Application of adaptively refined unstructured grids in DSMC to shock wave simulations

•Performance improvement strategies for an adaptively refined DSMC solver.•Implementation of thermal non-equilibrium models while optimizing memory usage in AMR.•Ideal strong scaling speed-up for up to 4096 processors.•Weak scaling efficiency of 87% for 8192 processors.•Application to 3-D simulation of shock-wave boundary layer interactions over a double wedge using 20,000 processors. An efficient, new DSMC framework based on AMR/octree unstructured grids is demonstrated for the modeling of near-continuum, strong shocks in hypersonic flows. The code is able to capture the different length scales in such flows through the use of a linearized representation of the unstructured grid using Morton-Z space filling curve for efficient access of collision cells. Strategies were developed to achieve a strong scaling of nearly ideal speed up to 4096 processors and 87% efficiency (weak scaling) for 8192 processors for a strong shock created by flow over a hemisphere. To achieve these very good scalings, algorithms were developed to weight the computational work of a processor by the use of profiled run time data, create maps to optimize processor point-to-point communications, and efficiently generate new DSMC particles every time step. Rigorous thermal non-equilibrium required for modeling high Mach number shocks was achieved through the accurate modeling of collision temperatures on a sampling grid designed to be compatible with the above approaches. The simulation of a nitrogen flow over a double wedge configuration for near-continuum conditions revealed complex hypersonic SWBLIs as well as three-dimensional gas-surface kinetic effects such as velocity and temperature slip. The simulations showed that three-dimensional effects are important in predicting the size of the separation bubble, which in turn, influences gas-surface measurements such as pressure and heat flux.