Ideas for enhancement and features

[JanisGeise]

Here is a list of ideas for improvement and potential features, which may be helpful in future versions of $S^3$. Feel free to comment if you have other suggestions.

• Collection of things that are a bottleneck in the current implementation:

• Parallelization of the re-numbering method

• Resolve memory issue for large meshes. Currently, the complete tree has to be hold in memory which is unnecessary and limits the applicability. The cells which are not part of the grid anymore can be removed internally to free-up memory during the refinement process. -> Requirement: more / better unit tests in s_cube script

• cache the interpolation coefficients of the KNN when exporting the CFD data to save runtime

• Collection of ideas for useful features:

• Restart option to continue the refinement process instead of having to repeat the complete process -> this makes only sense once the memory issue is resolved. Otherwise we would have to save the complete tree to disk, which takes long and requires a lot of space.

• Saving of checkpoints during the refinement in case the wall time limit is reached (on HPC) -> then use checkpoint for restart / continue refinement

• Option in export method to choose weather point-interpolated fields should be created and saved as well. Currently this is the default option, but for large meshes this takes up quite some disk space. In these cases, a cell-centered solution would suffice.

• Ideas for features:

• Option to append a field into an existing file. E.g. if a field was forgotten to be interpolated and exported. Currently, we can save these field in a separate HDF5 file, but it would be nicer to just add it to existing HDF5 file if it stems from the same case.

• handling of (partially) open STL files (limitation of PyVista)

• ways to handle multiple geometry objects of domain type, currently only exactly one domain is supported. To extend this to multiple domains, we would have to check if a cell is located in any of the other domains as well. Otherwise all cells will get masked out, because currently each geometry object is checked independently of each other.

• conical geometry objects (3D)

• triangles (2D) / prisms (3D) as geometry objects

• tetrahedron as geometry objects

• pyramids as geometry objects

• refactor unit tests for masking geometries to make it more readable/extendable, since the syntax is always the same

• add more unit tests for contents in s_cube script

• Documentation

• Update and clean-up documentation

• Add some jupyter notebooks with tutorials / examples