DOI 10.1007/s12273-008-8321-7
RESEARCH ARTICLE
1. Department of Architecture, Graduate School of Engineering, the University of Tokyo, Kato & Ooka Lab. CW403,
4-6-1 Komaba Meguro-ku, Tokyo 153-8505, Japan
2. Institute of Industrial Science, the University of Tokyo, Kato & Ooka Lab. CW403, 4-6-1 Komaba Meguro-ku,
Tokyo 153-8505, Japan
Abstract A number of research studies have employed a wide variety of test conditions to numerically
assess flow field around buildings. In such studies, the employed mesh system and its resolution played
a crucial role in determining the ultimate degree of solution accuracy. The objective of this study is to
better establish a quantitative assessment of the influence of cell geometry and mesh resolution on the
numerical predictions of large eddy simulation (LES) of the flow field around a single building. In order
to achieve this objective, a number of mesh styles including conventional hexahedral-based mesh,
tetrahedral-based mesh, and prism-based mesh have been considered. In addition, hybrid meshes of
tetrahedron and hexahedron cells are also considered. In each mesh style, four grid resolutions were
investigated: coarse, medium, fine, and very fine. Accuracy of the simulation has been assessed by
applying comparisons to the experimental data available in literature. Moreover, quantitative grid
convergence was calculated based on the grid convergence index, which accounts for the degree of grid
refinement. Future studies are needed to analytically evaluate the influence of cell geometry on the
solution accuracy and to more precisely evaluate other grid system options including various forms of
hybrid configuration meshes.
Keywords cell geometry, numerical accuracy, single building, grid convergence index
accuracy that is typically associated with conventional both the velocity magnitude and turbulent kinetic energy
hexahedral-based meshes, comparison and validation magnitude between the numerical values (uLES, kLES) and
studies are necessary. This is especially true for most CWE the experimental values (uEXP, kLES) at the same spatial
applications where tetrahedral-based and prism-based location (i) as follows:
meshes are frequently used in order to properly map
complex flow field arising from urban areas. However, ui ,LES ui ,EXP
u =
comparative investigations on the prediction of the ub
accuracy of CFD results using different types of cell (1)
geometries in the computational grid system are hardly ki ,LES ki ,EXP
k =
seen in the literature. Thus, in this paper, large eddy ub2 / 2
simulation (LES) of the flow field around a single building
in a boundary layer is conducted for many mesh systems where ub is the velocity magnitude at the inlet equivalent
under various grid resolutions. The differences among to the height of the building. The root-mean-square of the
these mesh systems and grid resolutions are investigated. relative errors for n locations in the computational domain
Section 2 of this paper presents procedures followed to is used to provide scalar measure of the error as follows:
estimate numerical errors in CFD solutions based on grid 1/ 2
i=n 2
i
convergence index (GCI). Section 3 gives computational
details and results of the LES of a single building in a rms = i =1 (2)
boundary layer and the effect of cell geometry. Finally, n
conclusions are reported in Section 4.
Grid convergence index (GCI ), which is based on the
2 Procedures for estimation of numerical errors
so-called Richardson extrapolation method (RE) (Richardson
1910) and has been suggested by Roache (1994), is
It is expected that higher numerical errors are likely to occur
calculated for each mesh style. The GCI method is an
in cases where non-orthogonal meshes are implemented.
acceptable and straight-forward method that has been
These errors should be investigated and cannot be ignored.
evaluated over several hundred CFD cases. To calculate
Selecting the most appropriate mesh style for a specific
GCI, successive refinements of each mesh style have been
task depends on the time available for mesh construction
considered. For each refinement, grid convergence is
and the required accuracy of the results. However, it is still
evaluated using a relative error measure of both velocity
next to impossible to estimate these errors without relying
and turbulent kinetic energy magnitudes between the
heavily on numerical experiments where the flow problem
coarse and fine solutions,
is solved using different mesh systems.
