Robert S. Laramee 1 1 Adventures in Scientific Visualization Robert S. Laramee, Visual and Interactive Computing Group, Department of Computer Science, Swansea University, UK

Robert S. Laramee 2 2 Overview  Introduction, Motivation  Adventures in Scientific Visualization Adventure 1: Mesh Driven Vector Field Clustering Adventure 2: A Streamline Similarity Metric Adventure 3: Visualization of Foam Simulation Data  Conclusions, Future Work, Acknowledgements

Robert S. Laramee 3 3 Introduction and Motivation  CFD simulations result in large, complex, multi-field data sets.  +Unstructured, adaptive resolution meshes pose challenges for visualization  Vector field clustering can provide details in selective areas of the domain that correspond to interesting areas  High resolution mesh regions are associated with important areas of the domain  50% of CFD process is spent on mesh generation [5]

Robert S. Laramee 4 4 Adventure 1: Mesh Driven Vector Field Clustering Main characteristics of this work are:  Novel, general, automatic, vector field clustering algorithm that incorporates properties of mesh and flow field  Visual emphasis is placed on more important regions of the domain  Large, adaptive resolution meshes are handled efficiently: no processing time is spent on occluded portions of the geometry  We introduce some novel glyph-based visualizations with use-controlled algorithm and visualization parameters

Robert S. Laramee 5 5 Method Overview

Robert S. Laramee 6 6 Hierarchical Clustering: A Mesh Driven Error Metric

Robert S. Laramee 7 7 Visualization with Glyphs

Robert S. Laramee 8 8 Results: Demo Video

Robert S. Laramee 9 Adventure 2: Similarity Measures for Enhancing Interactive Streamline Seeding Observations Many automatic streamline seeding algorithms are presented, few used in practice Commercial software, e.g., Tecplot, features interactive seeding rakes Little focus on seeding rakes in visualization Domain Experts may not be interested in entire spatio-temporal domain Our contribution: to enhance streamline seeding in conjunction with seeding rakes, + fast streamline similarity metric Requires interactive computation, evaluation with CFD domain experts

Robert S. Laramee 10 Method Overview

Robert S. Laramee 11 Streamline Similarity: Streamline Attributes Streamline Similarity metric starts with: curvature, torsion, and tortuosity  Curvature: quantifies deviation from a straight line  Torsion: measures how sharply curve twists out of plane of curvature  Tortuosity: length of curve / distance between start and end Each attribute magnitude is normalized, equal weight

Robert S. Laramee 12 Benefits and Advantages A fast streamline similarity test enables  Enhanced perception: reduced visual complexity, clutter, and occlusion  Interactive performance, application to unsteady flow  Automatic inter-seed spacing along seeding rake  Streamline filtering  Focus+Context visualization

Robert S. Laramee 13 Streamline Similarity: Streamline Signature Idea: Divide streamline into bins  Similarity metric computed for each bin  Streamline signature: sum of similarity metric over each bin  Introduce Chi 2 distance  A single scalar compares streamline signatures on per-bin basis

Robert S. Laramee 14 Streamline Similarity: Similarity Matrix Chi 2 test is performed for all streamline pairs  Forms a 2D matrix, M sim  Column and row for each streamline  Symmetric matrix entry stores Chi 2  Euclidean distance is an optional component of the distance metric (based on last point in each bin for performance)

Robert S. Laramee 15 Features and Results: Demo Video

Robert S. Laramee 16 Domain Expert Feedback Work carried out with two CFD experts from Marine Turbine Energy group, Swansea University  Emphasized need for interaction  Provides effective visual searching through enhanced perception  Engaging  Large regions of uninteresting flow (large spatial aspect ratios)

Robert S. Laramee 17 Adventure 3: Comparative Visualization and Analysis of Time-Dependent Foam Simulation Data  Introduction and Motivation for Foam Visualization  System Overview  Comparative Visualization – Two halves view – Reflection View – Linked Time and Event Synchronization – Kernal Density Estimation for Foam Topology – Velocity Glyphs and Streamlines  Results

Robert S. Laramee 18 Why Study Foam? Fire Safety Displaces oil from porous media Mineral flotation and separation Cleansing

Robert S. Laramee 19 Foam Properties Two-phase material: liquid and gas Complex behavior: Elastic solid at low stress Plastic solid as stress increases Liquid at high stress

Robert S. Laramee 20 Bubble Scale Research Challenges Triggers of various foam behaviors are difficult to infer. Foam attributes: position, size, pressure, velocity, topology Difficult to visualize general foam behavior: Time-dependent Large fluctuations in attribute values caused by dynamic topology of film network.

Robert S. Laramee 21 Standard Foam Visualizations Require modification of simulation code for computation of derived data. Lack ability to explore and analyze data through interaction. Slow, coarse level of detail Univariate Constriction simulation: average velocity over all time steps

Robert S. Laramee 22 System Overview (video)

Robert S. Laramee 23 Two Halves View Visualization of foam deformation, time-averaged over entire simulations.

Robert S. Laramee 24 Event Synchronization Visualizing average velocity, 30 time steps, synchronized orientation

Robert S. Laramee 25 Kernel Density Estimation of T1s Visualizing foam topology

Robert S. Laramee 26 Velocity Glyphs and Streamlines Visualizing time- averaged velocity (50 time-steps)

Robert S. Laramee 27 Visualizing Forces Simulation of Sedimenting Discs Pressure → blue-tan Network force → black Torque → red arrow The network force - contacting soap films pull normal to circumference with the force of surface tension. The pressure force - adjacent bubbles push against disc with pressure force.

Robert S. Laramee 28 Foam Visualization: Demo Video

Robert S. Laramee 29 Future Work and Acknowledgements Future Work: Extensions to time-dependent data and visualization of bubble paths in 3D and 4D Thank you for your attention Any Questions? Thanks to the following: Ken Brakke, Guoning Chen, Simon Cox, T Nick Croft, Tudor Davies, Edward Grundy, Chuck Hansen, Mark W Jones, Dan Lipsa, Rami Malki, Ian Masters, Tony McLoughlin, Zhenmin Peng, This research was funded, in part, by the EPSRC (EP/F002335/1) and The Welsh Research Institute for Visual Computing, (RIVIC, New Computer Technologies (NCT) Wales,

Robert S. Laramee 30 Question: What about performance? (Excellent question!)

Robert S. Laramee 31 Question: What if streamlines do not have same number of bins? (Good question) Smaller number of bins is used.

Robert S. Laramee 32 Question: What affect does bin size have on algorithm? (Good question) Affects streamline similarity metric.

Robert S. Laramee 33 Any Loss of Accuracy Due to Projection to Image Space?  Excellent Question!