Evolving Time Surfaces in a Virtual Stirred Tank - Semantic Scholar
Evolving Time Surfaces in a Virtual Stirred Tank Bidur Bohara, Farid Harhad, Werner Benger, Nathan Brener, S. Sitharama Iyengar, Bijaya B. Karki, Marcel Ritter, Kexi Liu, Brygg Ullmer, Nikhil Shetty, Vignesh Natesan, Carolina Cruz-Neira, Sumanta Acharya, Somnath Roy
WSCG 2010, Plzen, Czech Republic
LOUISIANA STATE UNIVERSITY
1
Outline • Overview of Integral Surfaces • Dataset & Data Model • Generation of Time Surfaces • Deployment to End User • Conclusion
LOUISIANA STATE UNIVERSITY
2
Motivation • extract the features of flow field from a very large time dependent CFD data set. • visualize the spatially evolving features of fluid flow in space-time domain • analysis tool for illustrating the movement (dispersions) of particles in a closed stirred tank system
LOUISIANA STATE UNIVERSITY
3
Motivation • stirred tanks – commonly used in industries • improvements in design can translate into billions of dollars, but better design comes with better understanding of mixing in tank • develop tool to analyze the mixing condition in such system
LOUISIANA STATE UNIVERSITY
4
Related Work • Stream surfaces, most commonly investigated technique for surfaces visualization • first algorithm given by Hultquist (1992) • Krishnan et. al (2009 IEEE), paper discusses the generation of Time and Streak surfaces in Large Time-varying data. • planer topology, as compared to our spherical topology
LOUISIANA STATE UNIVERSITY
5
Time Surfaces over Pathlines • Pathlines : most common approach to visualize the flow behavior in fluid system • Useful for few particles and for few time-steps.
(a) 23 particle system
(b) 516 particle system
• For large number of particles and longer time steps within enclosed flow system Pathlines => Spaghetti LOUISIANA STATE UNIVERSITY
6
Time Surfaces over Pathlines • analyzing flow field as evolving surfaces is more relevant in context of enclosed system
WHAT IS TIME SURFACE ? • evolving surfaces over time • higher-dimensional extensions of time lines ( evolution of seed line) • integration of surfaces over time, that can illustrate key flow characteristics
( such as dispersion of particle system in a flow)
LOUISIANA STATE UNIVERSITY
7
Stirred Tank Dataset • Provided by Dr. Sumanta Acharya and Somnath Roy, Department of Mechanical Engineering, Louisiana State University • 2088 curvilinear blocks • comprised of 3.1 million cells • velocity as vector field and pressure as scalar field
• each time slice has velocity for each grid point in all blocks • total of 5700 time steps; 500 GB binary data
LOUISIANA STATE UNIVERSITY
8
Stirred Tank Dataset CHALLENGES ?? • handling and processing of the - voluminous - multi-block - non-uniform curvilinear, dataset • to generate time surfaces and track set of particles in flow
LOUISIANA STATE UNIVERSITY
9
Data Model • based on concept of Fiber bundle data model • internal data structure with six levels, each composing of arrays representing properties of data set
Bundle
Time Slice
Grid
Topology
Representation
Field
• user only deals with • Bundle • Grid • Field
LOUISIANA STATE UNIVERSITY
10
Data Model For Stirred Tank data set and Time Surfaces
• Field
– coordinates, velocity, pressure, connectivity
• Grid
– collection of all Field entities (one Grid object per slice in our implementation)
• Bundle – entire dataset / sequences of Grid objects for all time slices
LOUISIANA STATE UNIVERSITY
11
Time Surface module • inputs : Data Bundle, Grid object with Seeding points and connectivity • output : Bundle as collection of Time Surfaces (Grids) F5 Stirred tank data. BUNDLE Seeding Input GRID
In : Bundle
Out : Grid
GRID Integration
Seeding Grid = Time Surface Grid
Out : Grid
In : Grid Vector Field Out : Field
In : Grid In : Field
Surface Computation Out : Grid
In : Grid Time Surfaces as BUNDLE Out : Bundle
LOUISIANA STATE UNIVERSITY
12
Particle Seeding & Advection • seeding input is the set of points and connectivity information among seed points • connectivity information is to create the triangle mesh for surface generation • the triangular mesh evolves over time, different to spanning surface out of line segment as in Hultquist’s approach
