# LW33 — mesh-stage investigation and R18 flow test

## Main finding

Enabling internal-point smoothing in the three existing pre-smoothing iterations reduces concavity flags from 20,662 (R15) to 2,823 (R18), an 86.3% reduction. Positive plane violations above cosine 1e-6 drop from 19,414 to 1,977. This is a substantial mesh improvement, but R18 still has 14 poor-determinant cells and fails two detailed mesh checks. It is not an accepted performance mesh.

| Mesh | Cells | Poor determinants | Concavity flags | Wet area with ≥3 layers | Failed checks |
|---|---:|---:|---:|---:|---:|
| R15 | 330,244 | 13 | 20,662 | 34.88% | 2 |
| R16 | 307,421 | 0 | 20,170 | 0.00% | 1 |
| R17 | 330,454 | 15 | 20,348 | 36.80% | 2 |
| R18 | 330,151 | 14 | 2,823 | 34.63% | 2 |

Layer coverage is area-weighted, selecting hull faces by centroid below the historical initial Z=0 waterline. It does not establish adequate y+ or resistance accuracy. The R16 zero-layer control deliberately has no wall layers and is not a resistance mesh.

## Controlled tests

- R16: R15 inputs with only addLayers disabled. Determinant check passes before layer generation, while 20,170 concavity flags remain. This comparison separates a layer-stage determinant problem from concavity already present in the snapped mesh.
- R17: R15 inputs with nFaceSplitInterval=5. Feature faces are split during snapping. Concavity flags fall slightly and three-layer coverage improves, but poor determinants increase from 13 to 15. It is not adopted as the preferred candidate.
- R18: R15 inputs with nSmoothInternal=3. The installed OpenFOAM v2412 snapParameters.C defaults this to zero; snappySnapDriver.C applies it to internal refinement-interface points during the existing three pre-smoothing iterations. Other snapping and layer settings are retained.

All three meshes use the same source hull STL SHA-256: e7fada1f1334c7998e43d9067d4e9a4d511f39464fe5ca6659a95d8d2f5554ef. The source STEP is unchanged. The generated surface/volume discretizations may differ; identical source geometry is not a claim of identical fluid meshes.

## Remaining concavity

R18 has 846 near-planar-only flags and 1,977 positive plane violations above cosine 1e-6, including 441 above 0.01. The independent calculation reproduces all 2,823 native flags using OpenFOAM area-weighted face centers and area normals. It does not waive the native check. Of the flagged R18 cells, 1,621 touch a hull patch; 2,493 have mean face-center Z below the historical waterline. These position counts are not volume-based phase classifications.

For comparison, only 2,123 of R15’s 20,662 flagged cells touch the hull. The internal-smoothing control therefore addressed a large interior-mesh component of the problem. Residual shape defects near the hull and layer termination remain repair targets.

## Recorded R18 flow result

Saved endpoint: **0.009986014 s**. Maximum saved air/water velocities: **29.00 / 16.45 m/s**. Stop reason: **450 second wall limit**. Exit code: 0. All final velocities finite: True.

The R18 case is freshly initialized on its own mesh. OpenFOAM v2412, serial interFoam, fixed hull, two-phase flow, prescribed towing speed 30 kn. Requested endpoint 0.01 s; 450-second compute budget; guards above 100 m/s global velocity or below 5-microsecond time step. Actual endpoint may differ due to adaptive stepping and checkpoint latency. Air maxima use water fraction <0.01, water maxima >0.99; the cockpit velocity curve is the global maximum. No flow run is claimed for the mesh-only R16/R17 controls.

The previous R14 continuation remains the longer control: saved through 0.020014109 s, with maximum water velocity 16.12 m/s, normal completion. This R18 test does not replace that history or establish longer-duration stability. Even 0.02 s is only about 3% of one hull transit (~0.64 s). Forces remain startup evidence, not converged resistance.

## Boat reference and limits

User reference: 4,100 kg loaded; one Volvo Penta D6-340; reported Volvo Penta H6; H8 only an unverified candidate; CG 2,750 mm with datum/direction unconfirmed. Drive/ratio and vertical CG remain unspecified. These tests retain the original 3,500 kg reference, initial waterline and inferred LCG 2.9405231959611386 m forward of aft keel. The new loading and CG have NOT been applied to these cases.

The instability repair remains open. Next: correct residual cell shape and layer-transition defects, establish adequate wall resolution, and demonstrate longer stable flow plus mesh/time-step and force convergence. Apply a confirmed loaded equilibrium before a new boat baseline, then compare with actual sea trials. No validated resistance, running trim, engine/propeller match, attainable speed, fuel consumption or wave response is supplied.

## Reproduction

The archive contains the R16–R18 mesh inputs and generated meshes, R18 initialized fields and final two saved field directories, full logs and histories, audit results and scripts. The R18 mesh is stored once under meshes/r18-internal-smoothing; copy its constant/polyMesh directory into cases/LW33-r18-internal-smoothing/constant/polyMesh after extraction. Preserve evidence and rerun in separate copies. OpenFOAM v2412 and Python with NumPy/VTK are not bundled. Preparation scripts retain historical source dependencies; the archived initialized case can be used directly with OpenFOAM. Read textual checkMesh failures even when the utility exits zero.