One Equation RANS Model with Wall Function Support #1299

hgopalan · 2024-10-17T12:52:46Z

Summary

This draft PR implements an one-equation RANS model based on the work of Axel et. al (https://link.springer.com/article/10.1023/A:1011560202388) for canonical ABL simulation. Terrain support is added similar to the Kosovic LES model. Wall function support is added through forcing term in the K-Equation. Not recommended for use as a low-Reynolds number RANS model with y+<30.

It is possible to convert the model into a hybrid RANS-LES model by changing the length scale function.

To be added:
Documentation

Pull request type

Draft

Please check the type of change introduced:

Bugfix
[X ] Feature
Code style update (formatting, renaming)
Refactoring (no functional changes, no api changes)
Build related changes
Documentation content changes
Other (please describe):

Checklist

The following is included:

new unit-test(s)
new regression test(s)
documentation for new capability

This PR was tested by running:

the unit tests
- on GPU
- on CPU
the regression tests
- on GPU
- on CPU

Additional background

Developed to provide a single interface for running wall-function aware RANS and LES using AMR-Wind.

amr-wind/turbulence/RANS/OneEqRANS.H

sayerhs · 2024-10-17T16:44:11Z

@hgopalan A note on the naming of the model. One-equation RANS model is usually used to refer to a family of models (https://turbmodels.larc.nasa.gov/), so naming this as one-eq RANS and the source term as Krans can be confusing to users. I would recommend using a more precise name to indicate the exact model here. If Axel et al. haven't provided a name, it might be better to refer to this model by the author names as Ganesh did with Sullivan and Moeng variants of ABL models.

sayerhs · 2024-10-17T16:46:42Z

amr-wind/equation_systems/tke/source_terms/Krans.cpp

+                (ustar * ustar / (0.556 * 0.556) + rans_b - tke_arr(i, j, k)) /
+                dt;
+        }
+        if (cell_blank == 1) {


In GPU mode, prefer to just multiply by cell_blank (i.e., drag_forcing = - Cd * m * tke_arr(i, j, k, 0) * cell_blank)

You'll most likely need a cast to to keep the linters happy.

sayerhs · 2024-10-17T16:48:16Z

amr-wind/equation_systems/tke/source_terms/Krans.cpp

+        amrex::Real dragforcing = 0;
+        amrex::Real ux = vel(i, j, k, 0);
+        amrex::Real uy = vel(i, j, k, 1);
+        const int cell_drag = (has_terrain) ? drag_arr(i, j, k) : 0;


If the cell_drag value is always 1 and the switch is has_terrain, why do we need drag_arr?

drag_arr purpose is to apply the wall function as a forcing term at first cell above the terrain. So added it to mark those cells.

sayerhs · 2024-10-17T16:48:37Z

amr-wind/equation_systems/tke/source_terms/Krans.cpp

+        amrex::Real ux = vel(i, j, k, 0);
+        amrex::Real uy = vel(i, j, k, 1);
+        const int cell_drag = (has_terrain) ? drag_arr(i, j, k) : 0;
+        const int cell_blank = (has_terrain) ? blank_arr(i, j, k) : 0;


Can't this be just blank_arr(i, j, k)?

blank_arr is available only if the terrain drag physics is enabled. Not sure if amrex::Array4() is always zero?

sayerhs · 2024-10-17T16:51:32Z

amr-wind/equation_systems/tke/source_terms/Krans.cpp

+    const bool has_terrain =
+        this->m_sim.repo().int_field_exists("terrain_blank");
+    const auto* const m_terrain_blank =
+        has_terrain ? &this->m_sim.repo().get_int_field("terrain_blank")
+                    : nullptr;
+    const auto* const m_terrain_drag =
+        has_terrain ? &this->m_sim.repo().get_int_field("terrain_drag")
+                    : nullptr;
+    const auto& blank_arr = has_terrain
+                                ? (*m_terrain_blank)(lev).const_array(mfi)
+                                : amrex::Array4<int>();
+    const auto& drag_arr = has_terrain ? (*m_terrain_drag)(lev).const_array(mfi)
+                                       : amrex::Array4<int>();


Would it be possible to simplify this?

…lation.

…ghts. Added a reg test.

marchdf

My main comment would be to figure out a way to simplify the code so that it is easier to follow. Maybe even totally different parallelfors depending on the path (with shared device function calls) or always defining terrain (default to 0) to minimize if conditions.

marchdf · 2024-10-21T15:00:41Z

amr-wind/equation_systems/tke/source_terms/CMakeLists.txt

@@ -1,4 +1,5 @@
 target_sources(${amr_wind_lib_name} PRIVATE
  KsgsM84Src.cpp
  KwSSTSrc.cpp
+  Krans.cpp


Agreed with @sayerhs. Lets have a more specific name for this.

