Transfer subchannel solution onto master 3D mesh

[andrsd]

I am building a simulation where the master app will be a 3D mesh encompassing a fuel assembly - both the coolant and the fuel pins. Subchannel will run as a sub-app (in the next iteration there will be more assemblies, i.e. multiple subchannel instances).

I need to transfer temperature (coolant and pins) and density (coolant) into the master app. Master app has temperature and density as const-monomials.

Which transfers do you recommend to use? Is there an example for such setup? (I am only seeing setups that transfer from subchannel into sub-app).

Thanks!

[GiudGiud]

Hi David

What is the mesh like in the parent app? Does it overlap with both the coolant channels and the pins?
Does it extend beyond the "1D" meshes in subchannel?

I think the answer will be MultiAppGeneralFieldNearestLocationTransfer.
ANL has a model with a similar setup I ll ask what they are doing.

[andrsd]

Hi Guillaume!

Think of this like if you inverted the transfer from subchannel onto the visualization mesh. The 3D mesh would be a box encompassing the assembly. It would have cylinders that would represent the fuel pins. Each cylinder would be positioned where the 1D fuel pin mesh is in subchannel and the coolant mesh would be overlapping the subchannel mesh.

Does it extend beyond the "1D" meshes in subchannel?

If you are asking about the axial direction, then no decision has been made there, yet. Most likely not.

I think the answer will be MultiAppGeneralFieldNearestLocationTransfer.

That one crossed my mind. IIRC, when going from nodal into elemental, nearest node transfer could create funny problems where it would grab the wrong value. I think in THM, we used to average values and then we would transfer that, since that was producing more stable results. Maybe, something similar would work here, too...

ANL has a model with a similar setup I'll ask what they are doing.

Thanks, the would help quite a bit.

[GiudGiud]

That one crossed my mind. IIRC, when going from nodal into elemental, nearest node transfer could create funny problems where it would grab the wrong value

It is definitely a concern. If you need averaging then a layered average user object for each assembly paired with the MultiAppGeneralFieldUserObjectTransfer should do it.
To send from a full-core mesh to individual assemblies (so the opposite direction), LayeredAverage won't do what you want. We have something else that can do it though.

However if the centroid of the elements you are transferring to are actually aligned with the nodes in subchannel (that are centered inside each pin & channel), radially we should be fine. With specific choices in axial discretization, we could be fine axially as well.

[andrsd]

Aligning nodes with centroids is an interesting idea. I will need to check if that could work for this setup.

Thanks for the discussion so far. I guess, it is time to go try stuff and see what works and what does not :-)

[andrsd]

Do doing something like this in the main app:

[MultiApps]
  [scm_assembly]
    type = FullSolveMultiApp
    input_files = 'assembly.i'
  []
[]

[Transfers]
    [T_avg_from_scm]
        type = MultiAppGeneralFieldUserObjectTransfer
        variable = T
        source_user_object = T_avg_uo
        from_multi_app = scm_assembly
        greedy_search = true
        use_bounding_boxes = false
    []
[]

and this in the sub-app:

[UserObjects]
    [T_avg_uo]
        type = NearestPointLayeredAverage
        variable = T
        block = sub_channel
        direction = z
        points = '...'
        num_layers = <...>
    []
[]

Seems to be doing the right thing for me.

[kyriv1980]

Do you need the user object?

I have an example where i do :

    [Tpin] # send pin surface temperature to pin model,
        type = MultiAppGeneralFieldNearestLocationTransfer
        to_multi_app = sub
        variable = Pin_surface_temperature
        source_variable = Tpin
        execute_on = 'timestep_end'
        greedy_search = true
    []

And that seems to work. (SCM is the main app)

[andrsd]

The user object turns the nodal variable into an elemental one. Does the transfer do that? From the discussion above, I understood that it does not.

[GiudGiud]

It does though we don't recommend doing it with a transfer. You should get a warning.
The problem when converting from node (nodal variable = linear lagrange here) to centroids (constant monomial) or back is that the transfer finds multiple points (=values) in the source mesh that are equidistant to each target point in the target mesh

[andrsd]

Ok. I think, I forgot about the bigger difference. Instead of taking a "random" value from the element neighborhood, I compute the average value for a particular slice and place it in the user object. Then, that's the value I transfer.

[GiudGiud]

@gangyang-anl if you want to pitch in