| The physical and engineering sciences are increasingly concerned with the study of complex, large-scale evolutionary phenomena. Such studies are often based on analyzing data generated from numerical simulations at very fine spatial and temporal resolutions. The size of these simulation data significantly challenges our ability to explore and comprehend the generated data. In this thesis, the problem of developing a feature-based approach to visualizing and mining large-scale simulation data is investigated. The premise of this thesis is that for features in simulation data, a feature-based framework requires three essential components: feature detection, feature verification, and feature characterization. Feature detection is a process that automatically locates and extracts features of interest from the simulation data. Feature verification is a process that distinguishes actual features from spurious artifacts in the detection results. Feature characterization is a process that computes and quantifies the relevant properties of extracted features as determined by domain experts. To demonstrate the essential nature of each component, we have developed and analyzed algorithms for each component, and applied them to swirling features, or vortices, in computational fluid dynamics simulation data. Although the individual algorithms we have developed for these components may not useful for other types of features, the overall feature-based framework can be applied to other types of features from other types of simulations. |
