right-hand side rule so that the cell boundary normal point outwards the cell.
The vector line algorithm consists of the following steps:
1-

find seed cell Ui for the given (xo,yo,zo)

2-

find seed point (uo,vo,wo), in Ui

3-

define isoparametric mapping A for Ui

4-

define Jacobian Js for each of the cell nodes for the mapped cell vector field gs=(J -1)svs.

5-

define isoparametric mapping A-1 for Ui

6-

integrate the vector field equation g applying the isoparametric mapping A-1. The integration is
performed by a fourth order Runge-Kutta method in the parametric cell space Ui. The
integration is continued until the vector line crosses a boundary of Ui.

7-

find the intersection of the vector line u with the cell boundary Ui. The intersection point
becomes the last point of the vector line u local to cell.
map the vector line u to the modeling space x=A(u).
find the neighboring connected cell Uj.

89-

If the connected cell Uj is found, reuse the intersection found in step 7 as the first
point of the vector line and replace Ui =Uj. Repeat steps 3-9.
If the connected cell Uj is not found the mesh boundary is reached and the vector
line algorithm stops.

115

The algorithm described above allows computing a vector line starting from some initial position xo in modeling
space. However, it is common to compute the vector line which reaches a given point in modeling space.
Usually, a vector line is computed over the complete mesh. For example, a trace that consists of the vector line
that reaches point xo together with the field line that starts from xo. To compute the field line that reaches xo one
simply computes a vector field line starting from xo but with a minus sign in the equation 1.4.5 - 1:
dx
= -[v(x)](t)
dt

1.4.5 - 8
with solutions x(t)=(x(t),y(t),z(t)). The overall result of the vector line algorithm consists of two distinct segments
of the vector field line, representing the forward and backward sweep respectively.
The vector line algorithm has following important issues:

identification of seed cells and points

integration step magnitude
integration break
mapping of parametric coordinates based on cell connectivity
Initial seed cell and seed point location are found with the point location algorithm, described in section 1.4.4.
Given the coordinate x0 of the seed point, its corresponding coordinates in parametric space are found. To
calculate u0 (uo,vo,wo) from x0 = A(u0), the nonlinear system of equations is solved using Newton-Raphsons
method. Since the mapping A is isoparametric, the system of equations has only one solution and NewtonRaphsons method, see page 111, can safely be used. The integration of the interpolated g within one cell Ui is
performed by a fourth order Runge-Kutta method, see page 119. It is carried out in the (u,v,w) coordinate system
from the first point in the new cell to where the integrated vector line point leaves the cell. An average velocity g
is defined from velocities defined in cell vertices as
N

g =

i =1

1.4.5 - 9
The integration step t is calculated for each cell so that approximately M steps are taken in the cell. The M
parameter can be interactively adjusted by the user then a step size t is given by:

t =

1
Mg
1.4.5 - 10
w
w
n2

e2

n3

n2

u
e1

e2

e3

e1

u
u
n0

e0

u
u
n1

n0

e0

u
n1

Figure 103: The cell boundaries parametric space in 2D

116

v
f1
u
v
w

f2

f3 u
v

u
f0
v

v
u

f4

f1
u

f2

u
f0

u
f3

v
f4
u

v
v

v
u

f2

f1

u
u

f0

f3 u

v
f5 v
u

v
u

f1
u
u
f0

v
w

f2

f4

v
f3 u
v

Figure 104: The cell boundaries parametric space in 3D

117

Figure 105: The map of a cell boundary point between connected cells in 3D
Experiments have confirmed good results with M=4 or M=5. The different reasons, for which the vector field
line algorithm (page 114) has to be stopped, are called integration breaks, and they are activated when:

vector line reaches the grid boundary (step 9)

number of vector line points have been exceeded

number of vector line points per cell have been exceeded

The Jacobian matrix J is singular (step 5), thus, the mesh contains a degenerate cell.

The point location and moving the parametric point from one cell to another is accomplished by applying the cell
connectivity and the imposed orientation of the cell faces defined for each cell type, see Figure 103 and Figure
104. The objective is to avoid the computation of the point parametric coordinates ones the vector line points are
mapped to the modeling space. As the vector line exit point is known applying the point location algorithm, the
parametric coordinates of the point on the cell boundary can be viewed from both connected cell. The mapping is
done in following steps as shown in the Figure 105 and Figure 106:

map the cell parametric coordinates to the boundary, which results in lowering the parametric
dimension.

swap in orientation of parametric coordinates between connected boundaries

map the cell boundary parametric coordinates to cell, which results in raising the parametric
dimension.

The important precondition is that the cell topology for the whole domain is completely defined, as described in
section 1.2.2.
The local parametric cell coordinates are transformed when passing the interface between two cells without the
necessity to find the interface point in the global grid coordinates avoiding Newton-Raphson method to find the
point location in (x,y,z)
n1

n1

n0

n3

P(u,v)

P(u)

P(u)

P(u,v)

u
n2
n0
n0

n1

n2

v
n0

n1

Figure 106: The map of a cell boundary point between connected cells in 2D
118

The mapping is based on following steps:

P(u, v)
I cell
out
T (u1 )
interface I cell
T (u2 )
interface II cell
in

A(T(u1,v1))

B (T ( u ))

II cell

T(u1, v1)

1. T (u) = A(T (u1 ,v1 ))

2. T (u ) = B (T (u ))
3. T (u 2 , v2 ) = C (T (u ))

The generic algorithm is based on A,B,C mapping, which are defined separately for each cell type involved. The
solution provides the required flexibility to support heterogeneous cells without complicating the generality of
the algorithm. The ability to change A,B,C at run time without the requirement to change the general algorithm
and the ability to allow the programs to be fine turned after it is up and running (interactive turning).

3D scheme:

(#cell type, #inter face)

T (u , v )

T (u ,v )

T (u 2 ,v2 , w2 )

(#cell type, #cell side)

A(T (u 1 , v1 , w1 ))
B(T (u ,v ))

C (T (u ,v))

Table 27: The mapping procedure of a cell boundary point between connected cells in 3D
In the first phase the point must be defined in (u, v) coordinates local to the face. This is handled with the A set of
functions, see Table 27, where w coordinates are always zero as they are part of the cell faces. If that is fulfilled
the intersection of the vector line with the cell edge end face is found. This algorithm includes the identification
of cell edge, which is used to identify the next cell and thus maintain the C0 continuation of the vector line. Each
cell which is holding the vector line has at least one interior and exit point. This is just not true for the cells in
which the vector field vanishes. For each cell type all the possible cases for entering and exiting of the vector line
are considered as detailed in the section 1.4.4 related to the point location algorithm.

Surface Vector Line

The surface vector line algorithm requires the same steps as the space vector line algorithm, in order to calculate
the surface particle traces. What makes it different is that the volume vector fields are projected to the tangential
plane of the surface points, which are defined for the created or inputted zones. The isoparametric mappings for
surface cells are derived from section 1.3.1 where maps between 3D modeling and 2D parametric space require
particular treatment, for the calculation of the inverse Jacobian matrix.

Runge-Kutta Method
The integration of ordinary differential equations is solved by the Runge-Kutta method [93, 94] The vector field
v is the RHS of the following equation:

119

dx
= v (x, t).
dt

1.4.5 - 11
The idea is to rewrite dx and dt as finite steps x and t and multiply the equation byt.

x = v(x, t) t
1.4.5 - 12
This is an algebraic equation for the change in x when the independent variable is stepped by one step size t. In
the limit to make step size very small a good approximation of the equation 1.4.5 - 11 is achieved. This
explanation results in the Euler method

xi+1 = xi + v(xi, ti) t

1.4.5 - 13
which advance the solution from ti to ti+1 ti + t and if expanded to power series O(t2) can be added to the
equation 1.4.5 - 13. If we consider the trial step at the middle of the interval

k1 = t v(xi, ti)
k2 = t v(xi+

t
k1
, ti+ )
2
2

xi+1 = xi + k2 + O(t2)
1.4.5 - 14
This is called the second-order Runge-Kutta method. As there are many ways to evaluate the right-hand side of
the equation 1.4.5 - 12, which all agree to the first order but may have different coefficients of higher order error
terms. With the Runge-Kutta method, the adding up the right combination of these coefficients the error terms
are eliminated order by order. The fourth-order Runge-Kutta method is defined as follows:

k1 = t v(xi, ti)
k2 = t v(xi+

t
k1
, ti+ )
2
2

k3 = t v(xi+

t
k2
, ti+ )
2
2

k4 = t v(xi+ k3, ti + t)
It requires four evaluations of v per step t. This is more efficient to the equation 1.4.5 - 14 if at least twice a
large step is possible with

xi+1 = xi +

1
[k1 +2( k 2 + k3 )+k4 ] + O(t5)
6

1.4.5 - 15
The higher order integration method is always superior to a lower order one, but when we consider the number
of arithmetic operations involved and the possibility of an adjustable integration step to achieve a desirable
accuracy, it is well known that the fourth-order Runge-Kutta method represents an optimize choice. This is the
reason why the fourth-order Runge-Kutta method is used for the calculation of volume and surface particle
traces algorithm, as described in this section 1.4.5.

120

2 Adaptation of Visualization Tools

Visualization tools for interactive quantitative and qualitative data analysis have to provide an effective and easy
to use functionality for the examination of fluid flows and their underlying physics. The interactive visualization
mimics the experiment in the laboratory. It requires to localize and to isolate regions where, for example, a shock
interaction or vortex generation flow pattern is present. If several visualization tools instruments are applied
together, they can intuitively stimulate new ways to analyze and discover fluid flow behaviors, especially valid in
the regions where experimental measurements were not taken.
The Representations are the graphical objects, discussed later on in this chapter, which are appearing on the
computer display and they are responsible for displaying geometry and different scalar and vector quantities. As
the thesis investigates a real-time interactive approach, considerable attention is given to the design and
development of numerical probes and their Graphical User Interface (GUI) setup. Although quite massive
calculations are involved for each invocation of such visualization tools, the achieved interactive feedback loop
between a user and the visualization system is quite satisfactory, even on todays ordinary PC hardware equipped
with a standard 3D graphics card.
During the visualization process, the user can restrict his/her analysis to a specific interactively selected Zone; see
section 1.1.4, to which different quantity fields could be related. The number of introduced quantities is left open
to a user input. In addition, representations are also available for the experimental data sets, which are commonly
used for performing comparisons with computed data sets.
The numerical probes are interactive diagnostic tools supporting scalar and vector representations, for example:
isolines and color contours for scalar fields
smoke/dye injection simulation of particles with no mass
bubble wire traces for velocity profiles
They are integrated in the visualization system interaction mechanism for providing information feedbacks
between displayed information, loaded input parameters, selected algorithms and the manipulated graphical
model. Some of the typical user interactions are:
mouse driven positioning of the starting point selection
menu driven selection of the numerical probe geometry (point, line or plane)
interactive color map modifications
mouse or menu driven selection of quantity threshold range,
arbitrary positioning of a section and cutting plane
The user interaction is made transparent through the use of mouse and keyboard input devices. The examples of
interactive user actions are the selection of an Active Surface or the manipulation of the Viewing Buttons, which
control the setup of the user viewing position and perspective angles. These interactions are in the direct relation
with the displayed geometries and quantity representations, and when applied they trigger an immediate visual
feedback of the visualization system.
Two different groups of algorithms are applied when the graphical representations are generated:
1. the algorithms involved in geometrical search
2. the algorithms that generate the image
The first group of algorithms is explained in detail in chapter 1 and the basic layer of this second group of
algorithms is today supported in hardware by the 3D graphics cards, which implement the OpenGL 3D API. The
OpenGL low level graphics library is encapsulated in the Graphics category, explained in the implementation
section 3.7.2.
The designed scientific visualization system manipulates structured and unstructured surfaces in a transparent
way, enabling that the same types of graphics representations are provided for both types of geometries. The
representations are classified in three main groups as geometry, scalars and vectors.

121

Geometry representations parameters and attributes are:

predefined boundary types identification
repetition: translation, rotation and mirroring,
wireframe,
hidden line/hidden surface removal,
flat, Gouraud and Phong lighting interpolation algorithm,
positioning of light sources,
material properties of the surface (transparency, texture mappings)
mesh quality assessment:
cell quality check
distorted cells based on edges or normal ratio criteria
small cells
Scalar representations are:
color contours
isolines
scalar section and grid line distribution in the Cartesian x-y plot
local value
local profile
traversal functionality for computational surfaces
cutting plane
iso-surface, clouds effects with transparency
Vector representations are:
vector field
vector section
local profile
cutting plane:
control of vector length: uniform or according to linear or log scale magnitude
uniform or colored according to a scalar quantity
particle path:
color and line type option
graphical and numerical control of the released location
Beside the classical field representation of isolines, contour shadings and thresholds, the numerical probes as
diagnostic tools, provide support for an interactive quantitative analysis of vector and scalar fields, and their
related geometries. Numerical probes are adapted for the localized user investigation, aided with mouse point
and click operation. The following numerical probes are designed:
local values
local isolines
sections with quantity distribution along curves (arbitrary or mesh lines)
local profiles
local vector lines

Numerical probes are displayed as point, line or plane object, which are interactively controlled to investigate
the displayed Zones. The objective is to restrict the analysis to the user selected zones, which are obviously not
the only part of the complete calculated domain. For example, the isoline is created on a surface and displayed.
Thus, only the isoline appears on the screen and the entire domain and reference surfaces involved in the isoline
generation could be hidden by the user. In such way the user interactively defines the visualized scene content,
and obtains focused, filtered and reduced set of displayed graphical information.

122

Figure 107: Local profile and Particle paths for a 2D airfoil

The local value probe displays a numerical value of the scalar field at the interactively selected point. The
interactive section line input results in the Cartesian plot showing the quantity distribution along the created
section. The local profile is a numerical probe which allows, to locally blow up a region, a direct example is the
boundary layer, and shows the quantity distribution in a specially designed Cartesian plot representation, which
is displayed at the location of the inputted section. As mentioned, this feature is valuable for the investigation of
flows in boundary layers, as the geometry scale along the section can be adjusted independently of the viewing
coordinate system in which the representation is posted. In Figure 107, local profiles are applied around the
airfoil together with few particle paths. It can be noted that the vectors are not tangential to the particle paths as
the local profiles show exploded region of the boundary layer.
All the previously mentioned probes are also available for 3D computations. In Figure 108 we can see the local
value tool to display density values combined with local isolines, streamlines in the form of stream-ribbons
especially adapted for 3D vorticity interrogations. The important approach in volume analysis of the structured
data set is the volume slicing by extracting surfaces of constant I, J or K indices. The volume traversal enables to
browse rapidly through the mesh and to localize interesting field features in an animated surface way. The
geometrical mode allows the visualization of surface geometry, while in the quantity mode the shaded scalar
contours are displayed. The possible switch between these two modes of interaction can be done without
interrupting the visualization process.
For all types of 3D data the volume investigation is based on the isosurface and cutting plane numerical probe.
The unstructured surface is the result in both cases. On that surface, other quantities can be interpolated and
analyzed using the same numerical probes and their representations (isolines, color contours, etc.) as available
for the surfaces provided by user input. The interactive user input for both tools is provided in numerical and
interactive mode. For the cutting plane tool the scrolling and rotation about an arbitrary axis are provided, and
for isosurface the interaction with the Colormap enables mouse click input of the isosurface scalar value.

123

Figure 108: Local values, Isolines and Particle paths representations in 3D

By combining different types of representations on the screen the graphic output can have qualitative and
quantitative meanings, as indicated for some representations in Table 28.

Qualitative

Quantitative

Stream lines

Vector arrows

Isosurface

Isolines

Contours

Plots

Table 28: Qualitative or Quantitative character of the Representations

2.1 Input Data Model

The input data model describes the organization of a numerical data expected to be specified by a user. The user
input consists of a mesh geometry and different number of quantities. The 1D, 2D or 3D mesh geometry is
composed of different cell types; as described in the Data Modeling section 1.1. The basic assumption is that the
mesh cells do not overlap. Each cell represents a unique definition of the approximated geometry space, it is
covering. Usually, such requirement is assured by the grid generation system responsible for the grid modeling.
The input data model of the visualization system has to be adapted to different usages. For example, the
imported geometry could be checked to be verified against the initial CAD model. Another example is, when
data from a CFD flow simulation are combined with data from FEA stress computation and needs to be
visualized in order to treat the fluid-structure interaction problems. In addition, measurements data could be
provided and comparison against the computed results could be requested.
The data models for computational methods can be decomposed in two large groups, and such categorization is
based on the way how the computed variables are related to grid cells or nodes:
124

1.
2.

The cell-centered input model defines the variables at cell centers.

The cell-vertex data model defines the variables at the cell nodes.

Because the output of each numerical simulation software is specific to its computational method (cell-centered
or cell-vertex) it is appropriate to have a unified input model for a scientific visualization system. In this thesis,
the cell-vertex model is chosen as the input data model, as it is more near to the data model of the applied
graphics model. Additional reason for selecting the cell-vertex model was that it was not obvious how to
extrapolate the input data to the domain boundaries when treating the cell-centered input model. The applied
mapping for converting the cell-centered data to cell vertices is defined by averaging the quantity values of the
neighbor cells surrounding each node, as explained in node topology, see section 1.2.4.
The designed input data model supports the multi-block decomposition, which is especially useful for the
treatment of complex geometry.
The quantity input data types are organized as:
field data
solid data
validation data
plot data

Field data are quantities given for the complete computational domain. Solid data are quantities confined to
solid boundaries, such as the heat transfer and skin friction coefficient. Validation data are coming from
experiments or other computations to facilitate the data comparison. Plot data are related to the arbitrary x-y
plots, such as convergence history. An arbitrary number of quantities are allowed for each data type.
The results of the computational analysis can be mapped to the designed input file system, which consists of a
combination of ASCII and binary files responsible storing different input data sets. The input file system is
organized through a small navigation file, which groups all the files related to the specific project. Different
navigation files can be created for the same data sets giving flexibility in organizing the visualization session.
The navigation file defines geometry dimension, topology type, number of domains and number of quantities for
each data type. It defines global transformation of overall geometry like mirror and repetition. For each domain,
the number of nodes and cells are given together with domain connectivity file indicating the types of defined
boundaries, as specified in section 1.2.5.
The quantity files are related to the defined quantity data set specified
per domain or per solid surfaces. Therefore, they are separated into
two categories: field and solid data files.

Figure 109: Main Quantity menu

The field data files are defined for the complete domain. For example,
the pressure or the velocity field around the space shuttle is defined
for each grid node.

The solid data are restricted only to the solid boundary segments. For
example, the skin friction on the solid surfaces of the space shuttle can represent a valid solid data input. Another
distinction is made between the computational data and the validation data. For example: comparisons have to
be made between experiments and the computed data, see Figure 109 where the pull down menu reflects the
mentioned distinction. In Figure 110, for the airfoil example, the field data pull right menu contains a list of
scalar and vector quantities defined in the computational space the solid data are also present under similar pullright menus, defined only the solid boundaries.

125

Figure 110: Field and Solid Quantity menu

The Validation Data menu for the airfoil example is shown in Figure 111. It contains all the validation data for
the project as specified in the input file.

Figure 111: Validation Data menus

The Plot Data is the fourth and the last defined quantity type. The selection of such data type allows creation of
simple Cartesian plot with arbitrary number of curves. The example is again from the same airfoil computation
showing convergence history, see Figure 112. This data type is introduced to give more possibilities for
presentations of data that are not directly linked to the project.

Figure 112: Plot Data menu

The mesh topology classified as structured or unstructured, requires different input files. The structured grid
topology has all mesh points lying on the intersection of two (or three) families of curves, which define the
curvilinear coordinate lines. They are represented by a set of integers as i,j or i,j,k depending upon the space
dimension 2D or 3D respectively, see Figure 113.

126

Figure 113: Structured topology for 2D and 3D geometry

The unstructured grid topology is formed by combination of different cell types, see Figure 114, where the grid
points cannot be identified with coordinate lines, but have to be indexed individually in a certain order. It
requires a more complex bookkeeping with respect to structured grids, because cell connectivity has to be
indexed separately for each cell-nodes relationship, while the structured cell connectivity can be explicitly used
from the grid indices.
In the input model the domains connectivity, see section 1.2.5, is defined and in addition the boundary conditions
are specified. The orientation of domain boundaries is determined and the connectivity rules are applied for the
matching neighboring segments in order that the consistent input setup is established. The boundary conditions
(BC) are defined on domain boundaries and depending on the project dimension associated to the 2D or 3D
segments. This information is mainly used for the following purposes:

when initializing the visualization session, by default the solid surfaces are displayed, as they
represent naturally the investigated problem (airfoil, airplane),
they can be identified and selected in this starting phase to get a first insight in the available data,
The periodic and connected boundary conditions are applied when particle trace algorithm is
performed for the multiblock data model.

Figure 114: Unstructured topology for 2D and 3D geometry

127

The standard notation for the description of boundary conditions, see Table 29, is applied on each segment
separately. Since the types of boundary conditions may vary according to the used flow model, a common set of
BC types was defined for the input data model, as follows:

BC types
Connection
Periodic
External
Singularity
Inlet
Solid
Outlet
Mirror

Input Abbreviations
CON
PER
EXT
SNG
INL
SOL
OUT
MIR

Table 29: Standard notation for boundary conditions

Defining the domains connectivity is a time consuming task, as the user needs to specify the relative positions of
the interfacing Segments. Normally this information is created in the grid generation phase [95], where the
connectivity of the boundary elements is specified along these interfaces according to the processed domain
topology taking into account their orientations, see section 1.2.5. The orientation parameter refers to the relative
position of grid points on two connected Segments (equal or reverse parametric coordinate direction along the
same arc length). Thus, for 2D grids, the connectivity of two boundary Segments (curves) is specified by
indicating the segment orientation. For 3D grids, the parametric local coordinate system and the orientation of
BC segments (surfaces) need to be specified. The segment corners nodes represent the possible locations at
which the origin of a parametric coordinate system could be located (considering corner 1 as the reference on the
first patch: 1-1, 1-2, 1-3 and 1-4). Thus, the segment connectivity can be specified in eight different ways. The
connected segment indices are also indicated using the standard notation, see section 1.2.2.

Figure 115: Domains connectivity for structured 3D multiblock grids

128

2.2 Surface model

The Surface model is a central organizational concept of the visualization system, as it defines a reference base
to access the created and displayed Representations in a Scene. The surface model is extended with the
representation container, which enables the surface model to keep track of the created representations, and to
incrementally associate the surface data during the visualization process. For example, the Geometry menu offers
the set of possible Actions as: select remove, create and delete the surface. Following the same idea, other
geometry models as points, curves and bodies are treated identically. The following explanations are based on
the Surface model and the related Representations. Each surface has two possible states:
Active: when its related representations can be invoked, such as geometry rendering or scalar fields
contours, and
Inactive: when a surface is present in the Scene, but the invoked representations are not applied to it.
In order to improve the user awareness of the active or inactive state of displayed surfaces, the user can select the
surface by pointing to it with the cursor. The mouse click will trigger the visualization system to localize the
surface and inform the user of its state. This feature enables the user to apply representations selectively to each
displayed surface. Such interactive way of modeling the scene by controlling the created representations for each
surface offers a large set of combinations to create appealing graphical presentations.
The surface solid boundaries are the initial representations displayed when the visualization is initialized in order
to provide an intuitive feeling for the inputted geometry, see Figure 116. These representations are depending on
the project dimension:

for 2D:

all boundaries defined in the boundary conditions data set.

for 3D:

only solid boundaries, if they dont exist, all the boundaries defined in the
boundary conditions data set.

Figure 116: Initial representations: boundaries in 2D and solid surface boundaries in 3D.

Figure 117: Geometry/Surface menu

The Geometry/Surface menu, see Figure 117, allows the creation,

removal or deletion of surfaces. The existing surfaces are organized in
two groups. The first group includes all the existing and created
surfaces in the project and this list is only one. The second group of
surfaces is defined for each View and these surfaces are building the
View scene. This mechanism allows that each surface can be selected

129

and reused in a multi-views visualization environment. However, the surface-representation-view relationship

keeps the displayed representations independent from each other. An operation performed on a surface in an
active view will not affect the same surface representations in the other views, thus the surface representations
may be different for each view.
The visualization tool for the extraction of surfaces from the structured topologies is making use of I, J or K
mesh indices. The tool is invoked by the Geometry/Surface/Create dialog-box; see Figure 118, which contains:

the surface name if saved, thus we create it

the selection of I, J or K surface index

the start index value of the surface to be first displayed

the Min-Max range of index values to display a part of the

surface(patch)

two different scrolling modes

the Save and Cancel operation

the multiblock mode for selecting individual domains

Figure 118: Create Surface dialog-box

The dialog-box allows the traversal of computational domains by scrolling through constant I, J or K surfaces
and if desired, to save these for further analysis. For the multidomain configuration, only the selected domains
are traversed. The traversal consists of displaying the surfaces in two modes: animated or step-by-step. The
displayed representations are:

Geometry: as a surface grid

Quantity: as shaded color contour, which is appropriate for quick localization of interesting 3D regions

Figure 119: Structured multiblock surface patch manipulation

130

Figure 120: Cutting plane and isosurface examples for structured and unstructured meshes
Figure 119 shows on the left side the Ahmed body multiple patches of the structured grid surfaces and the
respective scalar field contours and on the right side the Hermes space vehicle discretized with the structured 2block grid. The interesting aspect in both cases is the space traversal, which is performed with displaying
complete or partial surfaces; in addition, indicating the multidomain grid structure for geometry and quantity
representations. The possibility to create surface patches is defined within Min-Max range of the local I, J
surface indices. For example, the surface extraction can be reduced to cover a limited display area in order to
make other representations visible, as shown in Figure 119, where the surface patch was adapted to show the
particle traces in the background. The minimum surface patch size can be reduced to one cell. When the surface
or surface patch is created the surface boundaries are displayed, as part of the interactive system feedback, to
indicate that the surface creation was successful. A counterpart of such interactive visualization tool does not
exist for unstructured meshes.
The cutting plane and isosurface tools are the interactive tools, which are available for both structured and
unstructured topologies. It is important to mention that the related surface extraction process is involving all the
domains of a multidomain (multiblock) input. The created surfaces are automatically named with the domain
index and prefix, such as ISO for isosurfaces and CUT for cutting planes, thus preserving the relationship with
the domain from which they were extracted. In Figure 120 the example of simultaneous application of both tools
is shown on structured and unstructured mashes.
The Selection Surface dialog-box contains the surfaces associated with the view. As shown in Figure 121 the
created isosurface and cutting plane are highlighted as active and the automatic naming convention is applied, as
described in the previous section. The interactive process to remove or destroy the surface is equal to the surface
selection process. When the surface is destroyed all associated representations are removed in all the views. If
the surfaces are visible on the screen, the interactive surface selection can be performed with the mouse point
and click operation. Every visible surface in the view can be made active or inactive depending on its previous
state. The interactively selected surfaces are highlighted and the repeated action on the same surface will act as
an active/inactive toggle. For improving the interaction feedback, the surface name and the active state are
displayed in the monitoring area of the main window. In addition, the Selection Surface dialog-box is helpful
when reusing the same surfaces in a multi view visualization environment.
131

Figure 121: Surface dialog-box showing the cutting plane and isosurface instances

2.3 Geometry Representations

Figure 122. Geometry menu

The different geometry representations are depicted in the geometry pulldown menu, see Figure 122, as follows: Full mesh showing the surface grid,
Mesh boundary showing the surface boundary in 2D and boundary surfaces in
3D. Solid Boundary automatically displays only the boundaries/surfaces with
the imposed solid boundary condition. The geometry representations are
unified for structured and unstructured grids, as the basic graphics for both
topologies is defined by their nodes, edges and cells, see Figure 123.

Figure 123: Surface geometry with boundaries outlines

132

Figure 124: Geometry repetitions types: mirror, translation and rotation

Some geometry representations are based on the wireframe graphics composed of polylines, and others
representations are constructed of filled polygons reflecting the surface cells structure. The boundary and solid
boundary representations are used for fast identification of overall geometries as they are the lightest graphics
models. The Repetition functionality enables the duplication of the displayed geometry, such as: the translation
in a given direction, the rotation about and arbitrary axis or a mirror operation. The Figure 124 shows the three
possible examples of the geometry repetition. The rotation about an arbitrary axis and the translation of geometry
are especially useful for turbomachinery applications, where the blade to blade geometry can be duplicated for
the better understanding of the flow quantities behavior in the regions with periodic boundary conditions.

Figure 125: Render menu

The Render pull-down menu allows the display of the active surfaces
with Hidden line, Flat, Gouraud or Phong lighting interpolation, see
Figure 125. The hidden line representation assigns a uniform color to
all the surface area. The algorithm removes the invisible parts of the
surfaces.

The hidden line algorithm is combined

with grid wireframe representations to
display the cells shapes. If the flat
rendering is applied the surface color is
modified in accordance with the light
position and cell surface normal preserving
the uniform color inside the cells. The
Gouraud and Phong lighting interpolation
technique made the scene more realistic
because the lighting interpolation is
performed inside each cell according to
light position and cell vertex normal.
Figure 126 shows all the four rendering
possibilities for the space shuttle

Figure 126: Rendering of the space shuttle

133

2.4 Quantity Representations

The scalar and vector representation are treated both as representations and based on the selected quantity the
menu items are adjusted automatically, see Figure 127. The Representation menu is a content sensitive menu, as
its content depends on the type of the selected active quantity (scalar or vector). This is similar to the geometry,
where the Geometry menu depends on the input data dimension (2D or 3D) and grid topology (structured or
unstructured), see section 2.2.

(a)

(b)

Figure 127: Scalar and vector representation menus

The Scalar Representations are applied to the whole surface field as Isolines and Color Contours. Other
extracted representations are geometrical subspaces of the surface like the section (curve) and the local values
(point) types. The quantity representation becomes something which can be graphically displayed as pointmarker, curve-edge, surface-face or other textual information. All these graphics shape primitives are used
extensively in combination with colors. It is important to stress that setting of the color, as one of the
fundamental graphical attributes, has to be performed easily, especially in the case when interactively adapting
the colors mapping.
The Vector Representations are modeled as glyphs, in different directional arrow based shapes, defined with
color and shape parameters, which are associated respectively to the vector magnitude and direction. Such vector
glyphs, or icons, are scaled by a set of predefined values, which can be interactively manipulated. The two
important representations are the Vector Field representation which displays the vectors glyphs at predefined
surface points, and the Vector Line representation graphics displayed as curve. The application of such vector
representations depends directly on the analyzed fluid flow. For example, the Vector Field representation is not
appropriate to be applied where the grid points are dense and the vector magnitudes change rapidly. In such case,
the displayed vector arrows will fill up the displayed area and the presented graphics will be more disturbing
than helpful. In order to have better visibility of the underlying flow pattern the vectors distribution needs to be
coarser than the grid distribution, thus it is suggested that the limited number of Sections or Local Vectors are
interactively placed in the combination with particle traces starting in the problematic region, for example, where
the vortex core is located.
In this section, each Quantity representation will be tackled in a combination with other available interactive
tools, in order to indicate the best practice in how to use them in an appropriate and an efficient manner. As the
visualization system offers the possibility to combine several representations in the same view, in many of the
described examples their simultaneous usage is pointed out in order to explain the potential added value of such
visualization scenarios.

134

Figure 128.: Isolines representations for 2D and 3D geometries

2.4.1 Isolines
The Isolines interactive tool offers several options to
create isolines:
isolines are computed within specified
minimum-maximum range
giving the increment between two adjoin
isolines
specifying the number of isolines
In Figure 128 the Isolines are displayed together
with Local Value representations, in order to
enhance the relationship of displayed graphics with
numerical values.

Figure 129: Isolines menu item and dialog-box

The Isolines can be generated as a representation group using the described dialog-box or by the Local Isoline
tool, for the creation of individual isolines. This is performed by specifying a scalar value or by locating the
surface point, through which the isoline is requested to pass. For both invocations the parameters can be set
interactively or numerically. The firs approach enables interactive input, while the numerical one assures that the
exact numerical input of an isoline value or surface point coordinates are given. For the interactive approach, the
creation of an isoline by the mouse-click operation is done by pointing with the screen cursor inside the
colormap scale or by pointing the cursor over an active surface. The mouse click will trigger the interpolations:
of the selected value or of the selected point location, followed by the isoline generation algorithm for the
identified scalar value.

2.4.2 Quantity Fields and Thresholds

The Quantity Fields representation group brings together the scalar color contours and vector fields
representations, as they put on view all the inputted surface data. For each surface node the mapping functions
between the color and the quantity value are defined. The scalar contours are continuously coloring the surface
cells, while in the case of vector field representation the vector magnitude and color parameters are applied for
the vector glyph generation. The extension to such full field surface visualization is the partial display of the
quantity field specified through the threshold values.

135

Figure 130: Color Contours menu

The Color Contour representation is the most frequently used

representation used in case to gain an overall behavior view of the
scalar field. Each surface vertex is associated to a color which
corresponds to a scalar value at that point. The scalar field is
painted with a graduation of colors according to the selected color
mapping displayed through the Colormap, see Figure 131.

Figure 131: Color contours based on different rendering algorithms

The coloring options are reassembling the ones used for the Geometry Rendering where just one surface color is
applied to different rendering methods. In this case the number of colors is increased from one to the number of
colors used in the applied colormap. These options are equivalent to the geometrical ones and activated through
similar menu items, for example the ones with flat and smooth labels. The Flat Contour displays the scalar field
as collections of cells where a unique color is assigned to each cell. The assigned color corresponds to the
average value of the quantity in that cell. The Smooth Contour displays the scalar field as collections of vertices
where inside each cell the color is smoothly changing from one cell vertex to another. In this type of
representations the cell geometry is not visible. To see the cell wireframe geometry together with smooth
contours the geometry wireframe representation can be superimposed to it, see Figure 132.

Figure 132.: Threshold color contours

136

The extension to the surface Color Contour representation is the Threshold representation, which restricts the
coloring of the surface to a surface region depict by the Threshold range, see Figure 132. In the case of an airfoil
only the supersonic scalar values of a Mach number above 1 are displayed with the Threshold Color Contours
representation. In the subsonic region the Isolines representation is applied, and in the entire field the
quantitative information is displayed with some Local Values representations.

Figure 133: Vector Field dialog-box

The Vector Field representation draws the vector quantity at

the grid nodes of the selected surfaces. In the case of an
unstructured grid the vectors are displayed at each grid node.
The structured grid topology gives to the user the possibility to
display the Vector Field at different resolutions by defining the
numbers of vectors to be displayed in the I and J directions, see
Figure 133..

In Figure 134 the examples of unstructured and structured vector fields representations are presented. It can be
noted that the resolution of the structured field can be controlled with limiting the number of display vectors, thus
making more comprehensive the presentation of the examined vector field.

Figure 134: Structured and unstructured vector fields

The Threshold tool provides a convenient way to discard uninteresting parts of the vector field. The surface
regions in which the vectors magnitude is outside the threshold are discarded. Figure 135 shows the threshold
representation in two cross sections of the airplane wing where we isolated the vectors near the Mach 1 velocity
field.

Figure 135: Vector Thresholds on cutting planes

137

Figure 136: Local isolines, scalars and vectors assisted with coordinate axis tool

2.4.3 Point based numerical probes

The point based numerical probes allow interrogating scalar and vector fields at users defined points. The
outputs are the Local Scalar, Local Isoline and Local Vector representations. For improving the user positioning
within the viewing space and to verify his/her point locations input the Coordinate Axis tool provide the
necessary visual validation feedback, see Figure 136. The interactive approach supported with the point based
tool allows the use of the mouse as the point input device, or a string input, if exact numerical coordinates values
are requested. The selected point in the displayed view space doesnt need to be a surface point. The considered
line is perpendicular to the screen and passes through the selected point, thus the line defined with the cursor
point and the view normal will be applied in the intersection algorithm, as explained in section 1.4.4.

