BACK 25/05/2022

CoEC gives presentation on scalable unstructured adaptive mesh refinement at the 6th International Conference on Turbulence and Interactions


CoEC partner and senior researcher from CNRS-CORIA Vincent Moureau presented a paper titled “Scalable unstructured Adaptive Mesh Refinement and Accurate Conservative Level Set method for primary atomization” at the 6th International Conference on Turbulence and Interactions on 18 May 2022 in Isola d’Elba, Italy. The work described was a collaboration between researchers at CNRS-CORIA and SAFRAN Tech.


More about the presentation:


Despite the steady increase in super-computing power over the past two decades, Direct Numerical Simulation (DNS) of atomizing liquid jets is still out of reach in realistic operating conditions encountered in industrial applications. Adaptive mesh refinement (AMR) is an appealing technique to reach DNS at a lower CPU cost by concentrating dynamically the cells at the liquid/gas interface location. AMR has been originally designed for Cartesian grids (Berger, Colella, JCP 1989) but the recent progress in isotropic and anisotropic unstructured mesh adaptation (Dapogny et al., JCP 2014) has increased the attractivity of such method, especially in complex industrial flows. Mesh adaptation libraries such as MMG ( are computationally efficient, robust and with a good control of the cell quality. The cell quality control is a mandatory feature for interfacial two-phase flows, which require low-dissipation and low-dispersion schemes. However, AMR on distributed memory machines is difficult to parallelize and the use of unstructured grids strongly limits the accuracy of the numerical methods classically used for interfacial two-phase flows.


In the CoEC project, a hybrid parallel edge-cutting / moving interface method for scalable unstructured AMR illustrated in Fig. 1 is designed. This is an extension of the moving interface method proposed in (Bénard et al., IJNMF 2015) based on MMG. This efficient AMR technique is combined to the generalization of the Accurate Conservative Level Set (ACLS) method to unstructured grid (Janodet et al., JCP 2022). This strategy has been implemented in the YALES2 code ( and is currently applied to the modeling of primary atomization in various configurations. In these applications, the local adaptation enabled to reduce drastically the CPU cost compared to the fixed grid approach and to reach unprecedented mesh resolutions at the interface. More recently, this method has been applied by SAFRAN TECH to an industrial pressure-swirl atomizer presented in Fig. 2, and to an oil-scavenging system shown in Fig. 3. These large-scale and massively parallel applications show the impact of CoEC project to high-fidelity methods for interfacial two-phase flows.




Fig. 1: Principle of the parallel moving-interface/edge-cut adaptation method





Fig. 2: simulation of an industrial pressure-swirl atomizer with 390 million tetrahedra and 2176 cores of Topaze, TGCC, CEA. Courtesy J. Vernier, J. Leparoux, R. Mercier, SAFRAN TECH




Fig. 3: simulation of a rotating oil-scavenging system with up to 600 million tetrahedra and up to 6000 cores of Topaze, TGCC, CEA. Courtesy M. Cailler, SAFRAN TECH