WINDSOR: add AutoCFD5 Case 1 study (Windsor squareback)#1
Open
florian-simvia wants to merge 16 commits into
Open
WINDSOR: add AutoCFD5 Case 1 study (Windsor squareback)#1florian-simvia wants to merge 16 commits into
florian-simvia wants to merge 16 commits into
Conversation
New application case based on AutoCFD5 Case 1: external aerodynamics around the Windsor squareback at -2.5 deg yaw, on the ~6.3 M-cell c1g1 mesh. Two configurations on the same mesh: - RANS: steady k-omega SST, local time stepping, 3000 iterations - DDES: SST-DDES hybrid, dt = 8e-5 s, 3000 iterations_add, restarted from a converged RANS field. The DDES hybrid model is activated in cs_user_parameters.cpp; the base k-omega SST model lives in setup.xml. Aerodynamic coefficients (Cx, Cy, Cmz) are computed by the user routine cs_user_extra_operations.cpp -- not by the default boundary force log -- following the AutoCFD5 specification: normalised by the local dynamic pressure at probe [-2, 0, 1.3] m, rotated into the yawed vehicle frame by psi = -2.5 deg, and summed on the Windsor_Body / Windsor_Base / Windsor_Pins zones. DDES additionally reports a running time-average starting at t_avg_start = 0.5 s. The mesh file and the DDES restart field are too large to commit and will be fetched separately.
The mesh (c1g1.cgns, ~1.2 GB) and the DDES restart field (~806 MB) are
not committed to the repository. fetch_data.sh downloads them from an
external host into the expected locations:
WINDSOR/MESH/c1g1.cgns
WINDSOR/DDES_G1/RESTART/{main,auxiliary}.csc
The download URL is read from the WINDSOR_DATA_URL environment variable
or from the DATA_URL value at the top of the script. The placeholder
<REPLACE_WITH_HOSTED_URL> must be replaced once the artefacts are
uploaded (Zenodo, Simvia bucket, etc.).
Also adds MESH/README.md documenting mesh characteristics, boundary
zones, and the link back to the AutoCFD5 specification for geometry
regeneration.
Detailed application-case documentation covering: - Physical setup (geometry, yaw angle, Reynolds, boundary conditions, reference quantities) - Force coefficients pipeline implemented in cs_user_extra_operations.cpp (probe-based normalisation, rotation into the vehicle frame, output to force_coefficients.csv) - Numerical setup (k-omega SST, DDES activation, time stepping, restart workflow) - Indicative computational costs and parallelism guidance Also updates the top-level README so the BUNDLE benchmark and the WINDSOR application case appear side by side as the two studies provided by this repository.
The DDES case no longer reads a converged RANS restart. It is now initialised from a uniform free-stream field, the same way the RANS case is. This makes the case fully self-contained once the mesh is downloaded -- no large restart artefact to host. Changes: - DDES_G1/DATA/setup.xml: replace <iterations_add> by <iterations>, remove <restart path="RESTART"/> entry under <start_restart>. - DDES_G1/DATA/run.cfg: drop the restart_from: RESTART directive. - fetch_data.sh: drop the RESTART download (mesh only, ~1.2 GB). - README.md: update intro, quick start, force-coefficients note about t_avg_start (transient now longer), numerical-setup table, and repository layout to remove all RESTART references. Also drops the Computational cost section as no production wall times will be carried over.
Now that DDES no longer restarts from a converged RANS field, the
default time-averaging configuration must work from a uniform initial
field. Two coupled changes in DDES_G1:
- cs_user_extra_operations.cpp: express t_avg_start in convective
times rather than as a dimensional constant.
t_avg_start = n_tau_transient * L_body / U_inf
with L_body = 1.044 m, U_inf = 40 m/s and n_tau_transient = 10,
i.e. the lower bound of typical bluff-body DDES practice (skip
10-20 tau_c before averaging, accumulate 20+ tau_c afterwards).
- DATA/setup.xml: bump <iterations> from 3000 to 30000, giving ~2.4 s
of physical time (~92 tau_c). Averaging starts at iteration ~3263
(t = 10 tau_c) and accumulates ~82 tau_c of statistics over the run.
README updated accordingly: scaling rule, defaults, and the practical
guidance for further tuning.
The two cs_user_extra_operations.cpp files now share identical comment style for all common code paths: - same header docstring structure (case nature, frame-rotation rationale, probe-based normalisation note) - same "Reference quantities" block citing AutoCFD5 spec values - same inline unit annotations on s_ref / l_ref / rho - same detailed rotation-matrix comment - same labelled section dividers (probe interpolation, zone list, rotation to vehicle frame, CSV output) DDES keeps its specific additions (t_avg_start config block, t_cur/dt extraction, unique-owner election for the probe, running time-average, DDES log header) -- those are real functional differences. Also fixes a stale "URANS case" label inherited from an earlier setup: the case is DDES, both in the header docstring and in the bft_printf banner.
