Support direct import or conversion of multi-block structured CGNS turbomachinery meshes

[CHOSENX-GPU]

Summary

I would like to request improved support for multi-block structured CGNS meshes generated by turbomachinery mesh generators such as AutoGrid

The practical use case is a compressor or turbine mesh where each blade row is composed of multiple structured blocks in CGNS

For SU2 turbomachinery simulations, a natural workflow would be to treat each blade row as one computational zone, while the internal block interfaces inside the same blade row should become internal mesh connectivity, not external boundary markers or separate SU2 zones

Current limitation

Based on the current documentation, SU2 supports multi-zone computations, but a single multi-zone mesh file currently works only with the native SU2 mesh format using NZONE= and IZONE=

The CGNS input support appears to be limited mainly to unstructured single-zone CGNS files

This makes it difficult to directly use standard turbomachinery meshes generated by AutoGrid or similar tools, because these meshes are usually exported as structured multi-block CGNS files

Why this matters

Many compressor and turbine grids are generated as multi-block structured meshes

A single blade row may contain many blocks, such as passage blocks, O/H/C topology blocks, tip-clearance blocks, hub/shroud blocks, and periodic-interface blocks

For SU2, these blocks should usually be merged or interpreted as one blade-row computational zone

At present, users have to rely on external conversion workflows, for example Pointwise, Gmsh, custom CGNS scripts, or conversion to a native SU2 mesh

This adds a fragile preprocessing step and makes SU2 harder to use for realistic turbomachinery cases

Expected behavior

SU2 should ideally support one of the following workflows

1. Read a structured multi-block CGNS file and correctly assemble connected blocks within the same blade row

2. Convert a structured multi-block CGNS file into a native SU2 mesh while preserving physical boundary markers

3. Allow users to specify how CGNS zones or blocks should be grouped into SU2 computational zones

4. Treat internal block interfaces inside the same blade-row group as internal connectivity

5. Preserve physical boundaries such as inlet, outlet, blade, hub, shroud, periodic boundaries, and mixing-plane interfaces

Actual behavior

At present, a multi-block structured CGNS mesh generated by AutoGrid cannot be used directly in the expected blade-row based way

The user must first convert or reconstruct the mesh externally before SU2 can run the case

This is especially inconvenient for multi-row compressor and turbine calculations

Example workflow that would be useful

AutoGrid structured multi-block CGNS mesh

Blade row 1:
block_001
block_002
block_003
...

Blade row 2:
block_101
block_102
block_103
...

Desired SU2 interpretation:

SU2 zone 0 = all blocks belonging to blade row 1

SU2 zone 1 = all blocks belonging to blade row 2

Interfaces inside each blade row = internal connectivity

Interfaces between blade rows = SU2 zone interfaces or mixing-plane interfaces

Possible implementation direction

A useful option could be a CGNS zone grouping mechanism, for example through CGNS FamilyName, zone names, or a user-provided grouping map

For example:

CGNS_BLOCK_GROUPING= FAMILY
CGNS_GROUP_TO_SU2_ZONE= YES

or a mapping file such as:

row_0: block_001, block_002, block_003
row_1: block_101, block_102, block_103

The grouped blocks would then be converted internally into SU2-style unstructured connectivity for each computational zone

Context

This issue is motivated by compressor meshes generated by AutoGrid

The mesh is multi-block structured and organized by blade row

The target solver workflow is SU2 turbomachinery simulation with blade-row level zones and mixing-plane or zone interfaces

Question

Is support for structured multi-block CGNS turbomachinery meshes planned

If not, would the SU2 team consider accepting an enhancement for CGNS multi-block import and conversion into native SU2 multizone format

[bigfooted]

Hi @CHOSENX-GPU Sounds like a great idea to add more support for CGNS, especially for multizone/multiblock problems. I always struggle with my collection of python scripts to convert cgns meshes to su2 meshes.
I don't think there are currently people working on this, so if you have the time and expertise to work on some of the improvements feel free to start a pull request. I always recommend starting a pull request early (do not wait until it is completely finished), and split up the development in small pull requests.
Converting cgns directly and robustly into an su2 mesh sounds like a great addition!

[pcarruscag]

According to google, PyVista can do what you describe with 3 lines of code.
So please consider making this a python utility instead of a C++ meshreader in SU2 with a lot of code to merge points and connectivities.

[CHOSENX-GPU]

Hi @CHOSENX-GPU Sounds like a great idea to add more support for CGNS, especially for multizone/multiblock problems. I always struggle with my collection of python scripts to convert cgns meshes to su2 meshes. I don't think there are currently people working on this, so if you have the time and expertise to work on some of the improvements feel free to start a pull request. I always recommend starting a pull request early (do not wait until it is completely finished), and split up the development in small pull requests. Converting cgns directly and robustly into an su2 mesh sounds like a great addition!

Thanks for the clarification and encouragement

At the moment, I can already use Gmsh to merge and convert a CGNS multi-block structured turbomachinery mesh into a single-block mesh that can be used by SU2

I will first keep this as an external preprocessing workflow, and then investigate whether a small Python utility based on PyVista could provide a cleaner and more reproducible path before considering any SU2-side changes.

https://github.com/CHOSENX-GPU/cgns-to-su2-skills I developed a simple agent skill to convert mesh files from CGNS format to SU2 format.

[MicK7]

@CHOSENX-GPU The current SU2 CGNS IO does not support Families yet. It is a must have to handle zone grouping and easier setup of boundary conditions. I don't know if there is an equivalent in su2 to make the setup of case more compact in the configuration file. Thus this would be a nice add-on.
For the merging, I agree with @pcarruscag that it should be done in a separate workflow as it changes the topology of the data being read.
The only thing that could eventually be done in terms of IO is to convert structured zones to unstructured zones as it is an invertible transform (you can go back to original grid and write the solution in original space)

[CHOSENX-GPU]

According to google, PyVista can do what you describe with 3 lines of code. So please consider making this a python utility instead of a C++ meshreader in SU2 with a lot of code to merge points and connectivities.

Thank you for your advice! I have a success test now.