Skip to content

Ultravis66/cfMesh-OpenFOAM13

BranchesTags

Repository files navigation

cfMesh for OpenFOAM 13

v0.10 beta: active development on OpenFOAM Foundation v13. This fork focuses on OpenFOAM 13 compatibility, deterministic parallel behavior, boundary-layer robustness, and surface-conformance fixes for complex STL-driven Cartesian/polyhedral meshing. Rotor 37 is currently used as the primary stress-test case. Serial test runs have been deterministic in current Rotor 37 testing; parallel repeatability is substantially improved, with minor residual variance still under investigation.

NOTE: NASA Rotor 37 is used as a stress-test case because it combines tight blade/hub/shroud junctions, periodic sector boundaries, acute trailing-edge features, refinement transitions, and boundary-layer growth challenges. The intent is not to make the mesher Rotor-37-specific; Rotor 37 is used to expose general robustness issues that should benefit other external, internal, and turbomachinery CFD geometries. CFD validation is still limited and should be treated as preliminary.

Known remaining issues

Residual minor variance Major OMP races fixed in v0.8–v0.9. Serial (OMP=1) runs fully deterministic. OMP=32 variance substantially reduced — primary races (updateType, otherVrts_, frontPoints) fixed. Minor residual variance remains under investigation.

optimiseBoundaryVolumeOptimizer call-site bug (fixed v0.9) Original code passed true/false as iteration count to a function expecting (nIterations, nonShrinking). Early boundary motion was unconstrained instead of non-shrinking; late boundary fallback was doing zero iterations. Fixed to (1, true) and (1, false).

Visible surface protrusions at patch boundaries — Partially resolved in v0.10. Root cause: volume optimizer moves boundary points off STL surface with no constraint. Surface-constrained optimizer (Layer 3) now reduces visible-scale drift (>100 micron points) by ~46%. Residual drift on periodic patches remains under active development.

Runtime increase with surface-constrained optimizer — Serial octree pre-pass in partTetMesh::updateOrigMesh adds ~2x wall-clock time vs unconstrained. Can be disabled via constrainOptimizerBoundaryMotion false. Full parallel implementation planned.

Current limitations

  • Rotor 37 is a stress test, not a general validation suite.
  • CFD validation is preliminary.
  • Boundary-layer generation is much improved but still under active development near acute blade/endwall/periodic singularities.
  • Surface-constrained optimization reduces boundary drift but currently increases runtime.
  • A small number of negative-volume cells, bad pyramid faces, and non-orthogonality errors may remain in current development runs.
  • The mesher is not yet production-certified; users should run checkMesh, inspect boundary surfaces visually, and validate solver behavior for their own cases.

Original cfMesh developed by Franjo Juretic / Creative Fields, licensed under GPL v3.

Tested on NASA Rotor 37 two-passage turbomachinery sector geometry with boundary layers.

Branch status

Branch OpenFOAM Version Status
master OpenFOAM 13 Active development
openfoam12-archive OpenFOAM 12 Archived, no further updates

Quick start

Tested with: OpenFOAM Foundation v13, Ubuntu 22.04 / WSL2

source /opt/openfoam13/etc/bashrc
git clone https://github.com/Ultravis66/cfMesh-OpenFOAM13.git cfMesh
cd cfMesh
./Allwclean
./Allwmake

Verify the build:

ls $FOAM_USER_APPBIN | grep -E "cartesianMesh|tetMesh|pMesh"

Run the validation case:

cd tutorials/cartesianMesh/unitCube
cartesianMesh
checkMesh

What works

  • cartesianMesh Cartesian hex-dominant mesher, primary supported executable, tested on NASA Rotor 37
    • Current Rotor 37 results are promising but still under active development; this should not yet be treated as a production-certified mesher.
  • tetMesh Builds successfully, experimental, not fully validated
  • pMesh Builds successfully, experimental, not fully validated
  • libmeshLibrary.so Core mesh library

API changes from original cfMesh (compile fixes)

Original Updated
Pstream::blocking/scheduled/nonBlocking Pstream::commsTypes::blocking/scheduled/nonBlocking
eigenVector(T, scalar) eigenVectors(T) with direct component access
circumCentre()/circumRadius() circumSphere().first()/.second()
isFile() std::filesystem::exists() requires filesystem include and c++17
rm/isDir/rmDir requires OSspecific.H include in affected files
edgeMesh library include paths updated to use meshTools
unallocLabelList labelList