Figure 137 Range menu

In the Figure 136, for the velocity vector field the quantity range is
adjusted using the Range menu, see Figure 137, between 350 and 430
m/s, and we generate some Local Isolines representations with a precise

numerical input.
In addition, the isolines are validated with the Local Scalar values corresponding to the velocity magnitudes.
Finally, we have added the Local Vector representations, to show the correspondence with the newly created
Colormap.
The Range menu makes possible to modify the current range of the quantity or to reset it to the default state. This
option is useful for the comparison purposes between different projects, in order to synchronize the imposed
applied ranges. The range limits can be interactively changed utilizing the string input or the mouse point and
click operation within the Colormap display area. The user interaction is identical to the threshold range input
described in the previous section. The new setup up of the quantity range will affect the representation associated
with the related colormap, and consequently all the dependent representations will be updated inline with the
new color mapping. In Figure 136, it is visible that the colors of the isolines and vectors are in accordance with
the range indicated with the displayed Colormap in the right side of the figure.

138

2.4.4 Curve based numerical probes

There are three basic types of Representations coupled with the curve based probes. These curve based
representations are applying the geometries defined through:
1.

predefined curves

2.

sections (in addition, Cartesian plot for scalar fields) Isolines, Particle paths

3.

and Local Profiles

Figure 138: Cartesian Plot menu

The common curve representation is the extraction of structured curves

based on the I, J, K indexing, including the surface boundaries which can
be further refined, filtering out only the parts with solid boundary
condition, as shown in Figure 138. The Cartesian plot representation, see
Figure 139, is complex representations as it relates two views and their
respective curve representations: 1) in the main view the curve shape is
displayed and 2) in the Cartesian plot view the quantity distribution is
shown.

When the calculation of the scalar distribution along the extracted curves is called the view, called Cartesian plot
view is opened. The limits of the plot axis are automatically sized to the min-max limits of the quantity value
calculated from the sections present in the Cartesian plot. The Cartesian plot layout can be modified by the
Coordinate Axis Editor. For the Cartesian plot view, the view manipulations buttons can be used to adjust the
displayed space of curves geometry and scalar quantity range. For example, the zoom area operation can be used
to blow up a select a region of a Cartesian plot.

Figure 139: Scalar distribution along solid boundaries and sections

139

In Figure 139 are shown 2D Solid Boundary and Section representations for three different scalar quantities. For
the two plots, from left to right, the abscissa represents the x coordinate and the third plot is related to Section
representations where the quantity distribution is presented with arc length starting from the solid boundary. A
Section representation displays the geometry of the curve as a result of plane/surface intersection, see section
1.4.1. The Section input requests two points for defining section location. It is based on a mouse-cursor visual
feedback. After the input of the first section point is made, the red rubber band line attached to the cursor,
visually aids a user to define the section. The second mouse click triggers the section creation. In addition, to this
input type, a numerical input with precise section coordinates is possible through string based field using
keyboard. A special option is the input of a vertical section, which is defined by one point and the view-up
vector. Its generation is triggered by a mouse double click action at the cursor location. This approach avoids the
input error of two consecutive clicks at the same location, as in such case the section is not defined, and by
default the vertical section is made at that place.

Figure 140: Cartesian plot of the shock wave location in 3D

The Grid Line distribution is a specific representation for structured topologies. A special dialog-box, see Figure
140 in the upper left corner, is created to handle the interactive aspect for such selection. It can be noticed that
the black surface curve is the grid line, which is almost following the shock wave and has the highest Mach
number distribution as shown above in the Cartesian plot. The difference between the Apply and Accept button is
that Apply button is used to traverse the grid lines for temporary viewing, while Accept assign the curve to the
Cartesian plot. The next two plots in Figure 140 are related to 3D computation, and show the Section tool applied
to the Cutting Plane instance, created bellow the double ellipsoid. The coordinate system representations help the
user in orienting herself/himself within the 3D space and in addition, they verify that the sections are made, in
this example, along the x-axis. In 3D the section line input is the cutting plane, which is perpendicular to the
computer screen. The found intersections between of the cutting plane and the active surfaces appear as 3D
curves on each intersected surface. Their related scalar distributions are mapped to the Cartesian plot.
Figure 141 shows an example of a grid line distribution. To create such representation the user selects an
appropriate constant grid line index, commonly I or J, used for structured surfaces created from I, J or K constant
indices. The I and J indices appear in the dialog-box shown in the lower left corner, and have to be interpreted in
accordance with the rules explained in the boundary condition definition, see section 1.2.5.
The counterpart of the scalar distribution plots along curves or section are the vectors representations along the
constant I or J grid lines, with also the possibility to extract only part of an identified curve. The designed dialogbox is equal to the scalar counterpart and defines the same functionality. The Vector Filed Grid Line and Section
distributions are shown also in Figure 141.

140

Figure 141: Scalar and Vector curve based extractions in 2D

The section oriented input is also used for creating a set of Isolines or Particle traces, as shown in Figure 142.
The section span is breakdown in number of reference points through which the isolines or particle traces have to
pass. On the ship hull a set of three groups of isolines are created for the bulb area showing the pressure field,
while in the upper part three groups of Surface Particle traces are showing the shape of the wave generated by
the ship. What is interesting to notice is that by making the ship hull transparent, we have still the feeling of the
ship surface, while having the possibility to see the hidden part of the mirrored representations.
The algorithm behind the Section representation is based on the marching cell algorithm; see section 1.4.1, which
calculates the vectors at the intersected points of a surface with the plane, which is defined by the active view
normal and the inputted section. The same algorithm is used for the Local Profile representation. The difference
is that the vector representations do not need an additional Cartesian Plot to view the extracted scalars or vectors.
The Local Profile representation modifies the coordinate space metrics by allowing that the Local Profile
coordinate axes are stretchable, thus adjustable by the user. This characteristic allows the magnifying or blow
up effect of the investigated region resulting in a special Cartesian plot representation, which is inserted in the
active view at the inputted section location.

Figure 142: Isolines and Surface Particle Paths in 3D

141

Figure 143: Local profile representation for boundary layer analysis

Thus the Local Profile behaves like a virtual microscope, which is very appreciated by the users investigating the
boundary layer flow patterns. Figure 143 shows an example of the normalized velocity U component compared
with the validation data coming from experiments. The inputted line section localizes the region to be mapped to
the Cartesian plot representation plotted in the view itself. The validation data, as shown in the example, can be
combined with the local profile representation.
The user interaction is equal to the Section input, thus a red rubber band line appears as the visual feedback for
the profile positioning. The coordinate system is created based on the following assumptions:

When the two points are inside the active surface, the position of the origin is defined with the first
selected point.

If one of the points lies outside the surface boundary it is reset to the intersection point with the
surface boundary. The intersected point becomes the origin of the coordinate system.

If both points lies outside the boundary while the section passes through the surface, the point
which has the lowest y coordinate between the two intersections becomes the origin.

The quantity axis in the analyzed region is associated with the abscissa. It is drawn perpendicularly to the y axis.
The quantity axis parameters are modified through the Quantity Axis editor shown on the left side in Figure 143.
The analyzed distance is associated with ordinate axis. It starts from the origin of the coordinate system and ends
at the selected points. The distance axis is modified through the Analyzed Distance editor shown on the right side
in Figure 143, and it is divided into three sections for setting: the Analyzed distance, the Axis length and the
Plotting parameters. The Analyzed distance section modifies the real analyzed distance as follows:

Full Section option extends the analyzed region along the whole mesh. It is similar to the Cartesian
plot representation, but is locally plotted on the view.
Between selected points option analyzes the region between the two selected points, eventually resets
them to the mesh limits. This is the default representation.
On a given distance option requests the analyzed distance in user units. This item is essential as it
allows inspecting very small regions inside the boundary layers. If the given distance is larger than the
full section distance, it is reset to the maximum limit.

142

Figure 144: Vector Local Profiles

The Axis length section modifies the y axis length. It is the distance along which the distribution is plotted:

Not magnified option draws the axis from the origin and ends it at the analyzed distance. In this case
no blow up is performed and the quantity distribution is drawn in real scale.
Between selected points option draws the distance axis between the two selected points. If the first
point lies outside the boundary limits, it will be reset to the closest boundary intersection with the
section line.
In centimeters option requests the y axis length in cm to be printed for example on A4 page.
Analyzed x d option requests the magnify value. The y axis length is given by the product of the
analyzed distance and the magnify value. In this way the user can specify the number of times the
analyzed distance is magnified.
The plotting options section customizes the Cartesian plot scales and appearance. The Vector local profile is
used in the same way as the scalar local profile, with the exception that the quantity axis does not exist in this
representation, see Figure 144. What is interesting to note is that the particle paths are not tangential to the
vectors, which is correct, as the geometrical spaces differs.

2.4.5 Surface based numerical probes

Figure 145: Cutting Plane dialog-box

Surface based numerical probes are the Cutting Plane probe,

which extracts planar geometries and the Isosurface probe,
which extracts geometries with constant scalar value. It is
evident that both tools need 3D geometry and quantity field to
be defined. As indicated, the Isosurface tool is only applicable
to scalar fields, thus it cannot be applied to vector fields, while
the Cutting plane is operational for both types of quantities.
The interaction with the cutting plane tool is provided through
the dialog-box, see Figure 145.

143

Figure 146: Cutting planes representations in 3D

The cutting plane probe provides functionality to slice the 3D computational field in three different modes, as
indicate in the dialog-box: Polygon, Geometry and Quantity. The Polygon mode orients the cutting plane with a
simple solid based visual feedback within the viewed 3D space. The Geometry mode is more complex as it
performs the intersection algorithm throughout the full mesh geometry. The Polygon mode is computationally
the most demanding, as after performing the intersection algorithms, it needs to interpolate quantity fields
values for the new cutting plane and to finish create the smooth color contour, as explained in section 2.4.2. In
addition, the cutting plane dialog-box offers additional functionality for interactive plane positioning, as rotation
and translation buttons related to the stepping parameter for the controlled slicing of the 3D quantity fields. The
important button is the Save button which save the cutting plane for further analysis. After the cutting plane is
saved, it becomes available for interaction like all the other surfaces, see section 2.2. Figure 146 shows an
arbitrary cutting plane of the pressure field around the F-16 airplane. There are three cutting planes created as
constant X, Y and Z planes in which the pressure Contour Thresholds are defined with predefined Colormap
range to make also visible the plane solid shape with the pressure Isolines. The solid surface transparency was
applied in order that the volume perception is improved. Similar effect was achieved by displaying the geometry
in the light gray color, of the previously saved X constant cutting plane, visible in the background. The same
example was used for the vector Sections in the cockpit area in the Y constant cutting plane, see Figure 147.

Figure 147: Cutting plane with Vector representations

144

Figure 148: Several isosurface representations of the Temperature field around the airplane
The Iso-Surface probe is the second tool for the inspection of the 3D volumetric scalar analysis. The isosurface
calculation is based on the marching cube algorithm; see section 1.4.1, which result is an unstructured surface for
which the prescribed scalar value is constant. Figure 148, shows an example of the Temperature field isosurfaces
surfaces around an airplane, where temperature maxima are located in the region of the engine inlet and outlet.
The interesting aspect in this figure is the cut out of the isosurfaces in the mirror plane in order to see the
temperature field development. On the right side of the picture, with the use of transparency and gradual removal
of the airplane solid body, the complete shape of the temperature distribution could be analyzed. The interaction
with the Isosurface probe is supported by two menu items: one for creation and the other for saving of the
generated isosurface. The isosurface scalar value input is entered through the string input field, while the
interactive mouse based input is to move the cursor inside a colormap and with the mouse click select the desired
scalar value, which will trigger the creation of the isosurface. The isosurface algorithm involves the traversal of
the complete computational data input, thus it is computationally intensive for large 3D meshes. Ones the
interesting isosurface is found it can be saved for further manipulations, similar to the cutting plane, and become
one of the other active surfaces available for interactive manipulation. In Figure 149 the combined usage of
different probes is presented.

Figure 149: Combined use of Particle trace, Cutting plane and Isosurface probes

145

Figure 150: Vector Line menus and representations

2.4.6 Vector line numerical probes

The vector line numerical probes generate particle paths by integration algorithm in the vector quantity field, see
section 1.4.5, and obviously do not have their counterpart for the scalar fields analysis. The computed particle
trajectories are modeled as 2D or 3D polylines. As shown in Figure 150, the interactive process is controlled
through the Vector line pull-right menu, which contains the Local Vector Line, Local Section and Grid Line
options. Ones selected, they enable the user to enter the position of the starting (seed) point as the input to the
particle trace algorithm. The additional required parameters are specified through the Parameters dialog-box,
also shown in Figure 150. The user interactively inputs the particle path seed location using the mouse to point
the cursor to the desired location and the clicking triggers the particle path generation. In addition, the exact
numerical coordinates of the desired point can be inputted through String Input field. The explained point
selection is the same, as for the Local Value, Local Isoline or Local Vector representation. As in the case of
these point based tools, to speed up the input, the Section tool is applied to generate a set of seed points evenly
distributed between the two inputted section points. The interactive visual feedback is the rubber band line. The
difference in the user interaction is that the number of desired seed pints can be modified through the String
Input after each execution. For example in Figure 150, each group of colored vector lines is generated from the
Vector Line Section tool. In the same Figure 150, the central dialog-box is used for setting the seed from the
surface grid points, which is only valid for structured topologies because the grid points are linked to I and J
indices, as explained in the definition of structured topologies, see section 1.2.2.
The Vector Line Parameters dialog-box controls the algorithm input (number of global points or number of the
cell points, when computing the particle trace) and the appearance of displayed trajectories like the line color and
thickness. In the 3D computations there are two types of vector line algorithms which can be invoked, see
section 1.4.5. Their main difference is that they a confining the vector line to lie on the surface or to be
positioned inside the volume. The surface vector lines are sometimes associated with the skin-friction
phenomena, as shown in Figure 150 on the surface of the car or along the turbine hub. There is also useful to
draw streamlines in a turbine axial cross-section, as shown nicely the tip vortex location, in the same figure.

146

Figure 151 Surface and 3D Streamlines generation from a cutting plane surface
In Figure 151, the both kind of streamlines are generated for the airflow around the airplane. It is interesting to
notice that the combination of Local Vectors and Surface Streamlines representations show clearly the region of
the wing tip vortex, in a cross-section plane perpendicular to the airplane flight direction. In right side figure,
apart from the standard 3D streamlines strips, it is visible their intersection with a cutting plane instance, from
where they were started. The Vector lines strips where created with the Mono setup, where a unique color is
assigned to the group of vector lines. The Variable color option allows automatic drawing of vector lines, each
with different color. Vector lines applying such coloring scheme improve the visibility of the swirl flow
behavior, as shown in Figure 150 for the rear part of the car. The parameters for the control of the particle path
computation are introduced because there is a possibility that the vector field generates, for example, cyclic
streamlines which computation will never end. To avoid such problems the maximum number of points to be
calculated inside each cell is set by the Cell field. To control the number of integration steps inside each cell the
Cell average parameter guess the number of particle movements to traverse a cell. It is assumed that the cell
traversal will be done, if the particle performs the prescribed number of steps, with the average cell velocity. The
integration direction is shown in the Vector Parameters dialog-box, see Figure 150. In Figure 152, the Tk/Tcl
CFView GUI from Numeca is shown. The red-blue vector lines are created with the downstream integration,
while the yellow and violet with the backward integration from the structured surface nodes. The related dialogboxes, for the generation of the Vector lines representations which aid in the interactive setup, are shown.

Figure 152: Vector lines representations from structured surface points, with the required toolbox in action

147

2.5 User-Centered Approach

The development of an interactive visualization environment has to consider cognitive ergonomics or human
factor when designing the human-computer interaction (HCI) model. The cognitive ergonomics take care of the
user mental processes of perception, memory, reasoning, and reaction aspects, tied to the envisaged interfaces
and interaction techniques, required to be usable and receptive to the user's needs when he/her interacts with the
visualization system. The interactive environment of a scientific visualization system must have a simple userfriendly Graphical User Interface (GUI) having in mind the following reasons: first to be considered, later on to
become useful, and finally to ensure that the end users (CFD engineers or scientists) have an effective software
workspace to work with.
The end user requirements are the key element in modeling the GUI of a visualization system, and they need to
be continuously refined for each software development cycle. When the visualization system reaches its
operational state, the users observations and suggestions are considered with highest priority. Such software
development approach is user-centered as the user directly participates in a development of the GUI model.
When applied iteratively, it makes users contributions effective, and helps the developers to focus on the
pertinent software features by filtering the good design ideas from the bed ones. The important outcome of such
approach is that the involved users are satisfied, as they can experience that their input is continuously taken into
consideration. The users requirements specification has to include enough information to enable the software
developer to propose few solutions for the required functionality in order to have a better feeling for the
envisaged interaction to be designed. This impose an additional constrain to the adopted designed methodology,
as the identified users requirements has to checked for consistency. In the procedure-oriented classical
approach, the required functionality is well known from the initial development phase, as the input and output
operations are static and remain unchanged during the software development. In the interaction-oriented usercentered approach the system functionality is expected to change in an incremental and evolutionary manner.
Thus the scientific visualization system design has to be adapted for a fast prototyping, allowing that the
continuous usability tests are performed in order that the GUI model becomes as intuitive as possible.

Classical approach/procedural systems

User-centered approach/interactive systems

formal problem analysis

analysis of possible user interactions

algorithmic development

development of interactive environment

adding user interface

adding algorithms supporting functionality

Table 30: Comparison of classical and user- centered approach

The design of an interactive system implies the shifts in concepts: from data to visual information processing,
from formal analysis to interactive modeling, from algorithmic to direct manipulation, from problem solvers to
context sensitive environment and from passive support systems to intelligent cooperative system, see Table 30.
In the previous chapter the visualization tools and their possibilities were presented, and some interactive aspects
of the visualization process were mentioned. The interaction modeling is specially related to the described
numerical probes, which require an extensive and highly interactive GUI, in order to be used efficiently. The
following list of requirements is considered, when the interactive model of the visualization system is designed:

The visualization system has to provide efficient handling for large and heterogeneous data sets in a
multi window environment.

148

It has to be 3D graphics enabled in order to provide interactive an efficient: cursor control, visual
feedback and status monitoring.

The visualization tools and numerical probes have to perform fast in order to assure acceptable
interaction response for the variety of visualization tasks, see chapter 2, for example when Isosurfaces
or Streamlines need to be calculated and visualized.

The transparency in manipulating data 2D and/or 3D, structured and/or unstructured, scalars and/or
vectors, original input and/or derived quantities.

The variety of input data formats for 2D and/or 3D, structured and/or unstructured, multiblock
decomposition and connectivity specification, cell-vertex and cell-centered solutions with arbitrary
number of input quantities.

The variety of output data formats for generating image formats as bitmap, postscript and any other
graphics file format output.

A macro subsystem has to automate the user interaction.

A help subsystem has to assist the user to use it correctly.

Integrated databases access to remote data sources.

The designed interactive model supports the users active involvement in visual processing tasks. The main idea
underlying the interactive visualization activity consists in letting the system reacting in a concrete and flexible
manner to the users interactive request. The requirement is that the user controls the visualization experiment in
a quick and productive manner, and that the graphical output of the investigated data takes advantage of the
graphics acceleration hardware, adapted to the interactive manipulation of 3D models. The graphical
presentations have to be compatible with the user mental representations and have to allow the communication
and exchange of such results with minimum effort. If the system functions are not transparent to the user, the
user will either reject the system or he/she will use it incorrectly. This is always an underestimated objective in
the interactive system development planning, as commonly, the software designers apply the procedure-oriented
approach, where the major role is the identification of a well-defined problem around a set of supporting
algorithms.
The interactivity model is the central principle in the user-centered approach, as it establishes the system
dynamics and offers possibility to support variety of users scenarios, by combining a finite set of available
interactive components. The objective is to create intuitive visual human-computer interactions, which follows
the specified concepts and operations to be given to the user and, at the same time that they are adapted to a
specific user knowledge and experience in treating the simulated problem. The interactive environment has to
support effective man-machine cooperation, mainly based on the software capability to complement the user
activity with suitable feedback, as follows:

the recalculation mechanism, which operates on the functional relationship between data introduced
by the user and the application itself,
the model consistency check, which guarantees system robustness at every stage of interaction,
the immediate visual monitoring of the user actions, with useful confirmation indications.
When using an interactive visualization system, the end user can perceive the existing software objects. The
clear example is the visualization system GUI composed of many different interaction tools, where each of these
components can be invoked and manipulated separately. Thus each of them might be identified as Object, for the
end user thing to employ and for the developer thing to develop. This relationship made possible a focused
interaction between the two experts groups, as Object represents a tangible reality for both of them, users and

149

developers. The both groups work on a same object, but they are doing it from two different points of views. The
added value in this process is that both groups are contributing to the development of the same object. Such
approach is essential to the Object-Oriented Methodology (OOM) and tightly relates the user-centered GUI
design with Object-Oriented Programming (OOP).
An interaction modeling is an important activity in a design process of a scientific visualization system. In the
following section the applied modeling principles are described, followed by sections, which specify the details
of interaction behavior (dynamic) and interaction data (static) aspects.

2.6 Interaction Modeling

The objective is to establish a set of common abstractions that allows the modeling of an interaction process. The
adopted modeling approach covers the users interaction on the displayed objects and the dynamic modification
of the related objects parameters. Their design is based on the following characteristics:

the cognitive characteristics must improve the user intuition for specific activity,
the perceptual characteristics applies color, depth, perspective and motion to improve the visual
feedback,
the ergonomic characteristics are concerned with the system usability (easy-to-use) and the learning
phase through feedback monitoring of user actions and help facilities.
The effective use of GUI relies on the intellectual, perceptual and sensorial activities of the user. The GUI design
requires consultation with cognitive scientist, psychologist, ergonomic expert, graphics designer, artist and
application expert to work together in order to understand the complexly of the users tasks and how the user
reacts when performing them. The interaction process needs to be translated into the specification describing a
requested interaction: what the user has to do, and what the system has to respond.
The first phase in GUI design is to learn how the user thinks about the task and how he expects the work to be
done (cognitive issue). The understanding of the user thinking process is the most difficult part during the first
phase of GUI design. It consists of several trials-and-errors cycles, which are usually carried out with a set of
small prototypes.
In the second phase (perceptual issue) the integrated prototype is constructed in order to find an appropriate GUI
that supports the user-centered model, as the integrated GUI invokes functions, gives instructions, and allows
control, and present results without intruding in the visualization process.
The adoption of the system is linked to the third phase (ergonomic issues) that enhances the user-system
interaction space. The elements are the on-line context-sensitive help and the good hard copy print output
possibilities. The main objective is to aid the user in developing the workflow for his/her visualization task to
and to keep him/her aware of the involved data sets and applied algorithms.
The entities, which model the GUI architecture, are: Object, Event, Place, Time, State and Action. See Figure
153, where the ERM diagram of the interaction process is shown. In this diagram there are three important
relationships:
1.

Action in Time results in an Event (Action, Time).

2.

Object is defined with its State and Place.

3.

Object can perform Action, resulting in Object (Action) and

Action can be performed Object, resulting in Action (Object).

150

System

User

Event

S
Operation

Status

Action

Object

Time

Place

Figure 153: ERM of the interaction process

The user Action produces an Event that triggers a system Action. The link between System and User Action is
Event. Each user Action invocation must be uniquely interpreted by the system at every point of the interactive
process.
The analysis of the user requirements, applying the user-centered approach, showed that a multi-window
environment with menus and mouse controls covers most of the user interaction functionality and ease the access
to the application. In order that the user can performed interactions with the system, the following generic types
of user actions are considered to be triggered by the selected input devices, as:

menus
direct manipulation
key binding
macros

(mouse selection)
(mouse picking)
(keyboard)
(file)

In Figure 154, the menu structure is presented hierarchically in order to keep organized the actions around the
similar content. The menu organization follows, from left-to-right, the visualization process: The Project menu is
concerned with data input and output data formats. Geometry menu contains the mesh representation, followed
by different Rendering possibilities. Than the Quantity menu selects scalar or vector field quantity, and the
Representation menu offers creation of different representations possibilities. The View and Update menu are
related to the setup of the viewing and presentation parameters. However, the menu structure is flat, which
means that a respective menu items can be invoked at any level of interaction. The pull down and pull right
menus, together with dialog-boxes enable the user to localize quickly the specific input.
However, the command sequence has certain amount of intelligence build into it. Only the commands that are
consistent with the previous one will be executed. In addition, some commands can be triggered through the
mouse-cursor point and click interaction; as sometimes, it is easier that the user performs a direct mouse
manipulation than that he/her has to traverse all the menus hierarchy to find the envisaged command to execute.
For example to select a surface, it is easier to do it by the mouse picking than to remember the textual name,
which needs to be selected trough the dialog-box interface. The users control of the application by menus or by
mouse has to result in a consistent visual feedback through input or output self-explanatory status monitoring.
The GUI should provide a user-friendly point-and-click interface based on the use of mouse and keyboard,
having access to variety of sliders, buttons, text entries, dialog boxes and menus for a variety of users inputs.

151

Figure 154: The menu structure

As mentioned in several occasions, the interactive procedures need to be ergonomic and effective. The
fundamental user request for an interactive process is the user ability to instruct the application with point-andclick mouse actions and that the application reacts quickly by updating its display. A mouse movement is
connected with the cursor position on the screen. By moving the mouse across the GUI layout the user activates
menu items, selects active points and graphical objects. When the mouse is moved, the area that contains the
cursor becomes active and all the mouse actions are affecting this selection area, for example a View.

Figure 155: CFView GUI layout

152

The developed interface of CFView, see Figure 155, contains GUI components organized with menus, icons and
dialogue boxes. Viewing operations and interactive interrogations of the computational fields are ensured by the
cursor and view manipulation buttons. The general GUI layout is subdivided into different areas, as shown in
Figure 155:

MENU BAR area,

containing TIME, DATE, MEMORY

TOOLBAR area
QUICK ACCESS area
the middle GRAPHICS AREA and
the bottom area, subdivided into following regions:
1. message area
2. string input
3. viewing buttons
4. view monitor
5. cursor monitor
All the mentioned areas are 2D GUI components, except the graphics area, which is the part of the screen,
where the graphics objects appear. The graphics area displays and manipulates the 3D graphics content, which
is managed through specialized graphics objects called Views. One or more views may appear simultaneously
within the graphics area. These views can be positioned arbitrarily, thus, they can be moved, sized and can
overlap each other, see Figure 156.The generation of more than one type of view was analyzed and the
generation of three types of views is adopted:

2D views that are primarily related to 2D projects,

3D views which are related to 3D projects and

Plot views that are used to display data in a Cartesian plot form.

Figure 156: Different view types

153

Figure 157: Evolution of GUI

This GUI model was presented at the AIAA conference in 1992 [96], and at time it was the first of that kind. The
history showed that still today, see Figure 157, seems to be an appropriate and efficient GUI model for scientific
visualization software; the industrial example is the Numeca CFView software [14]. It is interesting to note that
the GUI was ported to variety of UNIX platforms and its successful reimplementation in Tk/Tcl by Numeca
made it even more portable to the Windows and UNIX platforms.
We continue with the explanation of the GUI model, which components are controlling the graphics area and its
multi-view structure.

154

KEY-BINDING
NUMERICAL OR MOUSE
POINT-AND-CLICK INPUT

PULL-RIGHT MENU INVOCATION

DIALOG-BOX INVOCATION

Figure 158: Reminders for different interactive components

Each view is allowed to have a different active state displayed in the view monitor. The views are graphic
windows which enable the user to rotate, zoom and pan using a mouse. The use of multiple views with
corresponding status displayed by the view monitor enhances the presentation of complex geometry and
problematical phenomena. The user is allowed to create additional views by opening a new file or by explicitly
requesting a new view for the same data. Special case is a Cartesian plot view, which is created whenever a new
quantity distribution is requested. Depending on the view type, some menu items and viewing operations may be
disabled (i.e., X and Y rotations are not allowed for 2D views). As visible in the Figure 156, some menu items in
the top menu bar a grayed, which means that for the selected view, only the normally displayed items are active.
This fact indicates that the menu system is content sensitive, which means that it adapts its content to the current
situation of a user interaction.
A menu item invocation can be also handled by typing the key bindings assigned to the menu items and editing
options starting from the menu bar, while the graphics area and viewing buttons have its own mouse bindings.
The menu choices have keyboard short-cuts that allow experience users to quickly invoke different commands.
In addition the menu items contain reminders that come with the menu item name, as indicated in Figure 158.
The designed GUI is context-sensitive, mouse-operated and intuitive and with the available options such GUI
reduces the learning phase to operate the visualization system. The additional configuration file allows the user
to set up the desired preferences, regarding the initial setup of the default parameters.

2.7 Viewing Space and Navigation

Although the scientific visualization system allows any number of views to appear in the graphics area, only the
view selected by a user can be manipulated. Such view is named Active View and the user interactions are
associated to it. The convenient way to the make a view an Active View is to select it with the mouse point-andclick operation. The user positions the cursor on the top of any view and then by clicking the mouse button the
view becomes active. The view is placed on the top of the view stack that contains all the views present in the
graphics area. The newly selected view becomes the Active View and all the menus and view buttons operations
become related to it. The Active View is recognized by red color highlighting its border, see Figure 156.
The viewing space is represents as a cube, which is aligned with the Cartesian coordinate system. Another cube
encompasses all the graphical objects present in the view scene. This cube defines the fitting space and it is used
for the following reason:

155

Figure 159: Cube model for sizing the viewing space

to evaluate the original view, when the view is created,
to fit the graphical objects to the maximum viewing space, without changing the view orientation,
to reset the view orientation, whenever the user has lost the content of the view.
Whenever a new graphical object, for example a surface is added to a view, a view automatically sizes the fitting
space in order that all the graphical objects present in the view are enclosed in it, see Figure 159. The original
view orientation defines the orientation of the view coordinate system according to the fitting space and the view
limits. These view limits define the portion of the view space which is displayed, called a view scene. The
original view displays all the graphical objects with the default view orientation.
All graphic objects lying outside the viewing space are clipped away, thus not displayed. Two clipping planes,
the front and back clipping planes have an important application in 3D. These planes are parallel to the plane of
the screen; and they delimit the viewing space in front and in the rear of the graphical objects that are clipped
away. Parts of the objects may become invisible if these planes are very close to the graphical objects. This
feature is very interesting for looking inside an opaque graphical object. The user can control this feature through
a view button that allows moving these planes continuously forwards or backwards, as shown in Figure 160.

Figure 160: Clipping planes in viewing space

156

Figure 161: Coordinates system and 3D mouse-cursor input

When moving the cursor in 3D, it is not always evident, which interactive point a user gives as input, as the
cursor is not indicating a point but a line perpendicular to the display screen. In order to resolve such situation,
an appropriate way is to display the location of a 3D selected point with the coordinate lines and the sphere
marker, see Figure 161, which visually defines an Active Point representation. The Active Point is by default
constraint to be inside the viewing space, as there is nothing to be selected outside it. In Figure 161 the viewing
space is described with the coordinate system displayed in black and in addition, there is a viewing coordinate
system, which is helping the user to stay oriented when navigating in 3D space. The possibility to input the exact
coordinates is provided with a simple dialog-box, shown in the same figure, which is also is reactive to the
cursor position. This approach is also used during the interactive picking on a surface. In this case, a line passing
through the active point and perpendicular to the screen is defined and used to find its intersection with the
surface. As indicated in the above explanation a view can be decorated with more types of coordinate systems.
The global coordinate system, has its origin always located at the lowest corner of the fitting space and is
graduated, while the other one is the local coordinate system and has its origin at the current active point and is
displayed without graduations. It defines the X, Y and Z rotation axis for the viewing operations. Another usage
of the Active Point is to help the setting of the local coordinate system around which the viewing operations are
defined.
A view can be provided by two types of projections for displaying three-dimensional objects: orthogonal and
perspective (see figure 2). Orthogonal projections are typically used to represent the metric properties of an
object and to show the exact shape of a side parallel to the plane of the screen. The perspective projection gives a
more realistic representation of an object as seen from an observer at a specific position. This position, called the
reference point, can be used to change the appearance of the graphical objects on the screen. The objects will
appear more distorted when the point is placed closer to them. On the other hand, if the reference point is far
from the objects, the projection will look like an orthogonal projection.

157

Figure 162: View projection types

A viewing buttons provide a convenient way to interactively navigate in the 3D space and they are part of the
GUI model, see Figure 163.

Figure 163: Viewing buttons

X, Y and Z are the projection buttons that allow setting the camera position with setting a view normal parallel to
X, Y or Z coordinates axis directions. The mouse-click operation triggers the setup action, also related to the last
two buttons, the Fit Button which fits the scene to the view and affects the camera position parameters: target,
width and height, see Figure 164, and the Origin button which brings the active view in its default viewing state
updating all the camera parameters. The other buttons interaction is associated with mouse press and drag
operations, and it modifies interactively the camera parameters in relation to the mouse movement, which is
proportional to the mouse dragging direction and amplitude. The Scrolling Button allows translating the camera
in a given direction, while the camera direction remains unchanged. Follows the Dynamic Viewing Button, which
allows to rotate and translate the camera position around the target as well as to fix the rotation center, which
model is shown in Figure 165.

Figure 164: Camera model and its viewing space

158

Rotation Buttons allow rotating the camera about the principal coordinate directions X, Y or Z for 3D views, but
it is deactivated for Cartesian Plots views. The Roll Button allows rolling the camera around the view normal
and affects the view up vector direction. The Zoom In/Out Button allows to interactively zooming in and out,
affecting the camera width and height parameters. The Zoom Area Button allows specifying a rectangular area of
the active view to fit to the whole view display area and affects the camera position, target, width and height of
the camera parameters, as shown in Figure 165.

Figure 165: Camera parameters and virtual sphere used for camera rotation

2.8 Visualization Scenarios

The visualization scenario is the set of interactive operations performed by the user, when creating visualization
representations for selected quantity fields. The generated display layout is arranged and presented with a set of
user defined views, realized through a kind of procedural template, usually based on standardized, common, or
scientifically accepted output formats. The presented GUI supports such multi-view configuration, which allows
simultaneous display and manipulation of heterogeneous data sets during the same visualization session. This
feature is particularly useful for comparative data analysis when presenting information through standardized
visualization layouts. In addition, the possibility to present different quantity fields in the same viewing space
facilitates the treatment of multidisciplinary problems, where for example the fluid-structure interaction is
analyzed.
The important issue to achieve this objective is the possibility to scale the geometry in order to unify the applied
units (meter, feet) between different projects. The same approach has to be adopted for the investigated
quantities in order that the quantity units are given in the same units space.
The unification of the content and the application of same visualization tools, still do not provide that the
displayed information is fully comparable. The use of the viewing parameters with a set of editors to adjust the
appearance of the displayed graphics makes possible these final adjustments in order that the displayed data are
equally processed. ones we have defined a visualization scenario with the layout configuration, with the unified
data sets, with the same visualization representations, with the same viewing parameters and with the same
graphics equalizing the decoration and appearance styles, we are in the position to look to different numerical
simulations in a standardize way.

