In most flows, large gradients of flow properties are generally confined to some local portions of the whole spatial domain. In these regions, particularly fine grids are required to achieve a sufficiently accurate solution. In general, it is impractical to adopt an uniformly fine mesh over the entire spatial domain in an effort to meet accuracy constraints, for such a practice tends to give rise to serious resource problems, not only in terms of memory but also because CPU times tend to rise quadratically or even cubicly with mesh size. The route adopted here, applicable to structured-grid environment, is via a local grid refinement strategy. This involves the selective insertion of additional grid nodes or cells in sub-domains. The insertion process may be pre-defined, but the true potential of mesh refinement can only be realised in combination with flow-related sensors which dictate the position and refinement level on the basis of solution-error, local rates of change, or other flow properties. For complex geometries to be computed, the method has been incorporated into a multi-block numerical framework. Of particular interest are the type of sensors, their critical values, at which refinement is initiated, and the role of the interpolation practices at coarse-mesh/fine-mesh boundaries in maintaining uniform accuracy. A more extensive account of the present work may be found in ref.
|Publication status||Published - 1996 Dec 1|
|Event||Proceedings of the 1996 7th UMIST Colloquium on Computational Fluid Dynamics - Manchester, United Kingdom|
Duration: 1996 May 2 → 1996 May 3
|Other||Proceedings of the 1996 7th UMIST Colloquium on Computational Fluid Dynamics|
|Period||96-05-02 → 96-05-03|
All Science Journal Classification (ASJC) codes