Runtime fixes

Bug Fix
face::centre() garbage for cfMesh face ordering phantom inversions Replace with vertex average everywhere
face::normal() garbage wrong BL extrusion direction Replace with triangle fan normal everywhere
Inversion check on unmapped octree mesh infinite loop Map vertices onto surface BEFORE preMapVertices()
preMapVertices() internal untangleSurface hangs Use preMapVertices(0)
findLeafContainingVertex throws char const* Replace throw with return -1
newPositionSurfaceOptimizer garbage vertex positions Displacement bound + vecY near-zero guard
pointFieldPMG::setSize uninitialized over-allocation Zero-initialize over-allocated region
Octree points uninitialized Zero-initialize octreePointsPtr_
checkFacePatchesGeometry unbounded loop Iteration limit (nGeomIter < 5)
findNearestSurfacePoint 100-iteration hang Reduce to 10 iterations
optimizePoint FPE from zero division Guard zero divisions in surfaceOptimizer
untangleSurface too many global iterations Limit to 2 global iterations
Garbage denormalized points in output Sanitize before write
face::areaAndCentre hangs in OF13 Replace with vertex average/triangle fan normal
omp_get_num_procs() over-threads OMP sections Replace with omp_get_max_threads()
meshOctreeCube::leavesInBox infinite recursion Add size guard and self-check
meshSurfaceMapperCornersAndEdges 20-iteration hang Reduce to 5 iterations
surfaceOptimizer 100-iteration hang Reduce to 10 iterations
addLayerForAllPatches O(n²) BL quality rollback hangs Removed
mapEdgeNodes OMP nondeterminism intermittent hangs and bad face explosions at full thread count Deterministic parallelBndNodes build: compute in parallel, move and sync in serial
smoothLaplacianFC apply loop OMP race concurrent writes to shared mesh point storage Serialize apply loop; compute remains parallel
smoothSurfaceOptimizer apply loop OMP race Serialize apply loop; compute remains parallel
nodeDisplacementLaplacianParallel constructs meshSurfaceEngineModifier per point inside loop Hoist modifier outside loop; serialize apply
nodeDisplacementLaplacianFCParallel apply loop OMP race Serialize apply loop
edgeNodeDisplacementParallel apply loop OMP race + std::map::operator[] mutating map during parallel read Serialize apply loop; replace operator[] with find()
lp.coordinates() / lp.pointLabel() divide-by-zero in parallel Laplacian smoothers Guard pointLabel() <= 0; fallback to original position where needed
globalToLocal[key] unguarded on received parallel data possible silent map insertion / invalid lookup Add globalToLocal.found() guards with correct packet consumption in receive loops
calculateBoundaryOwners uses patch 0 start offset regardless of activePatch_ Use boundaries[activePatch_].patchStart() when activePatch_ >= 0
calculateBoundaryFacePatches iterates all patches when activePatch_ >= 0 out-of-bounds write risk Add activePatch_ branch; fill selected patch label only
moveBoundaryVertex dereferences faceNormalsPtr_ assuming non-null when pointNormalsPtr_ is set Guard with pointNormalsPtr_ && faceNormalsPtr_
Face normal/centre computation has no size guard on degenerate faces bf.size() < 3 gives zero normal; bf.empty() gives zero centre
bp[e.start()] / bp[point] used without -1 guard throughout edge addressing Add bps < 0 / bpI < 0 continue guards
checkAndFixCellGroupsAtBndVertices performs topology surgery under OMP dynamic scheduling nondeterministic face connectivity Serialize repair loop when modifying topology
checkEdgeFaceConnections / checkFaceGroupsAtBndVertices missing clearAddressingData() after removeCells() Add mesh_.clearAddressingData() after each removeCells()
calcGlobalBoundaryPointLabels receive loops use remote face/point indices without bounds checks Guard face index, point index, and bp[] result before use
findLeafContainingVertex throws raw char const* for points near octree boundary — process terminates on fine meshes Replace FatalErrorIn + throw with WarningIn + return -1; all callers already handle -1 gracefully
Near-coincident STL vertices below mesh resolution cause protrusions at sharp trailing-edge corners meshDict-controlled geometry preprocessing pass: octree-based vertex weld before octree generation (geometryPreprocessing { active true; weldTolerance 1e-4; })
BL+BL+neutral 3-way corners (blade/shroud/periodic, blade/hub/periodic) extruded with averaged bisector normal causing protrusions Acute corner classifier measures patch-normal dot product; acute corners (dot < threshold) get 0.02 layerScale + projection guard; mild corners get 0.15 + ring taper
partTetMesh::updateVerticesSMP updateType = CELLCENTRE/FACECENTRE race on shared tet centre nodes
partTriMesh::updateVerticesSMP / updateVertices updateType = FACECENTRE/SMOOTH race
boundaryLayersCreateVertices concurrent std::map find/insert/[] on otherVrts_ during OMP parallel BL vertex creation Serialize otherVrts_ access; add FatalError guard for missing map entries
extrudeLayer frontPoints[f[pI]] = FRONTVERTEX race on shared point indices
meshSurfaceEngineModifier updateFaces/updateBndPoint shared boolList marking races Serialize marking loops
optimiseBoundaryVolumeOptimizer call-site bug — true/false passed as nIterations not nonShrinking Fix to (1,true) and (1,false)
tetMeshOptimisation::optimiseUsingMeshUntangler missing helper[tetI]=true — false convergence declaration Add missing assignment
Untangle stagnation deadlock at small residual bad-face count Detect stuck iterations, switch to unconstrained fallback after 4 unchanged iterations with ≤8 bad faces
mapVerticesOntoSurface unconstrained global nearest-triangle projection causes visible protrusions at patch boundaries Patch-constrained projection: validate returned patch against pointPatches; re-project to correct patch if mismatch
Volume optimizer moves boundary points up to 1.3mm off STL surface with no constraint Surface-constrained partTetMesh::updateOrigMesh: limited-displacement single-patch boundary re-projection during optimizer write-back; current implementation uses a serial octree query phase with parallel write-back.
volumeOptimizerEvaluateGradients FatalError on degenerate zero-volume tets Replace fatal with counted skip + end-of-function report
finalUntangleRejected_ flag incorrectly set when untangle skipped due to pre-existing negVol Skip rollback check when untangle did not run; BL proceeds normally