Let me finish the other changes and before final commit, change the name.

marchdf · 2024-10-21T15:03:09Z

amr-wind/equation_systems/tke/source_terms/Krans.cpp

+        }
+        dissip_arr(i, j, k) = std::pow(Cmu, 3) *
+                              std::pow(tke_arr(i, j, k), 1.5) /
+                              (tlscale_arr(i, j, k) + 1e-3);


the + 1e-3 feels arbitrary? is that to avoid a division by zero? If so then it feels very large. There was another PR where we suggested the right way to do this with numeric_limits.

marchdf · 2024-10-21T15:04:17Z

amr-wind/equation_systems/tke/source_terms/Krans.cpp

+            amrex::Real uz = vel(i, j, k, 2);
+            const amrex::Real m = std::sqrt(ux * ux + uy * uy + uz * uz);
+            const amrex::Real Cd =
+                std::min(10 / (dx[2] * m + 1e-5), 100 / dx[2]);


1e-5 -> see my comment about 1e-3. Also, let's see if we can abstract out/generalize the direction with an axis variable so we are not hardcoding for z-dir.

marchdf · 2024-10-21T15:04:39Z

amr-wind/turbulence/RANS/OneEqRANS.H

+    TurbulenceModel::CoeffsDictType model_coeffs() const override;
+
+private:
+    Field& m_vel;


I bet a bunch of these can be const refs

marchdf · 2024-10-21T15:04:50Z

amr-wind/equation_systems/tke/source_terms/Krans.H

+        const amrex::Array4<amrex::Real>& src_term) const override;
+
+private:
+    Field& m_turb_lscale;


I bet a bunch of these can be const refs

marchdf · 2024-10-21T15:05:56Z

amr-wind/turbulence/RANS/OneEqRANS.cpp

+                          gradT_arr(i, j, k, 1) * gravity[1] +
+                          gradT_arr(i, j, k, 2) * gravity[2]) *
+                        beta;
+                    amrex::Real epsilon = std::pow(Cmu, 3) *


marchdf · 2024-10-21T15:06:53Z

amr-wind/turbulence/RANS/OneEqRANS.cpp

+                        beta;
+                    amrex::Real epsilon = std::pow(Cmu, 3) *
+                                          std::pow(tke_arr(i, j, k), 1.5) /
+                                          tlscale_arr(i, j, k);


here you don't have that + 1e-3. Maybe instead of adjusting it everywhere, and forgetting sometimes, it is worth just having a clip step on tlscale at some point so you know it is never going to be 0?

marchdf · 2024-10-21T15:08:07Z

amr-wind/turbulence/RANS/OneEqRANS.cpp

+                    const amrex::Real lscale_b =
+                        Cb_stable *
+                        std::sqrt(
+                            tke_arr(i, j, k) / std::max(stratification, 1e-10));


another arbitrary small number.

First commit of the one equation RANS model.

c73bdce

hgopalan added the enhancement New feature or request label Oct 17, 2024

hgopalan assigned sayerhs, marchdf, mbkuhn and moprak-nrel Oct 17, 2024

Clang Formatting

9bef8ee

github-advanced-security bot found potential problems Oct 17, 2024

View reviewed changes

amr-wind/turbulence/RANS/OneEqRANS.H Fixed Show fixed Hide fixed

Security issue correction.

82092b6

hgopalan unassigned sayerhs Oct 17, 2024

hgopalan unassigned marchdf, mbkuhn and moprak-nrel Oct 17, 2024

hgopalan requested review from marchdf, mbkuhn and moprak-nrel October 17, 2024 16:51

sayerhs reviewed Oct 17, 2024

View reviewed changes

hgopalan added 5 commits October 17, 2024 15:57

Enabling non-netCDF wind profile for RANS simulation. Speeds up calcu…

763729e

…lation.

I keep forgetting clang.

0d55083

Add a profile for TKE consistent with Canonical ABL

7f44ec6

Fix a bug in TKE initialization. Fix an issue with not using wind hei…

ed8c7c9

…ghts. Added a reg test.

Unit Test and Reg Test CMake Add

6714ae9

hgopalan marked this pull request as ready for review October 18, 2024 21:37

Removing spurious comment

f221d8e

marchdf requested changes Oct 21, 2024

View reviewed changes

hgopalan added 2 commits October 21, 2024 12:04

Clean up 1e-XX with cleaner code

7d6a7b6

Clang Formatting

ae16bc2

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

One Equation RANS Model with Wall Function Support #1299

One Equation RANS Model with Wall Function Support #1299

hgopalan commented Oct 17, 2024 •

edited

Loading

sayerhs commented Oct 17, 2024 •

edited

Loading

sayerhs Oct 17, 2024

sayerhs Oct 17, 2024

hgopalan Oct 17, 2024

sayerhs Oct 17, 2024

hgopalan Oct 17, 2024

sayerhs Oct 17, 2024

marchdf left a comment

marchdf Oct 21, 2024

hgopalan Oct 21, 2024

marchdf Oct 21, 2024

marchdf Oct 21, 2024

marchdf Oct 21, 2024

marchdf Oct 21, 2024

marchdf Oct 21, 2024

marchdf Oct 21, 2024

marchdf Oct 21, 2024

One Equation RANS Model with Wall Function Support #1299

Are you sure you want to change the base?

One Equation RANS Model with Wall Function Support #1299

Conversation

hgopalan commented Oct 17, 2024 • edited Loading

Summary

Pull request type

Checklist

Additional background

sayerhs commented Oct 17, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

marchdf left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

hgopalan commented Oct 17, 2024 •

edited

Loading

sayerhs commented Oct 17, 2024 •

edited

Loading