159

Figure 166: Symbolic calculator for the definition of new field quantities
For such purposes, the Symbolic Calculator is an important element of the visualization system which can
involve available quantities to define an algebraic expression, as shown in Figure 166. The Derived quantities
are calculated from the user defined symbolic expression, and they can be field derived quantities (defined in the
whole computational domain) or surface derived quantities (defined only on surfaces). Based on a standard set of
computed quantities: static pressure, static temperature, density, absolute velocity vector field or relative velocity
vector field; different new common Thermodynamical Quantities can be computed, like Mach number, Total
Pressure or Internal Energy. In addition, differential operators, such as Gradient, Divergence or Curl, can be
applied to the existing field quantities, thus resulting in the creation of a new scalar or vector quantity field,
depending on the resulting quantity type resulting from each of the available differential operators, see section
1.3.2.
This process can be automated with a set of macro instructions, which can be invoked through the Macro
subsystem. The Macro subsystem allows the user to record the interactive actions he/her is performing, in order
that later on or next time the user is accessing the file, to automatically reply the performed action. This ability
increases the user efficiency to investigate similar cases without repeating all over again the same set of actions.
Thus, a user is actually becoming a high level programmer, and the macro script is the visualization programs,
which can be used to define different Visualization scenarios.
In the following two examples, see Figure 167, of the standardized output created for the EUROVAL project
[97], the two compared contributions are from Deutsche Airbus and Dornier. The airfoil computations were
validated against the computational data. The grid mesh and pressure field are presented for the whole 2D flow
field around the airfoil, while pressure coefficient, skin friction and displacement thickness distribution along
solid boundary are presented as three Cartesian plots, validated against experimental data. As there were many
European partners contributing, the generated visualization scenario macro has automated the generation of such
presentations. The second example is the visualization scenario for 2D bump test case, as shown in Figure 168,
where the vectors profiles in the boundary layer and the shock wave details with isoline and local values are
presented. The two contributed computations are from University of Manchester (UMIST) and Vrije Universiteit
Brussel (VUB).

160

Figure 167: EUROVAL visualization scenario for the airfoil test case

161

Figure 168: EUROVAL visualization scenario for the Delery and ONERA bump

162

As we can realize, the variety of presentation possibilities offered by the visualization system is large, and it is
necessary to carefully experiment with available possibilities to come up with an appropriate setup. As it was
explained, ones the visualization scenario is found the user interaction with the system can be quite fast and
straight forward, and even automated, for fast outputs as shown for the two EUROVAL examples.
The graphical primitives appearances influence the user visual perception of the displayed data. In Figure 169,
the setting of Node in blue, Edge in red and Face primitive to be transparent, with gray grid lines for reference
are shown. For comparison purpose, the use of these attributes can help to distinguish different data sets, when
presented in the same view.
Additional possibility is the superposing of different views in a transparent mode, which gives an impression of
one view. Such layering mechanism is interesting, as allows generating independent views, and than like putting
transparencies one over other the comparison can be performed, as shown in Figure 170.

Figure 169Setting of graphical primitives

Figure 170: Superposing different views

163

Modifying the colors associated to the scalar quantity can sometimes reveal new insight when analyzing the
scalar fields. Together with the graphical primitives, where each of them can have different colormap associated,
represents a powerful way to present data. This makes possible to manipulate in the same view different
quantities, still keeping the necessary visual distinction for keeping them separate.

Figure 171: Different graphical primitives showing the same scalar field

164

Figure 172: Different colormap of the same scalar field

165

The power element of the Symbolic calculator is to generate new geometries, which are analytical shapes like
sphere, and others. In the Figure 173, some examples are generated and the available quantity fields can be
analyzed in these curvilinear spaces, as the superset of the cutting plane mechanism.

Figure 173: Analytical surfaces generation for comparison purposes

All the mentioned possibilities make the visualization system a versatile tool for comparison purposes.

166

3 Object-Oriented Software Development

This Chapter describes how the Object-Oriented Methodology (OOM) was applied to develop a scientific
visualization system at all phases of the software engineering life-cycle, from problem statement, to analysis,
design, implementation and testing/validation. In contrast to traditional methodologies, like structured systems
analysis and design methodology [98], where functional decomposition is the primary concern and data are
identified later, OOM focuses on the objects and their related data, then on the algorithms built around them. The
present work demonstrates that OOM provides an approach to software development that those stemming from
the traditional methodologies, especially for building software that is expected to evolve, as is the case for SV
systems. The power of OOM becomes evident when facing complex applications which are tricky to develop
using the functional analysis approach. Yet, any structured description of a problem has value, and users are
often casting their views in functional terms (functional requirements). The role of the software developer is then
to integrate these functional requests into the object-oriented model/specification.

3.1 Software Engineering Model

The building of complex systems -- such as scientific visualization systems -- is made possible by breaking down
the whole problem into smaller, tractable and manageable problems that can be worked out by the developer. A
model is an abstract construct that formalizes the design of a system and provides a sound foundation for
implementing it. Several models can be used to represent different views of a system; complementary models
pursue a specific purpose and are used to capture the crucial aspects of the system of interest.
In the software analysis phase we identify and depict the essential aspects of the application, leaving out all
implementation issues. The analysis model is application-domain oriented: it includes descriptions of the objects
tangible to the user. The application-domain model is an input to the design model, where objects will be
designed for the selected computer platform, without consideration of coding and programming aspects. Finally,
the design model will be implemented in a code written in a selected programming language (in our case, in
C++).

ANALYSIS
The analysis begins with the problem statement expressed by the user.

Problem Statement:

The analysis model is a concise, precise abstraction of what the desired system
must do, not of how it will do it. The analysis model is to be understood, reviewed
and agreed upon by application domain experts (who are not computer-science
experts/programmers). This process leads to the extension of the user model data.

DESIGN
System design:

System designs include system partitioning into subsystems.

Object design:

Object design augments the content of the analysis model. Design decisions
include: specifying algorithms, assigning functionality to objects, introducing
internal objects to avoid re-computation, and optimization. The emphasis is on
essential object properties, so as to force the developer to construct cleaner, more
generic and re-usable objects.

IMPLEMENTATION
The overall system architecture defined by the design model is a tradeoff between the analytical model and the
target computer platform. Some classes whose properties do not derive from the real world -- for instance Set or

167

Vectors which support specific algorithms -- are introduced as an auxiliary part in the design model. The
implementation style must enhance readability, reusability and maintainability of the source code.
The models and the source code constitute together the software solution; they are the answer to the question
WHY is the software system created? When developing a visualization system, several types of OOM models
and diagrams are used to address three basic questions:

1.

WHAT?
The static model describes objects and their relationships by entity-relationship diagrams, a visual
representation of the objects in a system: their identity, their relationships to other objects, their attributes,
and their operations. The static model provides a reference framework into which to place the dynamic
model and the functional model. The object model describes classes arranged into hierarchies that share
common structures and behaviors. Classes define the attributes and the operations which each object has and
performs/undergoes.

2.

WHEN?
The dynamic model describes the interactive and control aspects of the system by state diagrams. This
model describes the time-dependent system characteristics and the sequences of operations regardless of the
nature or mechanism of the operation. Actions in the state diagram correspond to functions in the functional
model. Events in the state diagram become class operations in the object model.

3.

HOW?
The functional model describes data transformations in the system by data flow diagrams. The functional
model captures the systems functionality. Functions are invoked as actions in the dynamic model and are
shown as operations on objects in the object model.

The static model represents the reference base, because it describes what is changing or transforming before
describing when and/or how changes are done. Successful software engineering requires a number of very
different technical skills to satisfy research and industry needs. They include the ability to do analysis, system
design, programming design, coding, integration, testing and maintenance. OOM requires self-consistency and
sense of purpose. The experience with OOM is not only based on a set of techniques but also on their
interactions. That means that all of them work well together - much akin to the mathematical identification of
simplicity and beauty. In order to use various notations and techniques in a balanced way, OOM focuses to
achieve elegant design and implementation, which outcome is expected to be a comprehensive, implement-able,
efficient, maintainable and extendable code.
Software engineering is the compilation of different activities, which have to follow and overlap each other in
the cyclic and iterative manner. The class design should be a separate activity from the class implementation.
Thinking only of a class design in an abstract manner can lead to a design that is impossible to implement. The
balancing of these two thoughts is to the responsibility of an OO software engineer. The iterative process of the
methodology improvement and the iterative process of software development are advancing in parallel.
Consequently, the developed methodology extends beyond a single software project. Therefore it is not always
possible to determine the effects of methodology crafting at the starts of the project, as the methodology evolves
over the time in response to the acquired experience and this process is usually accompanies with development
problems.

168

Figure 174: Comparison of the traditional and object-oriented software development life-cycle
The software engineering methodology is an integrated combination of concepts, guidelines, steps and
deliverables in the context of an underlying process description, includes not only graphical notations but also
textual descriptions and documentation standards. The methodology encompasses a large integrated set of
development techniques and tools resulting in procedures as
Debugging codes and algorithms to support development,
Simulation results for industrial applications,
New visualization patterns to accomplish research.
Techniques are developed to improve software analysis and design possibilities:
Management techniques comprehends planning, organizational structure (hierarchy of abstraction
levels), deliverables (their description and timing) and quality control.
Design techniques provide means for verifying the design before coding.
The library management provides support to the OO approach by search and query possibilities of existing
reusable classes, when developing new classes. Tools supporting OO methodology provide notation, browsing
and annotation capabilities. Ideally, the OO design tools should provide mechanisms for navigation from higher
level OO diagrams to code and back to the design. This functionality is called forward and reverse software
engineering relating the developed code with design diagrams. Software engineering must improve both:
software quality (the product) and software production (the process). As B. Mayer says [4], there exists different
quality factors, as follows:

ROBUSTNESS

is the ability of software products to exactly perform their tasks, as defined by

requirements and specification.
is the ability of software products to function even in abnormal conditions.

EXTENSIBILITY

is the ease with which software products may be adapted to changes of specification.

REUSABILITY

is the ability of software products to be reused, in whole or in part, for a new application.

COMPATIBILITY

is the ease with which software products may be combined with others.

EFFICIENCY

is the skilful use of hardware resources, processors, external and internal memories and
communication devices, minimizing the resources and improving the performance.
is the ease with which products may be transferred to various hardware and software
platforms.
is the ease of preparing acceptance procedures, particularly test data, and procedures for
detecting failures and tracing them to errors during the validation and operation phases.
is the ease of learning how to use software systems, operating them, preparing input
data, interpreting results and recovering from usage errors.

CORRECTNESS

PORTABILITY
VERIFIABILITY
EASE OF USE

Table 31: Software quality factors

169

3.2 Object Oriented Concepts

The software development based on Object Oriented Methodology OOM stimulates the software designer to
declare what a selected software component has to do, while the application of software components assure how
the expected functionality is/will be accomplished. The software design defines a collection of software
components the objects --, each of which encapsulates a specific part of the designed know-how to be
implemented. When some of the needed software components are available, the task of the designer is
considerably facilitated, since he/she can incorporate them in his/her design. Before starting implementing
objects, one must analyze all objects and make a trade-off between ones that exist and the ones that need to be
implemented.
Object modeling in OOM is expressed through the objects declaration and definition specification, which
provides the level of data abstraction and encapsulation necessary to be defined for considering the object reuse
[99, 100]. The designers skill is revealed in the way he/she re-uses existing objects in combination with the
newly-created ones that need to be developed to suit the software design specification. Re-using (validated)
software elements is one of the fundamental principles/benefits of OOM: this contributes to improving the
quality of the software by reducing the amount of programming and the risk of errors. Indeed, objects which are
re-used are validated in many independent application contexts and continuously improved. Creating software
objects is a difficult, time-consuming and error-prone task, because the designer must take into account
application-domain requirements, which are usually broader than single-application requirements; clearly, being
able to reuse existing objects (code) is a plus in any software development project.
At the start of a software project, it is unlikely that all objects of interest are identified, complete and available.
Building an object-based application, creating new objects and re-using existing objects are slow and costly
processes -- compared to the traditional ways of building procedural software. One of the objectives of this thesis
is to describe and encourage the use of OOM, which prescribes the necessary and sufficient conditions for betterquality software production.
Software may be seen as a mass of tiny, uninteresting detailed elements which interact over a wide range of
operational conditions; no written documentation can describe (complex) software completely (what, why or
how), and this is true for complex systems like SV software. When developing software, the designer commonly
combines two contradictory methods, namely synthesis and analysis. For example, when writing a text chapter,
he/she applies synthesis; when reviewing the text, he/she relies on analysis; all together, writing and reviewing
form an iterative process which continuously combines synthesis and analysis. When writing software code, the
objective is to produce a text (the source code) which is sufficiently readable to be analyzed. To be able to reuse components, the source codes need to be sufficiently readable and precise for object selection and re-use,
whilst retaining enough flexibility for possible software extension. The application of OOM implies that the
knowledge encapsulated in objects grows constantly, which enables the emergence of new systems, platforms
and applications.
The selection of tangible objects is clearly a key factor for successful OO software design. Nothing definitive
can be said about the right way to choose objects: this is an acquired skill. Selecting different collections of
objects would lead to different bases for developing and extending a given application. Where new objects are
required, they are usually created from existing correct ones, which prevent error-prone situations (such as
unpredictable/erroneous inputs) and ensure the correct trapping and treatment of anomalous situations. This
significantly unloads the designers from basic concerns. In contrast to OOM, procedural methods are direct and
focused on the processing aspects; their strength is their ability to deliver high-performance software and their
weakness is in their low level of productivity and in the fact that they produce software that is ill-suited to
evolution/modification. Implicit in OOM is the ability to check the logical correctness of the software, which
workload is of the order of magnitude larger than the workload to improve software performance. OO software
170

tested under a large variety of operating conditions is expected (very likely) to work correctly in a new
situation/application.
To summarize: OOM was selected as the appropriate methodology to develop our VS system because OOM:

allows the developer to model real-world problems abstractly in terms of objects,

leads the developer to think about the problem of interest in terms of the application itself, not in
terms of computer concepts; it helps the developer to think without using programming concepts,
requires the full understanding of the application requirements and promotes a design that is clean
and implementation-independent,
defers all implementation details until later stages, which preserves coding flexibility,
builds upon a set of concepts and graphical notations which is used at all stages of the software
development process, in contrast to other methodologies which use different notations at each
development stage. The concepts and notations provide useful documentation during software
construction,
provides practical and effective instruments to develop, integrate, maintain and enhance software.
The fundamental concept of OOM is clearly the object, an entity which combines both data structure and
behavior. An advantage is that the application-domain and computer-domain objects are designed and
implemented using the same concepts and notations. In an OO approach, the software is organized as a
structured discrete collection of objects which incorporate state and behavior information. That is in contrast to
conventional functional-oriented FO approaches (also called procedural or structural approaches) where state
and behavior are uncoupled or at best loosely-coupled. The OO approach places greater emphasis on data
structures and lesser emphasis on procedural structures than the FO approach. The term object-oriented (OO) is
an adjective employed to qualify all phases of the software development cycle, as in OO analysis, OO design
and OO implementation [101]. In all development phases, we will be specifying the problem/solution domain
in terms of OO concepts such as:

Objects and Classes,

Messages and Methods,
Inheritance,
Polymorphism.
In the OO approach, all entities of the problem/solution domain are represented as objects. Although the OO
approach is based on a few simple concepts, applying them is not seen as being simple, probably because most
developers have inherited a procedural (or functional) way of thinking, a natural by-product of the algorithmic
logic applied in all engineering disciplines.
A short overview of the object-oriented concepts is given in the following sections.

3.2.1 Objects and Classes

An object is an entity possessing a state and behavior, see Figure 175, which describe its nature and functionality
[100, 102]. Objects can have analogies in real-world things (e.g. house, ship, and aircraft) or in mathematical
models (e.g. number, point, surface). When similar objects are grouped together, the Abstract Data Type (ADT)
can be defined, to encapsulate a set of common methods and states, accessed and modified only through a
message passing mechanism. In addition, the hierarchical structure of the ADT model allows for extensibility.
The Object Identity means that ADT are implemented into discrete and distinguishable objects, for example, a
surface of a mesh or a node of a cell. Each such object instance has its own intrinsic identity. Thus, two objects
are distinct even if all their attributes values are identical. In real world objects simply physically exist, while in
software, they exist if they can be uniquely accessed and referenced: by an address, array index or unique value
of an attribute.

171

MESSAGES
METHODS

ATTRIBUTES
STATE

BEHAVIOUR
INTERFACE

Figure 175: Object concept

Objects that are inside an application are identifiable, can refer too, or contain other objects. Object identity
concept is implemented with data types and pointers in OO programming languages, foreign keys in data bases
management systems, and file names in operating systems. The object identity is a unique way to access an
object by:
1. Memory reference / address
2. User-specified name
3. Identifier keys in the collection.
1. Memory reference / address
The Memory Address of the object is an external application mechanism to relate to an object, and therefore it is
application dependent. Identity has to provide a unique identification of the object and, therefore, must be an
independent feature internal to an object. Without object identity it is impossible to assign self-contained objects
to the class attributes or instance variables, moreover to let the same object be part of multiple objects. An object
identity makes possible to distinguish one object from another.
2. Object naming
The objects are identified inside the application by names. The problem rises when a specific object has more
names, thus it becomes more complicated to find out if a name identify the same object.
3. Identifier keys in the collection.
An object can be identified by a unique identity key name. For example, to access the instance of the color
contour representation regarding the scalar surface one can identify it with the name contour shaded. The
dictionary class that is used to collect all the scalar surface representations supports the functionality to extract
the object associated with the contour shaded label. This identity key name is uniquely associated with each
object inside the dictionary. The variability of the key name is not allowed because that can become a serious
problem when extending or replacing names in a consistent manner by giving the possibility to access the objects
in more than one way. The distinction of the objects content versus the object identity is fundamental in object
oriented programming, for example, when performing the following basic tests:

The identity test checks weather two objects are the same ones.
The equality test checks weather the content of the two examined objects is equal.

The state of the object is represented by a set of values assigned to its attributes, also called instance variables.
For example, the point can have coordinate values (5, 7). These values represent the state of the point objects.
The behavior of the object is represented by the set of methods (procedures, operations, functions), which
operate on its state. For example, the point can be moved. The method "move" represents the point behavior and
affects the point state (coordinate values).

172

ABSTRACT DATA TYPE

SPECIFICATION
(INTERFACE-MESSAGES)
SYNTAX

SEMANTICS

IMPLEMENTATION
REPRESENTATION
(ATTRIBUTES-STATE)

ALGORITHMS
(METHODS)

Figure 176: Abstract data type structure

The classification means that objects with the same data structure (attribute, state variables) and behavior
(messages, operations) are grouped into classes. A class is an abstraction that describes relevant properties and
ignores the rest. Each object is an instance of its class. Each instance has its own attribute value but shares the
same attribute names and operations with other class instances. The same classes design can be used without a
change in notation, through the complete software development lifecycle; although at different development
phases the encompassed details can vary. A class is a group or classification of objects that share the same state
and behavior. The class has to provide the abstract part of knowledge, with relatively stable meaning in various
contexts, tangible to the end users (e.g. CFD engineers). The class should be a clean implementation of ADT,
defined through the processes of data abstraction. Data abstraction is the process of specification and
implementation of ADT, see Figure 176. The specification defines what the ADT is able to do, which is in the
contrast with the implementation defining how the detailed implementation is done.
Data Abstraction consists on focusing on the essential, inherent aspects of an object, before deciding on how to
implement it, in order to avoid premature commitments to details. The abstraction process highlights the reasons
of the object existence, by determining what is important and what not. ADT attributes exist only through the
provided sets of messages offered by the ADT interface mechanism, which clearly separated the designer and the
user concern of the implemented ADT.
The term object is often used to identify a class instance. An object is an instance of one and only one class. The
objects of the same class share the same behavior. Data abstraction permits each class to be designed,
implemented, understood and modified independently, localizing the effect of any change made. Each object is:
is the instance of a class
has a state, made up of the values to its instance variables.
The state is captured in the instance variables, while the behavior is captured in the methods. They can be
accessed through messages directed to the object instances, and on which the object instances react. This is
assured through the object identity, which is established when the object is created. The state of the object can
change during the application session, but the identity remains unchanged throughout its life time. Identity is
supported in C++ with pointers, as unique identifiers associated with each and every object. In C++ language
syntax private and public keywords are used to separate the interface from the implementation parts of the
class declaration. The interface is the part which a class user can access, while the implementation is the part
which implements the interface functionality. Normally a class declaration is describing the class interface. In
C++ there is a possibility to deny access to some parts of the declaration and, at the same time, to speed up the
execution of some simple and frequently used functionality can be achieved by the in-lining mechanism.

3.2.2 Messages and Methods

An ADT can be seen as an object instance, whose state is only accessible through the class interface, see Figure
175. The class interface consists of the list of messages, that the object is capable to process. A message
represents a condensed notation for invoking the appropriate method. For example, an integer data type in

173

FORTRAN is an ADT as for example, to modify its integer value, an assignment operator (message) has to be
invoked, (e.g. I=5).
The message/method separation is present in ADT specification and implementation, see Figure 176. The
message syntax specifies the rules how to invoke the class methods. The message semantics specifies the
actions, which simulate the class behavior. The detailed definition of the message semantics is essential for the
method implementation of the class (ADT) [100]. The application of a message to an object is called sending the
message. The object class must find an appropriate method to handle the message. The message passing
mechanism allows method invocations, consisting of the object identification followed by the message. The
message can include any other parameter needed in method execution (e.g. aircraft-fly, point-move(x,y)),
therefore the message passing mechanism provides a consistent way of communication between objects, see
Figure 177.

Encapsulation or Information Hiding consists of separating the external aspects of an object, which are
accessible to other objects, from the internal implementation details of the objects, which are hidden from other
objects. The implementation of an object can be modified without the need to modify the application that uses it.
Thus, the encapsulation restricts the propagation of the side effects of small modifications. OOPL makes
encapsulation more powerful and cleaner than conventional languages that separate data structure and behavior.
Information hiding, or encapsulation, is the principle of hiding internal data representation and details of
implementation of ADT, allowing the access through a predefined interface. The interface, represented by a
limited number of messages, reduces the interdependences between objects, as each one can only access other
objects through their interface (messages). The interface is ensuring that some of the class attributes and methods
can not be corrupted from outside. An object has three types of properties:

a set of messages, interface, to which the object can respond,

a set of attributes, instance variables, which only the object can access,
a set of methods, which are invoked when message is send to the object.
MESSAGES
METHODS
MOVE
RESET
POINT
X,Y
X

STATE
BEHAVIOUR
INTERFACE

Figure 177: Point object

3.2.3 Inheritance
Inheritance is the sharing of attributes and operations among classes based on hierarchical relationship. Each
subclass incorporates or inherits the properties of its subclass and adds its own unique properties. The ability to
factor out common properties of several classes into a common super-class can greatly reduce repetition within
designs and programs and is one of the main advantages of an OO system. The sharing of code using inheritance
is one of the main advantages of OOPL. In the procedural approach we would have two separated hierarchies:
one for data structure and another for procedures structure. In the OO approach we have one unified hierarchy.

174

More important, then eliminating the code redundancy is the conceptual clarity in recognition that different
methods are related to the same data structure, what largely reduce the number of cases which need to be
specified and implemented.

Inheritance is a partial ordering of classes when the relationship of inclusions is applied among some of their
properties. Ordering is usually achieved hierarchically from generic abstract super-classes at the root to subclasses of greater specialization and tangibility, see Figure 178.
A new subclass is constructed from an existing one, conceptually related super class, by inheritance, specifying
only the difference between them. The inheritance can extend or restrict the features of the super class in order to
create the specialized subclass. Inheritance therefore represents the realization of the
generalization/specialization principle used to create new subclasses incrementally from existing less specialized
super classes.
A class can have multiple number of subclasses and super classes. If the class is having only one super class we
are speaking about single inheritance, see Figure 178a. The more complex inheritance is the multiple
inheritance, see Figure 178b, when a class is allowed to have more super classes. Two different approaches to
apply inheritance are based on ADT decomposition, see Figure 176:

specification inheritance (dynamic binding) is used as a mechanism of sharing messages,

common interface (syntax and semantics specification) between conceptually related classes
when implementation is not comprehended,
implementation inheritance (static binding) is used as a mechanism of sharing state, attributes
(data representation) and methods (algorithms) implementation of objects.
An inheriting object typically acquires the implementation of its super class and can specialize by adding the
appropriate functionality. Specification inheritance is more restrictive then implementation inheritance and
results in conceptually tight hierarchies. These forms of inheritance have contradictory objectives in many ways,
and consequently it is dangerous to attempt their simultaneous use.
C++ makes it possible to construct systems using both inheritance mechanisms. To efficiently use the hierarchy,
mechanism programmers need to be able to work effectively along the top layers of the hierarchy without going
into derived classes. This process enforces the selection of the appropriated class operation at run-time, which is
in contradiction with data type checks and performance. OOP enables that the run-time support is activated
where the compile-time support is not possible.
generalization
Object A

Object A

Object B

Object C

Object B

Object C
specialization

[a] single inheritance

[b] multiple inheritance

Figure 178: Single and multiple inheritance

175

The run-time identification together with inheritance provides a form of polymorphism that gives to the
designers and programmers a flexible construction methodology to create software components, which are
reflecting the Object Oriented Analysis and Design concept with the straight forward counterparts in OOP with
C++ as follows:

Inheritance mechanism allowing multiple base classes

Calling mechanism through virtual functions invocation
Virtual base classes enabling: Encapsulation, Type Checking and Data hiding.

Good software design is the result of well trained and insightful thought process of its architects, and inheritance
is one of the well adapted mechanisms to promote software reuse.

3.2.4 Polymorphism
Another, equally important OO concept is Polymorphism, which is much less understood and appreciated than
Inheritance. The term Polymorphism means to have many forms, and in software construction Polymorphism
means that the same message can be sent to different class instances. As mentioned previously, a specific
implementation of a message is a method. If a message is polymorphic, it may have more than one method
implementing it. An operation is an action or a transformation that an object performs or it is subjected to.
Polymorphism is the ability to define different kind of objects (classes), which support a common interface
(messages). Thus, objects with quite different behavior may expose the same interface (e.g. aircraft-move, pointmove). There are two types of polymorphism in OO approach, see Figure 179:

polymorphic object, where receivers of the message can be objects from different classes,
polymorphic message, where the same message can be invoked with different number and/or
types of arguments.
Example for polymorphic object: different classes as Aircraft and Point have the same message move inherited
from the super class Object. Both classes Point and Aircraft redefine the method move, which is executed when
the message move is sent to them. Here the specification inheritance (dynamic binding) mechanism is related to
polymorphism. Polymorphism is essential for the implementation of a loosely coupled collection of objects,
which classes are not known until they are identified by the program at run-time. Thus, the message move can
be applied to a collection of objects without knowing if the object is Point or a whole Aircraft.
Example for polymorphic message: class Point can have two messages. First, move(x), is a one argument
message and second, move(x,y), is a two argument message. In the first case the point is moved just in x
direction, while in the second case the point is moved in x-y direction. Thus, it is obvious, that the application of
the same message has two different behaviors.

move
Aircraft

move (x)

move

move (x,y)
Point

Point

[a] polymorphic objects

[b] polymorphic message

Figure 179: Polymorphism

176

3.3 Object Oriented Programming in C++

Any OOP language must provide the data abstraction, inheritance and polymorphism mechanisms required by
the OOP paradigm [36, 103, 104]. Each of these OO concepts may be used independently, but when applied
simultaneously they nicely complement each other. When creating an object, one aims not only at fulfilling the
requirements of an application, but at establishing the complete functional set that justifies the existence of the
object independently of the application. An OOP language like C++ enables the developer to implement a class
by specifying the interface and to encapsulate the algorithmic knowledge within its class methods.
Any implementation of an interactive visualization system comprise of the following elements:
A set of abstract data types that represent information corresponding to certain real or abstract
constructs/concepts;
A set of operations on these data types;
The C++ implementation that allows interaction of these data types.
Scientific VS has to implement general-geometric objects such as curves, surfaces and bodies, organized within
geometrical category, which groups together one of the key data-types. The geometrical category provides
operations that create new instances, which are queried, modified and processed by these classes.
Using C++ as a programming language yields the following benefits:
High efficiency: C++ is much more efficient than other OOPL, for example Smalltalk.
Industry-standard, production code is performing, as codes written in C or FORTRAN.
C++ code is easier to maintain.
Using C++ imposes a discipline on the programmer who must specify his/her implementation as a set of classes.
The programmer is encouraged to find clean ways to solve problems when using C++. Also, C++ code
organization simplifies the work of the programmer who has to maintain or upgrade the code.
In itself, the possibility of re-using CFView classes encapsulating various functionality sets was a sufficient
reason to move to C++. In CFView, every significant element of the users interface part is mapped to an object,
which implementation is driven by the GUI object architecture based on Model-View-Controller design, as
explained later on in section 3.6. These objects are spread throughout the CFView code and they do simplify
many implementation problems. A Category is a collection of classes that capture abstractions for a specific
layer, because it is important to cultivate these key abstractions, as they represent the interfacing mechanism
between classes, why its specific method implementation might change more frequently. Objects are natural and
comfortable to use even when other benefits are not obvious.
The inheritance property suggests that it might be possible to turn each specific class into a hierarchy in which
the general abstraction is captured in the base class. It is commonly observed that building good inheritance
hierarchies requires some forethought, while retrofitting of a generalization can be painful. The application
framework is a set of classes that provide basic components for the application assembling.
When a modification has to be done to C++ code, most programmers tend to stop and think. This may mean that
the change will take a little longer than expected with other programming languages but a main benefit is the
resulting code clarity, which is retained and rewards a programmer, especially the next time he/her has to look at
the affected code.
The quality of software architecture is preserved in the abstractions captured in the C++ language classes.
The Naming convention includes:
1. class naming,
2. member function,
3. order of arguments,
4. abbreviations of common words.

177

The objectives of the naming convention are:

1. to eliminate name conflict during linking,
2. to provide quick identification of class and class member, (data or function),
3. to provide syntactic cues to distinguish between functions,
4. to provide a way for deriving the function name from a desired functionality.
The naming is based on three concepts:
1. an upper case letter for class names,
2. a lower case letter for class data members, and
3. a lower letter for class member function:
The name of a member function might involve three parts:
1) <result object>

<name>

2) <object involved>

<name>

3) <functionality descriptor>

<verb>

The second part <object involved>identifies an object, which is manipulated and sometimes returned, for
example the field Find member function represents a use of such naming convention. The object name is omitted
in the member functions, as the function operates on an instance itself. The order of arguments in the member
functions calls for input arguments to appear before output arguments. The <functionality descriptor> describes
the operators like Find, Insert and Make. As earlier stated, these rules are used to facilitate the reuse of the
designed class.

OO programming did not begin with the C++ language: it started in the 1960s with the development of
SIMULA, a language developed at the Norwegian Computer Center for the purpose of simulating real-world
processes. SIMULA pioneered the concepts of classes, objects and abstract data-types. This was followed by
LISP/Flavors and SMALLTALK in the 1970s, and several other object-oriented implementations. However,
these were languages mainly used in the research environments. It was not until the late 80s that OOP begun to
gain momentum and C++ started to be recognized as the powerful language that is still today.
The ancestor of OOP languages, SIMULA 67, was conceived for developing simulation software. The first
complete, stand-alone object-oriented development systems were built in the early 70s by XEROX at its Palo
Alto Research Center. The main aim of Xeroxs first research on SMALLTALK systems was to improve the
communication between human beings and computers in the Dynabook project [31].

Artificial intelligence research -- especially expert systems -- had also a strong influence on object-orientation.
The developers of the time were guided by concepts like those of Marvin Minsky, first described in his framepaper and later summarized in his The Society of Mind [105]. Minsky and other authors explained the basic
mechanisms of human problem-solving using frames, a concept well-suited to computer implementation. This
explains the similarities between classes in systems like SMALLTALK [31] and LOOPS [106], and units in KEE
[107].
Programming methodologies are used to structure the implementation process. The current state-of-the-art
suggests that OOP results in greater benefits in implementation than other methodologies (e.g. structured
programming), namely:
simplification of the programming process,
increase in software productivity,
delivery of software code of higher quality.

OOP is a set of techniques for designing applications that are reusable, extensible, scalable and portable across
different computers platforms. The purpose of designing re-usable code is to lower the time (and cost) of
producing software. Experience shows that, no matter how well specified the application requirements are, some

178

requirements are bound to change and this results in the need for further development. The requirements change
frequently for the user interface and for the system environment/platform, while they tend to remain quite stable
for the underlying algorithms. Designing extensible code facilitates the changes that are necessary to meet
modified requirements. Designing reusable code ensures that code used in today's applications can be applied in
tomorrow's applications with minimum modifications.
To implement an abstract solution well-specified in terms of OO concepts and constructs, we need a
programming language with rich semantics, i.e. which can directly express the solution in a simple syntax. OOP
is a methodology for writing and packaging software using an OO programming language. C++ is an extension
of the C language which includes OOP support features; it was developed at AT&T in the early 1980s [30, 108,
109]. OO concepts have been well integrated into C++, and a very little amount of new syntax has to be learned
when moving from C to C++.
The fundamental OOP features supported by C++ are:
data abstraction and encapsulation,
message passing,
inheritance.
In software engineering, and particularly in OOP, it is important to manage the complexity of implementation;
OOP requires a software-development environment which comprises the following components:
text editor,
class, data and file browser,
compiler and linker,
debugger,
class libraries.
The programmers creativity is best enabled when the development environment is accessible through an
integrated user interface giving total control over all aspects of the implementation process. Such an environment
also constrains the developers moves and code must be produced and changed in a disciplined, consistent way.

OOP requires browsing techniques in order to identify the classes and their interactions. Software productivity
and quality are significantly improved in a development environment that can display class relationship
information and is able to present just enough necessary and sufficient data for the developer to manage the
complexity of the code and of the coding process. The complexity is usually is layered through class categories
and hidden through the inheritance mechanism. The real benefit to the programmer is that, code changes are
bounded a class when single functional change is done. To locate the code influenced by a change, the class
including the functionality is the basis for the scope of the change. If the class is a derived one, the tool must be
able to show all the functions that it has access to, not just the ones that it defines. An integrated development
environment must help to direct the developer on what he/she wants to do rather than on how to do it.
A class library is composed of header files and a library. The header files with extension .h can be located in
the include directory. They contain all the information needed by the programmer for using the class. The
files that contain the implementation of class methods (member functions) are archived in the library in compiled
form. The library can be located in the lib directory with the .a extension.

3.4 Example of Object-Oriented Software Development

It is obvious that all these concepts and techniques can not be grasped, mastered and successfully applied
without practice. The developers experience is very important in considering all the factors that influence the
implementation process and in balancing them to achieve optimum trade-off. The example below -- visualization
of a streamline -- illustrates how OOP is applied to software requirements analysis and development.
179

3.4.1 Requirements, Analysis and Design Example

Requirements analysis typically begins with a narrative description of the problem domain. We consider here
the problem of how to visualize a streamline. This example is simplified to best explain the different modeling
techniques used to tackle the problem.
The problem is to visualize the streamline that passes through the point that the user has picked on the screen
with a mouse click. The streamline can be associated to the pressure distribution. The user can ask for the
streamline to be displayed; he/she can ask for the streamline to be colored to reflect the pressure values along the
streamline.
Narrative problem descriptions are the source of many difficulties in OO analysis because the identification of
the objects and their behaviors are not explicit; several standard software modeling tools must be applied to
correctly identify and describe the objects of interest, including:
GRAPHICAL MODELING TOOLS:

Data flow diagram (DFD), for functional modeling (see Figure 180),
Entity-relationships diagram (ERD), for data modeling (see Figure 181),
State-transition diagram (STD), for the modeling of interactive system behavior (see Figure 182);

TEXTUAL MODELING TOOLS:

Data dictionary, for data specification.

Process specification.