Boundary Layer Generation Improvements

Patch-based BL control

Patches with nLayers 0 such as inlet, outlet, and periodic patches are excluded from BL generation. Patches with nLayers 0, such as inlet, outlet, and periodic patches, are excluded from BL generation. The original behavior relied on hardcoded patch-name assumptions in several paths; this fork reads patch-layer intent from meshDict boundaryLayers/patchBoundaryLayers. Three code paths were updated: findPatchesToBeTreatedTogether, addLayerForPatch, and addLayerForAllPatches./patchBoundaryLayers. Three code paths fixed: findPatchesToBeTreatedTogether, addLayerForPatch, addLayerForAllPatches`.

BL/BL junction topology infrastructure (C1)

Persistent metadata for sharp BL/BL junction points is now carried through the boundary-layer generation pipeline:

  • blblJunctionPoints_ labelHashSet on boundaryLayers capturing boundary points on sharp BL/BL feature curves.
  • Junction points use layerScale_ = 0.0 to suppress spurious extrusion at the singular feature line.
  • Sharp BL/BL junctions are detected using patch-normal angle comparison rather than hardcoded patch names.

This provides groundwork for a future transition-topology implementation at sharp BL/BL intersections. The remaining planned work is explicit wedge/pyramid/O-grid style topology emission for junction regions where ordinary prism extrusion becomes ill-conditioned.

layerScale ramp field

Smooth suppression zone at BL/no-BL patch boundaries:

Zone Scale
Transition edge 0.0
Ring 1 0.25
Ring 2 0.50
Ring 3 0.75
Ring 4 1.00

Previously ring1 was incorrectly 0.0, creating a flat zone and steep gradient that caused warts at patch junctions.

Surface quality system (v0.1–v0.5)

A topology-aware surface projection and repair pipeline, developed incrementally on top of the original cfMesh surface mapping code.

Architecture fix patch-identity-first projection

The original cfMesh pipeline ran mapMeshToSurface (global surface projection) before extractPatches (patch identity assignment). This meant every boundary point was projected using a global nearest-triangle search with no patch constraint causing points near concave multi-patch corners to snap to the wrong patch surface, producing visible protrusions.

