Multi-block high order DNS code development for jet in turbulent cross-flow simulation
Castagna, J. and Yao, Y. (2012) Multi-block high order DNS code development for jet in turbulent cross-flow simulation. Journal of Algorithms and Computational Technology, 6 (4). pp. 593-622. ISSN 1748-3018 Available from: http://eprints.uwe.ac.uk/17553
Full text not available from this repository
Publisher's URL: http://dx.doi.org/10.1260/1748-3018.6.4.593
A multi-block high-order finite-difference direct numerical simulation (DNS) code has been developed for studying turbulent flows over complex geometries.A global mapping technique is adopted and applied to a multi-block domain arrangement,in which each block is consisted of a structured mesh and the adjacent blocks are overlapped and matched with each other along the common interfaces. The physical bounds of interfaces are determined through a pre-processing, and a pre-compiler is developed to reduce the computational cost of expensive Jacobian calculations. The multi-block DNS code has been validated for benchmark test cases, including a free-stream preservation on three-dimensional wavy grids and a pulse signal traveling through an interface with different grid density. The parallel efficiency has been evaluated on various HPC platforms with very good scalability achieved up to 1000+ processors. Further code application has been carried out for a square jet flow interacting with a fully developed turbulent cross-flow. The conditions are the Reynolds number $Re_\infty$ = 1000, based on the free-stream quantities and the jet exit width, and a jet to a cross-flow velocity ratio of $R$ = 0.5, 1.5, respectively. A counter-rotating vortex pair has been captured downstream of the jet exit and it was also found that the streamwise and the spanwise mean velocity profiles are in reasonable good agreements with available experimental measurements at $Re_\infty$ = 4700. However, some discrepancies are observed in other flow statistics such as the normal mean velocity and the kinetic energy distributions, and they are probably due to the low Reynolds number effect of the simulation.