Massively Parallel Large-Eddy Simulations of Primary Atomization on Adaptive Unstructured Meshes. Interface capturing algorithm and multiscale coupling perspectives

International audienceThis work presents a robust and efficient procedure to simulate turbulent incompressible interfacial liquid-gas flows on massively-distributed dynamically-adapted unstructured meshes in complex geometries. The present strategy extends the Accurate Conservative Level Set (ACLS) / Ghost-Fluid interface-capturing framework of Desjardins et al. (2008) and Chiodi et al. (2017) to unstructured grids, and combines it with an isotropic Adaptive Mesh Refinement (AMR) technique for triangular and tetrahedral meshes. The computational cost of the ACLS method is reduced by using a narrow-band to compute level set variables only in a restricted region around the liquid-gas interface. In the ACLS method, the interface is defined as the isosurface of a hyperbolic tangent function, which is transported by the fluid, and then reshaped using a reinitialization equation. Several forms of this reinitialization exist: the original form proposed by Desjardins et al. involves numerical estimation of the hyperbolic tangent gradient, which is susceptible to induce spurious deformation of the interface, especially on unstructured meshes. Chiodi et al. proposed a new form, which much better preserves the interface shape. The implementation of this new equation on unstructured grids is not straightforward and thus requires special attention. In this work, a robust implementation of this new form for unstructured meshes is proposed and implemented in the YALES2 low-Mach flow solver. In order to compute interface normals and curvature, the signed-distance function is reconstructed in parallel at nodes in the narrow band using the second-order Geometric-Projection Marker Method (GPMM) of Janodet et al. (2019). Spatial convergence and physical meaning of the overall procedure are demonstrated through classical interface transport tests and canonical two-phase flow simulations, respectively. Then, to point out the large computational gain using dynamic mesh refinement, two Large-Eddy Simulations (LES) of atomizing liquid jets are presented, namely a water jet in quiescent air from a low-pressure compound nozzle and a high-pressure kerosene jet in air crossflow. Both simulations are validated against experiments, demonstrating the potential of the overall procedure to accurately and efficiently handle primary atomization with large-density ratios using unstructured grids. Eventually, the ongoing developments and perspectives of a multiscale coupling of the Eulerian interface-capturing ACLS/AMR technique with a Lagrangian Point-Particle (LPP) modeling of the small droplets are discussed