Fix: extractPatches now runs before mapMeshToSurface, establishing patch identity on the raw Cartesian hex mesh before projection. A second extractPatches call after projection corrects misassignments from the unprojected hex geometry. This addressed one class of patch-boundary protrusions. Later diagnostics showed that a larger remaining source was boundary-point drift during volume optimization, addressed in v0.10 by the surface-constrained optimizer.

Boundary point topology classifier

Boundary points are classified into topology roles before constrained projection:

Class Description
SINGLE_PATCH Interior surface point standard projection
TWO_PATCH_EDGE Feature curve point constrained to edge
MULTI_PATCH_CORNER 3+ patch junction constrained to corner endpoint
BL_NOBL BL/no-BL transition locked
NONMANIFOLD Non-manifold geometry locked and reported

VTK diagnostic output available for each class for ParaView inspection.

Feature curve extraction and corner snap

Dual-mode feature curve extraction (face-patch data when available, pointPatches fallback for early pipeline stages). Extracts ~10,260 feature edge segments. 24/24 multi-patch corner points snap correctly to nearest incident feature curve endpoint.

Pure quality evaluator proposedMoveIsValid

Reusable per-point quality gate used by all movement logic. Evaluates a proposed new position purely geometrically no mesh mutation, no const_cast, thread-safe by construction.

Checks per incident face:

  1. Movement distance cap (relative to local cell size)
  2. Face area (reject near-zero collapse)
  3. Pyramid height (relative to face area scale-invariant)
  4. Skewness proxy (rejects moves that worsen local face skewness beyond threshold)

Constrained surface smoothing

Patch-aware Laplacian smoothing pass after corner/edge snap. Two-phase update (compute from frozen geometry, apply in fixed order):

  • SINGLE_PATCH points: Laplacian average of same-patch neighbors → re-project to own patch → proposedMoveIsValid gate
  • TWO_PATCH_EDGE and MULTI_PATCH_CORNER: locked
  • BL_NOBL: locked

Targeted repair of validity-rejected points

After global projection, points that failed the validity check are collected. Their 2-ring neighborhood is expanded, classified, and repaired using constrained patch-aware projection. Typically repairs ~1,250 of ~1,500 neighborhood points per run, targeting the blade/periodic junction regions where protrusions concentrate.

Bisection backtracking (experimental, gated off)

Binary search along the snap vector for points that fail full projection. Gated behind useBisectionBacktracking = false pending stronger quality bundle (skewness + non-ortho comparison against old position required before enabling).

Version

Tag Description
v0.1-beta-stable Pipeline reorder + relative validity thresholds
v0.2 Skewness proxy added to quality evaluator
v0.35 Validity-rejected point VTK diagnostics
v0.4 Targeted repair of validity-rejected points
v0.5 Pure quality evaluator no const_cast, thread-safe, tight variance
v0.6 Partial deterministic optimizer — 8 unsafe OMP loops serialized
v0.7 BL/neutral edge detection + BL-side projection constraint + VTK diagnostics
v0.7.1 BL/neutral extrusion constraint + crossing clamp + layer rollback scaffold
v0.8 Full codebase audit — 60+ bug fixes: OMP serialization, globalToLocal guards, activePatch_ bugs, face size guards, bounds guards throughout
v0.8.1 feature/bl-transition-topology
v0.8.2 feature/bl-transition-topology

BL/no-BL transition edge constraint system

Generic protection system for points where BL patches meet no-BL patches (wall meets inlet/outlet/periodic). Fully geometry-agnostic driven entirely by nLayersForPatch_ from meshDict, no patch names hardcoded.