These modeling tools identify different characteristics of the same object; these must uniquely represent the
object and prove the necessity of the objects existence. The DFD is a graphical representation of the functional
decomposition of the system in terms of processes and data; it consists of (see Figure 180):
processes, shown as "bubbles", (e.g. calculate streamline),

data flows, shown as curved lines that interconnect the processes, (e.g. streamline geometry),

data stores, shown as parallel lines and which exists as files or databases, (e.g. velocity field),

terminators, shown as rectangular boxes. They show the external devices with which the system
communicates (e.g. screen, mouse).

Figure 180: DFD of the streamline example

180

The data flow diagram identifies the major functional components but does not provide any details on these
components. The textual modeling tools -- the data dictionary and the process specification -- respectively
provide details on the data structures and data transformations. For example, the Point data dictionary and the
Process 3 specification are as follows:
Point:
point = coordinate x + coordinate y
coordinate x = real number
coordinate y = real number
real number = [-10-6 - +106]
Process
1.
2.
3.
4.

3:
Find the cell of the first streamline point.
Interpolate the pressure.
Search for the next streamline point inside the cell.
If point found, interpolate the pressure,
else continue the search through neighbor cells.
5. repeat actions 3-4 for all streamline points.

The DFD is a useful tool for modeling the functions but it says little or nothing about data relationships. The
data-stores and the terminators in the DFD show the existence of one or more groups of data. One needs to know
in detail what data is contained in the data-stores and terminators, and what relationships exist between them.
The ERD tool is used to model these aspects of the system and is well-suited to perform an OO analysis of the
data and of their relationships.
An ERD comprises as main components (see Figure 181):

entities, shown as rectangular boxes, each representing one or more attributes, (e.g. Point, Curve),

relationships, shown as diamond-shaped boxes, each representing a set of connections, or

associations, (e.g. HAS, Streamline HAS Nodes),

attributes, shown as ellipses. They cannot contain entities, (e.g. Point coordinates).

Figure 181: The partial ERD of the streamline example.

Each relationship is expressed as a multiplicity statement, which can be one of three possibilities:
one-to-one

1<>1,

one-to-many

1<>M,

many-to-many

M<>M.

In Figure 181, one Streamline has many Nodes. The relationship is HAS, and the multiplicity of the relationship
is one-to-many, 1<>M.
A third aspect of the system that needs to be described is its time-dependent (real-time, interactive) behavior.
This behavior can be modeled by a state-transition diagram using sequences which show the order in which data
will be accessed and functions performed. To model the streamline example with a STD (see Figure 182), we
need to add the conditions that cause changes of state and the actions that the system will take in response to the
changes of state (e.g. click of mouse button -> get coordinates). A condition is an event in the external
environment that the system can detect (e.g. mouse movement, clicking of mouse button).

181

Figure 182: STD of the streamline example

For example in Figure 182, the system is waiting for point coordinates. Pressing the mouse button will cause the
system to change from the state "waiting for coordinates" to the state "waiting for choice".
The top-down decomposition, applied to the streamline example, shows the direct correspondences between data
(functions) and object states (behaviors). These correspondences (data - object states) and (functions - object
behaviors) -- help us in classifying and understanding the system components and in finding/defining the objects.
An OO analysis that uses the set of the presented modeling tools permits to identify and capture objects that are
systematically defined as part of a desired solution and thus they can be checked in advance (before
implementation occurs); which is somehow in contrast to a subjective and possibly inconsistent "users view" of a
problem to be solved.
The OO analysis has led to defining a set of objects and relationships that correctly depicts the real world and
meets the user requirements. These objects must perform various (types of) operations and must interact;
conventions must be defined to enable such interactions by means of messages.
OO analysis consists of these basic steps:

identify the objects,

identify the attributes and the messages of the objects,

identify the relationships between the objects.

Generally, the objects are identified as NOUNS in the process specification, and also as data in the data dictionary
(e.g. Point). The most appropriate way to find the objects is from the ERD, because of the (usual) one-to-one
correspondences between entities and objects. In this case, entities have also to be present in the process
specification and data dictionary. In the ERD of the streamline example, we can extract the objects: Point,
Quantity, Node and Streamline.
The object behaviors can be identified in the DFD. For example, the streamline can be calculated from the
velocity field. Hence, the Velocity Field must have a method for calculating the streamline when it receives the
Streamline message with the Point argument. The identification of the object behavior can be found in the
STD. For example, the Mouse object must return the Point object when the mouse button is pressed.
Three basic relationships may be identified between objects:

has

- indicates that one object contains another one,

is a kind of
Point),

- indicates that one object is a specialization of another one, (e.g. Node is a kind of

uses
- one object interacts with another (e.g. Velocity Field uses Point to calculate
Streamline).

182

(e.g. Curve has Points)

For the streamline example, a preliminary class specification would be:

CLASS NAME:

ATTRIBUTES:

MESSAGE:

MESSAGE RETURN:

Mouse

point

buttonDownEvent()

Point

VelocityField

velocities, cells

streamline(Point)

Streamline

Streamline

points

geometryStructure()

Structure

PressureField

nodes, cells

streamlineStructure()

Structure

Screen

display (Structure)

The essential part of the OO analysis phase is the documentation, which include the specification of the objects
with their attributes and messages. The methods have to be described in sufficient detail to ensure that the
application requirements are complete and consistent. These classes represent the problem space, the first layer
of abstractions.
Design can start as soon as we have the analysis of the problem. It is important to ensure that the analysis
specification can realistically be implemented with the available software development tools. OO design is a
bottom-up design: the lower-level classes can be designed before the high-level classes. To take advantage of
reusability, it is natural to separate the classes identified during analysis in two groups:

the classes to be developed

new components,

the classes to be reused

existing components (class libraries).

It is difficult to design an entire workable system without prototyping some parts of it and exploring alternative
solutions based on existing class libraries. The class libraries relate to distinct computer-system areas -- e.g.
graphics, user interface, mathematical, etc. -- and constitute frameworks that could be integrated to form a
workable solution. If we assume that the class libraries Continuum, InterViews and PHIGS are available,
the classes that must be designed are Streamline, Velocity Field and Pressure Field. OO design relies on two
graphical modeling tools for identifying the new classes, namely:

the class hierarchy diagram,

the class attributes diagram.

Figure 183: Class hierarchy diagram

The class hierarchy diagram (see Figure 183) of a class library provides an immediate means of positioning a
new class within the library. For example, in the class library Continuum, we have the class Scalar Curve
which is very well-suited for Streamline implementation; also, the classes Pressure Field and Velocity Field
are instances of the Scalar Field and Vector Field. Hence, we can start with the implementation phase and
prototype the application.
Because the classes of the streamline example are existing in the above class hierarchy diagram, the example of
the class attribute diagram (see Figure 184) is given for the involved Continuum classes.

183

ScalarCurve

Node

Field

Cell

Figure 184: Class attribute diagram

The design phase can stop when all the objects that have been defined are simple enough, that is to say that
they do not require any further decomposition. Thus, we have composed all the required objects from the
existing classes and the implementation can follow.

3.4.2 Data Abstraction and Encapsulation

The key concept in C++, as required by OOP paradigm, is the concept of class. The class implements the state
and behavior of an abstract data type (ADT) [110]. Data abstraction is supported through the class mechanism.
Data abstraction refers to the separation of the class interface from the class implementation. Abstracting
(separating) the class interface from its implementation also means that the class implementation can change
without affecting the client code that uses it. Data abstraction is normally used for the definition of primitive data
types (e.g. INTEGER, REAL in FORTRAN or int, float in C). For example, when using floating point arithmetic in C,
a programmer does not need to know HOW floating point numbers are represented internally, nor HOW floatingpoint arithmetic is performed: he/she merely needs to use the data type:
float a=1,b=2;
a+b;
sqrt(a);

With the C++ syntax, a class is specified as a set of data (variables) together with a set of functions. The data
describes the object state and the functions describe its behavior. A class is a user-defined data type, giving the
user the capability to define new data types. For example, the following is the definition of the Point class:
class Point
{
private:
// REPRESENTATION - object state
float xx;
// x coordinate
float yy;
// y coordinate
float zz;
// z coordinate
public:
// INTERFACE - messages
void move
// increase coordinates
(float dx,
// x direction
float dy,
// y direction
float dz);
// z direction
float x();
// return x coordinate
void
x
// modify x coordinate
(float v)
// new x coordinate
{xx=v;};
};

In C++, the class definition mechanism allows to declare variables called data members (e.g. float xx) and
functions called member functions (e.g. move). The class definition is a little bit different from the ADT
decomposition; because it puts together the class representation and interface (see Figure 176).
The access to actual data is controlled by admitting only the use of the messages associated with the object; this
prevents indiscriminate and unstructured changes to the state of the data. The data members are said to be
encapsulated since the only way to get at them is by use of the member functions (invoke messages) associated
with the methods implementation (e.g. move).
void Point::move

// move method implementation

(float dx, float dy, float dz)
{xx+=dx; yy+=dy; zz+=dz;};

184

The message passing mechanism forms a sort of software shell around the class. The C++ keywords private,
protected and public support the technique of encapsulation by allowing the programmer to control access
to the class members. In the class definition, member data and functions can be specified as public, protected or
private:
public members, typically functions, define the class interface,
protected members, typically functions, define the class interface when
inheritance is applied,
private members, typically data types, can only be accessed inside member
functions implementation (methods). They reflect the state of the object.
The distinction between public and private members separates the implementation of classes from their use.
As long as a class interface remains unchanged, the implementation of a class may be modified without affecting
the client code that uses it. The client programmer, user of the class is prohibited from accessing the private part.
By default, class members are private to that class. If we have to change point coordinates, we have to use
appropriate member functions, for example x to update the x coordinate. There are three ways to identify the
object in C++:
Point
Point&
Point*

a;
b=a;
c=&b;

a.x(10);
b.x(10);
c->x(10);

// by name
// by reference
// by pointer

In this example, the three points a, b and c represent the same object. In the class Point definition, the
implementation of x member function is also present. By default such function implementation is expanded as
inline function. Thus, there is no function call overhead associated with its use.
The following definition extends the Point class with two new member functions read and write allowing the
input and output of the Point instance coordinates:
class Point
{
private:
// REPRESENTATION - object state
float xx;
// x coordinate
float yy;
// y coordinate
float zz;
// z coordinate
public:
// INTERFACE - messages
void move
// increase coordinates
(float dx,
// x direction
float dy,
// y direction
float dz);
// z direction
float x();
// return x coordinate
void
x
// modify x coordinate
(float v)
// new x coordinate
{xx=v;};
ostream& write // output
(ostream&);
istream& read // input
(istream&)
...
};

In the definition of the Point class, we have introduced 2 new classes istream and ostream which are
standard classes in the C++ I/O stream library. Instances of the output stream cout and input stream cin are
defined by default. We can use their << and >> operators without caring about their implementation.
The implementation of the read and write member functions is as follows:
istream& Point::read(istream& s)
{return s>>xx>>yy>>zz;};
ostream& Point::write(ostream& s)
{return s<<"("<<xx<<","<<yy<<","<<zz<<")";};

In the body of the member function implementation, any undeclared variables or functions are taken to be
member variables or functions of the class (e.g. xx,yy,zz).
185

The default implementations of the << and >> operators are illegal too for the Point class, thus they have to be
implemented in terms of read and write operators.
ostream& operator<<(ostream& s, Point & p)
{return p.write(s);};
istream& operator>>(istream& s, Point & p)
{return p.read(s);};

In order to improve performance in the use of such functions, they may be defined inline.
In the following example member functions are called using the normal member selection operator (e.g. -> or .).
An object or pointer to an object must always be specified when one of the Point member functions is called.
Point a;
Point *b=&a;
a.read(cin);
b->write(cout);

These input and output member functions may be called from anywhere within the scope of the declaration of
the Point instance. The same functionality can be achieved with the new definition of the << and >> operators.
Point a;
Point *b=&a;
cin>>a;
cout<<*b;

Operator functions may be members of a class. For example, a += operator could be declared as a member
function in the class definition by:
Point& operator += (Point );

and the implementation of the operator as:

Point& Point::operator+=(Point p)
{xx+=p.x(); yy+=p.y(); zz+=p.z(); return *this;};

Often, member functions require state information to perform their operation. Since every instance may be in a
different state, the state information is stored in the data members. Thus, the member functions provide access to
the data members, whilst these support the functionalities of the member functions.
Some of the most important member functions are constructors and destructors (see class Curve). One can
imagine that a class prescribes that the layout of allocated memory should be a set of contiguous data fields.
Constructors and only one destructor per class are member functions which allocate and de-allocate memory
during the lifetime of the object.
Memory allocation during initialization can be:

automatic
dynamic
static
In the following example, the three types of initialization are shown:
main(){
...
{

};
...

// start of block
Point a;
Point *b=new Point;
static Point c;
...
//end of block

// automatic
// dynamic
// static

};

automatic objects are allocated on the program stack. Their lifetime is the execution time of the
smallest block,
186

 dynamic objects are allocated in free storage. The programmer explicitly controls the lifetime of
dynamic objects by applying the new and deleted operators.
static objects are allocated statically: storage is reserved for them during program execution.
The dynamic control of memory allocation allows the programmer to write a parameterized program. Such a
program, when running, will allocate the amount of memory just required in free storage depending on the data
being processed. That is in contrast with FORTRAN code where memory must be allocated in advance for
handling the maximum data size (FORTRAN ALWAYS allocates the maximum specified memory).
The following example defines the class Curve representation in C++:
class Curve{
private:
Point* store;

// REPRESENTATION - object state

// pointer to array of
// points
int
size; // number of points
public:
// INTERFACE - messages
Curve
// constructor
(int sz=10)
// default number of points
{size=sz; store=new Point[size];};
~Curve()
// destructor
{delete[size]store;};
...
};

This example uses several features of C++:

a default argument for the constructor function Curve(int sz=10),
a destructor function ~Curve(),
the free store operators new and delete to allocate storage.

3.4.3 Inheritance and Polymorphism

Inheritance is called derivation in C++. Classes can be arranged in a hierarchy in C++: a sub-class (of a class) is
called a derived class, while the super-class is called a base class. Derived classes may in turn have classes
derived from them. Hence a tree-structured set of classes can be built, known as a single inheritance.
For example, the ScalarCurve class can be derived from the Curve class and the Node class from the Point
class. Conceptually, both derivations are related to the specific quantity defined for the new derived classes. The
Node derived class extends the definition of the Point base class, so that all the members of the Point class
are also members of the Node class. The implementation is as follows:
class Point {...};
class Node: Point
{
private: float val;
...
}

// quantity value

Instances of base and derived classes are objects of different sizes because they are defined with different class
members. A derived object can be converted into a base class without the use of an explicit cast, i.e. without an
explicit request for conversion by the programmer. This standard conversion is applied in initialization,
assignment, comparison, etc. For example, a Node object can always be used as a Point object.
Node b;
Point& a=b;

Standard conversions allow the base classes to be used to implement general-purpose member functions, which
may be invoked through base class references without being aware of the objects exact derived class.

187

Any derived class inherits from its base class its representation (including the external interface messages) and
methods implementation. The derived class can modify the methods implementation and add new members to
the class definition, therefore:
the internal representation can be extended, or
the public interface can be extended (or restricted).
The same member function name can be used for a base class and one (or more) derived classes. These
polymorphic member functions can be individually tailored for each derived class by defining a function as
virtual in the base class. Virtual functions provide a form of dynamic binding (runtime type-checking).
This mechanism works together with the derivation mechanism and allows programmers to derive classes
without need to modify any member function of the base class. Virtual functions allow the flexibility of dynamic
binding while keeping the type checking of the member functions signatures.
class Point {
private:
// REPRESENTATION - object state
float xx;
// x coordinate
float yy;
// y coordinate
float zz;
// z coordinate
public:
// INTERFACE - messages
...
virtual ostream& write // output
(ostream&);
...
};
class Node: public Point {
private:
// REPRESENTATION - object state
float val;
// quantity value
public:
// INTERFACE - messages
...
ostream& write // output
(ostream&);
...
};
main(){
Point* store[2];
// set of objects
store[0]=new Point(...) // Point initialization
store[1]=new Node(...) // Node initialization
// output for both objects
for(int i=0;i<2;i++) store[i]->write(cout);
...
};

In the above example, the member function (message) write will invoke two different write implementations
(methods), respectively of the Point and Node classes.
When a member function is declared virtual in a class, all the objects of that class are labeled with the type
information as they are created. This virtual declaration adds some extra storage to the class. Any member
functions may be virtual, except constructors. Operator member functions including destructors may be virtual.
From the behavior point of view it is frequently useful to identify the common features provided by more than
one class. Most classes need to be categorized in more than one way. The mechanism of multiple inheritance
allows to extend the features of a derived class using more than one single base class. For example, the class
Node could be implemented differently if for example, the following Scalar class would exist.
class Scalar {
private:
float val;
...
};

// REPRESENTATION - object state

// quantity value

class Node: public Point, public Scalar

{};

188

This version of the class Node is composed of the Point and Scalar class giving the features of both classes to the
Node class without any additional coding.
Besides virtual member functions (polymorphic object), C++ supports function and operator overloading
(polymorphic message). Several member functions and operators can coexist having the same function name or
operator symbol. The compiler uses the member functions arguments types to determine which implementation
of a function or operator to use.
class Point {
...
float x();
// return x coordinate
void
x
// modify x coordinate
(float v)
// new x coordinate
{xx=v;};
...
};

In this example, member function x is type of polymorphic message.

3.4.4 Templates and Exception Handling

In special cases, when inheritance cannot be applied, a generic class mechanism based on class templates can be
used. It supports the automatic generation of individual classes, built upon the predefined code structure defined
by the templates library.
For example, the method count, which counts the number of elements in a collection, can be defined before
specializing it for the PointCollection or SurfaceCollection classes. In fact, generic classes are meaningful only
in typed languages, where each class is declared as belonging to a certain homogeneous collection. Hence one
can determine whether a method is correctly invoked simply by looking at the source code (static type checking)
or by checking it at execution time (dynamic type checking). Dynamic type checking can also be achieved by
inheritance. Generic classes are very often used in larger systems and are considered as one of important
mechanisms to eliminated/reduced performance penalties associated with the use of inheritance.

Exception handling provides a consistent method to define what happens and prescribe a suitable system
response when a class client misuses an object. This is a central issue when developing interactive visualization
systems since one cannot avoid (unwanted) situations of implementation/ or coding anomalies. Exceptions refer
to program states corresponding to run-time errors like: hardware failures, operating system failures, input errors,
system resource shortages (e.g. failure to allocate a request for memory), etc. Exceptions are control structures
that provide a mechanism for handling such errors. Not unsurprisingly, OOM handles exceptions by adding
them to class methods where they naturally belong; this requires the class designer to consider all possible
exception situations and systematically define the methods for error handling.

3.4.5 Dynamic Memory Management

The concept of shared and non-shared objects is central to memory management; this includes the use of the
control and free operations. In order to safely delete a shared object, one must ensure that the delete operation
will not corrupt other objects which may have been using it. When objects are created, they can be used in
different contexts, and multiple references to the same object can exist within an application.
A set of references guarantees that an object is free when it is no longer needed. In order to support the
referencing mechanism of shared objects, the shared objects are separated into:
controlled shared object, and
free shared object.
Shared objects are typically used when sub-objects are constructed as specific parts of a larger one. The shared
object ensures that the sub-objects are maintained in memory until needed. For example, the boundary surfaces

189

are defined as shared objects, since more than one region (cell-zone, segment) can be defined from existing
boundaries (shared objects), which must be retained until all zone cells (segments), which are depending on them
exist. Special types of shared objects are:
pointers to functions
exception objects
error states
library signals and long jump facilities
The applied solution to this shared objects referencing problem is implemented in the base class which manage
the reference counter and through virtual destructor redirect allocation/reallocation of object memory to the
derived classes. Derived classes implement appropriate memory allocation algorithms, when the related derived
class instance is created or deleted.
OOM includes a few other concepts like concurrency, persistence and garbage collection. They are mentioned
here for completeness:

Concurrency refers to the concept of operations carried out in parallel as opposed to sequentially
(computers naturally operate sequentially). Concurrent systems consist of processors that operate,
communicate and synchronize so as to perform a task as efficiently and rapidly as possible. For
example, computational processes can be distributed and executed on several processors (distributed
environment) that run on different platforms (parallel/vector/grid clusters); hence, a Process class can
be designed to encapsulate the sequence of dispatch actions that supports the execution of parallel
protocols.
Persistence refers the activating/deactivating mechanism which permits to automatically convert
(arbitrary) objects to symbolic representations. The object representations are stored as files, and the
objects can be regenerated from these files by inverse transformation. Thus, persistence provides the
methods for saving and restoring objects between different sessions.
Garbage collection refers to the automatic management of certain types of objects. A time comes when
(previously created and used) objects are no longer needed. The system must ensure that un-needed
objects are deleted ones (and once only) and can never be accessed afterwards. This is the purpose of
garbage collection.

190

3.5 Mapping Entity-Relationship model to ObjectOriented model

The analysis of the user requirements is the first step towards a more detailed software description where the
trade-of between suggested visualization system functionality and the one that will be designed, is elaborated.
The practice shows that the software development is not efficient, if the analysis is not carefully considered. To
facilitate the design of the OO data model ERM was applied, as applied systematically in chapter 1. ERM was
found useful for the description of the logical aspects of the visualization system, especially when requirements
are discussed with the end users. The important element in the analysis phase is the establishment of the Naming
Convention, which has to be respected through all the software development phases. Nouns are typically used to
define the names (labels) of the attribute types, while verbs are usually used to name relationship types in ERM.
For the creation of the naming standard of the attribute types, the following two approaches are considered:

1.

To define both verb and noun for each attribute type, describing the association, as data or process
depending on the context. It is obvious that such approach introduces flexibility, but also increases the
complexity of the model, as it is not uniquely defining the related attribute type.

2.

To define only noun or only verb to uniquely define the attribute type. This is more restricted approach but it
results in a more precise software specification.

The second choice was applied, where attribute types are uniquely defined with a verb or a noun: the entity type,
associated with noun and the relationship, associated with verb, while attributes for an entity or a relationship are
or nouns or verbs. With applying ERM, we grouped the relationships and attributes within defined entities in
order to progress in the design of OO classes. In this modeling process entities are mapped to classes. The side
effect of this process is also the creation of auxiliary classes, which were not directly identified in ERM, but are
found necessary to support the envisaged software functionality. The primary goal of the analysis model is to put
in evidence the end user concepts and prepare a foundation for the classes design.

Attribute Type (AT) defines a collection of identifiable associations relating objects of one type to objects of
another type. For example in ERM, Boundary is AT of the Zone entity modeled as an attribute of the Zone class,
called Boundary. This Boundary object has the attribute, which is associating a Boundary object with one or
more Zone objects, which are bounded with this boundary.
Scenarios provide outlines of the user activities that are defining the system behavior. They provide more
focused information related to the system semantics and serve as the validation elements of the software design.
Scenarios provide means to elaborate the system functionality, modeled usually through collaboration of the
respective classes. For example, the class View often participates in several scenarios: for rendering object on the
screen, interacting with the windowing system when resizing, making icons and updating the display. The most
complex scenarios cut across large parts of the software system architecture, touching most of the classes
constituting the application. The scientific visualization system is an interactive application, with variety of such
scenarios driven by external events. Each defined visualization tool is constructed around a specific scenario,
which is found sufficiently independent, thus it can be developed independently, although in practice, it could
have some semantic connections with other scenarios.
ERM defines the static data structure consisting of entities, their relationships and related constraints. This
modeling technique was extensively used in Chapter 1, to define the Visualization System data model. In OOM,
the data structure is defined within classes. Entities and relationships are modeled as classes and there is no need
for the specification of specific entity key attributes. The entity attributes are defined also as classes, which are
modeled, or as an individual class or as collections of classes. The following distinctions need to be considered
before the mapping is made:

191

Relationships are represented explicitly in ERM, but in OOM they are designed as the class references.
In ERM the relationship model expresses the semantics, cardinalities, and dependencies between entity
types.

Key attributes uniquely distinguish entities in ERM, which is not the case in OOM as the class instances
have their run-time identifier. However, these key attributes can be of use, if the persistence or the
search on such objects has to be implemented, based on an association with an indexing or labeling
scheme.

Methods are not present the ERM notation. When there are constraints that cannot be specified
declaratively within a class, we model such constrains with methods in order to support their
verification. The methods become an integrated part of the class design.

The steps to follow when ER model is mapped to OO model:

Entities and relationships are mapped to classes.

Methods are added to classes.

Additional methods are designed to support constrains, which cannot be specified declaratively.

Entity type is mapped one-to-one to a class. The entity type hierarchy is mapped to the OO class
hierarchy. Common entities are mapped to a specific subclass applying the multiple inheritance
mechanism. A composite attribute is mapped to its component attributes as a collection of attributes.

Binary relationship is modeled as object reference within a designed class. The bi-directional reference
facilitates the bi-directional navigation to support such inverse referencing. If the cardinality of a
relationship is greater than one, it is appropriate to model it as a collection of objects references. If a
binary relationship type has one of more attributes, then it is appropriate to design a new class to handle
these attributes and keep track of objects references of the two related classes, for which the bidirectional communication is required.

Constraints can be specified for entity and relationship types. The multiple inheritance mechanism is a
possibility to enforce the constraint of overlapping or disjoint entities in an entity type hierarchy.

Cardinality constraints in a relationship type are implemented in methods for constructing or destroying
objects and methods for inserting/deleting members into/from a collection of referenced objects. If
these cardinality constraints exists a dependency constraint for the relationship type, should be
implemented when constructing and destroying such objects.

3.6 Model-View-Controller Design

The visualization system design is built around an event-driven, multi-window, iconic GUI. It separates the event
management (Control layer), the appearance of the application data (the View layer) and actual modeling of the
application data and processes (the Model layer). This type of framework, known as the Model-View-Controller
(MVC), comes from the Smalltalk programming environment [31]. MVC is a factorization of the applications
interface such that the system architecture is decomposed in independent components that separate application
model from user interface model. MVC is appropriate for designing the visualization systems GUI which
integrates the specialized interfaces of many visualization tools.
The main MVC decomposition of the VS architecture distinguishes three components:
1.

MODEL

what application does logically?

192

Supports functionality, which consists of the visualization algorithms and CFD data
management elements. It handles the internal data and operations, which are isolated from
screen display, keyboard input and mouse action.
2.

how application displays itself visually?

VIEW

Represents the graphics display presentation, responsible for the displayed graphics as the
feedback to all the user interactions. It includes sub-layers with direct procedure calls to
window management and graphics kernels functionality.
3.

CONTROLLER

how application handles input?

The mediating components between the CFD data model and the viewing interface. It controls
the user interaction with the model, including the system limitations that constrain the system
operations in particular circumstances. It provides initial creation of the interactive environment
and maps the user input into the application.

MODEL
2

6
4

VIEW

CONTROLLER

Figure 185: The MVC model with six basic relationships

The MVC model is applied to the overall system architecture and to the individual visualization tools that
supporting the user interaction model. In Figure 185, six MVC relationships are applied to design the
visualization system, especially concerned with the user interaction modeling, as follows:
1. In the model->view relationship, the Model informs the View that its state has changed. This
dependency relationship tells the View to update itself.
2. The view->model relationship allows the View to request from the Model the data necessary for
updating itself.
3. The view-> controller relationship allows a View to activate a Controller that depends upon its state.
4. The controller->view relationship allows the modification of the View due to an input. It is not
necessary to identify the controller through 1 and 5. This is useful for View manipulations when the
Model of the view does not change (for example in the case of zooming or when rotating a scene).
5. The controller->model relationship allows the modification of the model due to an input.
6. The model->controller relationship allows a Model to activate/deactivate a Controller. For example,
a controller may need to be de-activated while the Model is under computation.
This design, which entails no central control, allows coexistent objects to communicate outside of the main
program execution, but any modification to a model might trigger updates of its views through requests by the
associated controllers. Thus, the dependency mechanism is localized per MVC objects triad. The MVC
193

framework can be applied to individual classes, but also as a group of collaborating classes. The envisaged
architecture organizes the software in interchangeable layers, which are perfectly in line with the applied MVC
framework. Such separation allows software updates to be performed without need for re-implementing the main
application structure, and allows for the easier integration of components -- for example, when creating different
layouts of the GUI. This separation also permits the creation of specialized layers only applicable to specific
hardware platforms.
We distinguish the following layers in the VS architecture:
1. The layer that handles the input data (and uncoupled from any graphical output), for example a
display screen or printer device.
2. The layer which is tightly linked to the graphics engine, i.e., the explicit invocation of graphics
kernel functionality.
3. The layer which coordinates the application model and the GUI.
These features make object-oriented systems different from traditional ones in which these layers are mixed
together. The possibility of independently changing the View layer or the Controller layer makes software
maintenance easier, including adaptations to different display or input devices. For example, keyboard (or
mouse) input devices can be changed altered without affecting the application structure.
Experienced designers solve problems by re-using proven, successful solutions or solution patterns: they can be
re-applied and need not be rediscovered. This is why the visualization system reuses the MVC solution and its
mechanism of dependency, which allows a change in a model to be broadcast to all objects concerned and to be
reflected in the multiple views.
The MVC pattern consists of three kinds of objects:
1. the model, which is the application object,
2. the view giving screen presentation, and
3. the controller which defines the way the user interface reacts to user input and model output.

Model and View are coupled through the subscribe/notify protocol. A view reflects the appearance of the model.
Whenever model data change, the model notifies its dependent views; in response, each view must update itself
by accessing the modified values and updating its appearance on the screen. This approach enables the creation
of multiple views of a model. The model contains data that can describe several representations. Views are nested
and MVC defines the way a view responds to a user command given via the associated input device. The viewing
control panel in CFView is designed with a set of buttons modeled as a complex view, which contains the views
of the related buttons. The views are contained in, and managed by the Window.
The interaction mechanism is encapsulated in a Controller object. The Controller class hierarchy supports the
design of a new Controller as a variant of an existing one. A View encapsulates the interaction mechanism trough
the interface of the controller subclass. The implementation can be modified by replacing the controller instance
with another one. It is possible to change a View and a Controller at run-time. The sharing of a mouse, keyboard
or monitor by several visualization tools demands communication and cooperation. Controllers must cooperate
to ensure that the proper controller is selected to interpret an event generated via an interaction component,
which contains (is attached) to the user-mouse-cursor interaction.
The Model has a communication link to the View because the latter depends upon the Models state. Each Model
has its own set of dependent Views and notifies them upon a change. In principle, Views can recreate themselves
from the Model. Each View registers itself as a dependent component of the Model, sets its controller and sets its
View instance variables. When a View is destroyed, it removes itself from the model, controller and sub-views.
Views are designed to be nested. The top View is the root of its sub-Views. Inside a top-View, are the sub-Views
194

and their associated Controllers. A single control-thread is maintained by the cooperation of the Controllers
attached to the various Views. Control has to result in only one Controller selected to interpret the user input: the
aim is to identify the one that contains the cursor. The identification of the cursor position is computed through
the traversal of the associated Views.
The Model is an instance of a class MForm and consists of application data. A change to a Model is a situation
which requires a modification of the View, an instance of VForm. MForm messages enable VForm to update by
querying the Model data. The VForm::update is selective as determined by several parameters of the model. The
VForm::display controls the graphical appearance of the Model. It is important to note that no element of the
interface (view/controller) is coded in the application (model). If the interface is modified, the model is not
influenced by this change. The MVC framework is established using the 3 basic classes MForm, VForm, CForm,
see Figure 186 and follows the example that describes the three specialized classes for the treatment of the
Surface model.

Figure 186: MVC framework for Surface manipulation

MForm is completely application-dependent and consists of key application data. In our surface example this is
the surface appearance. VForm is the class which contains the attributes that set the appearance of the surface on
the screen, for example the wireframe thickness parameter and its color. CForm is the class which controls the
input devices -- such as dialog boxes used for selecting the color. MForm does not define the user interface
directly; hence any message related to View, display of surface and to controller, modifying the surface
appearance is avoided inside it. The trigger mechanism designed through the dependency mechanism informs the

195

models dependents of its change. VForm and CForm requests the Model for the necessary data, in the surface
example, the color variable is accessible to them and MForm receives the message color sent by the VForm and
CForm (interface components). The result of the user interaction is new surface appearance and a change of the
internal state of MForm. CForm sends the message for the thickness update to the MForm, which provides
methods for updating the model data. The CFSurface is the direct subclass of CForm and it is specialized for
controlling the surface appearance. The CFSurface is the subclass of the Controller and handle the user input of
the dialog box that accepts the new thickness parameter. Message passing mechanism requires the link from the
controller to the model. Thus, the Controller cannot be predefined by the MVC framework, except the generic
methods: Model and View. The controller concept is slightly extended with call-back messages, which are
directly sent by a View to the Model ones the Controller has set the View reference. The design components of
the MVC model are shown in Figure 186 for the Surface example and consist of the following three parts:
1. The surface is the application object, and it represents the Model.
2. The displayed surface on the screen is the View of the surface.
3. The input handling for the surface parameters is done through the Controller.

3.7 Visualization System Architecture

The MVC framework is applied for the visualization system architecture, which has its root Controller, View and
Model components. The VS architecture can always be decomposed in these three components organized in a
hierarchical structure, in which the explained surface example represents one leaf of that MVC composition
framework with VS application at its root. The VS architecture is shown in Figure 187, where the main software
layers are identified. The software layers re-group the classes in the following categories:

Input model category:

o Geometry group: base class Zone and its derivates for modeling of the structured and
unstructured topologies
o Continuum group: base class Field from which the Scalar and Vector fields are derived
o CFD group:

base class CFD object from which the input data classes are derived
composed of Project, Mesh, Domain, Boundary and Segment classes.

3D Model category:

consists of classes that filter the Input data model and define different runtime models for 3D graphics, derived from the base class MForm for 3D
scene building and 3D windows manipulation.

3D View category:

consists of classes that defined the appearance of the 3D graphics models and
windows, derived from the base classes VForm. This category is enriched
with AForm classes, which are responsible for controlling the appearance
parameters of the displayed models.

3D Controller category:

consists of classes that defined the interaction control of the 3D graphics

models and windows, derived from the base classes CForm and CWindow.

2D GUI category:

consist of base classes Button, Dialog and Menus which builds the 2D GUI
layout. They are applied in as standalone components, as they are reused from
existing libraries like Interviews and Tcl/Tk are.

2D & 3D Event handling: is an important category of classes which synchronize Events, which the
system generates from user interactions, and received from 2D GUI or 3D
controller category.
The layers are described in the following Sections.

