From Human Computer Interaction Laboratory
General
The amount of data produced by finite element calculation and the size of the models exceeds the visualization capabilities of commodity computers by an order of magnitude. Several techniques that preprocess and optimize large datasets for visualization are known. The aim of this project is development of visualization engine for large FEM models and simulation results.
Details
The main techniques that are adopted for visualization of large datasets are scenegraphs and out-of-core. OpenSceneGraph as one of the leading scenegraph libraries is currently used. By this choice the system benefits from great performance of OSG, built-in optimizations (several culling methods, LOD, Vertex Buffer Objects, etc.) and implementation of newest OpenGL extensions. The lowest level of visualization is let on OSG and the system focuses on optimal scene hierarchy creation and management.
When a system needs to process data larger than main memory, it has to be done sequentially and partial results are stored out-of-core. Operating system functions for memory mapping may be used for external data access. We use an STXXL library: an implementation of the C++ standard template library STL for external memory (out-of-core) computations, i.e., STXXL implements containers and algorithms that can process huge volumes of data that only fit on disks.
Architecture of the system is based on plug-ins. The main component (the core) includes interfaces for several plug-in types like model database module, model visualization module, etc. Therefore more out-of-core and in-core modules may be combined, what is useful for testing and benchmarks.
Current Status
Currently the in-core version of FEM model viewer is done. This includes in-core model database and in-core model visualization. Application is in commercial use and is not part of this project.
Several out-of-core systems were studied and analyzed with respect to needs of finite element models visualization and interaction. The most suitable found approaches are Quick VDR ( http://www.cs.unc.edu/~geom/QVDR/ ) and Adaptive TetraPuzzles ( http://www.crs4.it/vic/data/papers/sig2004-tetrapuzzles.pdf ).
We adopt the algorithms of scene partitioning, preprocessing and scenegraph hierarchy from Adaptive TetraPuzzles approach with respect to needs of finite elements.
\r\nOut-of-core model plugin is based on STXXL and is created and tested. The out-of-core model visualization plugin is now in development.