LOUISIANA STATE UNIVERSITY
13
Seeding Input • seeds points – set of particles lying on the surface of sphere
Seeding input as points
Seeding input with triangular mesh
LOUISIANA STATE UNIVERSITY
14
Triangular Mesh Refinement • over integration time the triangular mesh of surface enlarges resulting unsmooth evolving surfaces. • adaptive surface refinement approach is mandatory for high quality output Surface Refinement Criteria: • Edge Length • Triangle Area
a
a d
d
f
c b
e
c
b
LOUISIANA STATE UNIVERSITY
15
Time Surfaces
Evolution of two spheres at time slices 0, 50, 100, 125, & 150 from left-top to bottom. [Top View] LOUISIANA STATE UNIVERSITY
16
Out of Core Memory Management • partial access and handling of input data • Slice-wise sequential access and Block-wise random access • each time slice with corresponding Grid Object is processed only once ( Slice-wise)
Time surface computed from vector field in 2088 fragments(curvilinear grids) covering the Stirred tank. LOUISIANA STATE UNIVERSITY
17
Out of Core Memory Management • each Grid object consists of vector field information of 2088 blocks • accessing only those blocks that the particles touches at particular time rather than all 2088 blocks( Block-wise)
Only the Fragments that affect the evolution of the time surface are loaded in the memory LOUISIANA STATE UNIVERSITY
18
Timing Analysis
Timing Analysis for Threshold = 0.01
• 150 timesteps, 12 GB Stirred tank data • 64 bit quadcore workstation with 64 GB RAM
• for increasing number of points block-wise memory access reduces per point LOUISIANA STATE UNIVERSITY
19
Deployment to End Users •VISH (http://sciviz.cct.lsu.edu/projects/vish/) provides the feature to decouple the user interface from the visualization application. • significant portion of interaction through “viz tangibles” • interaction control message triggered by physical events are sent to VISH • uses Cartouches (RFID tagged interaction cards)
User physically manipulating VISH Application through Viz Tangibles LOUISIANA STATE UNIVERSITY
20
Viz Tangibles • interaction with both data and operations by appropriate cartouches • Current implementation:
• Viewpoint Controls • Parameter Adjustment Controls
LOUISIANA STATE UNIVERSITY
21
Conclusion • an advancement towards exploring the time dependent feature of the fluid flow by generating Time Surfaces of the flow. • using software framework VISH and Fiber Bundle data model • Slice-wise sequential access and Block-wise random access of input data • adaptive refinement of evolving Time Surfaces • use of Viz Tangibles(Cartouches) as interfacing tool
LOUISIANA STATE UNIVERSITY
22
THANK YOU!!!
LOUISIANA STATE UNIVERSITY
23
WSCG 2010, Plzen, Czech Republic
LOUISIANA STATE UNIVERSITY
1
Outline • Overview of Integral Surfaces • Dataset & Data Model • Generation of Time Surfaces • Deployment to End User • Conclusion
LOUISIANA STATE UNIVERSITY
2
Motivation • extract the features of flow field from a very large time dependent CFD data set. • visualize the spatially evolving features of fluid flow in space-time domain • analysis tool for illustrating the movement (dispersions) of particles in a closed stirred tank system
LOUISIANA STATE UNIVERSITY
3
Motivation • stirred tanks – commonly used in industries • improvements in design can translate into billions of dollars, but better design comes with better understanding of mixing in tank • develop tool to analyze the mixing condition in such system
LOUISIANA STATE UNIVERSITY
4
Related Work • Stream surfaces, most commonly investigated technique for surfaces visualization • first algorithm given by Hultquist (1992) • Krishnan et. al (2009 IEEE), paper discusses the generation of Time and Streak surfaces in Large Time-varying data. • planer topology, as compared to our spherical topology
LOUISIANA STATE UNIVERSITY
5
Time Surfaces over Pathlines • Pathlines : most common approach to visualize the flow behavior in fluid system • Useful for few particles and for few time-steps.