196

Figure 187: Visualization system architecture

197

3.7.1 Model Layer CFD, Continuum and Geometry classes

The Input Model layer is constructed from CFD, Continuum and Geometry category of classes. The CFD classes
organize the user input, which is modeled with geometry classes for structured, unstructured, and hybrid (their
combination) topologies, and together with Continuum classes treats scalar and vector fields in 2D and 3D
geometries. The CFD classes are modeling the problem domain including functionality for:
Data management
I/O operations
Used in visualization algorithms
The input model consists of Project, Mesh, Domain, Boundary and Segment classes. Except the Project class
they can be all specialized in different topologies (structured, unstructured) and defined in different dimensions
(1D, 2D or 3D). The main idea regarding this decomposition is that each project can have multiple meshes, each
mesh multiple domains, each domain multiple boundaries, and on each boundary multiple segments can identify
different boundary conditions. For example, the solid boundaries are extracted at the initialization of the
visualization session by this classification mechanism. The ObjectCFD class is the base class of this hierarchy
and contains the indexing and labeling parameters to support such identification process. When the input file is
read in, these set of classes take care to check the input correctness. They are the data container for all the
visualization algorithms as they group the geometry and continuum classes to store the coordinates and quantity
field data.
The geometry data are process with the classes derived from the base Zone class. They are Point, Curve, Surface
and Body as modeled in chapter 1, which are then included in Quantity based classes such as Scalar Surface and
Vector Curve. All these specialized application classes are including the foundation classes as Dictionary,
Collection or Matrix classes for the storage of the numerical arrays and labeling lists.
For example, the Surface class defines the state and behavior for all subclasses like: SurfaceS or SurfaceU,
defining respectively structured and unstructured geometry. The Surface class defines the interface that the
graphics classes can access data for creating the surface Representations. With such design the manipulation of
heterogeneous set of structured and unstructured surfaces will be invoked through the same interface message in
order to refresh their graphical appearance. The designed common behavior at the abstract class level enforces
the software components consistency in treating a variety of different implementations applying the same syntax.
In addition, the memory storage requirements are reduced because the grid topologies store a cell-node
relationship and all in between topologies (edge, faces) are calculated when requested.
As discussed, the ObjectCFD class is the generalization of the basic common data sets coming from CFD
computation including geometry and related quantity data. The Field class associates geometry with a quantity
through the Zone class. The ObjectCFD is usually composed of many quantities, primitive variables or derived
quantities which are user selected for visualization. These values imply that the same geometry supports
different number of quantity fields. For example, the density, momentum and the energy are calculated at the
same grid nodes. The redundancy of geometrical information becomes obvious, when one is storing each field
with duplicating its geometry. The ObjectCFD class is avoiding such a possibility by application of the
referencing or sharable object mechanism.
For example, the Domain is always bounded by its boundary. The Boundary knowledge gives different
opportunities for optimizing the search, when traversing the geometry or the quantity in order to extract different
SubZones. SubZones can be extracted by indexing: 0, 1, 2, ...n; or quantity name can be applied for their
identification. At insertion, all the processed ObjectCFD instances are checked in order to eliminate duplicates,
as only one instance of the object can exist, to refer to key defined by an index and quantity name. The
ObjectCFD class defines the automatic naming convention, which labels the ObjectCFD instances with a unique

198

identification string. The most complex coding involves the specialized parts of the geometrical space
decomposition modeled with the Segment class defining the boundary condition applied to that geometry. The
input model is designed to support flexibility for processing of multiple components of each of the mentioned
classes. For example, the labeling for of a segment could look as follows:
M1.D2.B3.S4

which means that we are identifying the segment 4 of the boundary 3 in the domain 2 of mesh 1. Each boundary
consists of different BC segments, which can be further connected or not. In order to model such cascading
relationship, the base ObjectCFD class, has a Set attribute to handle such parent-child relationship. These classes
are also referencing the Representations, which are created during the interactive session, such as cutting planes
and isosurface instances. For these instances the specific quantity field is only created, if the user selects that
quantity for investigation; on which bases the creation of such field creation is triggered. The regeneration
mechanism is implemented through the inheritance mechanism, which supports the invocation of a specific
regeneration algorithm at the requested level, where the parameters for their creation are known. The recursive
search is possible through the mentioned parent-child relationship.

Figure 188: Hierarchy of Geometry classes

3.7.2 View Layer 3D Graphics Category

The View Layer is composed of classes which are designed to encapsulate 3D graphics, available through
application programming interfaces (API) for accessing 2D and 3D graphics functionality. 3D Graphics
category standardize a wide range of graphics features, which are available through different implementation of
graphics standards such as PHIGS, PEX, HOOPS and OpenGL implementations, see Figure 189. This graphics
layer assures 3D graphics portability to a wide variety of computer platforms ranging from personal computers
(PC) to workstations, to supercomputers. The View Layer classes are responsible to generate 3D graphics, which
is constructed from data received from the Model Layer and sometimes, additional parameters from the attribute
199

classes (part of the View Layer), which are manipulated through the Controller layer in order to provide
appearance parameters of the selected graphics mode. The View Layer provides a visualization system with more
adapted functionality for building 3D scenes, navigation and windowing operations. It supports 3D graphics
content composition, display, and interaction. The graphical primitives for building the Representations are
characters, lines, polygons or other graphics shapes available from the underlying 3D graphics engine. The 3D
graphics classes for the modeling of an interactive visualization system are expected to support the following
capabilities:
geometric and raster primitives,
RGBA or color index mode,
display list or immediate mode(tradeoff between editing and performance),

3D rendering:
lighting,
shading,
hidden surface, hidden line removal (HLHSR), (depth buffer, z-buffer),
transparency (alpha-blending),
special effects:
anti-alaising,
texture mapping,
atmospheric effects (fog, smoke, haze),
feedback and selection,
stencil planes,
accumulation buffer,
compatibility, inter-operability and conformance-compilation with different OS platforms like:
DEC, IBM, SGI, HP, SUN, PC-Windows, PC-Mac etc.

Figure 189: 3D View Layer

200

The graphics API accesses the graphics hardware to render 2D and 3D objects directly into a frame buffer. These
objects are defined as sequences of vertices (geometric objects) or pixels (images).

Graphics primitives are:

points,
polylines,
polygons,
bitmaps.
They are defined by a group of one or more vertices (nodes). A vertex defines a point, an end point of the
polyline or a corner of a polygon, where two edges meet. The vertex may consist, apart from the geometrical
coordinates, of: color, normal, texture coordinates and edge flags.

Fundamental operations involve:

transformation matrices,
lighting equation coefficients,
anti-aliasing methods,
pixel update operators.
Rasterization is the process by which a primitive is connected to a 2 D image. Each point of this image contains
the information about color, depth and texture. Frame buffer consists of logical buffers for:
color,
depth,
stencil,
accumulation.
Primitives

Rasterization

Coordinate transformation Pixel operation

Frame buffer operation

Coloring and lighting

Texture Mapping Evaluators

Clipping

Fog

Display list

Transparency is best implemented by using blend function. Incoming (source) alpha is correctly thought as a
material opacity ranging from 1.0 (representing the complete opacity) to the complete transparency. In RGB is
mode is appropriate, while in Color index mode is ignored.
Graphics primitives can be used to create forms as shown in Table 32. Graphics attributes of such Forms consists
of variety of colors and rendering styles. They are applied to localize and analyze rapidly geometric entities.

Geometry
text
points
curves,
surfaces,

Graphics Primitives
string
marker
polyline
polygons, fill area facet

Table 32: Graphics primitives for different geometries and text

Graphics attributes are:
Form
color - color, - color map,
paint option
culling,
refinement.

View
part,
unpost,
display,
flush,

201

Window
+
+
+
+
+
+
+
+

Display
resize,
clear,
destroy,
pan,
zoom,
reset,
update,
conform.

Window attributes are:

color,
ambiance light vector
z clip depth,
eye begin, end.
up vector
window view
view box
view plane
The 3D Graphics Category of classes is the layer that supports the display of representation objects created by
the visualization system. It provides a support for
window creation,
manipulation of window properties,
editing of graphical attributes associated to geometric entities.
A separate window has to be attached to the graphics object to select it for manipulation and the update has to be
sent to the graphic group. Such behavior requires a tighter integration between the graphic group and the graphic
part to handle presentation and interactivity. Graphical groups and their parts coordinate their actions in order to
achieve the correct display. These actions are split into three parts:
rendering,
layout negotiation (scaling, sorting),
screen update mechanism.
The important operation is the request for a Form updates on the screen, when being part of a group. The group
defines the order in which to traverse the group parts independently from the operation being performed. One
characteristic operation is the finding of the Minimum fitting box. Form objects do not update the screen directly,
but they request the redraw operation on them, which put them in a ready state in order that the update can be
performed, as the updating of a complete screen involves redrawing of all displayed objects (views). If the Form
is not present in the exposed region, it will be not drawn. Deferred updates effects are:
1. the clipped region can affect a Form several times and in a direct update mode, it would be needed
to be refresh twice.
2. double-buffering optimization,
3. clipping of complex objects.

3.7.3 Controller Layer 3D and GUI input processing

The Controller layer is responsible for processing the user interaction with the visualization system. The
Controller classes are modeling the control of the input devices like mouse and keyboard. The 3D and GUI
objects, which are displayed, might be selectable through the user interaction, and in that case they have an
associated controller for handling the expected user request. The important controller classes are CWindow,
CView and CSurface, as shown in Figure 190.
Controllers are grabbing the user input and through the input handling mechanism, they receive the necessary
information to trigger Actions, which usually access respective models and update their views on the screen. An
example for CWindow control is the management of views, while CView controls the displayed surfaces.
CSurface controls different rendering modes of a specific selected surface. These controllers are related to the
3D graphics engine, while in parallel we also need to control the input coming from the 2D GUI components,
like menus and buttons. They also provide the interface between input devices (for example: keyboard, mouse
and timers) and the associated model and view layers. They do not only deal with different input devices but
they also provide higher level activity, such as mouse tracking, scheduling of users requests, window resizing
and movement event. The controller layer objects define the protocols of the users interaction with the model
layer, and they provide methods between the externally generated events and the display elements modeled by
the view layer. The developed GUI components, as presented in chapter 2, where designed through several
202

iterations before the final design was reached. They control the data input, and provide feedback on the user
interaction, including errors reporting. The feedback information is extremely important as it keeps the user
informed about the system activities. The GUI toolkit contains general components for designing the user
interface, which cover the following functionality:
presenting menus,
parsing commands
reading free-format numeric input,
handling text input,
presenting alerts and dialog boxes,
displaying windows,
presenting help, and
outputting user provoked errors
The GUI combines all the three kinds of menu components: menubar, pulldown and pullright menus. The menu
items can have shortcut commands associated to the keyboard input for speed-up the interactions of experienced
users. The visualization system is a command-driven system and operates by sequentially processing the
triggered commands and their associated parameters, executed through the Action hierarchy of classes. The GUI
classes have been designed to support a common interface to 2D GUI elements, which implicitly take care of the
window system platform by wrapping the needed functionality to assure:

Intersection of functionality it is common to all the windowing systems.

Union of functionality all the windowing system functionality is encompassed.
Window as a base class supports different kinds of users windows, such as:
application window,
icon,
warning and
dialog.

Figure 190: Class hierarchy of the controller classes

203

System is a main controller manager that synchronizes the input- output control of the application window
consisting of three parts:
1. menu bar and time header,
2. working area for placing controllers,
3. monitor area with toolbox controlling the current active view and automatically assigning view
manipulation tools.
The System class coordinates these two different events handling mechanisms in the unified form by taking the
case of synchronization and event processing order. This activity is tightly linked with the update mechanism and
especially important for 3D graphics updates. The System delegates to the Display class the responsibility to
activate updates on the specified views. The user operation is not associated directly to a particular user interface,
because it can be triggered from multiple user interfaces. The idea behind is that the interface will changes in the
future, and it is expected that operations are changing with a slower rate. The objective is to access the required
functionality without creating many dependencies between the designed operations and the user interface classes,
which is an important element in modeling the undo and redo functionality to ease the interactive work. The
Menu Item is responsible to handle the user generated event, which contains the user request. The Event is
captured by the event Dispatcher, which sequentially process them. The Menu Item is associated with Action,
which is executed when the user selects a Menu Item and it is responsible to carry out the user request. The user
request involves an operation associated with different combinations of objects and operations. The
parameterization in the Menu Item is done through the object Action. The Action abstract class provides an
interface for issuing a request. The basic interface consists of a unified execute message, which is propagated
through the inheritance mechanism, till the requested specialization is found. The Pulldown menu is an example
of the class that triggers an Action in response to the button down event. The buttons and other visual interaction
components are derived from the Controller - Input Handler classes, which associate Actions in the same way as
done for the Menu Item class.

Figure 191: Event/Action coupling

The undo ability is the characteristic of the action that can cancel its own effect. The addition of the unexecute
function to the Action interface allows such features. Sometimes the undo ability is determined at run-time. The
final step in supporting the arbitrary-level of undo/redo is defined in the Action history, which is traced through
the Macro class. The number of undo levels is limited to the command history. The command pattern prescribes
a uniform interface for issuing the undo request. A command may delegate all, part or none of the requests to
other objects. The orientation towards the visible objects results in more error tolerance because every controller
class contains operations that make reuse in its own framework. The connection of the interaction object with
the model is supported with call-back messages. The Controller class is slightly extended with call-back concept
applied for the controller messages. The messages are directly sent by controller to the model, as a reaction to
the action. The model receives the message, and triggers its views to update.

204

3.8 Software Development Environment

3.8.1 Software Reusability
The developed software application and class library provide reusable components to support the software
development of a scientific visualization system based on the C++ programming language. The software design
itself can be reused in building new visualization application, in another object-oriented programming language,
like Java, as it shows clearly the visualization components and their interactions. The generic layer of the
designed application consists of abstract classes, defining functionality for each major component and its related
interface protocol in terms of messages. Abstract base classes are representing common data and procedures
associated to particular entity. The abstract class message protocol defines also relationships with other classes.
Messages protocols of the implemented classes might be interchanged with the ones declaring the same
message interface. This is particularly interesting for the GUI components, when adapting the visualization
system for other application domains.

3.8.2 Software Portability

The 3D rendering techniques are platform dependent and the performance issue needs to be carefully tested. The
graphics library offers computer and device independence, but it needs to be verified on different computer
systems the involved output devices:
graphics hardware

terminals

plotters

laser pointers

Portability is an important condition in order that the visualization system is running on different UNIX based
workstations and PC (Windows, Mac). Therefore, when CFView was being designed to be run on a variety of
different graphics systems we need to consider the following elements:
Number of available colors.
Line width support and availability of line type extension.
Polygon fill - maximum number of vertices availability, of pattern and hatch fill ability, to fill
certain classes of polygons.
Text (hardware character sizes and characteristics)
Selective erase
Size of display (input echo area).
Interactive input support
Double buffering,
Z - buffering,
Lighting & shading.

3.8.3 Integrated Development Environment

An object-oriented Integrated Development Environment (IDE) needs to be put in place in order to improve the
visibility of the source code and control of its evolvement, for example Eclipse or Microsoft Visual C++ in order
to handle the application building (compiling, debugging and testing). An application, as the one presented in
this thesis, consists of several hundred modules, which need to be systematically compiled with different
options. Such IDE keeps track of all the necessary files and their interdependencies, as the software development
project consist of many files, which are depending on each other, and organized, in different directories

205

structures. Some files are source code (headers *.h, and source *.c) while others are object files *.o, produced by
the compiler and executable ones by the linker. The last one depends directly on the source code.
An O-O development environment supports the sense of design. An application is decomposed in terms of
classes and their interactions. They are coded in C++ code and linked among several directories, which contains
a number of files, which needs careful and disciplined organization and naming convention. The source code is
divided in two major parts. The include-files are named <class name>.H and define the most abstract layer of
the application. They describe the class interface protocol: class declaration defined in terms of messages. The
source-files, which define the definition of the messages, are named as <class name>.C.
In addition, the development environment must have a class-hierarchy browser which ties together all the C++
classes and maybe some native C or FORTRAN code. The visual presentation of the class hierarchy is an
important aid to the developer, to make him/her easy to comprehend, the applied (sometimes complex)
inheritance and polymorphic model. A class browser allows to access and to edit, in highly structured way the
class files and the executables. It needs to provide menu-driven operations to import existing classes into a
project create new classes and delete existing classes from an application. The facility to visually create new
number functions, data members and friends, to examine, modify, delete existing members, to change the
visibility of any member and to change a member function to be / static, inline / normal or friend represents the
indispensable tool for modern software development. The makefile manages the mechanics of building the
application by keeping track of many included files required in C++ program. The automatic construction and
maintenance of makefile which controls the dependencies of the compile and link process is essential to be
integrated in IDE. The benefits from the use of such development environment are:
significantly reduced design and development time due to high degree of sensibility, extensibility
and ready to use components,
significantly reduced program size,
enforced consistency of the GUI which have feel-and-look characteristics built in,
enforced portability because all system and device dependencies are eliminated through the property
of encapsulation and unified protocols.
A computer program is a complex artifact usually presented as text. Visual presentation of code is a method of
presenting program structure visually. The program visualization is concerned with the following aspects:
techniques used to display a program visually,
how visualization affects the software development and maintenance,
general features of program visualization tools.
The graph representation is standard visual technique to present the program structure. The graph nodes
represent program entities such as classes and functions, while graph curves represent the relationship between
these entities. For example: Inheritance and function calls. Using different line styles or colors several
representations can overlay on the same display or window or image. The C++ language provides a rich set of
syntax, which can be visualized as graph nodes. That includes classes, class templates, class members, functions,
function templates, variables (both local and global) and files together with the relationship between them.

Inheritance which different types of visibility:

public
private
protected
Containment or Aggregation. - One class contains another class as:
data number,
instance,
pointer,
reference.
206

 Instantiation via template

Friend members
Function call: relationship between class variable nodes:
direct function call (is a relationship when a function directly calls another function ),
indirect function call (it exist when a function calls another via a function pointer stored in
the pointer),
virtual function call (exists when a function calls a virtual function of defined class,
global & local variable,
Uses and modifies. This relationship exists between variable nodes as structure nodes and function
nodes. The arc indicates that the function either references or changes the value of the variable.
When the relationship exists between a function and a structure, it indicates that the field of the
structure is either referenced or modified.
Definition of variable, function and class
Declaration of include dependencies
The program visualization techniques provide a high level view of program structure. The benefits of such
approach help reusing of the elements and improving by adding libraries, to be integrated as basic structure of
the application. The IDE tool has to consist of:

source code versioning

class hierarchy and class relationship diagrams for browsing and picking the source code
data flow diagrams
editor with syntax identification through different font types and colors
output of subparts of the code with associated diagrams in appropriated graphics format
performance benchmarks and quality tests.

Figure 192: Eclipse Integrated Development Environment

207

Conclusion and Future Developments

Innovation in visualization systems poses the simultaneous challenges of: building better, faster and cheaper
computer-aided solutions to ever more complex scientific, engineering and other multi-disciplinary problems;
developing sophisticated methodologies and algorithms; harnessing the power of upcoming technologies; reusing and leveraging the power of legacy systems and solutions; and working in increasingly shorter design and
production cycles.
The work presented in this thesis has been addressing, over many years, the problem of advancing the state-ofthe-art of scientific visualization systems using innovative software development methods and programming
techniques, specifically object-oriented methodologies and tools. We have succeeded in demonstrating that
object-oriented approaches and techniques are appropriate for designing and building interactive visualization
systems that meet all the requirements placed on them by scientific disciplines: correctness, accuracy, flexibility,
performance and by software engineering: compatibility, reusability, portability. In particular, we have shown
how a high degree of interactivity and user-friendliness can be achieved through the use of class libraries
encapsulating graphics and interfacing. More importantly, we have provided evidence that scientific
visualization has deeply changed the very nature of the investigative process itself by allowing the researcher to
explore and view the physical world in an intuitive, interactive and deeply illuminating manner.
We have illustrated our approach with examples taken from the genesis and development of CFView, an
advanced interactive visualization tool capable of handling complex scientific and engineering problems, in
various geometries and in virtually any application/investigation area. CFView -- and its several prototyping
variants produced over the years -- is the result of many iterative developments in continuous interaction with
engineers and scientists, mostly but not exclusively at the VUB [71, 111-124]. Many PhD theses have been
produced in the context of the research, design, development, prototyping, testing and packaging work that took
place in relation to the CFView system.
The process of transforming user requests into software functionality has been analyzed and mastered at all
detail levels, leading to a Graphical User Interface very-well adapted to the requirements of CFD engineering
techniques, as demonstrated by the commercial success of the CFView product marketed by NUMECA. The
software architecture, a main result of the present work, has proven to be flexible enough to cope with the
increasingly demanding visualization needs of advanced industrial turbo-machinery applications in ongoing
development at NUMECA.
System

Model

Data

Computer
Graphics

Interactive
Visualization

Quantity

User
Interface

Operators

Continuum Mechanics
Computational Geometry
Combinatorial Topology

Geometry
Continuum

Topological
Operation

Figure 193: Knowledge domains involved in interactive visualization

208

Geometrical
Operation

In order to provide for the interactive use of many different functionalities, a visualization system needs to
integrate several types of components. The graphical user interface, for example, which provides user
interactivity, relies on a dynamic windowing environment together with several I/O facilities and sets of
graphical primitives; the visualization operators are another example of software built upon code originating
from computational geometry [125] and combinatorial topology [126]. The number and the diversity of the
components/routines that must be handled and their tight couplings add considerable complexity to the task of
developing and maintaining visualization software over an extended period of time. To deal with this
complexity, our software was developed using object-oriented programming, allowing different data types and
routines to be encapsulated in objects. These objects are grouped in classes covering different application
domains. The classes themselves are organized into class hierarchies, which match both the internal knowledge
domain and link domains in-between, see Figure 193. Using OOP, one can systematically and naturally
decompose complex software into manageable modules.
As shown in Figure 193, interactive visualization calls upon various knowledge areas and conceptual models,
namely: application data model, computational geometry, combinatorial topology and computer graphics. The
application model is meaningful to the scientists who want to be able to identify, extract and view regions of
interest in the application model. Computational geometry and combinatorial topology provide the set of
operators needed to carry out the visualization process; these operations are specialized mathematical
transformations and mappings that operate on the application data sets. Finally, computer graphics provide for
the concepts that relate to the user interface and for the graphical primitives that display data on the monitor
screens.

Scientific Visualization has become a feature intrinsic in the scientific investigation and discovery process,
indeed a methodological instrument of scientific research. SV instruments help users to gain novel, often
revealing insights into complex phenomena. The investigator controls SV tools to jump from phenomenon
overview to detail analysis then back from details to overview, with as many iterations as required to gain a
better understanding of the phenomenon under study. Interactive visualization, with its ability to integrate and
synthesize many rendering/viewing techniques and manipulations, improves the perceptual capabilities of the
user in a simple, intuitive and self-explanatory manner.
The main features of the software we have developed include:
Integrated structured and unstructured geometry treatment for efficient visualization of numericallygenerated data around complex 2D/3D geometries,
Visualization environment with interactive user control,
Development of a highly portable graphical user interface (for easy access to integrated system
components),
Creation of class libraries, as reusable and extensible components for different software products,
We have investigated visualization methods covering:
Transparent 2D/3D interactive visualization tools with user-controlled parameters.
Interactive 2D/3D qualitative tools, local values, sections, arbitrary cutting planes.
Interactive techniques simulating smoke injection (particle paths).
The interactive techniques comprise:
Interactive creation of non-computational grid surfaces (cut planes, iso-surfaces) where analysis
tools can be applied (e.g. sections, isolines),
Interactive identification of block topology and connectivity with boundary conditions,
Analysis of multiple quantities related to the same geometry in the same or side-by-side view,
Data comparison in a multi-view, multi-project environment.
The visualization tools were integrated in the GUI in order to enable seamless communication and control
between different geometries and quantity representations.

209

In the following section, we describe the impact that our visualization system has had on 5 EC and IWT R&D
projects.

PASHA Parallel CFView improving the performance of

visualization algorithms
Parallel CFView, developed under the ESPRIT III-7074 PASHA project, is a heterogeneous, distributed
SIMD/MIMD software system whose concept is illustrated in Figure 194. The system is divided in 3
independent components:
Parallel
CFView

Interface
Framework

Parallel
Applications

Parallel

PVM
- SCSI
- S-bus
- Other...

Network

Parallel
Server

Graphical
Workstation
Network

Application

Parallel
Machines
Bus

Figure 194: Conceptual overview of the SIMD/MIMD Parallel CFView system

1.

The CFView visualization package, which runs on a high-end graphical workstation.

2.

Parallel Applications that run on parallel machines. These applications can be viewed as a collection of
stand-alone parallel programs performing mapping operations on CFD data.

3.

The Interface Framework, which governs the communication between CFView and the Parallel
Applications. The Interface Framework is a distributed program which for its largest part runs on a
Parallel Server machine, but with extensions to the graphical workstation and the parallel computers.
Among its tasks are communication management, parallel server control, networking and
synchronization. It provides generic functionalities for transparent access to the parallel applications in
a heterogeneous and distributed environment.

The four algorithms in an SIMD (as well as in an MIMD) implementation of Parallel CFView system [127] are:
1.

The parallel cutting plane algorithm uses a geometrical mesh, a scalar quantity and the equation of a
plane to calculate a collection of triangles and scalar quantities. The triangles represent the triangulated
intersections of the plane with the mesh. The scalar data are the interpolated values of the scalar quantity
at the vertices of the triangles.

2.

The parallel isosurface algorithm uses a geometrical mesh, a scalar quantity and an isovalue. It
calculates a collection of triangles. The triangles represent the triangulated intersections of the
isosurface with the given isovalue with the mesh.

210

3.

The parallel particle-tracing algorithm uses a geometrical mesh, a vector quantity and the initial
positions of a number of particles in the mesh. It calculates the particle paths of the given particles. The
paths are represented as a sequence of particle positions and associated time steps.

4.

The parallel vorticity-tracing algorithm uses a geometrical mesh, a vector quantity and the coordinates
of a number of points in the mesh. It calculates the vorticity vector lines associated with the vector
quantity that pass through the given points. The lines are represented as a sequence of positions.

Algorithm

#triangles

Cutting Plane

5189

Isosurface

10150

System

Send Data

Sequential
SIMD
MIMD
Sequential
SIMD
MIMD

---3.66
3.66
---3.66
3.66

Load Data

---5.82
28.86
---5.82
28.86

Execute

6.00
3.63
4.03
27.47
6.62
4.96

Table 33: Average times (s) for Sequential, SIMD and MIMD implementations of
Cutting Plane and Isosurface algorithms (wall-clock time)
In order to evaluate the performance of the heterogeneous and distributed approach for Parallel CFView, a
limited benchmarking analysis was conducted. We compared the performances of the SIMD and the MIMD
implementations with the stand-alone, sequential CFView system. The SIMD parallel machine used was a CPP
DAP 510C-16 with 1024 bit processors and 16MB RAM (shared). The MIMD parallel computer was a Parsytec
GCel-1/32 with 32 T-805 processors each having 4 MB of RAM. Both parallel machines were connected
(through SCSI and S-bus respectively) to the Parallel Server machine which is a SUN SparcStation10. For the
graphical workstation, we used an HP9000/735 which communicates with the Parallel Server over an Ethernet
LAN. The results of the measurements are given in Table 33. All times are in seconds, averaged over 20 runs.
The table shows the average execution times for both algorithms on the different systems.
For the SIMD and MIMD Parallel CFView implementations, the average time needed to send the data (mesh and
scalar quantity) from the graphical workstation to the Parallel Server is given (Send Data), as well as the average
time needed for loading the data from the Parallel Server onto the parallel machines (Load Data).
The average execution time (Execute) for the parallel implementations includes (i) the sending from the
graphical workstation to the parallel machines of the algorithmic parameters (i.e. the equation of the cutting
plane or the isovalue); (ii) the execution of the algorithm on the parallel machines; (iii) the retrieval of the result
data from the parallel machines to the Parallel Server; and (iv) the sending of the result data from the Parallel
Server to the graphical workstation. For the stand-alone, sequential implementation of CFView, only the average
execution time is shown. The number of triangles (averaged over the runs) generated by the algorithms is given.
The total execution time, given in Table 33, for the parallel implementations is the sum of the three timings
(Send, Load, and Execute). However, by making use of the caching mechanism in the Interface Framework, data
send and load needs be done once only, in the beginning. After that, only the new equation of the cutting plane
or the new iso-value for the isosurface has to be transmitted to the parallel machines, after which execution can
take off. Hence, in a realistic situation, only the times listed in the last column (Execute) are relevant for
comparison.
The results revealed how massively-parallel computers (SIMD as well as MIMD) can be used as powerful
computational back-ends in a heterogeneous and distributed environment. A performance analysis of Parallel
CFView showed that both types of parallel machines are about equally fast. The total execution times on the
SIMD implementations are sensitive to the amount of computation required, whereas the execution times on the
MIMD implementations are dependent upon the amount of data routed between the processors. The overheads
induced by the interface framework are seen to require only a minor fraction of the global execution times. This

211

indicates that the heterogeneous and distributed approach of Parallel CFView is indeed viable since it performs
significantly better than its stand-alone, sequential counterpart. This is especially true for computationallyexpensive operations, such as isosurface calculations on problems with large data volumes. Since the
heterogeneous and distributed nature of the system allows the transparent use of remote parallel machines on
various hardware platforms.

Alice QFView towards the transparent visualization of

numerical and experimental data sets
The development of QFView in the ESPRIT-IV ALICE project (EP-28168) extended the authors research
towards using the World Wide Web for designing and building up distributed, collaborative scientific
environments [47, 128]. QFView was developed in a web-oriented client-server architecture (e.g. Java, JDBC)
which allowed openness and modularity, as well as improved flexibility and integration of the visualization
components (current and future). A core element was the creation of a central database where very large data sets
were imported, classified and stored for re-use. The distributed nature of QFView allows the user to extract,
visualize and compare data from the central database using World Wide Web access. QFView integrates EFD
and CFD data processing (e.g. flow field mappings with flow field visualization).
QFView supports the integrated use of visualization and animation tools which are integrated with the database
management system for data access and archiving. For example, in a possible configuration of QFView
components over the Internet, the J2EE application servers [129] and the Database Management System servers
(DBMS) [130] are located in one site, whilst the GUI client applications run on a server in another location (see
Figure 195 (a)). The meta-database is based on a J2EE distributed architecture whose execution logic is stored
and executed at the EJB container level [131], and which is accessed from the Java based graphical user interface
(GUI) via the HTTP protocol. The main advantage of the Internet is the possibility to store and access data
(images, input files, documents, etc.) from any URL in the world. Both EFD and CFD applications generate huge
quantities of data which can be used to populate the database. The user needs to access, extract, visualize and
compare the required quantities; these functions as illustrated in Figure 196. The QFView system is composed of
3 major elements:
(1)

An EJB container with all the metadata management rules to manipulate metadata and the relational
database used to store the metadata information. The EJB container acts as security proxy for the data in
the relational database.

(a) distributed environment

(b) web-based data management component)

Figure 195: QFView an Internet based archiving and visualization environment

212

COMPARISON TOOLS
experiment vs.
experiment

experiment vs
simulation

simulation vs.
simulation

FLUID FLOW DATABASE

Measurement
device

Measurement
data

Database

CFD
simulation data

CFD code

ANALYSIS TOOLS
statistical
analysis

quantification
tools

image
visualisation
animation

Figure 196: The QFView framework

(2)

(3)

A Thin GUI Java client is used for remote data entry, data organization and plug-in visualization.
GUI clients must be installed at the end-user location, either at application installation time or by
automatic download (Zero Administration Client).
URL accessed data (images, video, data files, etc.) can be placed at any URL site.

QFView organizes, stores, retrieves and classifies the data generated by experiments and simulations with an
easy-to-use GUI. The data management component, see Figure 195 b, offers a user-friendly web-enabled frontend to populate and maintain the metadata repository. The user can accomplish the following tasks using this
thin GUI Java client:

Start the GUI Java client application, create a new folder (measurement), define metadata information
for the folder, such as keywords, physical characteristics, etc. It is important to emphasize that the GUI
Java client is connected to the EJB server using the HTTP protocol, and all the information entered by
the user are automatically stored in the relational database.

Organize the data into a hierarchy of predefined or newly- created tree- like nodes; user can also
execute a data search procedure, combine documents, and do several other operations on the input
folder.

Create and define new raw data such as XML files, images, input files, etc. for a particular folder by
specifying either a local or a remote location of the data.

Define old raw data (XML files, text files, videos, etc.) by specifying either a local or a remote location
of the data.

Start the visualization plug-in application on the selected raw data.

Essential Database Tools present in QFView are:

a)
b)
c)
d)
e)

Data Entry
Folder and Document insertion
Data Classification
Full Text Search
Meta Model and Data Organization

213

Figure 197: VUB Burner Experiment

These tools were used to assist experimental setup, to improve data acquisition and systematic manipulation of
the results of the VUBs double-annular jet modeling (Figure 197, frame 1 from left) of the flow in a prototype
combustion chamber. The requirement was to analyze the flow from LSV data (frame 2), PIV data (frame 3),
LDV data (frame 4) and CFD calculation data (frame 5). The LDV (frame 4) clearly shows the mean flow field,
which can be compared with CFD (frame 5). Qualitatively they produce a similar topology but the prediction is
not accurate. For correct prediction of the flow, it is necessary to take into account the effect of coherent
structures (frame 3), which strongly influence the combustion process [132].
The QFView system coupled to the unified numerical/experimental database enables:

the users to reduce the time and effort they put into setting up their experiments and validating the
results of their simulations,

the technology providers to develop a new products [133] capable of meeting evolving and increasingly
demanding industry requirements.

The users have observed that QFView provided them with a means not only for archiving and manipulating
datasets, but also for organizing their entire work-flow. The impact of visualization systems like QFView on the
investigative process itself opens the way to entirely novel ways of working for researchers in experimental and
computational fluid dynamics. Integrated, distributed, collaborative visualization environments give the
possibility of and indeed point to the need of reorganizing research methods and workflows; the various
experiments conducted with QFView in ALICE have given glimpses of the exciting advances that one may
expect from such systems in the coming years.

QNET-CFD on quality and thrust in industrial applications

QNET-CFD was a Thematic Network on Quality and Trust for the industrial applications of CFD [134]. It
provided European industries with a knowledge base of high quality application challenges (reviewed and
approved CFD results) and best practice guidelines. QNET-CFD was part of the EC R&D GROWTH program.
Eight newsletters see Figure 198 and four workshops on Quality & Trust in CFD were coordinated by VUB
during the project.
The main objectives of the project were:

To create the Knowledge Base to collect Application Challenges and Underlying Flow Regimes from
the trusted sources and made them available to the Network members;

To publish 8 issues of the Network Newsletter to report on the Networks activities

To maintain an open Web-site to publish the Networks work progress and achievements

To organize 4 Annual Workshops, a key instrument for disseminating material on advances and
achievements in Quality & Trust, validation techniques and uncertainty analysis.

214

Figure 198: The eight QNET-CFD newsletters

The project showed that making trusted and validated computational and experimental data sets widely available
is of great importance to the EU scientific community; also, that the integrated processing of CFD and EFD data
requires appropriate access and manipulation tools. As visible on the cover pages of the QNET-CFD
newsletters, scientific visualization has become the most effective and striking way of analyzing presenting
scientific data; obviously, new SV tools will be required for interactive visual access, manipulation, selection and
processing in the future, and such tools can be built upon the basic QFView architecture pioneered in the ALICE
project.

LASCOT Visualization as decision-making aid

The LASCOT project [55] is part of the EUREKA/ITEA initiative. The Information Technology European
Advancement (ITEA) program for research and development in middleware is jointly promoted by the Public
Authorities in all EU Members States and some large European industrial companies.
The goal of LASCOT was to design, develop and demonstrate the potential benefits of distributed collaborative
decision-support technology to the future cyber-enterprise in the global economy; the LASCOT demonstrator
was to:
Support access to traditional information systems and to Web data;
Enable situation assessment, and provide decision-support facilities as well as simulation and validation
facilities to support business decisions;
Include current, enhanced-as-required security tools;
Make use of visualization technology for critical tasks such as decision-making and knowledge
management;
Produce an on line learning application to facilitate the embedding of the platform, by the users.

