Friday, 26th January 2001, 16:30 hrs
Unstructured Mesh Solvers for Atmospheric Dispersion Problems:
Mesh Adaptation, Parallelism and Visualization.
Prof. Martin Berzins
Computational PDES Unit
School of Computer Studies
The University of Leeds
|Keywords :||3D finite elements, atmospheric pollution, unstructured meshes, adaptive model, parallel computing|
Presentation will feature joint work with A.Tomlin, L.J.K. Durbeck, S.Ghorai and P.Selwood.
It has been shown that the accuracy of solution for atmospheric pollution dispersion problems is highly dependent on the computational mesh and in particular the degree of resolution. Coarse meshes cannot resolve the underlying structure and uniformly fine meshes are prohibitively expensive for reactive flow problems with a large number of chemical species. A solution to this problem is to provide extra resolution of the mesh where large solution errors or steep concentration gradients exist, leaving a coarse resolution elsewhere. In this way computational resources are utilised where they provide significant gains in accuracy. This talk presents a 3-D finite volume reactive flow model based on a transient adaptive unstructured mesh. The use of tetrahedral mesh elements allows fully 3-D adaptivity and the flexibility to enable the code to handle complex structures arising from source terms of very different spatial scales. The underlying algorithm makes use of positivity preserving finite volume methods, fast iterative solvers, mesh adaptation and parallel computing.
The use of unstructured meshes raises the issue of whether or not the mesh is appropriate. This will be addressed by the use of advanced visualization techniques. Examples will be described for a number of different pollution dispersion problems covering a range of meteorological conditions. Results will demonstrate that the adaptive model is capable of achieving accuracy close to that of fixed high resolution meshes at a fraction of the computational cost.
|About the Speaker :||
Prof. Berzins is Co-director of the Computational PDEs Unit. PDE
is a research and consultancy unit within the School of
Computing, University of Leeds, providing PDE (Partial
Differential Equation) problem solving expertise and software to
industry and to academic research groups.
Research Interests of Prof. Berzins are Adaptive Numerical Methods and software, Unstructured Tetrahedral Meshes, Reacting Flows, Time Dependent PDEs, Parallel Algorithms, Computational Fluid and Solid Mechanics Applications.
Last updated 06th August 2001. Maintained by M.Molinari.