PLEASE NOTE: Deadline moved to Aug. 15, 8AM CET
Ensemble simulation can be used to examine aleatoric uncertainty in simulation models that contain stochastic effects. For this purpose, a simulation experiment is performed many times to generate an ensemble of realizations of model. The object of the 2016 Scientific Visualization Contests is the visualization of an ensemble of three-dimensional, transient fluid flows obtained through particle simulation with stochastic effects at multiple levels of resolution.
In this ensemble, the behavior of so-called viscous fingers is of primary interest. The six-dimensional nature and size of the data is the main challenge for visualization. Effective browsing, summarization, and data reduction strategies are needed to obtain meaningful insight into the data.
Please note: this website will be updated with more information in the near future. Make sure to check back!
The Finite Pointset Method (FPM) is a general approach for the numerical solution of problems in fluid dynamics and continuum mechanics. In this approach the medium is represented by a cloud of numerical points, each endowed with the relevant local properties of the medium such as velocity, pressure, and temperature. The points can move with the medium, as in the Lagrangian approach to fluid dynamics [1] or they may be fixed in space while the transport of the physical quantities has to be determined explicitly, as in the Eulerian approach. A mixed Lagrangian-Eulerian approach may also be used.
FPM is a meshfree method and therefore easily adapted to domains with complex and/or evolving geometries and moving phase boundaries (such as a liquid splashing into a container, or the blowing of a glass bottle) without the software complexity that would be required to handle those features with topological data structures. They can be useful in non-linear problems involving viscous fluids, heat and mass transfer, linear and non-linear elastic or plastic deformations, etc.
[1] J. Kuhnert (2014) Meshfree numerical schemes for time dependent problems in fluid and continuum mechanics. In: S. Sudarshan (Ed.), Advances in PDE modeling and computation, Ane Books, New Delhi, pp. 119-136.
To generate the ensemble the Lagrangian approach is used, i.e. the points move with fluid velocity. The simulation setup for each ensemble member is as follows: A cylinder is filled with pure water. At the top of the cylinder an unlimited supply of salt exists. Due to diffusion salt is washed out locally from the top. Since the salt solution has a higher density than pure water the highly concentrated fluid sinks down in the cylinder. The areas of high concentration are called viscous fingers. When and where they appear is not deterministic. The described simulation setup can be used to determine mean solution rates for the given type of salt in pure water.
Stochastic effects in the ensemble occur due to the simulation setup itself (which can also be observed in experiments) as well as the resolution of the FPM point cloud. The denser the point cloud the finer is the resolution of the viscous fingers as well as the more accurate is the measurement of the mean solution rate. Refinement of the point cloud is important to study the convergence of the simulation results, i.e. the mean solution rate. This helps to prove the validity of the simulation results themselves. An additional source of stochastic effects is the use of the k-epsilon turbulence model. Due to the non-deterministic behavior of the simulation multiple runs produce sub-ensembles with the same point cloud resolution. The full ensemble is generated through simulation with several resolution levels.
Evolution of concentration (color coded) using 250K particles.
The basic simulation setup consists of a cylindrical flow domain that contains water. At the top of the cylinder, modeled by a corresponding boundary condition, is a solid body of salt that is dissolved by the water. Each simulation of the provided ensemble is based on this setup. Due to the transient nature of the solution process, approximately 100 time steps are provided for each simulation to provide significant temporal resolution. However, as the simulation code uses adaptive time stepping, the sequence of steps in time differs between simulations.
The simulation is run for each of three levels of resolution, using 250,000, 650,000, and 1,900,000 points to discretize the underlying model. It should be pointed out that in FPM simulations, the points should not necessarily be interpreted as particles moving with the flow of water. However, in the provided data, this is mostly the case.
To obtain insight into statistical variance between viscous fingers properties, 50 runs are provided per resolution level. Thus, the there is a total of 150 simulations available.
The data can be downloaded from the San Diego Supercomputing Center cloud. Please see here for detailed instructions.
The final data size will be approximately 400GB. Simulations will be available for download in groups of five. Please note that data will be made available in several batches until the end of January, 2016.
The central problem of the contest is to visualize the evolution of viscous fingers across time and multiple resolutions, and study the variation of this evolution across the provided ensemble. Within these constraints, you are free to come up with visualization methods as you see fit.
The first task is to create a basic framework to visualize, organize, and browse the ensemble data to provide a first overview of the dataset.
Viscous fingers can be identified by considering the concentration of the salt solution – fingers are connected regions of the dataset that exhibit increased concentration. The central question of this task is, how fingers can be quickly identified and visualized at a specific point in time for a particular ensemble member.
Questions to be answered include but are not limited to:
To understand the temporal evolution of the concentration and viscous fingers, a visualization focused directly on visualization of this question. A straightforward approach to this may consist of analyzing and visualizing the changing properties of viscous fingers, such as volume, growth rate, position, movement speed, etc. These may be represented statistically, but could also be presented in correspondence to individual fingers – the point-based nature of the dataset appears to offer opportunities for tracking the viscous fingers across time. Based on the quantification of the evolution of the fingers’ properties, the possibility for finding and visualizing fingers with specific properties is desirable. For this purpose, linked views would appear a good candidate.
Questions to be answered include but are not limited to:
While Task 3 is concerned with analyzing and visualizing properties of individual ensemble members, the focus of this task is to summarize these properties across the entire ensemble. An important aspect is to understand the variation of properties – is the ensemble largely uniform with respect to the properties of viscous fingers, or are there significant deviations or even multiple classes for datasets of the same resolution? Of particular interest are outliers that exhibit significantly different properties in a statistical sense. Having multiple resolutions begs the question if there are properties of fingers that only occur dependent on the point cloud resolution. This can give insight in how far simulations with low resolutions are valid. To achieve comprehensible visualizations, summarization (statistical or otherwise) appears mandated. Again, it should be possible to focus visualization on ensemble members that are identified through the summarization, e.g. outliers.