215

Figure 199: The LASCOT application

The scenario that was retained to demonstrate the LASCOT system is illustrated in Figure 200; it shows the
various actors who are aided by the LASCOT system in monitoring a crisis situation and in making decisions.
The project called for 3D visualization research and for the development of 3D presentation tools capable of
treating general-purpose and highly conceptual information in an appropriate and readily understandable manner.
In this project, our research was focused on the visualization and manipulation of graphical content in a
distributed network environment. Graphical middleware and 3D desktop prototypes [135] were specialized for
situational awareness, see Figure 199. A state-of-the-art review did not identify any publicly available largescale distributed application of this kind. The existing proprietary solutions rely on conventional technologies
and are limited to 2D representation. Our challenge was to apply the latest technologies, such as Java3D, X3D
and SOAP, compatible with average computer graphics hardware and to demonstrate a solution allowing: data
flow from heterogeneous sources; interoperability across different operating systems and 3D visual
representations to enhance the end-users interaction.
We applied the Model-View-Controller (MVC) paradigm to enhance the interactivity of our 3D software
components for: visualization, monitoring and exchange of dynamic information, including spatial and timedependent data, see Figure 199. The software development included the integration and customization of
different visualization components based on 3D Computer Graphics (Java3D) and Web (X3D, SOAP)
technologies and applying the object-oriented approach based on Xj3D.
Cutting-edge 3D graphics technologies where integrated, including the Java-based X3D browser; we used Xj3D
to visualize data from various external sources using our graphical middleware. While present software
components provide highly flexible interactions and data-flows, the coupling between these components is
designed to be very loose. Thus, the components can be upgraded (or even replaced) independently from each
other, without loss of functionality. With SOAP messaging, inter-components communication is completely
independent of software platforms and communication transmission layers. In our approach, Java components
co-exist with Microsoft .NET front-end, as well as back-end implementations. This approach allows improving
software development of 3D collaborative and visualization tools. Future development of an appropriate
ontology could significantly improve the distributed visualization framework.
216

City hall
City

Hospital

Firebrigade
Boat

Boat

The Actors/
Stakeholders
that have
partial
information
inside their
information
systems

River
Administration
Factory

Factory
Meteo
Service

XML
Information
Objects

XML
Information
Objects

XML
Information
Objects

XML
Information
Objects

XML
Information
Objects

XML
Information
Objects

XML
Information
Objects

XML
Information
Objects

XML
Information
Objects

Actors/Information
Providers providing input
to the LASCOT System via
SOAP/XML Middleware

LASCOT Dynamic
Business Process Engine
runs the business process
workflows based on
Decision Scenarios

Decision Scenarios
create/drive the
business process
workflows

Common
LASCOT XML
Information
Objects

Graphical Middleware that

tranforms the Common
LASCOT XML information
objects into data sets
suitable for the
visualization engine

Translation
Format usable for
Visualization
Engine

Visualization Engine

Figure 200: The LASCOT scenario

217

Visualization Engine
presents information as 2/3D
visual objects and alerts
problem areas to the crisis
manager and other
stakeholders

Data Model
Persistence

Data Model
Exchange

Data Model
Viewer

Data Model
Viewer
Graphical
Middlewar

Figure 201: The security SERKET scenario

SERKET Security situation awareness

The SERKET project [136] intends to explore a solution to the issue of security in public areas and events by
developing an innovative system whereby dispersed data from a variety of different devices are automatically
correlated, analyzed and presented to security personnel as the right information at the right time. The aim is to
design and to develop an open-software platform that can be deployed at low cost.
3D software development in SERKET is centered on the visualization and presentation engine, with special
attention on the application of X3D (and XML) standards. The graphical middleware must integrate, correlate,
combine, annotate and visualize sensor data and related metadata (the application context is airport security).
Using sensor data analyzed by other processing and data fusion components, the graphical middleware will build
3D scenes which represent the objects detected by the sensors and the operational status of the sensors at their
locations. Objects in the 3D scenes will be annotated with metadata and/or with links to metadata describing the
security context in relation to the displayed 3D objects. The 3D rendition of the situation must provide an
unambiguous, highly understandable overview of the situation to the user, who should be able to switch between
different levels of viewing details and select, at each level, desired viewpoints (the locations of the video
cameras define the available viewpoints). The 3D model of situation-security awareness will be parameterized in
space and time as shown in Figure 201.
The solution will be based on the J2EE Application Server platform (JBoss). The interface between the
components will be based on SOAP messaging. 3D objects and scenes will be using the X3D file format (an
extended XML-based format for storing graphics). The XML mark-up language will be used for data exchange.
X3D can be extended to include meta-data, thus non-graphical information related to the created graphical
object. The development for the graphical middleware visualization components will be done in Java, using the
Eclipse environment. The 3D SERKET desktop will enable access to and treatment of security data from
heterogeneous sources and will support:
3D models with features
Level of Detail Selection
Navigation and Interaction
Figure 202 illustrates the main functionality of the SERKET application; clearly, 3D interaction and 3D models
are key elements to a new generation of visualization software which must now treat and be able to display
general-purpose and abstract information such as security.

218

OVERVIEW
SMART
TOURING

PTZSnap-shots

Alarmlist

3D model

Camera calibrations

Zones

Detailed
Event View

Video & metaControl

Radar data

Viewpoints
Alarms

ShowCamera(...),
Render a transparent cone to indicate
from where a (virtual) camera is viewing.

Showing alarms in the alarm list.

Show alarmdata : Showing PTZ
snapshots: small video sequences that
run endlessly showing a snap-shot of one
of the targets, zoomed-in from one of the
PTZ cameras.

ShowSource(...),
Switch to and render a live video stream in a
predefinedwindow.

LiveVideo

GotoViewPoint("myViewPoint")
Zoom in on the 3D scene and show the
scene in a helicopter view.

Show live video merged in the

3D scene.
(Based on info in metadata stream)

ShowCEPEventPositions(T/F)
Mark the positions of the CEP events in the event
database in the alarm level colors.
(Blue, yellow, orange, red)

ShowRadarImages(...)
Render the raw radar images in the
scene.

ShowZone(T/F)
Render the zones.
Display a label with status info (recording,
armed, scenario, number of events, etc)

ShowSensorEvents(...)
Request a service to stream to a metadatastream and render targets (person, car,
bicycle or unknown) and their tracks.

Figure 202: The SERKET application

Future developments
Web3D for Collaborative Analysis and Visualization
Todays computer graphics technology enables users to view spatial data in a true-to-life spatial representation,
to the point that non-technical people can comprehend complex data with easiness and comfort. Visualization
tools and data are now the most common media for people -- both technical and non-technical -- to exchange
information. Yet, not everyone has access to visualization tools, although access to the Internet keeps increasing
as the public becomes more at ease with the idea of using the Internet-based services. To meet all demands for
visualization and make the best use of existing and upcoming ICT, we need methodologies and tools that can be
applied collaboratively to collect, inventory, process and visualize 3D data provided by possibly thousands of
servers in the world. A main requirement is that such Web3D-enabled visualization will occur interactively, on
demand and in real-time (seeing is believing).
The challenges are enormous and research will be about creating new visualization tools and combining them
with existing ones to 3D computer graphics and Web-based technologies, and about developing collaborative,
interactive 3D analysis and visualization methodologies for building real-time virtual reality (VR) worlds. We
need visualization aids capable of combining and interpreting Web-based spatial and temporal data to assist us in
the analysis of shared information (supporting collaboration) over the Internet. The two main user activities on
the Web -- content retrieval and content viewing -- will be profoundly modified in this Web3D approach by the
Client Spatial Interpreter (CSI), a new component which will enable the users to perform spatial analyses of
contents retrieved and viewed with the help of Graphical Middleware (GM). A first step will be to reconstruct
spatial data with a conventional PC and interactive 3D visualization using object-oriented computer graphics.
The next step will be to extend the previously mentioned application for collaborative analysis and visualization,
219

with the use of VR technologies, of Web3D, VRML/X3D and Java3D, Internet Map Server (IMS), Server Pages
(ASP and JSP) and Spatial Database Engine (SDE) and Graphical Middleware (GM). These technologies need to
be coupled with application knowledge and to use the 3D web for viewing it, as 3D spatial analysis can be
carried out by an Internet user in real times with the data coming from different servers.
Problem Statement: Although Spatial Analysis and Modeling can be achieved on the desktop using
Visualization Software tools, the development of a Methodology for Web3D Collaborative Analysis and
Visualization promises to improve the way we carry out analysis by providing real-time support for visualization
of spatial features and quantities.
Research Themes: Web3D, Object-Oriented Computer Graphics, Collaborative Analysis and Visualization,
Client Spatial Interpreter, Graphical Middleware and Software Engineering.

European Scientific Computing Organization

An ESCO proposal [137] was recently submitted to the 7th European Framework Program (FP7) which will
encompass scientific visualization research work. ESCO will not cover all disciplines of physics (fluid dynamics,
crash, biology, geophysics, chemistry) nor all application domains (space, health, security, environment,
energy) but will focus on cross-disciplinary and cross-domain issues. Fluid dynamics will be coupled to crash
simulations; Transport and Environment questions will be tackled together, etc. Fields of improvement have
been identified around 4 themes:

Integrated Modeling Environments (IMEs): Scientists largely use commercial, homemade or opensource Integrated Modeling Environments (Matlab, Scilab, Tent, Salome). Providing new
functionalities to IMEs and creating critical masses of users and professional services through
technology providers is a key to success for the European scientific computing software industry.

Scientific visualization: 2D/3D data acquisition and treatment are day-to-day business for many
scientists. Yet there is still a lack of local computer resources for the pre- or post-processing of large
data sets, a situation that could be solved by remote interaction. Collaboration with other scientists to
compare models or results (a significant part of scientists job) could also be made easier.

Scientific software-packaging tools: To ensure that an application reaches a critical mass of users, it is
mandatory for the developer to build and package it on a multi-platform basis. Preparing packaging and
testing environments on several operating systems (Linux, Windows, UNIX, MacOS) is competenceand resource-consuming. Effective tools for packaging should be provided to developers.

Service-oriented interoperability middleware components: To improve the interoperability of

scientific software, it is proposed to develop an open set of middleware components to allow easy and
standardized interfacing and access to modeling/computational tools, data exchanges, control
commands definition, locally and remotely. Hence, the aim is to offer an open and extensive system of
inter-compliant scientific tools (simulation codes and data processing) and make them easily available
and useable locally and through the Web. This effort should help the users reap the benefits of state-ofart software and of up-to-date middleware. One technical solution to be explored is the ServiceOriented Approach (SOA). The SOA concept allows the development of composite software solutions,
adaptable to tailored applications and easily customizable. The possibility of services orchestration
helps to identify modeling processes and favors the access to advanced tools by non-specialists.

Development trends in Interactive Visualization systems

The intensive interchange of design information and of test results and the distribution of tasks between the
aerospace industry, the universities and the research institutes should be enabled on an international basis over a
network like the Internet. Mechanisms for data encryption and security must be built into the configuration of the
220

system. The goal is to allow several/many users to simultaneously visualize information from various data
sources on a large display wall, in multiple high-resolution 3D windows, in a situation that permits collaborative
decision-making. The emerging augmented reality systems provide several advantages over conventional
desktops [138]. Virtual reality environments may provide features such as true stereoscopy, 3D interaction, and
individual/customized viewpoints for multiple users, enabling complete natural collaboration at an affordable
cost.

Figure 203: Visualization of 3D Model

Figure 204: Components of a 3D Model

Visualization and Simulation

Visualization includes functionalities such as: accurate representation of a 3D model, its decomposition in
components and realistic simulation of its physical behavior. Visualization differs from simulation in the sense
that, while both are dynamic (and seen in animation sequences), simulation implies real-time computation with
some amount of user control over the dynamics of the model (settings of parameters). In order to create an
object-oriented framework for supporting the dynamic virtual collaboration, an interactive platform could be
constructed with virtual worlds using specialized 3D objects that contain their own interaction information. A
rigid body simulator can be used to calculate actor and object movements. If all object interactions are
application-independent, a single scheme only is required to handle all interactions in the virtual world. Inverse
kinematics could be used to increase the interaction possibilities and realism in collaborative virtual
environments. This would result in a higher feeling of presence for connected users and allow for the easy, onthe-fly creation of new interactions. In order to allow interactivity the network load must be kept as low as
possible.

Figure 206: Representation of a Measurement

Figure 205: Graphical and Textual Annotations

221

Collaboration and User Interaction

The expected benefits of Collaborative Visualization (CV) are as follows [139, 140]; CV could:

Enable collaborative decision-making

Improve the insight into complex problems [141]

Significantly reduce production and labor costs

Streamline project analysis, design, engineering, and testing

Eliminate the impracticality, hazard, or high-cost encountered in physical environments

Demonstrate products, processes, and plans with highly-realistic simulations

Enable the intuitive exploration and analysis of relationships between variables

Allow interactive analyses of high-resolution, time-varying data sets (of theoretically unlimited size -although limited in practice by available computational resources)

Collaboration functionalities include annotation, measuring, symbols and metaphors, version management and
hierarchical classification, possibly through the use of ontology [142]. The implementation of new types of
visualization techniques, such as disc trees, may overcome difficulties associated with the traditional
representations of decision trees, such as visual clutter and occlusion by elements in the foreground (cone tree
example) [143].
Without associating to the information the probability of certainty, analysis of the visualization would be
incomplete, and could lead to inaccurate or incorrect conclusions. 3D reconfigurable disc trees could be used to
provide users with the information visualization, together with uncertainty (e.g. expressed with different color
attributes).

Figure 207: Cone Trees

Figure 208: Reconfigurable Disc Trees

Human Machine Interface

An appropriate level of human-computer interaction needs to be implemented [144]. A first issue to be resolved
concerns the level of intervention required in a particular context. The system would possibly provide a contextsensitive structure that enables users (otherwise unfamiliar with the system) to navigate through it. Three levels
of intervention can be specified:
1. Simplest: to replicate a path the user is guided along a deterministic sequence of steps.
2. More complex: present alternatives to the user who selects from among them.
3. Sophisticated: to provide a critique of the process.
A second issue to address is configurability: different individuals, as well as different groups, have different
views of a problem and its representation. One way to consider configurability is to specify a set of generic
operations that would encompass the types of operations, information access and user tasks that need to be

222

supported by the system. If this list is compiled it can be structured hierarchically to provide several depths of
intervention. This presumes that an analysis of the tasks required to accomplish goals has been conducted. A
third issue lies with the need to handle different user roles -- such as facilitator or mediator; the system could
be designed to support and facilitate the tasks of various roles. One of the typical participants roles, for example,
is to understand a problem so that it can be defined and alternatives solutions generated and evaluated.
Input can be provided by (remote) sensors and other special equipment. The system could use sensor models and
3D scenic models to integrate video and image data from different sources [145]. Dynamic multi-texture
projections enable real-time updating and painting of scenes to reflect the latest scenic data. Dynamic controls,
including viewpoint as well as image inclusion, blending, and projection parameters, would permit interactive,
real-time visualization of events. Mobile devices (PDA) could be used, as an alternative to mouse and keyboard,
as user input devices.

Figure 209:- Mobile Device Controlling Virtual Worlds

Figure 210: Mobile Application Over Internet

Figure 211: Alternative User Interaction Devices

Figure 212: Handheld Devices

A large multi-tiled display wall, driven by a system for parallel rendering running on clusters of workstations
(e.g. Chromium [146]) can adequately satisfy the requirements of an output device for an advance visualization
system. Several examples are given in Figure 214 to Figure 221.

Figure 213: New generation of miniature computers and multi touch-screen inputs
223

Figure 214: 3D Model of Machine on Display Wall

Figure 215: Scientific Visualization with Chromium

Figure 216: Example of Augmented Reality

Figure 217 :NASA Space Station on Display Wall

Figure 218: Collaborative Visualization

Figure 219: 6xLCD Based Display Unit

Figure 220: Parallel Rendering

Figure 221: 3D Model of Visualization Lab

224

References
[1]

G. K. Batchelor, An introduction to fluid dynamics: CUP, 1973.

[2]

H. S. Lamb, Hydrodynamics, 6th ed.: Cambridge University Press, 1993.

[3]

B. McCormick, T. DeFanti and M. Brown, "Visualization in Scientific Computing," in ACM

SIGGRAPH, New York, 1987.

[4]

M. Gbel, H. Mller, and B. Urban, Visualization in scientific computing. Vienna ; New York:
Springer-Verlag, 1995.

[5]

K. Gaither, "Visualization's role in analyzing computational fluid dynamics data," Computer Graphics

and Applications, IEEE, vol. 24, pp. 13-15, 2004.

[6]

J. D. Foley, Computer graphics : principles and practice, 3. ed.: Addison-Wesley Publ., 2006.

[7]

J. D. Foley and A. Van Dam, Fundamentals of interactive computer graphics. Reading: AddisonWesley, 1982.

[8]

P. Wenisch, A. Borrmann, E. Rank, C. v. Treeck, and O. Wenisch, "Collaborative and Interactive CFD
Simulation using High Performance Computers," 2006.

[9]

A. B. Hanneman and R. E. Henderson, "Visulaization, Interrogation, and Interpretation of Computed

Flow Fields Numerical Experiments," in AIAA Modeling and Simulation Technologies Conference,

Denver, CO: AIAA-2000-4089, 2000.

[10]

"Ensight, CEI Products Overview - extreme simulation software " in http://www.ensight.com/product-

overview.html: Computational Engineering International (CEI) develops, markets and supports software
for visualizing engineering and scientific data, 2007.
[11]

E. Duque, S. Legensky, C. Stone, and R. Carter, "Post-Processing Techniques for Large-Scale Unsteady
CFD Datasets " in 45th AIAA Aerospace Sciences Meeting and Exhibit Reno, Nevada, 2007.

[12]

S. M. Legensky, "Recent advances in unsteady flow visualization," in 13th AIAA Computational Fluid

Dynamics Conference Snowmass Village, CO, 1997.

[13]

D. E. Taflin, "TECTOOLS/CFD - A graphical interface toolkit for network-based CFD " in 36th

Aerospace Sciences Meeting and Exhibit Reno, NV, 1998.

[14]

"CFView a visualization system from Numeca," http://www.numeca.com, 2007.

[15]

P. P. Walatka, P. G. Buning, L. Pierce, and P. A. Elson, "PLOT3D User's Manua," NASA TM-101067
March 1990.

[16]

G. V. Bancroft, F. J. Merritt, T. C. Plessel, P. G. Kelaita, and K. R. Mccabe, "FAST - A multiprocessed

environment for visualization of computational fluid dynamics " in 29th Aerospace Sciences Meeting
Reno, NV, 1991.

[17]

R. Haimes and M. Giles, "Visual3 - Interactive unsteady unstructured 3D visualization " in 29th

Aerospace Sciences Meeting Reno, NV, 1991.

[18]

H.-G. Pagendarm, "HIGHEND, A Visualization System for 3d Data with Special Support for
Postprocessing of Fluid Dynamics Data," in Visualization in Scientific Computing, 1994, pp. 87-98.

[19]

"ParaView Parallel Visualization Application," in http://www.paraview.org, 2004.

[20]

B. J. Whitlock, "Visualization with VisIt," California, Lawrence Livermore National Laboratory:

http://www.llnl.gov/visit/home.html, 2005.

225

[21]

D. Vucinic, M. Pottiez, V. Sotiaux, and C. Hirsch, "CFView - An Advanced Interactive Visualization

System based on Object-Oriented Approach," in AIAA 30th Aerospace Sciences Meeting Reno, Nevada,
1992.

[22]

J. Walton, "NAG's IRIS Explorer," in Visualization Handbook, C. R. J. a. C. D. Hansen, Ed.: Academic

Press, 2003.

[23]

C. Upson, "Scientific visualization environments for the computational sciences," in COMPCON Spring

'89. Thirty-Fourth IEEE Computer Society International Conference: Intellectual Leverage, Digest of
Papers., 1989, pp. 322-327.
[24]

D. Foulser, "IRIS Explorer: A Framework for Investigation," Computer Graphics, vol. 29(2), pp. 13-16,
1995.

[25]

"OpenDX is the open source software version of IBM's Visualization Data Explorer,"
http://www.opendx.org/, 2007.

[26]

"PV-WAVE, GUI Application Developer's Guide," USA: Visual Numerics Inc., 1996.

[27]

W. Schroeder, K. W. Martin, and B. Lorensen, The visualization toolkit, 2nd ed. Upper Saddle River,
NJ: Prentice Hall PTR, 1998.

[28]

W. Hibbard, "VisAD: Connecting people to computations and people to people " in Computer Graphics

32, 1998, pp. 10-12.

[29]

"Fluent for Catia V5, Rapid Flow Modeling for PLM," http://www.fluentforcatia.com/ffc_brochure.pdf,
2006.

[30]

B. J. Cox and A. J. Novobilski, Object-oriented programming : an evolutionary approach, 2nd ed.

Reading, Mass.: Addison-Wesley Pub. Co., 1991.

[31]

A. Goldberg and D. Robson, Smalltalk-80 : the language. Reading, Mass.: Addison-Wesley, 1989.

[32]

B. Meyer, Reusable software : the Base object-oriented component libraries. Hemel Hempstead:
Prentice Hall, 1994.

[33]

B. Meyer, Eiffel : the language. New York: Prentice Hall, 1992.

[34]

B. Meyer, Object-oriented software construction. London: Prentice-Hall International, 1988.

[35]

L. J. Pinson and R. S. Wiener, Objective-C : object-oriented programming techniques. Reading, Mass.:

Addison-Wesley, 1991.

[36]

B. Stroustrup, The C++ Programming Language, Special Edition ed.: Addison Wesley, 1997.

[37]

G. D. Reis and B. Stroustrup, "Specifying C++ concepts," in Conference record of the 33rd ACM

SIGPLAN-SIGACT symposium on Principles of programming languages Charleston, South Carolina,

USA: ACM Press, 2006.
[38]

B. Stroustrup, "Why C++ is not just an object-oriented programming language," in Addendum to the

proceedings of the 10th annual conference on Object-oriented programming systems, languages, and
applications (Addendum) Austin, Texas, United States: ACM Press, 1995.
[39]

R. Wiener, "Watch your language!," Software, IEEE, vol. 15, pp. 55-56, 1998.

[40]

D. Vucinic and C. Hirsch, "Computational Flow Visualization System at VUB (CFView 1.0)," in VKI

Lecture Series on Computer Graphics and Flow Visualization in CFD, Brussels, Belgium, 1989.
[41]

D. Vucinic, "Object Oriented Programming for Computer Graphics and Flow Visualization," in VKI

Lecture Series on Computer Graphics and Flow Visualization in CFD, von Karman Institute for Fluid
Dynamics, Brussels, Belgium, 1991.

226

[42]

J.-A. Dsidri, R. Glowinski, and J. Priaux, Hypersonic Flows for Reentry Problems: Survey Lectures

and Test cases for Analysis vol. 1. Antibes, France, 22-25 January, 1990. : Springer-Verlag, Heidelberg,
1990.
[43]

K. D. Torreele J., Vucinic D., van den Berghe C.S., Graat J., Hirsch Ch., "Parallel CFView : a
SIMD/MIMD CFD Visualisation System in a Heterogeous and Distributed Environment," in

International Conference on Massively Parallel Processing, Delft, The Netherlands, 1994.

[44]

"PAGEIN - Trans-European Testbed for Aerospace Applications," http://visu-www.onera.fr/PAGEIN/,

1996.

[45]

"Europe: Building Confidence in Parallel HPC," IEEE Computational Science and Engineering, vol.
vol. 01, p. p. 75, Winter, 1994. 1994.

[46]

B. E. Grijspeerdt K, Rammant J P, "LCLMS, an advanced database environment for the development of

multimedia courses " in Computers in the practice of building and civil engineering, Worldwide ECCE

symposium Finland, 1997.

[47]

D. Vucinic, M. R. Barone, B. Snder, B. K. Hazarika, and G. Tanzini, "QFView - an Internet Based

Archiving and Visualization System," in 39th Aerospace Sciences Meeting & Exhibit Reno, Nevada,
2001.

[48]

M. Gharib, "Perspective: the experimentalist and the problem of turbulence in the age of
supercomputers," Journal of Fluids Engineering, 118-2 (1996), 233-242., 1996.

[49]

B. K. Hazarika, D. Vucinic, F. Schmitt, and C. Hirsch, "Analysis of Toroidal Vortex Unsteadiness and
Turbulence in a Confined Double Annular Jet," in AIAA 39th Aerospace Sciences Meeting & Exhibit
Reno, Nevada, 2001.

[50]

D. Vucinic and B. K. Hazarika, "Integrated Approach to Computational and Experimental Flow

Visualization of a Double Annular Confined Jet," Journal of Visualization, vol. Vol.4, No. 3, 2001.

[51]

F. G. Schmitt, D. Vucinic, and C. Hirsch, "The Confined Double Annular Jet Application Challenge,"
in 3rd QNET-CFD Newsletter, 2002.

[52]

D. Vucinic, B. K. Hazarika, and C. Dinescu, "Visualization and PIV Measurements of the

Axisymmectric In-Cylinder Flows," in ATT Congress and Exhibition Barcelona, Spain, 2001.

[53]

K. Grijspeerdt, B. K. Hazarika, and D. Vucinic, "Application of computational fluid dynamics to model

the hydrodynamic of plate heat exchangers for milk processing," Journal of Food Engineering, vol. 57
(2003), pp. pp. 237-242, 2003.

[54]

K. Grijspeerdt, D. Vucinic, and C. Lacor, "Computational fluid dynamics modeling of the

hydrodynamics of plate heat exchangers for milk processing," in Computational Fluid Dynamics in

Food Processing, January 2007 ed, P. D.-W. Sun, Ed.: CRC Press, 2007, p. 25 pages.
[55]

"LASCOT project - home page," in http://www.bull.com/lascot/index.html, Bull, Ed., 2005.

[56]

"JOnAS: Java Open Application Server," in http://wiki.jonas.objectweb.org/xwiki/bin/view/Main/,

ObjectWeb, Ed., 2007.

[57]

M.-J. Jeong, K. W. Cho, and K.-Y. Kim, "e-AIRS: Aerospace Integrated Research Systems," in The

2007 International Symposium on Collaborative Technologies and Systems (CTS07) Orlando, Florida,
USA, 2007.
[58]

C. M. Stone and C. Holtery, "The JWST integrated modeling environment," 2004, pp. 4041-4047
Vol.6.

227

[59]

G. Martin, Extended Entity-Relationship Model: Fundamentals and Pragmatics: Springer-Verlag New

York, Inc., 1994.

[60]

C. Hirsch, Numerical computation of internal and external flows. Vol. 1, Fundamentals of numerical

discretization. Chichester: Wiley, 1988.

[61]

J. H. Gallier, Curves and surfaces in geometric modeling : theory and algorithms. San Francisco, Calif.:
Morgan Kaufmann Publishers, 2000.

[62]

M. A. Armstrong, Basic topology. New York Berlin: Springer-Vlg, 1983.

[63]

J. R. Munkres, Elements of algebraic topology. Cambridge, Mass.: Perseus, 1984.

[64]

F. Michael and S. Vadim, "B-rep SE: simplicially enhanced boundary representation," in Proceedings

of the ninth ACM symposium on Solid modeling and applications Genoa, Italy: Eurographics
Association, 2004.
[65]

M. Gopi and D. Manocha, "A unified approach for simplifying polygonal and spline models," in

Proceedings of the conference on Visualization '98 Research Triangle Park, North Carolina, United
States: IEEE Computer Society Press, 1998.
[66]

F. Helaman, R. Alyn, and C. Jordan, "Topological design of sculptured surfaces," in Proceedings of the

19th annual conference on Computer graphics and interactive techniques: ACM Press, 1992.
[67]

A. Paoluzzi, F. Bernardini, C. Cattani, and V. Ferrucci, "Dimension-independent modeling with

simplicial complexes," ACM Trans. Graph., vol. 12, pp. 56-102, 1993.

[68]

O. C. Zienkiewicz, R. L. Taylor, and J. Z. Zhu, The finite element method : its basis and fundamentals,
6. ed. Oxford: Elsevier Butterworth-Heinemann, 2005.

[69]

O. C. Zienkiewicz and R. L. Taylor, The finite element method, 4. ed. London: McGraw-Hill, 1989.

[70]

C. Hirsch, Numerical computation of internal and external flows. Vol. 2, Computational methods for

inviscid and viscous flows. Chichester: Wiley, 1990.

[71]

A. Sturmayer, "Evolution of a 3D Structured Navier-Stokes Solver towards Advance Turbomachinery

Applications," in Department of Fluid Mechanics, PhD Thesis: Vrije Universiteit Brussel, 2004.

[72]

A. N. S. Emad and K. K. Ali, "A new methodology for extracting manufacturing features from CAD
system," Comput. Ind. Eng., vol. 51, pp. 389-415, 2006.

[73]

K. Lutz, "Designing a data structure for polyhedral surfaces," in Proceedings of the fourteenth annual

symposium on Computational geometry Minneapolis, Minnesota, United States: ACM Press, 1998.
[74]

S. R. Ala, "Design methodology of boundary data structures," in Proceedings of the first ACM

symposium on Solid modeling foundations and CAD/CAM applications Austin, Texas, United States:
ACM Press, 1991.
[75]

D. Vucinic, J. Decuyper, D. Keymeulen, and C. Hirsch, "Interactive Visualization techniques in CFD,"

in VIDEA First International Conference in Visualisation & Intelligent Design in Engineering and

Architecture Southhamton, United Kingdom, 1993, pp. 331-347.

[76]

T. Lewiner, H. Lopes, A. W. Vieira, and G. Tavares, "Efficient implementation of Marching Cubes'

cases with topological guarantees," Journal of Graphics Tools 8, 2, 1--15. , 2003.

[77]

R. Aris, Vectors, tensors, and the basic equations of fluid mechanics. New York: Dover Publications,
1989.

[78]

W. H. Press, Numerical recipes in C++ : the art of scientific computing, 2. ed. Cambridge: Cambridge
Univ. Press, 2002.

228

[79]

W. T. Vetterling, Numerical recipes example book (C++), 2. ed ed. Cambridge: Cambridge University
Press, 2002.

[80]

C. T. J. Dodson and T. Poston, Tensor geometry : the geometric viewpoint and its uses, 2. ed. Berlin ;
New York: Springer-Vlg, 1991.

[81]

G. H. Golub and C. F. Van Loan, Matrix computations, 3rd ed. Baltimore: Johns Hopkins University
Press, 1996.

[82]

W. E. Lorensen and H. E. Cline, "Marching cubes: A high resolution 3D surface construction

algorithm," Computer Graphics, vol. 21:4, pp. 163-169, 1987.

[83]

M. d. Berg, Computational geometry : algorithms and applications, 2., rev. ed. Berlin: Springer, 2000.

[84]

J. E. Goodman and J. O'Rourke, Handbook of discrete and computational geometry. Boca Raton:
Chapman & Hall, 2004.

[85]

J. D. Foley, Computer graphics : principles and practice, 2. ed. Reading: Addison-Wesley, 1990.

[86]

P. Eliasson, J. Oppelstrup, and A. Rizzi, "STREAM 3D: Computer Graphics Program For Streamline
Visualisation," Adv. Eng. Software, vol. Vol. 11, No. 4., pp. 162-168, 1989.

[87]

S. SHIRAYAMA, "Visualization of vector fields in flow analysis," in 29th Aerospace Sciences Meeting
Reno, NV: AIAA-1991-801, 8 p., 1991.

[88]

P. G. Buning and J. L. STEGER, "Graphics and flow visualization in computational fluid dynamics " in

7th Computational Fluid Dynamics Conference, Cincinnati, OH, 1985, pp. 162-170.
[89]

C. S. Yih, "Stream Functions in 3-Dimensional flows," La Houlle Blanche, vol. No. 3, 1957.

[90]

D. N. Kenwright and G. D. Mallinson, "A 3-D streamline tracking algorithm using dual stream
functions," 1992, pp. 62-68.

[91]

R. Haimes, "pV3 - A distributed system for large-scale unsteady CFD visualization " in 32nd Aerospace

Sciences Meeting and Exhibit, , Reno, NV, Jan 10-13, : AIAA-1994-321, 1994
[92]

T. Strid, A. Rizzi, and J. Oppelstrup, "Development and use of some flow visualization algorithms,"
von Karman Institute for Fluid Dynamics, Brussels, Belgium, 1989.

[93]

W. H. Press, Numerical recipes in C : the art of scientific computing, 2. ed. Cambridge: Cambridge
Univ. Press, 1992.

[94]

W. H. Press, Numerical recipes : example Book (C), 2. ed. Cambridge: Cambridge Univ. Press, 1993.

[95]

C. Dener, "Interactive Grid Generation System," in Department of Fluid Mechanics. vol. PhD Brussels:
Vrije Universiteit Brussel, 1992.

[96]

P. M. Vucinic D., Sotiaux V., Hirsch Ch., "CFView - An Advanced Interactive Visualization System
based on Object-Oriented Approach," in AIAA 30th Aerospace Sciences Meeting Reno, Nevada, 1992.

[97]

W. Haase, F. Bradsma, E. Elsholz, M. Leschziner, and D. Schwamborn, EUROVAL - an european

initiative on validation of CFD codes (results of the EC/BRITE-EURAM project EUROVAL, 19901992) Vieweg, Braunschweig, Germany, 1993.
[98]

E. Yourdon and L. L. Constantine, Structured design : fundamentals of a discipline of computer

program and systems design. Englewood Cliffs, N.J.: Prentice Hall, 1979.
[99]

B. Mayer, Object-Oriented Software Construction: Prentice-Hall, Englewoods Cliffs, NJ., 1988.

[100]

B.Liskov and J. Guttag, Abstraction and Specification in Program Development: Mc Graw-Hill, 1986.

[101]

I. Jacobson and S. Bylund, The road to the unified software development process. Cambridge, New
York: Cambridge University Press, SIGS Books, 2000.

229

[102]

F. L. Friedman and E. B. Koffman, Problem solving, abstraction, and design using C++, 5th ed.
Boston: Pearson Addison-Wesley, 2007.

[103]

A. Koenig and B. E. Moo, Ruminations on C++ : a decade of programming insight and experience.
Reading, Mass.: Addison-Wesley, 1997.

[104]

S. B. Lippman, J. Lajoie, and B. E. Moo, C++ primer, 4th ed. Upper Saddle River, NJ: AddisonWesley, 2005.

[105]

M. L. Minsky, The society of mind. New York, N.Y.: Simon and Schuster, 1986.

[106]

Bobrow and Stefik, LOOPS (Xerox) Lisp Object-Oriented Programming System: "The LOOPS
Manual", Xerox Corp, 1983.

[107]

C. V. Ramamoorthy and P. C. Sheu, "Object-oriented systems," Expert, IEEE [see also IEEE Intelligent

Systems and Their Applications], vol. 3, pp. 9-15, 1988.

[108]

K. Ponnambalam and T. Alguindigue, A C++ primer for engineers : an object-oriented approach. New
York: McGraw-Hill Co., 1997.

[109]

D. Silver, "Object-oriented visualization," Computer Graphics and Applications, IEEE, vol. 15, pp. 5462, 1995.

[110]

K. E. Gorlen, P. S. Plexico, and S. M. Orlow, Data abstraction and object-oriented programming in

C++. Chichester: Wiley, 1990.

[111]

S. Kang, "Investigation on the three dimensional flow within a compressor cascade with and withouth
clearance," in Department of Fluid Mechanics, PhD Thesis: Vrije Universiteit Brussel, 1993.

[112]

Z. W. Zhu, "Multigrid operations and analysis for complex aerodynamics," in Department of Fluid

Mechanics, PhD Thesis: Vrije Universiteit Brussel, 1996.

[113]

P. Alavilli, "Numerical simulations of hypersonic flows and associated systems in chemical and thermal
nonequilibrium," in Department of Fluid Mechanics, PhD Thesis: Vrije Universiteit Brussel, 1997.

[114]

E. Shang, "Investigation towards a new algebraic turbulence model," in Department of Fluid

Mechanics, PhD Thesis: Vrije Universiteit Brussel, 1997.

[115]

P. V. Ransbeeck, "Multidimensional Upwind Algorithms for the Euler / Navier-Stokes Equations on

Structured Grids," in Department of Fluid Mechanics, PhD Thesis: Vrije Universiteit Brussel, 1997.

