In the present study, a DSMC technique is developed using a cell-based data structure introdued by Dietrich & Boyd. The method utilize a localized data structure based on computational cells to achive efficient computational performance especially on PC clusters. Dynamic load balancing is applied by decomposing the computational domain with a weight factor accounting for number of particles, number of cells, and boundary faces such that high parallel efficiency can be achived during computation. Grid adaptation is also applied to improve the resolution of solution and to reduce the grid dependency. It was demonstrated that accurate solution can be obtained starting from coarse initial grids.
The method was applied to two-dimensional supersonic leading-edge flow, axi-symmetric Rothe`s nozzle flow, and open hollow cylinder flare flow. The results are validated by comparing with other simulation result and experimental data.