In Rocstar, each physics module operates on some type of mesh. An outstanding issue in integrated rocket simulations is the degradation of mesh quality due to the changing geometry resulting from consumption of propellant by burning, which causes the solid region to shrink and the fluid region to expand, and compresses or inflates their respective meshes. This degradation can lead to excessively small time steps when an element becomes poorly shaped, or even outright failure when an element becomes inverted. Some simple mesh motion algorithms are built into our physics modules. For example, simple Laplacian smoothing is used for unstructured meshes, and a combination of linear transfinite interpolation (TFI)[8] with Laplacian smoothing is used for structured meshes in Rocflo. These simple schemes are insufficient when the meshes undergo major deformation or distortion. To address this issue, we take a three-tiered approach, in increasing order of aggressiveness: mesh smoothing, mesh repair, and global remeshing.
Mesh smoothing copes with gradual changes in the mesh. We provide a combination of in-house tools and integration of external packages. Our in-house effort focuses on parallel, feature-aware surface mesh optimization, and provides novel parallel algorithms for mixed meshes with both triangles and quadrilaterals. To smooth volume meshes, we utilize the serial MESQUITE package[9] from Sandia National Laboratories, which also works for mixed meshes, and we parallelized it by leveraging our across-pane communication abstractions.
If the mesh deforms more substantially, then mesh smoothing becomes inadequate and more aggressive mesh repair or even global remeshing may be required, although the latter is too expensive to perform very frequently. For these more drastic measures, we currently focus on only tetrahedral meshes, and leverage third-party tools off-line, including Yams and TetMesh from Simulog and MeshSim from Simmetrix, but we have work in progress to integrate MeshSim into our framework for on-line use. Remeshing requires that data be mapped from the old mesh onto the new mesh, for which we have developed parallel algorithms to transfer both node- and cell-centered data accurately, built on top of the parallel collision detection package developed by Lawlor and Kalé.[10] Figure 5 shows an example where the deformed star grain is remeshed with the temperature field of the fluids volume transferred from the old to the new mesh.