(a) 23 particle system
(b) 516 particle system
• For large number of particles and longer time steps within enclosed flow system Pathlines => Spaghetti LOUISIANA STATE UNIVERSITY
6
Time Surfaces over Pathlines • analyzing flow field as evolving surfaces is more relevant in context of enclosed system
WHAT IS TIME SURFACE ? • evolving surfaces over time • higher-dimensional extensions of time lines ( evolution of seed line) • integration of surfaces over time, that can illustrate key flow characteristics
( such as dispersion of particle system in a flow)
LOUISIANA STATE UNIVERSITY
7
Stirred Tank Dataset • Provided by Dr. Sumanta Acharya and Somnath Roy, Department of Mechanical Engineering, Louisiana State University • 2088 curvilinear blocks • comprised of 3.1 million cells • velocity as vector field and pressure as scalar field
• each time slice has velocity for each grid point in all blocks • total of 5700 time steps; 500 GB binary data
LOUISIANA STATE UNIVERSITY
8
Stirred Tank Dataset CHALLENGES ?? • handling and processing of the - voluminous - multi-block - non-uniform curvilinear, dataset • to generate time surfaces and track set of particles in flow
LOUISIANA STATE UNIVERSITY
9
Data Model • based on concept of Fiber bundle data model • internal data structure with six levels, each composing of arrays representing properties of data set
Bundle
Time Slice
Grid
Topology
Representation
Field
• user only deals with • Bundle • Grid • Field
LOUISIANA STATE UNIVERSITY
10
Data Model For Stirred Tank data set and Time Surfaces
• Field
– coordinates, velocity, pressure, connectivity
• Grid
– collection of all Field entities (one Grid object per slice in our implementation)
• Bundle – entire dataset / sequences of Grid objects for all time slices
LOUISIANA STATE UNIVERSITY
11
Time Surface module • inputs : Data Bundle, Grid object with Seeding points and connectivity • output : Bundle as collection of Time Surfaces (Grids) F5 Stirred tank data. BUNDLE Seeding Input GRID
In : Bundle
Out : Grid
GRID Integration
Seeding Grid = Time Surface Grid
Out : Grid
In : Grid Vector Field Out : Field
In : Grid In : Field
Surface Computation Out : Grid
In : Grid Time Surfaces as BUNDLE Out : Bundle
LOUISIANA STATE UNIVERSITY
12
Particle Seeding & Advection • seeding input is the set of points and connectivity information among seed points • connectivity information is to create the triangle mesh for surface generation • the triangular mesh evolves over time, different to spanning surface out of line segment as in Hultquist’s approach
LOUISIANA STATE UNIVERSITY
13
Seeding Input • seeds points – set of particles lying on the surface of sphere
Seeding input as points
Seeding input with triangular mesh
LOUISIANA STATE UNIVERSITY
14
Triangular Mesh Refinement • over integration time the triangular mesh of surface enlarges resulting unsmooth evolving surfaces. • adaptive surface refinement approach is mandatory for high quality output Surface Refinement Criteria: • Edge Length • Triangle Area
a
a d
d
f
c b
e
c
b
LOUISIANA STATE UNIVERSITY
15
Time Surfaces
Evolution of two spheres at time slices 0, 50, 100, 125, & 150 from left-top to bottom. [Top View] LOUISIANA STATE UNIVERSITY
16
Out of Core Memory Management • partial access and handling of input data • Slice-wise sequential access and Block-wise random access • each time slice with corresponding Grid Object is processed only once ( Slice-wise)
Time surface computed from vector field in 2088 fragments(curvilinear grids) covering the Stirred tank. LOUISIANA STATE UNIVERSITY
17
Out of Core Memory Management • each Grid object consists of vector field information of 2088 blocks • accessing only those blocks that the particles touches at particular time rather than all 2088 blocks( Block-wise)
Only the Fragments that affect the evolution of the time surface are loaded in the memory LOUISIANA STATE UNIVERSITY
18
Timing Analysis
Timing Analysis for Threshold = 0.01
• 150 timesteps, 12 GB Stirred tank data • 64 bit quadcore workstation with 64 GB RAM
• for increasing number of points block-wise memory access reduces per point LOUISIANA STATE UNIVERSITY
19
Deployment to End Users •VISH (http://sciviz.cct.lsu.edu/projects/vish/) provides the feature to decouple the user interface from the visualization application. • significant portion of interaction through “viz tangibles” • interaction control message triggered by physical events are sent to VISH • uses Cartouches (RFID tagged interaction cards)
User physically manipulating VISH Application through Viz Tangibles LOUISIANA STATE UNIVERSITY
20
Viz Tangibles • interaction with both data and operations by appropriate cartouches • Current implementation:
• Viewpoint Controls • Parameter Adjustment Controls
LOUISIANA STATE UNIVERSITY
21
Conclusion • an advancement towards exploring the time dependent feature of the fluid flow by generating Time Surfaces of the flow. • using software framework VISH and Fiber Bundle data model • Slice-wise sequential access and Block-wise random access of input data • adaptive refinement of evolving Time Surfaces • use of Viz Tangibles(Cartouches) as interfacing tool
LOUISIANA STATE UNIVERSITY
22
THANK YOU!!!
LOUISIANA STATE UNIVERSITY
23