There is currently only one mesh (g1) shipped with the case, so tagging the case directories with the mesh level adds noise without disambiguating anything. Rename: WINDSOR/RANS_G1 -> WINDSOR/RANS WINDSOR/DDES_G1 -> WINDSOR/DDES and update all in-tree references (top-level README, fetch_data.sh, WINDSOR/MESH/README.md, WINDSOR/README.md including the numerical-setup table, code snippets, quick-start commands and repository layout).
The mesh is now served from the AutoCFD v3 S3 bucket: https://autocfdv3.s3.eu-west-1.amazonaws.com/test-cases/case1 Refactor the fetch() helper to accept separate remote and local paths (the bucket lays meshes out as meshes/c1g1.cgns whereas code_saturne expects MESH/c1g1.cgns at the study root). Drop the placeholder URL guard now that a real default is wired in -- the WINDSOR_DATA_URL env override is kept for users who want to test alternate mirrors. Verified the URL responds 200 OK (ETag 6c50b608..., 1219 MB).
Prevent accidental commits of: - Large mesh files fetched by WINDSOR/fetch_data.sh (*.cgns ~1.2 GB) - Preprocessor outputs and run output directories (RESU/, RESTART/) - code_saturne compile artefacts (*.o, *.so, cs_solver) - Editor and OS scratch files
Two polish items on the user-defined sources: 1. Add the standard EDF code_saturne GPL header (1998-2025 EDF S.A., GPL v2+) on all three WINDSOR SRC files, matching the convention used everywhere else in code_saturne (and in BUNDLE/GENERIC_SRC/). The previous "/* code_saturne version 9.1 */" stub is dropped. 2. Port the unique-owner election fix from DDES to RANS in cs_user_extra_operations.cpp. The reference probe at [-2, 0, 1.3] lies on the y = 0 symmetry plane, so two MPI ranks routinely end up equidistant to the same nearest cell. The previous logic declared both as owners, after which cs_parall_sum double-counted v_probe and p_probe -- inflating q_ref by ~4 and crushing every reported coefficient by the same factor. The fix elects a single owner (smallest rank id) via cs_parall_min on a candidate-rank, so the broadcast is always a single source.
The previous defaults (28 RANS / 96 DDES) reflected the production machine used during the AutoCFD5 study. For a publicly downloadable case the defaults should run out of the box on a typical multi-core workstation, so: RANS/DATA/run.cfg n_procs: 28 -> 8 DDES/DATA/run.cfg n_procs: 96 -> 16 Add a "Parallelism" section to WINDSOR/README.md explaining the defaults, recommended cluster targets (~64 / ~256 ranks given the 6.3 M-cell mesh) and the partitioner switch (Morton -> Scotch / ParMETIS) typically needed at large rank counts.
Add the sha256 of c1g1.cgns so fetch_data.sh fails loudly if the download is truncated, corrupted, or replaced upstream: 15c55862c720a1110fe65579e20f6c7cfc506fc2da78c596d46145bd7d5d2bca Computed locally from the canonical 1.2 GB file. The fetch() helper already supports an optional sha256 argument and runs sha256sum -c once the download finishes.
Short paragraphs covering what code_saturne is, what this repository is for, the per-case layout convention, and the requirements for running each study. The studies table is retained as the central navigation aid.
A 423 KB WebP render of the Windsor squareback body with a Q-criterion isosurface of the turbulent wake (from a DDES run) is inserted just after the intro paragraph as a visual anchor for the case. The image lives next to the README at WINDSOR/hero.webp; the Repository layout tree is updated accordingly.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Adds a new application case under
WINDSOR/reproducing the AutoCFD5 workshop Case 1: Windsor squareback body at 2.5° yaw, full scale. Two end-to-end setups on the same ~6.3 M-cell mesh (with wall law):RANS/- steady k-ω SST, local time stepping.DDES/- SST-DDES, started from a uniform free-stream (no RANS restart needed, case is self-contained once the mesh is downloaded).The mesh (~1.2 GB) is fetched by
fetch_data.shfrom the AutoCFD v3 S3 bucket with SHA-256 verification. Force coefficients are computed incs_user_extra_operations.cpp(probe-based normalisation, yaw rotation to vehicle frame, per-iteration CSV). DDES additionally reports a running mean over a window expressed in convective times.The top-level
README.mdis also updated to describe the repository.Test plan
./WINDSOR/fetch_data.shsucceeds, SHA-256 matches.code_saturne run --case RANSruns on the shipped defaults,force_coefficients.csvpopulated.code_saturne run --case DDESruns on the shipped defaults, running-mean columns activate around iteration ~3 263.