Coupled Reference Frames for Enhanced Visual Analysis of Scientific Data

摘要

Researchers often rely on large-scale scientific simulations to model complex systems. This is especially necessary in applications involving fluids and flows as they are often dominated by complex interactions. Traditionally, the data from such simulations can be represented in a number of different ways that each represents a unique frame of reference. The Eulerian reference frame stores measurements on a fixed spatial grid (scalar or vector fields) while the Lagrangian reference frame follows the motion of discrete parcels throughout the domain (particle data). These are often saved and analyzed separately depending on the hypotheses in question as each specification has its own set of advantages and disadvantages. While studying each representation on its own has been useful in the past, researchers are still limited to exploring a partial representation of the data, making it difficult to uncover subtle patterns. This dissertation research focuses on the development of new techniques that combine these representations for enhanced analysis and visualization schemes that can utilize the advantages of both formats simultaneously. By providing researchers with the tools necessary to explore their systems of study from multiple perspectives, new insights can be gained from today's datasets which are continually increasing in size and complexity.;This dissertation presents a number of different components which can be used in conjunction with one another to facilitate joint analysis and visualization tasks. First, a joint Eulerian-Lagrangian data management framework is presented which focuses on the fast and efficient organization and retrieval of large-scale data in both representations simultaneously. A number of fundamental joint operations are designed, such as a set of conditional queries and multi-resolution sampling schemes using both formats. Next, a trajectory-based feature tracking scheme is described, which uses corresponding particle data to improve the study of the evolution of volumetric features in both large-scale datasets and those with missing temporal or spatial regions. This dissertation then focuses on exploring spatio-temporal neighborhoods through both reference frames simultaneously. A user-guided extraction scheme as well as new methods of visually representing these joint features are presented. Lastly, this work concludes with a 4D segmentation scheme that can automatically extract numerous multifaceted features containing information from both reference frames. Collectively, these components enable a new means of data exploration previously unavailable to researchers and provide a stepping stone towards developing other sophisticated analysis techniques that can incorporate information from multiple data representations.