Application of a coupled CFD-DSMC approach for low-speed rarefied gas flows in a lid-driven cavity

Multi-scale methods that couple a macro-model, computational fluid dynamics (CFD), with a particle-based micro-model, direct simulation Monte Carlo (DSMC) method, are often required to model low-speed rarefied gas flows, like those in MEMS devices, due to the wide range of length and time scales involved. The main challenge in this coupling is the transfer of information from DSMC to the CFD. The DSMC method can produce significant statistical noise in estimating macroscopic properties at low speeds, which can lead to numerical instabilities in the coupling algorithm. This can be avoided by using more particles per cell or running over a larger sampling window. However, such strategies increase the computational cost. In this work, we propose a new CFD-DSMC coupling approach based on a Micro-Macro-Surrogate (MMS) method [Tatsios et al., arXiv:2311.14140], applied to steady-state lid-driven cavity flow. Using sparse Bayesian learning and radial basis functions, we observe our hybrid method is computationally efficient in removing statistical noise and finding corrections from the target DSMC, producing results in good agreement with benchmark data for a range of Mach and Knudsen numbers.