No standard method for evaluating the numerical ui ,coarse ui ,fine
uncertainty is currently accepted by the CFD community; u =
ui ,fine
however, several methods have been suggested by (3)
many authors where the intricacies, shortcomings, and ki ,coarse ki ,fine
generalization of numerical uncertainty have been widely k =
ki ,fine
investigated. For LES calculations, some methods were
suggested by Geurts and Frhlich (2002) for explicit LES, Relative error values are determined for n consistent
by Celik et al. (2005) and Klein (2005) for implicit LES. points located in the computational domain. The root-
The scope of this paper is devoted to the difference among mean-square of the relative error is used to provide an
different mesh styles without going further into the details initial scalar measure of grid convergence for the points
of the several methods used in determining the numerical considered. It is calculated as follows:
accuracy. The procedures followed in this study depend on
1/ 2
two approaches: first, the comparison to the experimental i=n 2
results available in the literature (the experiment performed i
rms = i =1 (4)
by Uehara et al. (2003) is chosen for comparison) and n
second, the comparison of grid convergence index for
each case.
The numerical errors based on the experimental results In order to incorporate the mesh size and refinement,
are calculated according to the relative error measure of GCI can be applied for the fine grid solution as
Build Simul (2008) 1: 251 260 253
Fig. 2 Configuration of different mesh styles at vertical symmetry plane for (a) hexahedral, (b) tetrahedral, (c) hybrid with near wall
tetrahedron cells, (d) hybrid with near wall hexahedron cells, and (e) prism grid system
conditions were used: The experimental profile of the that were an order of magnitude below all the other meshes
boundary layer is used at the inlet boundary, which is for all resolutions considered. The differences in error values
located 6b units upstream of the building (Uehara et al. were in order of magnitude 5% 11% compared to the
2003), while the inlet turbulence is neglected in order to tetrahedral-based mesh case. Second, the prism-based
simplify the solution and to keep the focus on the effect of mesh exhibits relatively lower values of errors compared
cell geometry. No-slip condition is set for all solid surface to the tetrahedral-based mesh. This may be attributed to
boundaries while free slip condition and periodic condition lower truncation error because the cell geometry is parallel
are adopted for upper and lateral boundaries, respectively. to the flow in at least one direction for the prism-based
At the downstream boundary, the convective boundary mesh. The error values of the hybrid grid systems were in
condition is used. The standard type of Smagorinsky model between those of the hexahedral and the tetrahedral cases
is used for LES analysis (which utilizes implicit filter). as a result of the composition of the mesh and the cells
The Smagorinsky constant, Cs, is set to be 0.1. used.
Fig. 3 Profiles of u for different locations at symmetry plane with comparison to experimental results (Uehara et al. 2003)
Build Simul (2008) 1: 251 260 257
Fig. 4 Locations of experimental measurements in the experiment performed by Uehara et al. (2003)
Fig. 6 Profiles of k for different locations at symmetry plane with comparison to experimental results (Uehara et al. 2003)
Build Simul (2008) 1: 251 260 259
5 Conclusions
means of large eddy simulation. Journal of Wind Engineering Shah KB, Ferziger JH (1997). A fluid mechanics view of wind
and Industrial Aerodynamics, 25(3): 291 305. engineering: Large eddy simulation of flow past a cubic obstacle.
Richardson LF (1910). The approximate arithmetical solution by Journal of Wind Engineering and Industrial Aerodynamics,
finite differences of physical problems involving differential 67&68: 211 224.
equations, with an application to the stresses in a masonry dam. Uehara K, Wakamatsu S, Ooka R (2003). Studies on critical Reynolds
Philosophical Transactions of the Royal Society of London, number indices for wind tunnel experiments on flow within
210(A): 307 357. urban areas. Boundary-Layer Methodology, 107: 353 370.
Roache PJ (1994). Perspective: A method for uniform reporting of Vinokur M (1980). On One-Dimensional Stretching Functions for
grid refinement studies. ASME Journal of Fluids Engineering, Finite-Difference Calculations, NASA-CR-3313.
116:405 413.