[116]

N. Hakimi, "Preconditioning Methods for time dependent Navier-Stokes equations - Application to

environmental and low speed flows," in Department of Fluid Mechanics, PhD Thesis: Vrije Universiteit
Brussel, 1997.

[117]

B. Lessani, "Large Eddy Simulation of Turbulence Flows," in Department of Fluid Mechanics, PhD

Thesis: Vrije Universiteit Brussel, 2003.

[118]

J. Ramboer, "Development of numerical tools for computational aeroacustics," in Department of Fluid

Mechanics, PhD Thesis: Vrije Universiteit Brussel, 2005.

[119]

K. Kovalev, "Unstructured hexahedra non-comformal mesh generation," in Department of Fluid

Mechanics, PhD Thesis: Vrije Universiteit Brussel, 2005.

[120]

O. U. Baran, "Control methodologies in unstructured hexahedral grid generation," in Department of

Fluid Mechanics, PhD Thesis: Vrije Universiteit Brussel, 2005.

[121]

S. Geerets, "Control methodologies in unstructured hexahedral grid generation," in Department of Fluid

Mechanics, PhD Thesis: Vrije Universiteit Brussel, 2006.

230

[122]

S. Smirnov, "A finate volume formulation of compact schemes with application to time dependent
Navier-Stokes equations," in Department of Fluid Mechanics, PhD Thesis: Vrije Universiteit Brussel,
2006.

[123]

T. Broeckhoven, "Large Eddy simulations of turbulent combustion: numerical study and applications,"
in Department of Fluid Mechanics, PhD Thesis: Vrije Universiteit Brussel, 2006.

[124]

M. Mulas and C. Hirsch, "Contribution in: EUROVAL a European Initiative on Validation of CFD
Codes," in EUROVAL an European Initiative on Validation of CFD Codes, Notes on Numerical Fluid

Mechanics. vol. 42, W. Haase, Ed.: Vieweg Verlag, 1993

[125]

F. P. Preparata and M. I. Shamos, Computational geometry : an introduction, Corr. and expanded 2nd
printing. ed. New York: Springer-Verlag, 1988.

[126]

M. Henle, A combinatorial introduction to topology. New York: Dover, 1994.

[127]

J. Torreele, D. Keymeulen, D. Vucinic, C. S. van den Berghe, J. Graat, and C. Hirsch, "Parallel CFView
: a SIMD/MIMD CFD Visualisation System in a Heterogeous and Distributed Environment," in

International Conference on Massively Parallel Processing, Delft, The Netherlands, 1994.

[128]

D. Vucinic, J. Favaro, B. Snder, I. Jenkinson, G. Tanzini, B. K. Hazarika, M. R. dAlcal, D.

Vicinanza, R. Greco, and A. Pasanisi, "Fast and convenient access to fluid dynamics data via the World
Wide Web," in ECCOMAS European Congress on Computational Methods in Applied Sciences and

Engineering 2000 Barcelona, Spain, 2000.

[129]

B. Shannon, Java 2 platform, enterprise edition : platform and component specifications. Boston ;:
Addison-Wesley, 2000.

[130]

S. Purba, High-performance Web databases : design, development, and deployment. Boca Raton, Fla.:
Auerbach, 2001.

[131]

A. Eberhart and S. Fischer, Java tools : using XML, EJB, CORBA, Servlets and SOAP. New York:
Wiley, 2002.

[132]

K. Akselvoll and P. Moin, "Large-eddy simulation of turbulent confined coannular jets " Journal of

Fluid Mechanics, vol. 315, pp. 387-411, 1996.

[133]

J. Favaro and e. al., "Strategic Analysis of Application Framework Investments," in Building

Application Frameworks: Object Oriented Foundations of Framework Design, 1999.

[134]

"QNET-CFD home page," http://www.qnet-cfd.net/, Ed., 2004.

[135]

D. Vucinic, D. Deen, E. Oanta, Z. Batarilo, and C. Lacor, "Distributed 3D Information Visualization,

Towards Integration of the dynamic 3D graphics and Web Services," in 1st International Conference

on Computer Graphics Theory and Applications Setbal, Portugal, 2006.

[136]

"SERKET project - home page,"

http://www.research.thalesgroup.com/software/cognitive_solutions/Serket/index.html, Ed.: Thales
Research & Technology, 2006.

[137]

"European Scientific Computing Organisation," in FP7 proposal in INFRA-2007-1.2.2: Deployment of

e-Infrastructures for scientific communities: ESCO consortium, 2007.

[138]

A. Fuhrmann, H. Loffelmann, D. Schmalstieg, and M. Gervautz, "Collaborative Visualization in

Augmented Reality," IEEE Computer Graphics and Applications, vol. 18, no. 4, pp. 54-59, 1998.

[139]

D. Santos, C. L. N. Cunha, and L. G. G. Landau, "Use of VRML in collaborative simulations for the
petroleum industry," in Simulation Symposium Proceedings, pages: 319-324, 2001.

231

[140]

G. Johnson, "Collaborative Visualization 101," ACM SIGGRAPH - Computer Graphics, pages 8-11,

volume 32, number 2, 1998.

[141]

Q. Shen, S. Uselton, and A. Pang, "Comparison of Wind Tunnel Experiments and Computational Fluid
Dynamics Simulations," in Journal of Visualization, volume 6, number 1, pp. 31-39, 2003.

[142]

G. Kwok-Chu-Ng, "Interactive Visualisation Techniques for Ontology Development," in University of

Manchester, Faculty of Science and Engineering, PhD Thesis, 2000.

[143]

C.-S. Jeong and A. Pang, "Reconfigurable Disc Trees for Visualizing Large Hierarchical Information
Space," IEEE Symposium on Information Visualization, pages 19-25. IEEE Visualization, 1998.

[144]

P. Jorissen, M. Wijnants, and W. Lamotte, "Dynamic Interactions in Physically Realistic Collaborative

Virtual Environments," in IEEE Transactions on Visualization and Computer Graphics, vol. 11, no. 6,

pp. 649-660 2005.

[145]

M. Kreuseler, N. Lopez, and H. Schumann, "A Scalable Framework for Information Visualization," in

IEEE Symposium on information Vizualization, INFOVIS. IEEE Computer Society, Washington, DC.,
2000.
[146]

G. Humphreys, M. Houston, Y.-R. Ng, R. Frank, S. Ahern, P. Kirchner, and J. T. Klosowski,

"Chromium: A Stream Processing Framework for Interactive Rendering on Clusters," in SIGGRAPH,
2002.

[147]

M. A. Linton, P. R. Calder, and J. M. Vlissides, "Interviews: A C++ Graphical Interface Toolkit,"

Technical Report: CSL-TR-88-358, Stanford 1988.

[148]

W. J. Schroeder, W. E. Lorensen, G. D. Montanaro, and C. R. Volpe, "VISAGE: an object-oriented

scientific visualization system," in Proceedings of the 3rd conference on Visualization '92 Boston,
Massachusetts: IEEE Computer Society Press, 1992.

[149]

E. L. William and E. C. Harvey, "Marching cubes: A high resolution 3D surface construction

algorithm," in Proceedings of the 14th annual conference on Computer graphics and interactive

techniques: ACM Press, 1987.

[150]

C. Hirsch, J. Torreele, D. Keymeulen, D. Vucinic, and J. Decuyper, "Distributed Visualization in CFD "

SPEEDUP Journal, vol. Volume 8, Number 1, 1994.

[151]

D. Vucinic, J. Torreele, D. Keymeulen, and C. Hirsch, "Interactive Fluid Flow Visualization with
CFView in a Distributed Environment," in 6th Eurographics Workshop on Visualization in Scientific

Computing, Chia, Italy, 1995.

[152]

M. Brouns, "Numerical and experimental study of flows and deposition of aerosols in the upper human
airways," in Department of Fluid Mechanics, PhD Thesis: Vrije Universiteit Brussel, to be completed
in, 2007.

[153]

D. Vucinic, D. Deen, E. Oanta, Z. Batarilo, and C. Lacor, "Distributed 3D Information Visualization,

Towards Integration of the dynamic 3D graphics and Web Services," in VISAPP and GRAPP 2006,

CCIS 4, Springer-Verlag Berlin Heidelberg, 2007, pp. 155168.

[154]

"Interactive Visualization - A State-of-the-Art Survey," in Advanced Information and Knowledge

Processing vol. to be published, E. Zudilova-Seinstra, T. Adriaansen, and R. v. Liere, Eds.: Springer,

UK, 2007.
[155]

"ILOG JViews Visualization Products," in 1987-2007 ILOG, Inc.:

http://www.ilog.com/products/jviews/, 2007.

232

[156]

J. J. Garrett, "Ajax: A New Approach to Web Applications," in

http://www.adaptivepath.com/publications/essays/archives/000385.php: Adaptive Path, LLC, 2007.

[157]

"Eclipse - an open development platform," in Copyright The Eclipse Foundation,

http://www.eclipse.org/ ed, 2007.

[158]

I. A. Salomie, R. Deklerck, A. Munteanu, and J. Cornelis, "The MeshGrid Surface Representation,"

Tech. Rep. IRIS-TR-0082 http://www.etro.vub.ac.be/Publications/technical_reports.asp, Vrije
Universiteit Brussel 2002.

[159]

A. Markova, R. Deklerck, D. Cernea, A. Salomie, A. Munteanu, and P. Schelkens, "Addressing viewdependent decoding scenarios with MeshGrid," in 2nd Annual IEEE Benelux/DSP Valley Signal

Processing Symposium - SPS-DARTS, (Paper 21) pp.71-74, Antwerpen, Belgium, 2006.

[160]

I. A. Salomie, A. Munteanu, A. Gavrilescu, G. Lafruit, P. Schelkens, R. Deklerck, and J. Cornelis,

"Meshgrid - A Compact, Multi-scalable and Animation-Friendly Surface Representation," in IEEE

Transactions on Circuits and Systems for Video Technology, vol. 14, no. 7, pp. 950-966, 2004.
[161]

M. Bourges-Sevenier and E. S. Jang, "An introduction to the MPEG-4 animation framework

eXtension," Circuits and Systems for Video Technology, IEEE Transactions on, vol. 14, pp. 928-936,
2004.

[162]

"Advanced Scientific Computing Research," in US Department of Energy:

http://www.science.doe.gov/obp/FY_07_Budget/ASCR.pdf, 2007.

233

Appendixes
Lookup table and its C++ implementation for the pentahedron cell
Label
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63

0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1

Node Mask
0 0 0 0
0 0 0 0
0 0 0 1
0 0 0 1
0 0 1 0
0 0 1 0
0 0 1 1
0 0 1 1
0 1 0 0
0 1 0 0
0 1 0 1
0 1 0 1
0 1 1 0
0 1 1 0
0 1 1 1
0 1 1 1
1 0 0 0
1 0 0 0
1 0 0 1
1 0 0 1
1 0 1 0
1 0 1 0
1 0 1 1
1 0 1 1
1 1 0 0
1 1 0 0
1 1 0 1
1 1 0 1
1 1 1 0
1 1 1 0
1 1 1 1
1 1 1 1
0 0 0 0
0 0 0 0
0 0 0 1
0 0 0 1
0 0 1 0
0 0 1 0
0 0 1 1
0 0 1 1
0 1 0 0
0 1 0 0
0 1 0 1
0 1 0 1
0 1 1 0
0 1 1 0
0 1 1 1
0 1 1 1
1 0 0 0
1 0 0 0
1 0 0 1
1 0 0 1
1 0 1 0
1 0 1 0
1 0 1 1
1 0 1 1
1 1 0 0
1 1 0 0
1 1 0 1
1 1 0 1
1 1 1 0
1 1 1 0
1 1 1 1
1 1 1 1

0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1

nF nN Edges Intersected
0 0
1 3
0-3-2
1 3
0-1-4
1 4
1-4-3-2
1 3
1-2-5
1 4
0-3-5-1
1 4
0-2-5-4
1 3
3-5-4
1 3
3-6-8
1 4
0-6-8-2
2 6
0-4-1, 3-6-8
1 5
1-4-6-8-2
1 6
1-2-3-6-8-5
2 7
4-7-6, 0-3-5-1
1
1
2
1
1
1
2

4
3
6
4
4
5
7

4-6-8-5
4-7-6
0-3-2, 4-7-6
0-1-7-6
1-0-3-6
0-6-7-5-2
3-5-7-6, 0-4-1

3-5-7-6

0-4-7-8-2

1-7-8-2

1
1

3
3

5-7-8
5-8-7

1-2-8-7

1
1

4
4

3-8-7-4
3-6-7-5

0-1-4-6-7-5-3

1
2
1

4
6
3

0-6-7-1
0-2-3, 4-6-7
4-6-7

2
1
2
1
1
1
1
1
1
1
1
1
0

6
5
6
4
3
3
4
4
3
4
3
3
0

1-5-2, 3-8-6
1-2-8-6-4
0-1-4, 3-8-6
0-2-8-6
3-8-6
3-4-5 *
0-4-5-2
0-1-5-3
1-5-2
0-2-3-4
0-4-1
0-2-3
-

Table 34: The lookup table for the pentahedron

234

/*---------------------------------------------------------------------------*/
/*
CLASS Cell DEFINITION
*/
/*---------------------------------------------------------------------------*/
/*
V U B
*/
/*
Department of Fluid Mechanics
Oct 1993 */
/*
Dean Vucinic
*/
/*---------------------------------------------------------------------------*/
/*
HEADER FILES
*/
/*---------------------------------------------------------------------------*/
// ----------------------- Pentahedron: -------------------------------

UCharVec

Cell3N6::NodesE_[9]={UCharVec("[0
UCharVec("[1
UCharVec("[2
UCharVec("[0
UCharVec("[1
UCharVec("[2
UCharVec("[3
UCharVec("[4
UCharVec("[5

// nodes(edge)
1]"),
2]"),
0]"),
3]"),
4]"),
5]"),
4]"),
5]"),
3]")};

// nodes(face)
Cell3N6::NodesF_[5]={UCharVec("[0 2 1]"),
UCharVec("[2 0 3 5]"),
UCharVec("[0 1 4 3]"),
UCharVec("[1 2 5 4]"),
UCharVec("[3 4 5]")};
const Cell3N6::TP2T4 Cell3N6::X_[64]=
{
{ 0,0,{{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
0,{{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 1,1,{{3,{ 0, 3, 2,-1,-1,-1},{ 2, 1, 0,-1,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
1,{{{ 0, 3, 2},{ 2, 1, 0}},{{-1,-1,-1},{-1,-1,-1}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 2,1,{{3,{ 0, 1, 4,-1,-1,-1},{ 0, 3, 2,-1,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
1,{{{ 0, 1, 4},{ 0, 3, 2}},{{-1,-1,-1},{-1,-1,-1}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 3,1,{{4,{ 1, 4, 3, 2,-1,-1},{ 3, 2, 1, 0,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
2,{{{ 1, 4, 3},{ 3, 2, 6}},{{ 1, 3, 2},{ 5, 1, 0}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 4,1,{{3,{ 1, 2, 5,-1,-1,-1},{ 0, 1, 3,-1,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
1,{{{ 1, 2, 5},{ 0, 1, 3}},{{-1,-1,-1},{-1,-1,-1}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 5,1,{{4,{ 0, 3, 5, 1,-1,-1},{ 2, 1, 3, 0,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
2,{{{ 0, 3, 5},{ 2, 1, 6}},{{ 0, 5, 1},{ 5, 3, 0}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 6,1,{{4,{ 0, 2, 5, 4,-1,-1},{ 0, 1, 3, 2,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
2,{{{ 0, 2, 5},{ 0, 1, 6}},{{ 0, 5, 4},{ 5, 3, 2}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 7,1,{{3,{ 3, 5, 4,-1,-1,-1},{ 1, 3, 2,-1,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
1,{{{ 3, 5, 4},{ 1, 3, 2}},{{-1,-1,-1},{-1,-1,-1}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 8,1,{{3,{ 3, 6, 8,-1,-1,-1},{ 2, 4, 1,-1,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
1,{{{ 3, 6, 8},{ 2, 4, 1}},{{-1,-1,-1},{-1,-1,-1}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 9,1,{{4,{ 0, 6, 8, 2,-1,-1},{ 2, 4, 1, 0,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
2,{{{ 0, 6, 8},{ 2, 4, 6}},{{ 0, 8, 2},{ 5, 1, 0}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 10,1,{{6,{ 0, 1, 4, 6, 8, 3},{ 0, 3, 2, 4, 1, 2}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
4,{{{ 0, 1, 4},{ 0, 3, 6}},{{ 0, 4, 6},{ 5, 2, 7}},
{{ 0, 6, 8},{ 6, 4, 8}},{{ 0, 8, 3},{ 7, 1, 2}}}},
{ 11,1,{{5,{ 1, 4, 6, 8, 2,-1},{ 3, 2, 4, 1, 0,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
3,{{{ 1, 4, 6},{ 3, 2, 6}},{{ 1, 6, 8},{ 5, 4, 7}},
{{ 1, 8, 2},{ 6, 1, 0}},{{-1,-1,-1},{-1,-1,-1}}}},
UCharVec

235

{ 12,1,{{6,{ 1, 2, 3, 6, 8, 5},{ 0, 1, 2, 4, 1, 3}},

{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
4,{{{ 1, 2, 3},{ 0, 1, 6}},{{ 1, 3, 6},{ 5, 2, 7}},
{{ 1, 6, 8},{ 6, 4, 8}},{{ 1, 8, 5},{ 7, 1, 3}}}},
{ 13,1,{{5,{ 0, 6, 8, 5, 1,-1},{ 2, 4, 1, 3, 0,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
3,{{{ 0, 6, 8},{ 2, 4, 6}},{{ 0, 8, 5},{ 5, 1, 7}},
{{ 0, 5, 1},{ 6, 3, 0}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 14,2,{{3,{ 0, 2, 3,-1,-1,-1},{ 0, 1, 2,-1,-1,-1}},
{4,{ 4, 6, 8, 5,-1,-1},{ 2, 4, 1, 3,-1,-1}}},
3,{{{ 0, 2, 3},{ 0, 1, 2}},{{ 4, 6, 8},{ 2, 4, 6}},
{{ 4, 8, 5},{ 5, 1, 3}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 15,1,{{4,{ 4, 6, 8, 5,-1,-1},{ 2, 4, 1, 3,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
2,{{{ 4, 6, 8},{ 2, 4, 6}},{{ 4, 8, 5},{ 5, 1, 3}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 16,1,{{3,{ 4, 7, 6,-1,-1,-1},{ 3, 4, 2,-1,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
1,{{{ 4, 7, 6},{ 3, 4, 2}},{{-1,-1,-1},{-1,-1,-1}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 17,1,{{6,{ 0, 4, 7, 6, 3, 2},{ 2, 3, 4, 2, 1, 0}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
4,{{{ 0, 4, 7},{ 2, 3, 6}},{{ 0, 7, 6},{ 5, 4, 7}},
{{ 0, 6, 3},{ 6, 2, 8}},{{ 0, 3, 2},{ 7, 1, 0}}}},
{ 18,1,{{4,{ 0, 1, 7, 6,-1,-1},{ 0, 3, 4, 2,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
2,{{{ 0, 1, 7},{ 0, 3, 6}},{{ 0, 7, 6},{ 5, 4, 2}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 19,1,{{5,{ 1, 7, 6, 3, 2,-1},{ 3, 4, 2, 1, 0,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
3,{{{ 1, 7, 6},{ 3, 4, 6}},{{ 1, 6, 3},{ 5, 2, 7}},
{{ 1, 3, 2},{ 6, 1, 0}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 20,1,{{6,{ 1, 2, 5, 7, 6, 4},{ 0, 1, 3, 4, 2, 3}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
4,{{{ 1, 2, 5},{ 0, 1, 6}},{{ 1, 5, 7},{ 5, 3, 7}},
{{ 1, 7, 6},{ 6, 4, 8}},{{ 1, 6, 4},{ 7, 2, 3}}}},
{ 21,2,{{3,{ 0, 4, 1,-1,-1,-1},{ 2, 3, 0,-1,-1,-1}},
{4,{ 3, 5, 7, 6,-1,-1},{ 1, 3, 4, 2,-1,-1}}},
3,{{{ 0, 4, 1},{ 2, 3, 0}},{{ 3, 5, 7},{ 1, 3, 6}},
{{ 3, 7, 6},{ 5, 4, 2}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 22,1,{{5,{ 0, 2, 5, 7, 6,-1},{ 0, 1, 3, 4, 2,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
3,{{{ 0, 2, 5},{ 0, 1, 6}},{{ 0, 5, 7},{ 5, 3, 7}},
{{ 0, 7, 6},{ 6, 4, 2}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 23,1,{{4,{ 3, 5, 7, 6,-1,-1},{ 1, 3, 4, 2,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
2,{{{ 3, 5, 7},{ 1, 3, 6}},{{ 3, 7, 6},{ 5, 4, 2}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 24,1,{{4,{ 3, 4, 7, 8,-1,-1},{ 2, 3, 4, 1,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
2,{{{ 3, 4, 7},{ 2, 3, 6}},{{ 3, 7, 8},{ 5, 4, 1}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 25,1,{{5,{ 0, 4, 7, 8, 2,-1},{ 2, 3, 4, 1, 0,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
3,{{{ 0, 4, 7},{ 2, 3, 6}},{{ 0, 7, 8},{ 5, 4, 7}},
{{ 0, 8, 2},{ 6, 1, 0}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 26,1,{{5,{ 0, 1, 7, 8, 3,-1},{ 0, 3, 4, 1, 2,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
3,{{{ 0, 1, 7},{ 0, 3, 6}},{{ 0, 7, 8},{ 5, 4, 7}},
{{ 0, 8, 3},{ 6, 1, 2}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 27,1,{{4,{ 1, 7, 8, 2,-1,-1},{ 3, 4, 1, 0,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
2,{{{ 1, 7, 8},{ 3, 4, 6}},{{ 1, 8, 2},{ 5, 1, 0}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 28,2,{{4,{ 1, 2, 3, 4,-1,-1},{ 0, 1, 2, 3,-1,-1}},
{3,{ 5, 7, 8,-1,-1,-1},{ 3, 4, 1,-1,-1,-1}}},
3,{{{ 1, 2, 3},{ 0, 1, 6}},{{ 1, 3, 4},{ 5, 2, 3}},
{{ 5, 7, 8},{ 3, 4, 1}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 29,2,{{3,{ 0, 4, 1,-1,-1,-1},{ 2, 3, 0,-1,-1,-1}},
{3,{ 5, 7, 8,-1,-1,-1},{ 3, 4, 1,-1,-1,-1}}},
2,{{{ 0, 4, 1},{ 2, 3, 0}},{{ 5, 7, 8},{ 3, 4, 1}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 30,2,{{3,{ 0, 2, 3,-1,-1,-1},{ 0, 1, 2,-1,-1,-1}},
{3,{ 5, 7, 8,-1,-1,-1},{ 3, 4, 1,-1,-1,-1}}},
2,{{{ 0, 2, 3},{ 0, 1, 2}},{{ 5, 7, 8},{ 3, 4, 1}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 31,1,{{3,{ 7, 8, 5,-1,-1,-1},{ 4, 1, 3,-1,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},

236

1,{{{ 7, 8, 5},{ 4, 1, 3}},{{-1,-1,-1},{-1,-1,-1}},

{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
32,1,{{3,{ 5, 8, 7,-1,-1,-1},{ 1, 4, 3,-1,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
1,{{{ 5, 8, 7},{ 1, 4, 3}},{{-1,-1,-1},{-1,-1,-1}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
33,1,{{6,{ 0, 3, 8, 7, 5, 2},{ 2, 1, 4, 3, 1, 0}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
4,{{{ 0, 3, 8},{ 2, 1, 6}},{{ 0, 8, 7},{ 5, 4, 7}},
{{ 0, 7, 5},{ 6, 3, 8}},{{ 0, 5, 2},{ 7, 1, 0}}}},
34,1,{{6,{ 0, 1, 5, 8, 7, 4},{ 0, 3, 1, 4, 3, 2}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
4,{{{ 0, 1, 5},{ 0, 3, 6}},{{ 0, 5, 8},{ 5, 1, 7}},
{{ 0, 8, 7},{ 6, 4, 8}},{{ 0, 7, 4},{ 7, 3, 2}}}},
35,2,{{3,{ 1, 5, 2,-1,-1,-1},{ 3, 1, 0,-1,-1,-1}},
{4,{ 4, 3, 8, 7,-1,-1},{ 2, 1, 4, 3,-1,-1}}},
3,{{{ 1, 5, 2},{ 3, 1, 0}},{{ 4, 3, 8},{ 2, 1, 6}},
{{ 4, 8, 7},{ 5, 4, 3}},{{-1,-1,-1},{-1,-1,-1}}}},
36,1,{{4,{ 1, 2, 8, 7,-1,-1},{ 0, 1, 4, 3,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
2,{{{ 1, 2, 8},{ 0, 1, 6}},{{ 1, 8, 7},{ 5, 4, 3}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
37,1,{{5,{ 0, 3, 8, 7, 1,-1},{ 2, 1, 4, 3, 0,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
3,{{{ 0, 3, 8},{ 2, 1, 6}},{{ 0, 8, 7},{ 5, 4, 7}},
{{ 0, 7, 1},{ 6, 3, 0}},{{-1,-1,-1},{-1,-1,-1}}}},
38,1,{{5,{ 0, 2, 8, 7, 4,-1},{ 0, 1, 4, 3, 2,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
3,{{{ 0, 2, 8},{ 0, 1, 6}},{{ 0, 8, 7},{ 5, 4, 7}},
{{ 0, 7, 4},{ 6, 3, 2}},{{-1,-1,-1},{-1,-1,-1}}}},
39,1,{{4,{ 3, 8, 7, 4,-1,-1},{ 1, 4, 3, 2,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
2,{{{ 3, 8, 7},{ 1, 4, 6}},{{ 3, 7, 4},{ 5, 3, 2}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
40,1,{{4,{ 3, 6, 7, 5,-1,-1},{ 2, 4, 3, 1,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
2,{{{ 3, 6, 7},{ 2, 4, 6}},{{ 3, 7, 5},{ 5, 3, 1}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
41,1,{{5,{ 0, 6, 7, 5, 2,-1},{ 2, 4, 3, 1, 0,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
3,{{{ 0, 6, 7},{ 2, 4, 6}},{{ 0, 7, 5},{ 5, 3, 7}},
{{ 0, 5, 2},{ 6, 1, 0}},{{-1,-1,-1},{-1,-1,-1}}}},
42,2,{{4,{ 0, 1, 5, 3,-1,-1},{ 0, 3, 1, 2,-1,-1}},
{3,{ 4, 6, 7,-1,-1,-1},{ 2, 4, 3,-1,-1,-1}}},
3,{{{ 0, 1, 5},{ 0, 3, 6}},{{ 0, 5, 3},{ 5, 1, 2}},
{{ 4, 6, 7},{ 2, 4, 3}},{{-1,-1,-1},{-1,-1,-1}}}},
43,2,{{3,{ 1, 5, 2,-1,-1,-1},{ 3, 1, 0,-1,-1,-1}},
{3,{ 4, 6, 7,-1,-1,-1},{ 2, 4, 3,-1,-1,-1}}},
2,{{{ 1, 5, 2},{ 3, 1, 0}},{{ 4, 6, 7},{ 2, 4, 3}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
44,1,{{5,{ 1, 2, 3, 6, 7,-1},{ 0, 1, 2, 4, 3,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
3,{{{ 1, 2, 3},{ 0, 1, 6}},{{ 1, 3, 6},{ 5, 2, 7}},
{{ 1, 6, 7},{ 6, 4, 3}},{{-1,-1,-1},{-1,-1,-1}}}},
45,1,{{4,{ 0, 6, 7, 1,-1,-1},{ 2, 4, 3, 0,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
2,{{{ 0, 6, 7},{ 2, 4, 6}},{{ 0, 7, 1},{ 5, 3, 0}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
46,2,{{3,{ 0, 2, 3,-1,-1,-1},{ 0, 1, 2,-1,-1,-1}},
{3,{ 4, 6, 7,-1,-1,-1},{ 2, 4, 3,-1,-1,-1}}},
2,{{{ 0, 2, 3},{ 0, 1, 2}},{{ 4, 6, 7},{ 2, 4, 3}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
47,1,{{3,{ 4, 6, 7,-1,-1,-1},{ 2, 4, 3,-1,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
1,{{{ 4, 6, 7},{ 2, 4, 3}},{{-1,-1,-1},{-1,-1,-1}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
48,1,{{4,{ 4, 5, 8, 6,-1,-1},{ 3, 1, 4, 2,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
2,{{{ 4, 5, 8},{ 3, 1, 6}},{{ 4, 8, 6},{ 5, 4, 2}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
49,2,{{4,{ 0, 4, 5, 2,-1,-1},{ 2, 3, 1, 0,-1,-1}},
{3,{ 3, 8, 6,-1,-1,-1},{ 1, 4, 2,-1,-1,-1}}},
3,{{{ 0, 4, 5},{ 2, 3, 6}},{{ 0, 5, 2},{ 5, 1, 0}},
{{ 3, 8, 6},{ 1, 4, 2}},{{-1,-1,-1},{-1,-1,-1}}}},
50,1,{{5,{ 0, 1, 5, 8, 6,-1},{ 0, 3, 1, 4, 2,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
3,{{{ 0, 1, 5},{ 0, 3, 6}},{{ 0, 5, 8},{ 5, 1, 7}},
{{ 0, 8, 6},{ 6, 4, 2}},{{-1,-1,-1},{-1,-1,-1}}}},

237

{ 51,2,{{3,{ 1, 5, 2,-1,-1,-1},{ 3, 1, 0,-1,-1,-1}},

{3,{ 6, 3, 8,-1,-1,-1},{ 2, 1, 4,-1,-1,-1}}},
2,{{{ 1, 5, 2},{ 3, 1, 0}},{{ 6, 3, 8},{ 2, 1, 4}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 52,1,{{5,{ 1, 2, 8, 6, 4,-1},{ 0, 1, 4, 2, 3,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
3,{{{ 1, 2, 8},{ 0, 1, 6}},{{ 1, 8, 6},{ 5, 4, 7}},
{{ 1, 6, 4},{ 6, 2, 3}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 53,2,{{3,{ 0, 4, 1,-1,-1,-1},{ 2, 3, 0,-1,-1,-1}},
{3,{ 3, 8, 6,-1,-1,-1},{ 1, 4, 2,-1,-1,-1}}},
2,{{{ 0, 4, 1},{ 2, 3, 0}},{{ 3, 8, 6},{ 1, 4, 2}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 54,1,{{4,{ 0, 2, 8, 6,-1,-1},{ 0, 1, 4, 2,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
2,{{{ 0, 2, 8},{ 0, 1, 6}},{{ 0, 8, 6},{ 5, 4, 2}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 55,1,{{3,{ 3, 8, 6,-1,-1,-1},{ 1, 4, 2,-1,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
1,{{{ 3, 8, 6},{ 1, 4, 2}},{{-1,-1,-1},{-1,-1,-1}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 56,1,{{3,{ 3, 4, 5,-1,-1,-1},{ 2, 3, 1,-1,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
1,{{{ 3, 4, 5},{ 2, 3, 1}},{{-1,-1,-1},{-1,-1,-1}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 57,1,{{4,{ 0, 4, 5, 2,-1,-1},{ 2, 3, 1, 0,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
2,{{{ 0, 4, 5},{ 2, 3, 6}},{{ 0, 5, 2},{ 5, 1, 0}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 58,1,{{4,{ 0, 1, 5, 3,-1,-1},{ 0, 3, 1, 2,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
2,{{{ 0, 1, 5},{ 0, 3, 6}},{{ 0, 5, 3},{ 5, 1, 2}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 59,1,{{3,{ 1, 5, 2,-1,-1,-1},{ 3, 1, 0,-1,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
1,{{{ 1, 5, 2},{ 3, 1, 0}},{{-1,-1,-1},{-1,-1,-1}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 60,1,{{4,{ 1, 2, 3, 4,-1,-1},{ 0, 1, 2, 3,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
2,{{{ 1, 2, 3},{ 0, 1, 6}},{{ 1, 3, 4},{ 5, 2, 3}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 61,1,{{3,{ 0, 4, 1,-1,-1,-1},{ 2, 3, 0,-1,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
1,{{{ 0, 4, 1},{ 2, 3, 0}},{{-1,-1,-1},{-1,-1,-1}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 62,1,{{3,{ 0, 2, 3,-1,-1,-1},{ 0, 1, 2,-1,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
1,{{{ 0, 2, 3},{ 0, 1, 2}},{{-1,-1,-1},{-1,-1,-1}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}},
{ 63,0,{{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}},
{0,{-1,-1,-1,-1,-1,-1},{-1,-1,-1,-1,-1,-1}}},
0,{{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}},
{{-1,-1,-1},{-1,-1,-1}},{{-1,-1,-1},{-1,-1,-1}}}}};
/*---------------------------------------------------------------------------*/
/*
CLASS Cell END OF DEFINITION
*/
/*---------------------------------------------------------------------------*/

238

Research Projects Timeline

1988-1991

ESA/ESTEC HERMES EURANUS project

European Aerodynamic Numerical Simulator

1990-1992

EUROVAL - EC/BRITE- EURAM project

An European initiative on validation of CFD codes

1992-1994

PASHA EEC/ESPRIT project

Parallel Software Hardware applications

1992-1995

PAGEIN - EEC/RACE project

Pilot Application on a Gigabit European Integrated Network

1993-1995

ECARP - EEC/BRITE- EURAM project

Validation of CFD models

1996-1998

LCLMS IWT project

Live Code Learning Multimedia System

1995-1999

CFDice.- FWO project

An Integrated Computational Environment for CFD

1998-2001

ALICE ESPRIT project

QFView - Quantitative Flow Field Visualization

2000-2004

QNET-CFD FP5 GROWTH project

A thematic network for quality and trust in the industrial application of
Computational Fluid Dynamics

2004-2005

LASCOT EUREKA ITEA project

Large Scale Collaborative Decision Support Technology

2004-2007

NUSIC EU Tempus project

Numerical Simulation Curricula

2005-2008

NSP-ME EU Tempus project

Numerical Simulation Program in Mechanical Engineering

2006-2007

SERKET - EUREKA ITEA project

Security Keeps Threats away

Table 35: Research Projects Timeline

239

Authors publications at VUB

1989
Vucinic D., Hirsch Ch., (1989). Computational Flow Visualization System at VUB (CFView 1.0), VKI Lecture
Series on Computer Graphics and Flow Visualization in CFD, Brussels, Belgium September 1989.
1991
Hirsch Ch., Lacor C., Dener C., Vucinic D. (1991). An Integrated CFD System for 3D Turbomachinery
Applications. In AGARD-PEP, 77th Symposium on CFD Techniques for Propulsion Applications, San Antonio,
Texas, May 1991, AGARD CP 510, pp. 17-1, 17-15.
Vucinic D. (1991). Object Oriented Programming for Computer Graphics and Flow Visualization, invited lecture,
VKI Lecture Series on Computer Graphics and Flow Visualization in CFD, Brussels, Belgium September 1991.
1992
Vucinic D., Pottiez M., Sotiaux V., Hirsch Ch. (1992). CFView - An Advanced Interactive Visualization System
based on Object-Oriented Approach. AIAA-92-0072, in AIAA 30th Aerospace Sciences Meeting, Reno, Nevada,
January 1992.
Hirsch Ch., Lacor C., Dener C., Vucinic D. (1992). An Integrated CFD System for 3D Turbomachinery
Applications. AGARD-CP-510, Paper No. 17, 1992.
1993
Vucinic D., Decuyper J., Keymeulen D., Hirsch Ch. (1993). Interactive Visualization techniques in CFD. First
International Conference in Visualisation & Intelligent Design in Engineering and Architecture VIDEA, Elsevier
Science Publishers, 1993.

