# LW33 R05–R10: airflow and waterline instability

The calm-water performance baseline remains unvalidated. These are fixed-hull startup diagnostics on the unchanged LW33 CFD surface, with 3,500 kg reference loading and an imposed 30 kn. The D6-380 engine does not determine this imposed speed.

## What changed

R05b replaced the original six-layer prescription with three thicker, smoother relative-size layers. It improved standard coverage and reduced poorly conditioned cells from 3,813 to 94, but air speed increased from 33.0 to 77.1 m/s by the last saved checkpoint. It did not establish a stable baseline. The rejected first R05 candidate retained layers over only 0.206% of the initially wetted area and was not run.

R06 kept original layers on surface triangles whose centroid was below +0.35 m, removing layers higher on the hull. It retained 91.95% initial wetted coverage and 56.38% coverage with at least three layers. However it also retained the initial velocity hotspot at X6.0511, Y0.9643, Z0.1147 m. Air velocity escalated to 335.6 m/s before the case was stopped.

R07 restricts layers to triangles with every vertex below Z=-0.05 m relative to the initial waterline. The rest of the surface keeps identical wall boundary conditions but has no extruded wall layers. Because the assignment uses whole triangles, the unlayered region extends well below the waterline in places; 23.54% of initially wetted area lies in that region. It is a diagnostic compromise, not a final resistance mesh. Region assignment changes no underlying triangle coordinates or orientation; the source STEP is untouched.

## Recorded results

| Case | Last saved field time (s) | Maximum air velocity (m/s) | Maximum water velocity (m/s) |
|---|---:|---:|---:|
| R05b smoother layers | 0.002618182 | 77.06 | 16.72 |
| R06 upper-hull layers removed | 0.001303831 | 335.65 | 16.78 |
| R07 fully submerged layers | 0.005000000 | 31.06 | 16.63 |

R07 computed endpoint: 0.005 s; last reported timestep: 0.0002 s. Requested endpoint was 0.005 s, with a 600-second wall-time guard and an early-stop guard at timestep below 5 microseconds. A manual stop, if used, is documented in the accompanying flow log/notes. Maxima select alpha.water <0.01 for air and >0.99 for water. They are saved cell-centre maxima, not time-history maxima.

R07 mesh: 339,391 cells; 71.41% of initially wetted hull area has layers; 43.54% has at least three layers. Coverage is area weighted by face centroid below Z=0. Standard checkMesh and detailed allTopology/allGeometry reports are retained separately; a standard pass does not clear detailed quality warnings.

## What is still required

A stable short startup is only the first gate. The model still needs adequate coverage near the running waterline, a defensible wall treatment, longer physical-time runs with settled forces and heave/pitch, and mesh/timestep sensitivity checks. One hull-length transit at 30 kn is about 0.64 s. None of these short runs provides converged resistance, achievable speed, fuel use, or wave response. Confirmed CG/loading and measured sea-trial observations are needed for physical validation.

## Reproduction and runtime

The accompanying archive preserves R06 and R07 flow cases, meshes, initial/saved fields, force histories, meshing dictionaries, region STL files, logs, audits and scripts. Extract alongside the existing lw33-cfd-update project. Use OpenFOAM v2412. The R05 checkpoint is a separate archive. Scripts protect existing cases from overwrite.

A four-process MPI continuation was attempted for R05, but this environment denied required local sockets before any parallel timesteps ran. The runtime was restored to serial execution and continuation used the saved serial checkpoint. No approval change is required for the working serial workflow.

## R08–R10: precise boundary and restart controls

R08 subdivides original planar surface triangles at Z=-0.05 m to recover coverage lost by whole-triangle tagging. This increases triangle count from 250,106 to 251,316, with unchanged bounds, zero measured relative area change and relative signed-volume difference 3.16e-16. The source STEP remains unchanged.

The mesh has 341,626 cells. Initially wetted coverage with layers increases from R07's 71.41% to 78.90%; coverage with at least three layers is 46.36%. The explicitly unlayered region still represents 14.53% of initially wetted area. Detailed checks flag 332 poorly conditioned cells, 25,025 concave cells and 111 low-weight faces. These are not cleared by the standard checkMesh pass.

The original above-water velocity hotspot is suppressed, but a new water-velocity instability develops near X4.42, Y0.90–0.92, Z=-0.05 to -0.07 m, close to the layer termination. R08 was stopped before its requested 0.010-second endpoint.

R09 restarts from exactly the R08 0.003-second saved fields and sets surface tension to zero. R10 restarts from the same fields and instead changes only momentum advection from linearUpwind limitedGrad to upwind, retaining surface tension 0.072 N/m. Output cadence differs but does not alter the equations. R10's first timestep was checkpointed and the same configuration resumed. Both controls retain the velocity growth. This rules out either of those individual changes as a remedy; it does not uniquely prove a particular pressure/mesh mechanism.

| Case | Final saved field time (s) | Max air velocity (m/s) | Max water velocity (m/s) |
|---|---:|---:|---:|
| R08 | 0.004110547 | 31.45 | 615.70 |
| R09 | 0.003942602 | 31.53 | 130.46 |
| R10 | 0.003811509 | 31.58 | 148.73 |

Maximum-velocity histories are retained under postProcessing/speedEnvelope and in the cockpit. They cover only the recorded interval and include both phases. R08 logging began after startup; do not interpret the missing initial interval as zero velocity. R09/R10 stopped on a 100 m/s diagnostic guard; the next saved step can exceed that value.

## Engineering conclusion

The original early air-layer failure can be suppressed in short tests by keeping layers away from the air region. R07 reaches 0.005 s with air/water maxima 31.06/16.63 m/s, but its reduced layer coverage is a material accuracy limitation. Restoring coverage with a precise boundary introduces a severe water instability at that boundary. A reliable calm-water baseline has therefore NOT been established.

The next work must address layer termination geometry and its pressure/viscous discretisation, with local cell/face-quality checks and an adequate transition mesh. The small-determinant set does not contain the two observed R08 maximum-velocity cells (313925 and 55864), so that metric alone is not a demonstrated cause. Once bounded flow and adequate near-wall resolution coexist, longer convergence and sensitivity runs are required before any resistance, wave or engine predictions.

This archive includes R08–R10 cases and recorded checkpoints, with the identical mesh stored once in the R08 case. After extraction, run `python code/restore_r08_shared_mesh.py` from the lw33-cfd-update folder before running R09 or R10. It restores the shared mesh or verifies an existing one without overwriting mismatches. The archive includes the pre-failure 0.003-second state retained in the control cases. Earlier R05 and R06–R07 cases are in their accompanying archives. All runs are diagnostics, not sea-trial measurements.