Visualization and feature level comparisons in fluid flow
Vivek Kumar Verma, Alex Pang · 2001
Increase in computing power and advances in computational fluid dynamics (CFD) have enabled scientists to create increasingly accurate models and run simulations that generate huge datasets. Flow field data is computed using batch jobs on supercomputers but visualization demands interactivity. This demand on visualization has been somewhat of a burden in recent years and in an effort to use desktop tools for visualization, there has been a drive to generate visualizations using a simplified model of the original data and also a quest for fast visualization algorithms. Our research is concerned with developing fast visualization algorithms to study flow fields as well as the problem of studying differences in flow models that represent similar physical phenomenon. A simplified model often throws away information that could be crucial. Hence, there is a real need to develop tools that would help in understanding the differences in the original and approximated datasets. Other applications that can benefit from comparative analysis are comparison of experimental data with computational models, comparison of various computational models, comparison of different flow visualization methods, and the effect of changing the various parameters associated with them. The simplest of comparison approaches are image-level and data-level comparisons. Image level comparisons use visualization images and image processing to study the differences in the datasets. Data-level comparisons use the raw data for comparisons. Another approach to do comparison is to use feature-level comparisons. Feature level techniques compare the features extracted from the flow data. We have developed methods to compare some of the flow features commonly extracted from flow data like streamlines and streamribbons. These comparison methods can also be used to study differences in vortex cores. To answer the demand on the workstation to develop visualization methods, we have developed two new methods to visualize flow fields. The first is a fast texture synthesis method called PLIC that flexibly generates flow visualizations spanning the spectrum of streamline-like to LIC-like. Our PLIC method has several advantages over LIC in terms of computation time, image quality, and ease of handling time-varying data. The second is our proposal for a new approach to visualize streamlines by taking guidance from the features present in the flow itself. We call this method flow-guided streamline seeding. We expect that feature level comparisons will give useful insight into the differences in physical phenomenon represented by different flow datasets and hope that some day comparative visualization will be an integral part of all visualizations. We also hope that our research will provide useful tools to systematically compare the physical phenomenon under investigation as well as further the cause of developing fast visualization algorithms.