Mesh adaptation is crucial in numerical simulation, providing optimal resource allocation for accurately capturing physical phenomena. However, when applied to Computational Fluid Dynamics (CFD) problems with complex multi-scale properties, existing adaptation methods face huge challenges due to the high computational cost of solving auxiliary partial differential equations (PDEs) and the difficulty in aligning the flow features with mesh geometric features. In this work, an end-to-end data-driven mesh adaptation framework, Flow2Mesh, is proposed to address these challenges by adopting a hybrid modeling strategy to construct the mapping from pixelated flow-fields to graph-based meshes. It achieves a rapid and accurate one-step mesh adaptation via a perceptual feature network (PFN) and a mesh movement network (MMN). PFN extracts the global perceptual features from flow-fields to enhance flow feature representation and mesh resolution independence. In MMN, these features are utilized to deform the initial mesh to a topology-invariant adaptive mesh by a proposed physically driven mesh convolutional network. It considers the inherent mesh geometric information for efficient node feature aggregation and alignment of mesh density with a flow-field structure. To generate high-quality adaptive meshes, various mesh-related losses are designed to regularize the mesh movement and alleviate the mesh tangling. Experiments in CFD scenarios demonstrate the generalization of our model to different design parameters and mesh configurations. It takes three orders of magnitude less time to generate similar meshes than the PDE-based method. The results exhibit the potential of Flow2Mesh to be a flexible and reliable tool for rapid mesh adaptation in scientific and industrial fields.
1 January 2014 in Advances in Applied Mathematics and Mechanics
Immersed boundary method (IBM) can easily distinguish fluid and solid regions in the computational region, thereby the workload of complex grid generation can be reduced. To accurately characterize the solid geometry, a large number of cells are required near the solid surface. The h-adaptive algorithm is adopted to reduce the requirement for the number of cells. In addition, considering the inherent adaptability to the h-adaptive Cartesian grids of the discontinuous Galerkin method, a high-order discontinuous Galerkin solver with an IBM is developed. To validate the h-adaptive algorithm and the solver, three cases are tested, including the steady flow past the National Advisory Committee for Aeronautics 0012 airfoil, the steady flow past a cylinder, and the unsteady flow past a cylinder. Compared with the non-adaptive cases, the h-adaptive cases need smaller total number of cells, and the numerical accuracy is significantly improved with an increasing degree of mesh refinement.
The generation of hybrid anisotropic-isotropic meshes is a crucial step in computational simulations, as it allows for the accurate representation of both isotropic and anisotropic physical properties. However, existing methods used for mesh generation typically rely on user-defined mesh size distributions (MSDs), which can be labor-intensive and may not adapt well to complex geometries. Moreover, due to the highly nonlinear relationship, it is challenging to directly generate the corresponding hybrid anisotropic-isotropic mesh from the geometric shape. In this paper, we propose GAIMesh, a data-driven end-to-end geometry-aware anisotropic-isotropic hybrid mesh generation framework based on artificial intelligence, to conditionally map the geometric shapes to hybrid meshes. The whole framework consists of three parts: geometric feature network (GFN), mesh diffusion network (MDN), and mesh mapping network (MMN). The GFN extracts the geometric features from the rasterized representation of the geometry, which enhances the representational ability of the extracted geometric features and mesh resolution independence. The MDN utilizes diffusion models to progressively refine the MSD conditioned on geometric features, gradually optimizing it from an initial random distribution to the final optimal conditional distribution. The MMN maps the high-dimensional MSD to a distribution of mesh node positions in space, which effectively transforms it into mesh node layouts, completing the spatial mapping from geometry to hybrid meshes. The proposed GAIMesh framework is evaluated on various complex geometries, including general shapes, airfoils, and missiles, demonstrating its ability to directly generate geometrically aligned hybrid meshes from the input geometries. The results exhibit the potential of GAIMesh to be a flexible and reliable tool for automatic hybrid anisotropic−isotropic mesh generation in scientific and industrial fields.