Questions to be answered include but are not limited to:
It is desirable to provide the analysis and visualization capabilities from the first four tasks into a common visualization system. To achieve this, a framework and user interface to load, process, organize and visualize the ensemble data should be created. Specific questions arise from the nature and size of the data.
Specific questions of interest are:
In order to demonstrate your approach, you are expected to submit:
For the winning entry we expect the following additional requirements:
PLEASE NOTE: Deadline moved to Aug. 15, 8AM CET
As in recent years, the contest offers the opportunity to transfer cutting-edge visualization research to a specific, real-world application scenario. Beyond the mere achievement of having solved an inherently tricky problem, we put out a number of other incentives to join the contest.
We are thus delighted to announce that we are directly collaborating with the editorial staff of IEEE Computer Graphics and Applications (CG&A) in order to get the winning entry published as a peer-reviewed, full paper. The submission of the winning entry will be treated as an extended abstract for a CG&A submission. With winning the contest, it will successfully have completed the first of two review cycles. After the contest results have been announced, the winners will be asked to submit a revised and extended version of their submission. This paper will then undergo another formal review by CG&A.
The contest will be a visible activity at IEEE VIS 2016. Dependent on available disk space, we hope to publish all positively reviewed entries via the electronic conference proceedings. Additionally, contest winners will be recognized with a certificate and provided the opportunity to present their work during the IEEE VIS 2016 Contest Session. The winning entry will also have the opportunity to present a poster during the IEEE VIS 2016 regular poster session. To support these onsite activities, one complementary registration for IEEE VIS 2016 will be provided to the contest winner.
Further rewards for quality contributions to the contest are currently being explored and will be announced shortly.
A jury of domain experts and visualization researchers will carefully judge each submission. Since the main goal of the visualization contest is to promote the transfer of cutting-edge visualization research to concrete application domains, the rating will favor the domain experts' assessments by a weighting of 75:25. Hence, successful entries will first and foremost provide an insightful visualization that actually helps researchers gain insight from the presented data.
The jury consists of three domain experts and two visualization researchers. These are (in alphabetic order):
Christoph Garth (University of Kaiserslautern) is an assistant professor at the University of Kaiserslautern, Germany. His research focuses on vector field visualization, parallel visualization of very large datasets, material interface reconstruction, query-driven techniques and uncertainty visualization. Christoph received his PhD in 2007 from the University of Kaiserslautern working on the visualization of features in simulated fluid flows.
Berk Geveci (Kitware, Inc.) is the Senior Director of Scientific Computing at Kitware Inc. He is one of the lead developers of the Visualization Toolkit (VTK) and ParaView. His research interests include large scale parallel computing, computational dynamics, finite elements and visualization algorithms. Berk received his PhD in Mechanical Engineering and Mechanics from Lehigh University.
Bernd Hentschel (RWTH Aachen) is a senior researcher with the Virtual Reality Group, RWTH Aachen University, Germany. His research interests include the analysis of domain-specific features in large simulation data, parallel visualization algorithms, and immersive visualization. During his studies he closely collaborates with domain scientists in order to find ways to leverage the benefits of virtual reality-based user interfaces for complex visual data analysis problems. He holds a MSc. degree in computer science and a PhD, both from RWTH.
Jörg Kuhnert (Fraunhofer ITWM) is the founder of the Finite Pointset Method (FPM). He is the lead developer of this meshfree method. After studies in Mechanical Engineering (RWTH Aachen) and Numerical Mathematics (Oregon State University), he received his PhD in 1999 from the University of Kaiserslautern.
Isabel Michel (Fraunhofer ITWM) is part of the FPM developer team since 2012. Her main research interests are meshfree simulation methods, fluid dynamics, continuum mechanics, mathematical geosciences, and visualization of point data. Michel received her PhD in Mathematics in 2011 from the University of Kaiserslautern.
Theresa-Marie Rhyne is a recognized expert in the field of computer-generated visualization and a consultant who specializes in applying artistic color theories to visualization and digital media. In the 1990s, as a government contractor with Lockheed Martin Technical Services, she was the founding visualization leader of the US Environmental Protection Agency's Scientific Visualization Center. In the 2000s, she founded the Center for Visualization and Analytics and the Renaissance Computing Institute's Engagement Facility (renci@ncsu) at North Carolina State University. Rhyne is the editor of the Visualization Viewpoints Department for IEEE Computer Graphics & Applications Magazine. She received an MS in civil engineering from Stanford University and is a senior member of the IEEE Computer Society.
Simon Schröder (Fraunhofer ITWM) is also a member of the FPM developer team. Initially, he got his PhD in Computer Science in the area of Computer Graphics/Scientific Visualization in 2013 from the University of Kaiserslautern. Now, he applies this knowledge for handling the geometry in the simulation.
For questions regarding the contest, please do not hesitate to contact either of the chairs directly via scivis_contest(at)ieeevis.org.
For questions that are of general interest to all participants, particularly those concerning the input data and detection methods, we have set up a mailing list at scivis_contest_participants(at)ieeevis.org. Posting and access to the mailing lists' archive are limited to registered users. Please feel free to contact the chairs at scivis_contest(at)ieeevis.org for registration.
Contest Chair:
Contest Co-Chair: