Optimizing Particle-Based Flow Visualization

Within both the scientific and academic communities, the study and visualization of flow is an essential tool used to model fluid transport and mixing. Modern Computational Fluid Dynamics (CFD) simulations are used to successfully approximate real-world physics to reproduce dynamic flow behavior, however, from simulation to visualization, maintaining a high level of accuracy is fraught with obstacles.;In this thesis, we address two major challenges to visualizing fluid flow data accurately. The first challenge, encountered during the simulation stage, is the inability of supercomputers to allow simulations to be saved at their native resolution. The second challenge, confronted in the post-processing stage, involves combating the substantial and discouraging computational expense associated with highly accurate particle-based visualization techniques.;We begin by presenting a framework for detecting and accumulating flow characteristics during the particle advection process, aimed at optimizing the associated computational burden. Consequently, we provide evidence of a wealth of fluid flow knowledge that can be gained with little computational overhead. We then extend the detection of these flow transformation effects to the simulation stage, offering a novel compression algorithm designed to dynamically distribute the constrained budget set when writing simulation results to disk. The output focuses on flow deformation and rotation, thereby improving visualization accuracy in turbulent regions.;Next, we describe the versatility of interpolation as a substitute for costly and iterative numerical integration. We present a method for categorizing salient flow features using only a sub- set of advected particles to interpolate the remaining domain information. Overall, the number of advection steps is drastically cut down and our method is able to better visualize features of interest with fewer particles. Building on this work, we eliminate advection entirely and develop an inter- active exploration environment based solely on interpolating trajectory position information. After achieving positive results in the visualization stage, we further explore the Lagrangian representation - a dense sampling of the flow field with particle trajectories - as an alternative output to CFD simulations. Through a series of comprehensive tests, we verify that the Lagrangian representation is less sensitive than traditional output to the unavoidable temporal compression that occurs when writing to disk. Subsequently, we demonstrate that particle trajectories extracted during the post- processing stage are more accurate, with the added benefit of increased performance and less disk space usage necessary to store the simulation output itself.;The primary objective of our work is to advance the field of flow visualization, focusing on the challenges that arise when high accuracy is required. With CFD simulations growing ever more complex and many scientists within academia and industry turning to flow visualization, we provide solutions to help retain flow field accuracy while optimizing both the storage of simulation data on disk and the computational expense of particle-based visualization techniques.