Information for


Interactive visualization of multi-scale, multi-physics simulations

Volume visualization of the higher-order density field of fluid flow around a sphere obstacle (left) computed with our higher-order ray casting approach. The discontinuous Galerkin simulation that produced this data set (IAG) works with complex adaptive grid data structures and polynomial ansatz functions to be able to numerically discretize the physical domain appropriately and efficiently solve the problem (right).

This project is part of SimTech project network 5 (PN5). The research of this project continues previous work at VIS and VISUS in the field of interactive visualization of data generated by scientific simulations. In the following, one of the central topics of the project is presented: The interactive visualization of grid-based simulation data with complex unstructured domain discretizations. Such data is, for example, generated by computational fluid dynamics (CFD) simulations. An example is the simulation of fluid flow in the vicinity of a driving car. Previous work mainly concentrated on the interactive volume visualization of scalar field data, like the physical density field of the fluid, given on simple structured Cartesian grids or tetrahedral grids. Volume visualization tries to solve the problem of occlusion, which is inherent to spatial 3D data, with a semi-transparent representation and a classification of the interesting structures contained in the field data with an appropriate color mapping (see image). 

The ambitious goal of SimTech is the simulation of very complex physical phenomena. However, due to limited computational resources it is usually not possible to compute the physical processes at a very fine scale, like, for example, on the atomic scale. On a global spatial domain, this would even not be possible due to the immense amount of data involved. Therefore, research within SimTech focuses on multi-scale and multi-physics simulation methods that follow adaptive approaches. To reduce computational complexity, such simulations concentrate their effort on the critical regions within the simulation domain for which it is necessary to apply a fine resolution and to compute with high accuracy. For less important regions, in contrast, a coarser resolution or a simpler mathematical model of the underlying physics, like for example on the macroscopic instead of on the atomic scale, is often sufficient. 

Spatial grid-based discretizations often refine the simulation grid locally to achieve adaptivity. This can be seen in the illustration which shows the simulation grid together with the density field volume visualization of a fluid flow near a sphere. In the direct vicinity of the sphere, as well as behind the sphere, the effects of turbulence are relevant, resulting in the well-known Kármán vortex street. In those regions, the grid is refined strongly and polynomial ansatz functions of increased order are additionally used to further improve numerical accuracy. For the boundary regions, however, a coarse grid resolution with simple linear ansatz functions is adequate.

The interactive visualization of discontinuous Galerkin data of higher-order (Institute of Aerodynamics und Gas Dynamics, IAG) is challenge due to the complex adaptive nature of the data. Our ray casting approach shoots viewing rays starting from a virtual observer of the 3D volume visualization through the simulation domain. The field values found along those rays have to be processed by the algorithm. This is a non-trivial task for the adaptive unstructured grid data which can even feature elements with curved geometry. Moreover, the evaluation of the polynomial field solution along the rays is computationally extremely expensive. In contrast to the simulation side, our visualization algorithm additionally is subject to the requirement of real-time visualization: computing multiple images per second to allow an efficient exploration of the simulation results. To cope with these requirements, our algorithms are optimized for current many-core graphics hardware (GPUs) with their highly parallel architecture. In the case of our discontinuous Galerkin visualization technique, the computation is even distributed on multiple nodes of a GPU compute cluster.

The project also works on the development of novel feature extraction techniques for complex multi-field data sets, which help in the analysis of the correlations and interactions of multiple fields given on the same simulation domain, e.g., a density, a velocity, and a temperature field. Here, the developed algorithms also strongly benefit from parallel computation. Besides the development of novel visualization techniques for the multi-scale, multi-physics simulations carried out in PN5, the research of tightly integrated interfaces for simulation and visualization is another topic of the project. Such interfaces can alleviate computational steering and allow for efficient exchange of data between the simulation and the visualization side.