1994
Torreele J., Keymeulen D., Vucinic D., van den Berghe C.S., Graat J., Hirsch Ch. (1994). Parallel CFView : a
SIMD/MIMD CFD Visualization System in a Heterogonous and Distributed Environment. Published in
Proceedings of the International Conference on Massively Parallel Processing, Delft, The Netherlands, June 1994.
Vucinic A., Hirsch Ch., Vucinic D., Dener C., Dejhalle R. (1994). Blade Geometry and Pressure Distribution
Visualization by CFView Method. Stojarstvo vol. 36, No. 1, 2, pp. 45-48.

1995
Vucinic D., Torreele J., Keymeulen D. and Hirsch Ch., Interactive Fluid Flow Visualization with CFView in a
Distributed Environment, 6th Eurographics Workshop on Visualization in Scientific Computing, Chia, Italy,
1995.

2000
Vucinic D., Favaro J., Snder B., Jenkinson I., Tanzini G., Hazarika B. K., Ribera dAlcal M., Vicinanza D.,
Greco R.and Pasanisi A., Fast and convenient access to fluid dynamics data via the World Wide Web, European
Congress on Computational Methods in Applied Sciences and Engineering ECCOMAS 2000, (2000).
Vucinic D., PIV measurements and CFD Computations of the double annular confined jet experiment , Pivnet
T5/ERCOFTAC SIG 32 2nd Workshop on PIV, Lisbon July 7-8 2000.
B. K. Hazarika and D. Vucinic, Integrated Approach to Computational and Experimental Flow Visualization of
a Double Annular Confined Jet, 9th International symposium on flow visualization, Edinburgh 2000.

240

2001
Hazarika B.K., Vucinic D., Schmitt F. and Hirsch Ch. (2001). Analysis of Toroidal Vortex Unsteadiness and
Turbulence in a Confined Double Annular Jet. AIAA paper No. 2001-0146, AIAA 39th Aerospace Sciences
Meeting & Exhibit, 8-11 January 2001, Reno, Nevada.
Vucinic D., Barone M.R., Snder B., Hazarika B.K. and Tanzini G. (2001). QFView an Internet Based
Archiving and Visualization System. AIAA paper No. 2001-0917, 39th Aerospace Sciences Meeting & Exhibit,
8-11 January 2001, Reno, Nevada.
D. Vucinic and B. K. Hazarika, Integrated Approach to Computational and Experimental Flow Visualization of
a Double Annular Confined Jet, Journal of Visualization, Vol.4, No. 3, 2001.
D. Vucinic, B. K. Hazarika and C. Dinescu, Visualization and PIV Measurements of the Axisymmectric InCylinder Flows, ATT Congress and Exhibition, Paper No. 2001-01-3273, Barcelona, 2001.
Charles Hirsch, Dean Vucinic, editors of the QNET-CFD Newsletter:
1st QNET-CFD Newsletter, Vol. 1, No. 1, January 2001, published in 1600 copies.
2nd QNET-CFD Newsletter, Vol. 1, No. 2, July 2001, published in 1600 copies.
2002
F. G. Schmitt, D. Vucinic and Ch. Hirsch, The Confined Double Annular Jet Application Challenge, 3rd
QNET-CFD Newsletter, January 2002.
Charles Hirsch, Dean Vucinic, editors of the QNET-CFD Newsletter:
3rd QNET-CFD Newsletter, Vol. 1, No. 3, January 2002 published in 1600 copies.
4th QNET-CFD Newsletter, Vol. 1, No. 4, November 2002 published in 1600 copies.
2003
Grijspeerdt K., Hazarika B. and Vucinic D. (2003). Application of computational fluid dynamics to model the
hydrodynamic of plate heat exchangers for milk processing. Journal of Food Engineering 57 (2003), pp. 237-242.
Charles Hirsch, Dean Vucinic, editors of the QNET-CFD Newsletter:
6th QNET-CFD Newsletter, Vol. 2, No. 1, April 2003 published in 1600 copies.
6th QNET-CFD Newsletter, Vol. 2, No. 2, July 2003, published in 1600 copies.
7th QNET-CFD Newsletter, Vol. 2, No. 3, December 2003, published in 1600 copies.
2004
Charles Hirsch, Dean Vucinic, editors of the QNET-CFD Newsletter:
8th QNET-CFD Newsletter, Vol. 2, No. 4, July 2004, published in 1600 copies.
2006
Dean Vucinic, Danny Deen, Emil Oanta, Zvonimir Batarilo, Chris Lacor, Distributed 3D Information
Visualization, Towards Integration of the dynamic 3D graphics and Web Services, 9 pages, 1st International
Conference on Computer Graphics Theory and Applications, Setbal, Portugal, February 2006.
2007
Koen Grijspeerdt and Dean Vucinic, Chapter 20. Computational fluid dynamics modeling of the hydrodynamics
of plate heat exchangers for milk processing in the book "Computational Fluid Dynamics in Food Processing",
25 pages, edited by Professor Da-Wen Sun and published by CRC Press, January 2007.
Dean Vucinic, Danny Deen, Emil Oanta, Zvonimir Batarilo, Chris Lacor, "Distributed 3D Information
Visualization, Towards Integration of the dynamic 3D graphics and Web Services," in VISAPP and GRAPP
2006, CCIS 4, Springer-Verlag Berlin Heidelberg, 2007, pp. 155168.

241

Research Developments Timeline

The development time of this thesis spans almost two decades of the authors research and development
activities, started when he joined VUB in September 1988. Somehow, these two decades naturally split his work
in two distinctive periods:
1988-1997

is the period when the scientific visualization (SV) system was created, and further evolved to
the industrial application

1998-2007

is the period when different parts of the developed methodology and software where applied to
the IWT and EC projects

1988-1997 period
1988
The SV system software development was initialized with a very immature C++ programming language preprocessor outputting C code, which needed to be further compiled in order to be executed. The available
hardware was APOLLO workstations running a proprietary PHIGS graphic library (completely vendordependent environment). The initialization of the object-oriented software design was done for 2D structured and
unstructured mono block data model.
1989
In the summer of 1989 the first implementation of the CFView software was accomplished, which was presented
in September at the VKI Computer Graphics Lecture Series [40]. To the authors knowledge, it was the first time
ever object-oriented interactive SV software for fluid flow analysis had been developed. It is obvious that more
powerful implementations existed, but these were not implemented with an object-oriented approach. That
summer, Prof. Hirsch made a firm decision of allowing the author to continue working on his object-oriented
methodology for developing CFView.
1990
As OOM was starting to gain acceptance, the C++ compiler came out. It immediately showed that C++
performance was starting to be comparable to C and FORTRAN. The CFView architecture had become more
elaborated; the GUI and Graphics categories of classes had been modeled to enable the upgrade of CFView to
the X-Windows C++ InterViews library from Stanford [147] and to the vendor-independent graphics library
Figaro/PHIGS. The 3D structured mono-block data model was developed.
1991
As mentioned in the Introduction, this year was crucial in the CFView development: the main visualization
system architecture was defined and the applied methodology was presented at the VKI Computer Graphics
Lecture Series [41]. The 3D structured data model was extended to the multi-block data sets. 1991 was also the
year when we received the first Silicon Graphics workstation running a specialized graphics library. It was the
first time that 3D user interaction was performed with an acceptable system response time.
1992
In January 1992, CFView was presented at the AIAA meeting [21]; to the authors knowledge, it was the firsttime ever OO-application created for interactive fluid flow analysis. CFView was compared with visualization
systems from NASA [16], MIT [17] and DLR [18] which had been implemented in C and developed with
structured programming methodologies. Another system, similar in design to CFView, was Visage [148], came

242

out later on in 1992 at the IEEE meeting; Visage was implemented in C and developed at General Electric by the
group of researchers which developed in the late 90s the OO VTK library .[27] In 1992, the CFView data model
was extended to unstructured meshes based on tetrahedrons only. An upgrade to the new version of
Figaro/PHIGS was done, which improved the CFView graphics performance.
1993
As the CFView application became more complex, the development of a data model, as presented in this thesis,
was conceived for supporting transparent interactivity with structured and unstructured data sets. In this year, the
marching cube algorithm [82, 149] was extended and became a marching cell algorithm for the treatment of
unstructured meshes with heterogeneous cell types [75], enhanced with unambiguous topology for the resulting
extracted surfaces. The Interviews library was upgraded to version 3.1 in order to support the CFView Macro
capability. The Parallel CFView was under ongoing development.
1994
The 3D structured and unstructured multi-block data model was completed, including the particle trace algorithm
for all cell types. The Parallel CFView was released and presented [127, 150]. As there were portability
problems with Figaro/PHIGS, the HOOPS graphics library was selected as the new 3D graphics platform for
CFView.
1995
The development of the HOOPS graphics layer was ongoing and the Model-View-Controller design was applied
to replace the interactive parts of CFView [151]. This process resulted in a cleaner implementation which was
then prepared for porting onto the emerging PC Windows platforms enhanced with 3D graphics hardware
running OpenGL. The PC software porting and upgrade were done later on in NUMECA.
1996
Under the LCLMS project, the symbolic calculator for CFView was developed and investigations were carried
out to find an appropriate GUI platform, which had to be PC-Windows/UNIX portable. The result was the
Tcl/Tk library, which was found appropriate; it is still today the GUI platform for CFView.
1997
The CFView multimedia learning material was prototyped in the IWT LCLMS projects; the author continued to
advocate the use of his development methodology in the EU R&D-project arena, which attracted interest in R&D
partners and succeeded in bringing new EU-funded projects to the VUB.

1998-2007 period
1998
The development of QFView in the EC ALICE project extended the authors research towards applying the
World Wide Web concept for designing and building distributed, collaborative scientific environments [47, 128].
The CFView data model was used for the development of QFView, which enabled combined visualization of
CFD and EFD data sets. As described in the Introduction, 3 test cases were performed using the developed
approach. It is important to mention that it is in the same project that the first PIV measurement system was
established at the VUB (a premiere in the Belgian universities). This influenced the Departments research to the
integrated application of CFD and EFD in fluid flow analysis; today, the PIV measurements at VUB are
performed with pulse laser [152].

243

2001
QNET-CFD was a Thematic Network on Quality and Trust for the industrial applications of CFD [134]. The
authors contributions were more of a coordination and management nature, as he was entrusted with presenting
and reporting on project activity, as well as preparing the publishing and dissemination material for the fluid
flow knowledge base, involving web-site software development and maintenance.
2004
The LASCOT project [55] was about applying the Model-View-Controller (MVC) paradigm to enhance the
interactivity of our 3D software components for: visualizing, monitoring and exchanging dynamic information,
including space- and time-dependent data. The software development included the integration and customization
of different visualization components based on 3D Computer Graphics (Java3D) and Web (X3D, SOAP)
technologies, and applying the object-oriented approach based on Xj3D to improve decision- making situational
awareness [153].
2006
The SERKET project -- currently in progress -- focuses on the development of more-realistic X3D models for
information visualization of security applications.

Summarizing the Current State-of-the-Art

Today, scientific visualization constitutes an important research area in Information Technologies (IT) and also
in Engineering Sciences, with its main objective of representing data in a variety of pertinent visual forms. SV is
here to help scientists and analysts, to better understand the results of their research, and to allow them to
effectively convey these results to others by means of graphics, pictures and other visual ways (see also our
Introduction, Section SV Software - state of the art and our final chapter Conclusion and Future
Development).
The current SV systems do not only provide visualization tools: they also support the interfacing with complex
computational codes, which facilitates the integration with the computing demands of expert users [10].

Interactive Visualization seems to continuously gain research interest [154], as there is a need to empower users
with tools for extracting and visualizing important patterns in very large data sets. Unfortunately, for many
application domains, it is not yet clear which are the features of interest, how to define them, let alone how they
can be detected. There is a continuous need to develop new ways, which enable more intuitive user interactions,
as a crucial element for further enhancement and exploration work. The interactive visualization tools for fluid
flow analysis were discussed in Chapter 2 Adaptation of Visualization Tools; no doubt that ongoing research
will lead to the expansion and the improvement of the capabilities of the tools that were developed in the context
of the authors work.
Several current visualization development frameworks focus on providing customized graphical components for
high-end desktops [155] and compatible with Ajax [156] and Eclipse [157] Open Source development
environments. These components are based on the MVC paradigm; they deliver point-and-click interaction tools
based on specialized SDKs, which allow the design of intuitive graphical displays for new applications. As
discussed in Chapter 3, Object-Oriented Software Development will remain an integral part of to-morrows
software development process.
Animation has become an important element of modern SV systems, and new video coding technologies will
need to be taken into account in future developments. An example of this new technology is MESHGRID,
developed at the VUB/ETRO [158-160] and adopted by MPEG-4 AFX [161]. The model-based representation
provided by the MESHGRID technology could be used for modeling time-dependent surfaces (in association with
244

the iso-surface algorithm discussed in this thesis) including video compression. This representation combines a
regular 3D grid of points, called the reference-grid, with the wire-frame model of a surface. The extended
functionality of MESHGRID could provide a hierarchical, multi-resolution structure for animation purposes,
which allows the highly compact coding of mesh data; it could be considered for integration in a new generation
of SV software for fluid flow time-dependent analysis. Other MESHGRID elements that could be considered for
SV software are view dependency (not considered in this thesis) and Region of Interest (ROI) -- a concept which
can be associated to the multi-block (multi-domain) model of fluid flow data; it offers the possibility of
restricting visualization to a limited part of the whole data set. The MESHGRID data model seems appropriate for
implementation in distributed environments since it provides advanced data compression and data transfer
techniques that are important for the quality of service in animation enhanced software.
Concerns of performance are at the core of the development of VS systems distributed over desktops. Such
systems are capable of very large-scale parallel computation and of distributed rendering on large display walls
or immersive virtual environments. Today, modern graphics hardware (Graphics Processing Units -- GPU)
performs complex arithmetic at increasingly high speed. It can be envisaged that GPU-s can be used to execute
SV non-graphics algorithms, which offers a potential for increasing VS performance for large and complex data
sets without sacrificing interactivity.
Current trends in SV research and development help advance the state-of-the-art of computational science and
engineering by:

investigating high performance computing;

exploring collaborative approaches that encompass multi-disciplinary, team-oriented concepts

applying distributed software with visualization and virtual reality components that improve the
ergonomics of human-machine interfaces.

These aspects are further covered in the concluding Chapter Development trends in Interactive Visualization
systems.
The 2007 Advanced Scientific Computing Research program of the US Department of Energy [162] includes
SV research and development work in relation to advances in computer hardware (such as high-speed disk
storage systems, archival data storage systems) and high-performance visualization hardware. An example of a
high-performance computer network is the UltraScienceNet Network (USNET) Testbed, a 20 gigabit- persecond, highly-reconfigurable optical network that supports petabyte data transfer, remote computational
steering and collaborative high-end visualization. USNET provides capacities that range from 50 megabits-persecond to 20 gigabits-per-second. Such capability is in complete contrast with the Internet where the shared
connections are provided statically with a resulting bandwidth that is neither guaranteed nor stable.
Ongoing research in visualization tools for scientific simulation is exploring hardware configurations with
thousands of processors and developing data management software capable of handling terabyte-large data sets
extracted from petabyte-large data archives. It is predicted that large-scale, distributed, real-time scientific
visualization and collaboration tools will provide new ways of designing and carrying out scientific work, with
distributed R&D teams in geographically distant institutes, universities and industries accessing and sharing in
real-time extremely powerful computational and knowledge resources, and yet to be measured gains in
efficiency and productivity.

245

Performance Analysis
In order to assess the capability of the CFView system, a performance analysis was conducted in the EC PASHA
project [127, 150]. The main goal was to see how two parallel SIMD and MIMD implementations of CFView
would perform compared to its sequential implementation on a SISD stand-alone machine. The benchmarks
consisted of test cases specifically designed to provide a meaningful, reliable basis for comparing the
performances of the implementations. The test cases were theoretically-minded, not necessarily representative of
practical applications. These test cases were retained because they offered significant advantages over real-world
examples, as follows:
1.

The test cases were not biased towards any particular algorithms, computer systems or applications.
This property makes them good candidates for comparisons between different systems.

2.

The test cases were based on simple, algorithmic definitions. This made them readily available to
anyone wanting to perform similar experiments.

3.

The test cases were all different, so as to avoid the testing processes to focus upon the singularities of
one particular data set.

4.

Single test cases were devised to explore a wide spectrum of difficult, uncommon, challenging or
otherwise interesting characteristics. This enabled us to considerably speed up the testing, debugging
and optimization procedures.

5.

The test cases were designed to offer the user the possibility to control the complexity of the problem
(e.g. control over the size and complexity of iso-surfaces).

The test cases that were developed and used are described below. An overview of the testing environment is
then given, including a description of the hardware used, of the algorithms tested and of the measurements taken.
The results of the experiments are reported and summarized in tables. A discussion of the results is provided
showing the performance and characteristics of the systems and algorithms under test. Finally, some conclusions
are presented.

Theoretical Test Cases

Specifically designed test cases have been used to benchmark the distributed, SIMD and MIMD CFView
implementations. For each test case, one low- and one medium- volume cases were considered. Specifically,
for the cutting plane and the iso-surface experiments, one used a geometrical data set (the mesh) and a scalar
quantity data set (the scalar field). The dimensions of the low- and medium-volume cases were respectively of
20*20*20 (8000 vertices) and 50*50*50 (125000 vertices). Size-wise, the full test cases -- the mesh plus the
scalar field -- add up to some 128K (low) and 2MB (medium) of data.
The data were generated by the following piece of C-code:
extern random()

/* returns random number: 0 <= random() < 1 */

{
FILE *geo_file, *scal_file;
int i, j, k;
int max_i, max_j, max_k;
/* data for the low data volume case: 20*20*20 vertices */
max_i = 20;
max_j = 20;
max_k = 20;

/* number of vertices in dimension i */

/* number of vertices in dimension j */
/* number of vertices in dimension k */

*/ open the data files /*

geo_file = fopen("low-20.geo", "w");
scal_file = fopen("low-20.scal", "w");

246

/* random data generation */

for (k=0; k<max_k; k++){
for (j=0; j<max_j; j++){
for (i=0; i<max_i; i++){
fprintf( geo_file, "%f %f %f\n",
random()+(float) i,
random()+(float) j,
random()+(float) k
);
fprintf( scal_file, "%f\n", 2 * random() - 1 );
}
}
}

This code generates structured meshes with a random distortion and scalar fields with values that randomly vary
between -1 and 1. The mesh vertices coordinates are the (i,j,k) indices of the corresponding mesh cell. Each
coordinate is incremented with a positive random number strictly smaller than 1, which leads to the definition of
a structured mesh with strongly deformed cells.
The particle-tracing benchmarks were run with a helical vector-field data set defined over a regular (not
distorted), structured mesh. Here also, a low- and a medium-volume test cases were considered, respectively with
10*10*10 (1000 vertices) and 30*30*30 (= 27000 vertices). The helical field (u,v,w) was computed for every
vertex (x,y,z) using the spiral equation:

(u,v,w) = (ax-by, bx+ay, -2az+c)

with parameters set to: a=0; b=1; c=1. The full test cases (mesh and helical field) accounted for some 25K (low)
and 650K (medium) of data.

Hardware Environment
SISD computer:
HP 9000/735 (under HP-UX 9.0.1)
99 MHz clock speed
96 MB RAM
CRX-24Z graphics board
124 MIPS peak performance
40 MFLOPS peak performance
SIMD computer:
CPP DAP 510C-16
1024 bit-processors each with a 16-bit floating point coprocessor
16 MB RAM (shared)
140 MFLOPS peak performance
SCSI connection with front end
MIMD computer:
Parsytec GC-1/32 (under Parix)
32 processors (Inmos T8, RISC)
128 MB RAM (4 MB local memory per processor)
140 MFLOPS peak performance
S-bus connection with front end.
To be able to visualize the results of the computations, the parallel computers were linked to a front-end
workstation running the CFView interface. This work-station was the computer separately used for the SISD
benchmarks (see above). The SISD machine was connected to a Parallel Server via an Ethernet LAN (10
Mbits/s). The Parallel Server was running the largest part of the Interface Framework, built on top of PVM, see
Figure 194. It communicates with the SIMD and MIMD computers through its SCSI-bus and through its internal
S-bus respectively. For the Parallel Server machine the SUN workstation with the following characteristics was
used:

247

SUN SPARCstation 10/30GX

SuperSPARC processor
32 MB RAM
86 MIPS peak performance
10 MFLOPS peak performance

A schematic view of the environment used for the benchmarking is shown in Figure 222.

HP 9000/735
Graphical
Workstation

Ethernet
Local Area
Network
10 Mbit/s

Rendering,
Viewing, Filtering,
User Interface,
Controller, Data
Module and
local manipulation
of images

USER

SUN SparcStation 10
Parallel Server

SCSI

S-bus

SIMD

MIMD

CPP DAP
510C-16

Parsytec
GCel-1/32

Figure 222: Overview of the heterogeneous and distributed environment used for the theoretical benchmarks

Algorithms Parameters
The performance tests were done for 3 extraction algorithms: cutting-plane, iso-surface and particle-tracing. For
each algorithm, the tests were run with parameter ranges chosen to obtain data from very low computational load
(e.g. a plane that intersects only with a very small part of the mesh) to very high load (e.g. a plane that intersects
a very large part of the mesh).
For the cutting-plane algorithm, the intersection plane is defined by the normal (1,1,1) and the intersection with
the X-axis linearly varying between 0 and 2 times the x-dim of the mesh. Given the characteristics of the meshes
used, this means that the largest intersection will be found at x=1.5*x-dim.
For the iso-surface computations, a range of iso-values was chosen to generate an (approximately) linearlyvarying number of triangles (out of which the iso-surface is built) keeping their number in the range, which the
computers could handle.
For the particle-tracing benchmarks, the number of particles launched (at some fixed time t) was varied. In a
parallel configuration, multiple vector-field lines may be computed at the same time from a single data set: this is
achieved, essentially, by allowing the parallel processors to work simultaneously on different field lines. For the
CFView user, launching many particles at the same time (usually by distributing them evenly along a line
segment) is a common procedure. Typically, the user first looks for interesting regions of the model by studying
the traces of individually-placed particles. Then, the user positions a suite of particles in a region of interest.

248

(a)

(b)

Figure 223: The theoretical random-base meshes (a) 20x20x20 (b) 200x200x250
Being located fairly close to each other, the computed particles traces form a ribbon whose motion shows the
particularities of the flow field. Clearly, one expects that the (speed-up) effects of parallel computation to be
comparatively stronger at high particle numbers. By varying the number of particles from a few to very many,
one was able to explore the effects of parallelism with respect to sequential computation.
Figure 223 (a) illustrates the benchmarking mesh data (20x20x20 vertices) used in the PASHA project. Figure
223 (b) shows the 10-million-point mesh (~300MB data on disk), which was created (on the very week of
publication of this thesis) to demonstrate that CFView is capable of handling data sets of higher orders of
magnitude. The 3 visualization algorithms were executed on such large data sets. Figure 224 (a) shows a cutting
plane with particle traces illustrating the benchmarked helical vector field. Figure 224 (b) shows the iso-surface
extracted from the complex- and unconventional-geometry test case. The apparently-perfect spherical shape of
the iso-surface demonstrates the correctness of the extracted data. The clean helical geometry of the particle
traces is a visual indicator of the correctness of the algorithm applied here in trying conditions, namely to a
vector field defined on a distorted mesh with 200x200x250 vertices. For the record, these computations were run
on a Dell XPS-M2010 with Intel Dual Core 2 CPU @ 2.33 MHz and 2GB RAM.

(a)

(b)

Figure 224: Mesh size 200x200x250 (a) Cutting plane and Particle traces (b) Isosurface

249

Measurements Timing
When analyzing the performance of distributed computation, one must account for the (unavoidable) overhead
due to the network background tasks and the multiple processes that govern the calculations (e.g. their
synchronization). This is why time in the tests was measured using wall-clock time as CPU time was not
available. The time figures used include overhead due to the activity of various UNIX processes and to the
network load caused by system activities (swapping, saving, etc.).
Time was measured using the C function ftime with a resolution of about 1 millisecond. Measurements were
made on the individual workstations involved. The network times were either inferred from the measured times
(when possible) or computed through explicit handshaking between the communicating processes.
In order to obtain realistic performance figures, the measurements were performed in a typical usage situation,
i.e. with a normal network load, normal computer load, etc. In this approach, the part which includes
overhead in the timing can be considered as normal -- and indeed unavoidable for the user.
Because of the heterogeneous and distributed nature of the test implementation (SIMD/MIMD), timing a test run
requires combining several measurements obtained on sub-parts of the system. Every single run of a particular
algorithm was divided in a number of consecutive stages, and timing measured for each stage, namely:

PVMRcvGeo is the time it takes for the geometry data to go over the LAN from CFView to the parallel
server.

GCSndGeo or DAPSndGeo is the time it takes for the geometry data to go from the parallel server to
the parallel machines (Parsytec GC and DAP respectively).

PVMRcvScal is the time it takes for the scalar quantity data to go over the LAN from CFView to the
parallel server.

GCSndScal or DAPSndScal is the time it takes for the scalar quantity data to travel from the
parallel server to the parallel machines (Parsytec GC and DAP respectively).

PVMRcvParam is the time is takes for the parameter values (equation of the plane, the particular isovalue, the initial positions of the particles, etc.) to travel over the LAN from CFView to the parallel
server.

GCExe or DAPExe is the time it takes for the respective parallel machines to execute the given
algorithm using the data stored in their own memory.

GCRcvResults or DAPRcvResults is the time it takes for the respective parallel machines to
return the results of their computation (collections of triangles or streamlines) to the parallel server.

PVMSndResults is the time it takes for the computation results to travel over the LAN, from the
parallel server to CFView (running on a workstation).

Render is the time it takes for CFView to render the results on the screen of the workstation it is
running on.

Measurements Results
We present below the characteristic results of the benchmarks. A result table is given for each algorithm; it
shows averaged algorithm-execution times on the different systems for several test cases. Averages are
computed over 20 runs for each test case. All time values in the Tables below are in seconds.
For the CFView SIMD and MIMD implementations, the following average time values are given:

(Send Data): time needed to send the data (mesh and scalar quantity) from the Workstation to the Parallel
Server;

250

 (Load Data). time needed for loading the data from the Parallel Server onto the parallel machines;
(Execute): execution time on the parallel implementations, which includes:
(i)

sending of the algorithmic parameters from the Workstation to the Parallel Machines;

(ii)

execution of the algorithm on the parallel machines;

(iii)

sending the result data from the parallel machines to the Parallel Server; and

(iv)

sending the result data from the Parallel Server to the Workstation.

For the CFView SISD implementation, only the averaged execution time is meaningful and shown. The averaged
number of triangles generated by the algorithm and the number of particles traced are given where relevant. All
times shown are averaged over 20 different runs, with parameters varying (see Algorithmic parameters above).
The total execution time for the parallel implementations is the sum of the three timings: Send, Load, and
Execute. However, by making use of a caching mechanism in the Interface Framework, the sending and the
loading of the data needs to be done once only (in the beginning). Thereafter, data to be transmitted to the
parallel machines include only a (new) cutting-plane equation, an iso-value, or a set of initial particle positions,
after which the execution can take off. Hence, in a realistic situation, only the times in the (Execute) column are
relevant for comparison.
Mesh size
low

medium

#Triangles
915

10150

System

Send Data

SISD
SIMD
MIMD
SISD
SIMD
MIMD

---0.26
0.26
---3.66
3.66

Load Data

Execute

---0.48
1.96
---5.82
28.86

1.04
1.16
1.63
6.00
3.63
4.44

Table 36: Average times for Cutting Plane (wall-clock time in seconds)
Mesh size

#Triangles

low

11145

medium

10150

System

Send Data

SISD
SIMD
MIMD
SISD
SIMD
MIMD

---0.26
0.26
---3.66
3.66

Load Data
---0.48
1.96
---5.82
28.86

Execute
9.96
6.05
5.13
27.47
6.62
4.96

Table 37: Average times for Isosurface (wall-clock time in seconds)

For the Particle Trace measurements the initial positions of the particles were evenly distributed along an
imaginary line segment within the meshes. The particle trace input parameters were set as follows:
average number of integrations per cell:
10
maximum length of 1 particle path:
1000
Forward and Backward tracing
yes
The results of the experiments are summarized in Table 39. It shows the average execution times for the tracing
of N particles, with N = 2, 5, 10, 15... 50.
Mesh size

#Particles

low

25

medium

25

System

Send Data

SISD
SIMD
SISD
SIMD

---0.34
---1.79

Load Data
---0.35
---1.90

Table 38: Average times for Particle Trace (wall-clock time in seconds)

251

Execute
67.04
17.14
85.27
27.18

#particles
2
5
10
15
20
25
30
35
40
45
50
25

low data volume (10*10*10)

SISD
SIMD
5.81
4.43
14.12
8.76
38.04
11.67
36.70
15.11
66.28
15.79
71.28
16.80
85.78
19.05
82.13
21.08
99.26
23.27
115.01
25.04
122.99
27.58
Average
67.04
17.14

medium data volume (30*30*30)

SISD
SIMD
6.23
4.81
17.70
9.59
35.53
14.54
51.89
19.32
71.16
23.64
79.10
27.51
95.68
31.64
113.94
35.87
141.38
40.04
155.14
43.95
170.30
48.06
85.27

27.18

Table 39: Evolution of the execution times in seconds with the number of particles used
#Processors

MIMD Calculate

MIMD Retrieve

MIMD Send

2
4
8
12
16

1.9
1.49
1.33
1.26
1.21

1.47
1.47
1.49
1.51
1.54

0.39
0.40
0.39
0.39
0.41

Table 40: Execution times in seconds for Isosurface on MIMD for different machine configurations (wall-clock
time) with varying number of processors
Since the cutting plane and isosurface algorithms are conceptually very similar [151], comparing the
performance results for them makes sense from a users point of view. The timing values as seen by the user are
consolidated in Figure 225. The chart shows the averaged execution times for the cutting-plane and iso-surface
algorithms on the different machines. For the SIMD and MIMD implementations, the execution times are subdivided in three parts:
(i)
the actual algorithm-execution time on the given machine (Calculate);
(ii)
the time it takes to transfer the computation result data from the parallel machine to the
Parallel Server (Retrieve Results); and
(iii)
the time needed to send the result data from the Parallel Server to Parallel CFView
running on the Workstation (Send Results).
30

25
20

Send Results

15

Retrieve Results
Calculate

10
5

0
SISD

SIMD

MIMD

SISD

SIMD

MIMD

CUTTING PLANE

ISOSURFACE

Figure 225: Average execution times in seconds for the algorithms on the different machines (with caching
mechanism enabled for the parallel implementations).
252

Average Time [s]

18
16

SIMD Receive Results

14
SIMD Send Results

12

SIMD Calculate

10
8

MIMD Receive Results

MIMD Send Results

4
MIMD Calculate
2
28195

23456

18992

14582

10004

4834

912

476

40

Number of triangles

Figure 226: Average execution times in seconds for the SIMD and MIMD implementations of the isosurface
algorithm, with respect to the number of triangles generated (caching mechanism on)
As can be seen on the chart, the execution times for the parallel implementations are significantly shorter than
for the sequential one. This is especially visible for the iso-surface algorithm because it requires a sweep
through the complete mesh. Note that the machine running the sequential CFView was a state-of-the-art HP
9000/735workstation, at the time amongst the fastest in its category. This demonstrated that the parallel CFView
SIMD/MIMD implementations offered intrinsically more power than the sequential CFView.
A second remarkable observation related to the distribution of computational load on the parallel machines.
Although both implementations are approximately equally fast, it was found that the MIMD machine spent most
of its CPU time for transferring results data to the parallel server, whereas the SIMD machine was busier with
the actual computation. The reason for this is, of course, the overhead caused by having to route the data
between the different processors of the MIMD machine. Figure 226 shows this difference more clearly.
As seen from the chart above, the total execution time (as seen by the user) varies with the number of triangles
generated, that is, with the complexity of the computation. On the SIMD machine, the largest contributing factor
is calculation -- the time for computing the iso-surface. On the MIMD implementation, however, calculation
times (MIMD Calculate) are fairly independent of the number of triangles generated. The time required to move
the results from the MIMD machine to the Parallel Server (MIMD Retrieve Results) is, by contrast, largely
dependent on the number of triangles. As mentioned earlier, this behavior results from the need to route the
results of the computations by the different processors to the master processor (which governs the
communications with the parallel server, which is a time-consuming process.
An interesting observation can be made concerning the difference in performance between the parallel machines.
As long as the number of triangles generated is small, the SIMD machine outperforms the MIMD machine. But
when the task requires more and more computation (more intersections need to be calculated and more triangles
need to be constructed), the situation reverses. This phenomenon can be explained as follows: for implementing
the iso-surface algorithm in a fully parallel mode, one requires the ability to index the data set in parallel, a
feature which is not available on the SIMD machine. Therefore, part of the computation is implemented as a
serial loop over the edges of the intersected mesh cells. As the number of intersections (hence the number of
triangles) increases, the serial part of the algorithm tends to dominate the computation and slows down the
overall execution. One can also see in the charts that the local networking tasks (SIMD/MIMD Send Results),
which ensure the communication in the heterogeneous distributed environment, account for only a small fraction

253

of the total time. This suggests that a heterogeneous, distributed approach is feasible since it does not impose
unacceptable overhead on the overall performance of the system.
A comparison of the execution times for particle-tracing on the SIMD and SISD machines can be found in Figure
227. As one can infer from the chart, the execution times on the SIMD machine raise linearly with the number of
particles traced. The performance of the SISD machine quickly degrades as the number of particles increases;
mesh size does not seem to influence this behavior. The chart shows that the parallel implementation succeeds in
keeping execution times at an acceptable level, even for very large number of particles.
The SIMD implementation seems to be able to fully exploit the computing power that results from the
distribution of the particle-tracing calculations onto several independent parallel processors.
The benchmarking experiment on the SIMD and MIMD CFView systems has shown comparable performance for
the two parallel machines. The execution times were found to be comparatively more sensitive to the amount of
computation required on the SIMD machine, and relatively more dependent on the amount of data to be routed
between the different processors on the MIMD implementation. The overhead induced by the Interface
Framework (developed for communication between the different machines) was seen to contribute to a small
fraction only of the overall execution times.
It was also observed that the SIMD implementation really takes advantage of its multi-processor structure for
particle-tracing. Indeed, the SIMD machine was demonstrated to be the only useable machine for tracing
moderately-large numbers of particles.
The overall performance of the SIMD and MIMD Parallel CFView implementations were shown to be
significantly better than the performance of the SIMD sequential version, especially for computationallyintensive operations, such as iso-surface construction on problems with large data volumes or particle-tracing
with large numbers of particles.
Overall, the benchmarking of CFView has demonstrated that heterogeneous and distributed SV system
configurations were indeed a viable proposition. This opens the interesting prospect of transparently using VS
systems capable of harnessing the computing power of different computing platforms and taking advantage of
geographically-distant, parallel machines.

150

130

110
Low SISD
90
Low SIMD

time(s)
70

Medium SISD
Medium SIMD

50

30

10

-10

10 15 20 25 30 35 40 45 50

#particles

Figure 227: Execution times in seconds for particle tracing with respect to the number of particles

254

VUB Double Annular Jet Experiment in turbulent flow regime:

LDA (Laser Doppler Anemometry)
CFD with Baldwin-Lomax model
LSV (Laser Sheet Visualization) during transition
PIV (Particle Image Velocimetry) averaged from 100 frames

255

256