Interactive particle tracing for the exploration of flow fields in virtual environments

Schirski, Marc; Bischof, Christian (Thesis advisor)

Aachen / Publikationsserver der RWTH Aachen University (2008) [Dissertation / PhD Thesis]

Page(s): X, 139 S. : Ill., graph. Darst.

Abstract

Due to rising data sizes and growing complexity, the results of modern numerical simulations are increasingly difficult to understand. Thus, scientific visualization is vital for an in-depth comprehension of computational fluid dynamics data. Especially the use of Virtual Reality methodology for an interactive exploration is gaining more and more importance, as it significantly facilitates the analysis of complex, three- to four-dimensional flow phenomena. However, large data sizes and high computational complexity limit a straightforward application of flow visualization methodology for real-world datasets due to violations of vital real-time constraints. This thesis aims at providing means for an intuitive exploration of complex flow phenomena in immersive virtual environments – including large, unsteady datasets, which tend to frequently exceed the memory capabilities of modern workstations. Due to its high expressiveness, the main focus of this work lies on interactive particle tracing. By exploiting the respective capabilities of the individual components of today's heterogeneous visualization environments, a level of interactivity is achieved which lies beyond that of contemporary approaches. The theoretical foundation of this work consists of novel models for the interactive exploration process and contemporary visualization infrastructure, as well as an extension of the classical visualization pipeline. Based on an execution frequency analysis, computational tasks are distributed onto the components of a heterogeneous visualization environment comprising high performance computing (HPC) infrastructure and dedicated rendering workstations equipped with modern graphics processing units (GPUs). For interactive particle tracing, this results in HPC-based data reduction, followed by GPU-based computation and depiction of particle trajectories, thus performing a paradigm shift regarding the utilization of HPC infrastructure for interactive data analysis. The building blocks of such a system include an efficient method for demand-driven data reduction – in this case by resampling a given unstructured grid into a Cartesian grid. An efficient cell search method and multi-level parallelization allow for an excellent utilization of available HPC resources. In order to assess the approximation quality of the resampled flow field, error estimation and feedback are discussed as well. Full interactivity for the actual exploration is achieved by performing particle tracing on the visualization front-end using the GPU. While support for flow fields given on Cartesian grids is relatively straightforward, this thesis presents a method for storing flow data in the form of tetrahedral grids within graphics memory, thus maintaining an optimized domain discretization as used in the flow simulation phase. Handling time-dependent flow fields on Cartesian and static tetrahedral grids adds to the flexibility of the presented approaches. For an efficient depiction of particle information, this work introduces image-based rendering techniques for both instantaneous particles and particle traces. An emphasis is put on their applicability in immersive virtual environments. The presented methods have been integrated into the visualization framework ViSTA FlowLib. These implementations have been used for a thorough evaluation via performance measurements and an assessment of usability, applicability and expressiveness for an interactive exploration of various real-world datasets with different characteristics.

Identifier

  • URN: urn:nbn:de:hbz:82-opus-23901
  • REPORT NUMBER: RWTH-CONV-112644