Consistent and symmetry preserving data-driven interface reconstruction for the level-set method

Recently, machine learning has been used to substitute parts of conventional computational fluid dynamics (CFD) solvers, e.g., the cell face reconstruction in finite-volume method or curvature computation Volume-of-Fluid (VOF) method. The latter showed improvements terms accuracy for coarsely resolved interfaces, however at expense convergence and symmetry. In this work, a hybrid data-driven approach is proposed, addressing aforementioned shortcomings. We focus on interface (IR) level-set method, i.e., volume fraction apertures. IR, classification neural network decides based local resolution whether use linear IR IR. proposed improves interfaces recovers high resolutions, yielding first order overall convergence. Symmetry preserved by mirroring rotating input grid subsequently averaging predictions. model implemented into CFD solver demonstrated two-phase flows. Furthermore, we provide details floating-point symmetric implementation efficiency.