Support unstructured grid

[ashwinvis]

In #31, we attempt to add support for structured curvilinear meshes (incomplete) by adding indices as the dimensions, instead of x and y.

Here the question is how to open a unstructured mesh into a Xarray dataset. In this case, we cannot rely on indexing the mesh. We have an example file which fits this description.

https://github.com/eX-Mech/pymech-test-data/tree/main/nek/naca

Close to the airfoil we cannot rely on indexing to group together elements.

Possible approaches

We need to explore and see what is the best approach:

1. There is a uxarray package which could be repurposed to load unstructured data.

   • https://uxarray.readthedocs.io/en/latest/
   • Load field file using pymech.readnek
   • Load the EXODUS mesh file using xarray.open_dataset

2. Convert to vtk and use pyvista-xarray:

   • https://github.com/pyvista/pyvista-xarray
   • We have a function to convert to vtk, but it relies on mayavi and traitlets. Directly relying on pyvista might be better solution. https://pymech.readthedocs.io/en/latest/vtksuite.html#pymech.vtksuite.hexa2vtk

[ashwinvis]

Loading the EXODUS mesh file is easy. We can add supports for other meshes with help of meshio (https://github.com/nschloe/meshio) in the future if this works.

In [8]: mesh = xr.open_dataset("naca(10x4).e")

In [9]: mesh
Out[9]:
<xarray.Dataset> Size: 2MB
Dimensions:        (time_step: 0, num_el_blk: 1, num_side_sets: 5,
                    num_nodes: 72734, num_dim: 2, num_el_in_blk1: 23924,
                    num_nod_per_el1: 8, num_side_ss1: 80, num_df_ss1: 240,
                    num_side_ss2: 80, num_df_ss2: 240, num_side_ss3: 300,
                    num_df_ss3: 900, num_side_ss4: 300, num_df_ss4: 900,
                    num_side_ss5: 202, num_df_ss5: 606, num_qa_rec: 1, four: 4,
                    num_elem: 23924)
Dimensions without coordinates: time_step, num_el_blk, num_side_sets,
                                num_nodes, num_dim, num_el_in_blk1,
                                num_nod_per_el1, num_side_ss1, num_df_ss1,
                                num_side_ss2, num_df_ss2, num_side_ss3,
                                num_df_ss3, num_side_ss4, num_df_ss4,
                                num_side_ss5, num_df_ss5, num_qa_rec, four,
                                num_elem
Data variables: (12/30)
    time_whole     (time_step) float64 0B ...
    eb_status      (num_el_blk) int32 4B ...
    eb_prop1       (num_el_blk) int32 4B ...
    ss_status      (num_side_sets) int32 20B ...
    ss_prop1       (num_side_sets) int32 20B ...
    coordx         (num_nodes) float64 582kB ...
    ...             ...
    side_ss5       (num_side_ss5) int32 808B ...
    dist_fact_ss5  (num_df_ss5) float64 5kB ...
    qa_records     (num_qa_rec, four) |S33 132B ...
    elem_map       (num_elem) int32 96kB ...
    elem_num_map   (num_elem) int32 96kB ...
    node_num_map   (num_nodes) int32 291kB ...
Attributes:
    api_version:               8.03
    version:                   8.03
    floating_point_word_size:  8
    file_size:                 1
    maximum_name_length:       32
    int64_status:              0
    title:                     cubit(C:/Users/gaura/Desktop/bioinspired_aerod...

[ashwinvis]

@airwarriorg91 if you are interested and if you think it is not too complicated, I can assign this to you. You don't have to work through it all alone, so feel free to submit an incomplete pull-request and we can refine it together.

[gauravity30]

The UXarray approach seems promising. I make a PR and try it out. Fixing this might be a step for including support for Hybrid meshes like the naca one (https://github.com/eX-Mech/pymech-test-data/tree/main/nek/naca). The extension for other mesh formats can be extended later as NEK5000 supports exodus and GMSH mesh formats.