Albert, M. (2018):
Visualizing Big Data in Virtual Reality - Interactive Analysis of Large Scale Turbulence Simulations
Direct numerical simulations of turbulent fluid simulations produce time dependent scalar fields on grids of at least 1024^3 points, along with huge amounts of trajectories of tracer par- ticles which are advected with the fluid flow. Being able to explore the produced volumetric data in a three dimensional visualization might be a huge benefit over conventional two dimensional projections. Therefore a way to interactively explore these peta-scale data sets in virtual reality seemed to be a very appealing option for scientists researching turbulent fluid dynamics. The goal of this work is to visualize a specific large scale turbulent fluid simulation in virtual reality. In order to do so, an already existing virtual reality application for the HTC Vive was extended. To ensure the applicability of the work flow described in this thesis and of the extended application itself, data of a real turbulent fluid simulation are used. Due to the large size of the extracted isosurface graphic files, finding a way to visualize them, without reducing them to a quality so low that they are no longer fit for scientific visualization is the main challenge of this thesis. To accommodate for their size a multi threaded loading approach is used, where a second asynchronous thread loads new graphic files when needed. The thread communication is realized via the shared context functionality of OpenGL. The second aim is to provide users with new controller functionalities, displayed above the controller, which enable them to explore the data efficiently and to adjust the amount of information shown without restarting the application. To identify which kind of functions are useful, a scientist tested an early version of the application and specified which functionalities were missing or needed to be changed. The resulting application, based on NOMAD VR, provides the means to visualize large scale turbulent fluid simulations. Trajectories and various vectors of tracer particles are automatically calculated and visualized by the application, different controller functionalities allow to rescale the visualization, translate isosurfaces and to select single particles. The implemented asynchronous multi threaded loading approach works theoretically as long as the size of the few isosurface graphic files, which are loaded by the second thread, do not exceed one quarter of the graphic card memory.