Component Description
detectBLNoBlTransitionEdges() Lightweight pre-detection before edge extraction. Runs before any surface projection. No mesh modification.
blNoBlEdgePoints_ / blNoBlEdges_ Generic layer-termination edge and point sets
blNoBlPointPatch_ BL-side patch ownership per interface point
protectedPoints_ + protectedPointPatches_ Per-mapper protection: interface points redirected to findNearestSurfacePointInRegion on their own patch only, preventing projection across feature curves onto the wrong patch
Guards mapEdgeNodes, mapCorners, mapVerticesOntoSurface, mapVerticesOntoSurfacePatches
meshSurfaceEdgeExtractorNonTopo Overloaded constructor accepts and passes through protected point sets

meshDict controls

Parameter Default Description
blblFeatureAngleDeg 40 Junction angle threshold for BL/BL sharp curve detection (degrees)
blblCornerAcuteThreshold 0.3 Dot product threshold for acute corner classification. 0.5=60°, 0.3=73°, 0.0=90°
lockAcuteCornerPoints true Lock acute corners in final surface optimizer
acuteCornerRing0-6 0.0/0.0/0.05/0.15/0.35/0.60/1.00 layerScale taper seeded from acute corner points
acuteCornerCapLayers false Cap layer count near acute corners (experimental)
virtualTopologyExclusion false Enable local BL exclusion face-rings at acute junctions
virtualTopoRing0-2 0.05/0.15/0.40 layerScale for VT face-ring suppression zones
detectGaps false Enable gap/contact proximity detection
gapThreshold 5e-4 Proximity distance threshold in metres
gapPatchPairs none Explicit patch pairs to check for proximity
gapSuppressPatches none Patches to suppress BL on; other patches are protected
gapFaceRingExclusion true Enable face-ring BFS around gap points to suppress prism topology locally
gapFaceRing0-2Scale 0.02/0.05/0.20 layerScale for gap face-ring suppression zones
layerScaleRing1-6 0.10/0.25/0.45/0.65/0.80/1.00 BL/no-BL transition ramp scales
terminationPatches none Patches with no BL, such as inlet/outlet
postRefineBLOptimisation true Run surface optimizer after BL refinement
constrainOptimizerBoundaryMotion false Enable Layer 3 surface-constrained optimizer during volume optimization
postOptimizerReprojStepFraction 0.005 Fraction of drift corrected per incremental post-optimizer re-projection pass
postOptimizerReprojPasses 10 Maximum incremental post-optimizer re-projection passes
postBLSnap false Post-optimizer/pre-BL surface re-projection safety pass (experimental)
postBLSnapTolerance 1e-6 Minimum drift threshold for post-optimizer/pre-BL snap candidate selection (m)
postBLSnapAllowedPyramidIncrease 25 Absolute bad-pyramid increase allowed during snap validation
postBLSnapAllowedPyramidIncreaseFraction 0.001 Fractional bad-pyramid increase allowed during snap validation

BL/BL sharp-junction suppression

Points where two BL patches meet at a sharp angle (blade+hub, blade+shroud) are detected using patch normal angle comparison and suppressed. Prevents degenerate layer cells at blade-endwall junctions in turbomachinery geometry.

Surface-constrained optimizer (v0.10)

Layer 3 surface-constrained optimizer architecture: boundary points are constrained to their STL patch surface during volume optimization, preventing drift accumulation.

Architecture

The volume optimizer (optimizeMeshFV, untangleMeshFV, optimizeLowQualityFaces) previously moved boundary points freely in 3D with no surface awareness. Points could drift up to 1.3mm from the STL surface, causing visible protrusions that the downstream snap could not correct without inverting cells.

Fix: partTetMesh::updateOrigMesh now applies a limited-displacement re-projection for single-patch boundary points before committing optimizer results:

  1. Serial pre-pass: query octree for each drifted boundary point (drift > 1e-5m), compute limited target position (1% step toward STL)
  2. Parallel write-back: apply precomputed targets in OMP — no octree access in parallel path

Results on Rotor 37

  • Visible-scale drift (>100 micron points): 2863 → 1540 (46% reduction)
  • Max drift: 1.19mm → 0.52mm
  • BL coverage: 95–99% across all patches

refineFace graceful handling

refineBoundaryLayers::refineFace previously called FatalError when no split direction could be found for transition-zone faces. Now returns the original face unrefined instead of crashing.

new 3-gate rollback in development

Initial infrastructure exists for detecting degenerate layer-cell candidates and reducing local layer complexity, but the detection/rollback policy is still under development:

  • mixedDisplacementFaces_ member on boundaryLayers
  • forceSingleLayerFaces_ + forceSingleLayerAtFaces() on refineBoundaryLayers
  • Layer count smoothing on boundary-face adjacency graph (max jump of 1 between neighbors)
  • Phase 1 non-mutating diagnostic in progress

Blade TE singularity One residual high-aspect cell and one near-zero volume cell at the blade trailing-edge geometric singularity (knife-edge meeting shroud+periodic). This is a CAD/STL limitation — the feature is below mesh resolution. The geometry preprocessing weld mitigates but cannot fully resolve a true zero-thickness trailing edge.

