AUV VISCOUS-FLOW SCREENING — 6 OCTOBER 2026

STATUS: computational screening, not validated vehicle performance.
Read assessment.json before interpreting any force, wake or pressure result.
No physical measurement data are included. Tow/handling/probe CSV files are blank templates.

UNITS AND CASES
Coordinates in metres. Reference water speed 0.3 m/s, density 998 kg/m3.
SU2 INITIAL_VALUES output velocities are nondimensional: multiply by 0.3.
Turbulent kinetic energy: multiply by 0.09; pressure: multiply by 89.82 Pa.
Pressure_Coefficient is already nondimensional. Do NOT use Bernoulli speed-to-Cp in viscous flow.
Drag in N = CD * (0.5 * 998 * 0.3^2 * 0.02040375).
The fine covered field is the source of the published wake figures.

REPRODUCE
Install SU2 v8.5.0, exact upstream commit listed in software/local-build-provenance.json.
The local OpenMP call-site patch is included. Prefer serial if using unpatched upstream.
Meshing/analysis dependencies: Python 3.12, gmsh 4.15.2, NumPy, SciPy, trimesh, VTK, matplotlib, Pillow.
Each vehicle case contains its configuration, history, audit, surface result and solver log.
Original vehicle volume meshes and the full 3D volume field remain in the complete local record.
The public bundle includes input surfaces, mesher/settings, sampled plane/trajectory fields,
all force records and the small benchmark meshes. Regenerating a vehicle mesh need not be
bitwise identical to the original; use the complete archive for exact mesh reproduction.
Run SU2_CFD -t 1 case.cfg in a copied case directory. Keep original evidence unchanged.
All eight compared cases start from uniform flow with RESTART_SOL=NO.
Experimental medium-grid continuations are excluded because their exact initial checkpoint files were not preserved.
Copy each case before running; SU2 overwrites outputs.
Mesh regeneration uses mesh_body.py with the supplied faired-light/body.ply or bare-light/body.ply;
use h=0.0106666666667, 0.008, 0.006. Domain-scale 1.5 is the larger-domain comparison.
Threaded Gmsh regeneration can change mesh ordering/topology; supplied meshes are authoritative.
The source GLB is linked in the portfolio; its SHA256 is in each geometry.json.
Geometry processing files are included for provenance; the supplied input surfaces let you remesh without the original project layout.
The historical geometry script expects the original project layout and geometry-helper/geometry.py; it need not be rerun to use the supplied surfaces.

AUDIT
audit_solution.py independently integrates Cp and Cf on the surface and checks force history/y+.
plot_flow.py samples actual velocity and k; export_streamlines.py traces the 3D mean field.
verify_streamlines.py rejects paths leaving the fluid. compare_streamline_steps.py checks integration step sensitivity.
No resolved instantaneous turbulent eddies or powered propeller result is claimed.
The fine covered full volume field is retained in the complete local archive, not this compact public bundle.
The supplied sampled fields and surface solutions support inspection of the published numerical record.
Solver software is not redistributed here; the upstream LGPL license accompanies the small patch.

FROZEN HORIZONS
Each vehicle case also has case-frozen-horizon.cfg with ITER set to the actual saved iteration.
Use it to repeat the bounded screening horizon. Use case.cfg for the original maximum cap.
Thread ordering and platforms can alter convergence histories; matching one frozen value is not validation.
The full local record retains the exact meshes omitted from this public bundle.

GEOMETRY RESTORATION
The compact download uses voxel-geometry.npz instead of redundant body.ply files.
Run python restore_geometry.py to rebuild both surfaces into restored-inputs/.
Their original-byte hash check is in geometry-restoration-check.json.
Pass the restored body.ply files to mesh_body.py. The full local archive retains originals.
The compact bundle retains display-streamlines.npz; raw/half-step candidate arrays remain local.
