Advances in hierarchical adaptive mesh refinement with CODA

Kurzfassung

In many aerospace applications the flow field exhibits a wide spectrum of spatial scales, and the simultaneous presence of far-field boundaries, boundary layers, and shocks complicates the numerical treatment. Resolving every minute detail typically leads to labor-intensive grid fine-tuning or over-resolution and excessive CPU cost. Adaptive mesh refinement (AMR) offers an alternative by allowing the degrees of freedom to be increased where necessary, while keeping less-important regions coarse. In current industrial practice, AMR is often implemented as a metric-based refinement of purely tetrahedral meshes. By contrast, hierarchical AMR, although being more efficient, more scalable, able to be used for mixed-element meshes, and featuring rapid coarsening, is rarely employed for industrial applications, as just recently in part in reference. In additional, to exploit more benefits of hierarchical AMR, hanging nodes should remain unresolved, but no publications describe a corresponding robust, in-memory toolchain that also supports geometry-adapted refinement. Even though the technique shows promise on academic test cases and on simple flows of industrial relevance, its application to real-world aerospace problems remains an open question. This work presents advances in the use of AMR towards industrially relevant aircraft configurations. The iterative mesh adaptation workflow is implemented within the FlowSimulator framework. During each refinement iteration a) refinement indicators are computed with the computational fluid dynamics (CFD) software by ONERA, DLR and Airbus (CODA). Subsequently, b) FSAdaptationNG [5] subdivides a given number of cells identified by the indicator. Furthermore, if the mesh boundary shall converge to the true geometry, the subdivision of single cells is extended to corresponding whole cell stacks, and the loop incorporates two additional components: c) FSOCCT computes the updated surface node locations, which are then used to d) deform the volume mesh with FSMeshDeformation. The workflow is demonstrated on a range of aerodynamic test cases, from simple 2-D airfoils to full aircraft with engine exhausts (see Figs. 1 - 4). The influence of different refinement indicators on the convergence of the adaptation process is examined, spanning simple heuristic measures to sophisticated residual-based and goal-oriented ones as described in. Their impact on the required number of degrees of freedom and on overall runtime is quantified. Results show a substantial reduction of the CPU time for a given accuracy, with residual-based and goal-oriented indicators outperforming heuristic ones. Regarding geometry-preservation for linear meshes, the experiments confirm that omitting CAD reprojection causes aerodynamic forces to converge to values inconsistent with the true geometry. Finally, for geometries featuring concave intersections of convex surfaces (looking from the flow domain towards the body), challenges remain when clean prismatic layers are absent, because pure cell-stack refinement becomes insufficient. Overall, this analysis demonstrates the potential of hierarchical AMR workflows to reduce both computational cost and human effort for accurate CFD predictions for industrial aerospace applications.