BL/BL feature-edge rounding Blade/hub and blade/shroud feature curves can be slightly rounded by the surface smoother. Fix in progress: feature-curve locking during smooth passes.

What's new

Patch-role classification system

Patches are now classified into three roles driven by meshDict:

  • BL_PATCH wall patches with active boundary layers
  • TERMINATION_PATCH inlet/outlet (explicit via terminationPatches)
  • NEUTRAL_PATCH periodic/symmetry (no BL suppression triggered)
boundaryLayers
{
    terminationPatches (inlet outlet);
    ...
}

BL/neutral edge constraint (v0.7)

Blade/periodic junction points where a BL wall patch meets a periodic/symmetry patch are now fully constrained:

  • Projection: locked to BL-side patch only via findNearestSurfacePointInRegion
  • Extrusion normal: recomputed from BL-side faces only, preventing tilt toward periodic plane
  • Crossing clamp: extruded points that cross the periodic plane are clamped back to surface
  • VTK diagnostic: blNeutralEdgePoints_predetect.vtk written per run for ParaView inspection

Topology-aware BL suppression

Corner points (3+ patches) at BL/termination junctions are fully suppressed (layerScale_ = 0.0). Sharp two-patch BL/termination edges are suppressed based on patch normal angle (75° default threshold). Periodic/neutral patches never trigger suppression regardless of angle.

Parallel hardening and deterministic-optimizer work (v0.8–v0.9)

Many known race-sensitive OMP apply loops have been serialized or guarded across the meshLibrary. v0.9 adds: negativeNode tet-marking loops (primary source of OMP=32 nondeterminism), isExitFace/isBndPoint/removeBndPoint shared-write races, and 100+ globalToLocal guards across 30+ files. Serial (OMP=1) runs are now fully deterministic. Minor residual variance at OMP=32 under continued investigation.

Mesh quality on NASA Rotor 37 (development baseline)

Tested with boundary layers active. Current development testing commonly uses 3-layer blade boundary layers and 3–5 layer hub/shroud configurations depending on the experiment.

Metric v0.10 (current) v0.9
Cells ~1.77M ~1.77M
Average non-orthogonality ~12.5° TBD
Non-ortho errors 3–9 TBD
Negative volume cells 2–5 TBD
Bad pyramid faces 11–57 TBD
Max skewness ~8.5–11.6 TBD
Visible-scale drift (>100µm pts) ~1540 ~2863
Runtime (OMP=4) ~580s wall ~90s
Runtime (OMP=32) ~204s wall ~90s
Deterministic OMP=4 stable Improving

Version history

Tag Branch Description
v0.10 feature/bl-transition-topology Surface-constrained optimizer (Layer 3): patch-constrained projection in mapVerticesOntoSurface, limited-displacement partTetMesh::updateOrigMesh re-projection, two-phase serial/parallel write-back, post-optimizer incremental snap, pre-BL snap with displacement filter and pyramid-tolerant rollback. 46% reduction in visible-scale boundary drift.
v0.9 feature/bl-transition-topology Comprehensive parallel hardening: 100+ globalToLocal guards, partTetMesh/partTriMesh updateType races, otherVrts_ std::map race in BL vertex creation, frontPoints extrusion race, meshSurfaceEngineModifier marking races, optimiseBoundaryVolumeOptimizer call-site bug, MeshUntangler convergence fix, untangle stagnation detection. Serial runs fully deterministic; OMP=32 quality substantially improved.
v0.8.3 feature/bl-transition-topology Virtual topology face-ring BFS..., in a selected Rotor 37 test run: 0 bad pyramids, skew 9.77.
v0.8.2 feature/bl-transition-topology Geometry preprocessing..., in a selected Rotor 37 test run: zero bad pyramids, zero non-ortho errors.
v0.8 feature/bl-transition-topology Full codebase audit 60+ bug fixes: OMP apply-loop serialization, globalToLocal guards, activePatch_ owner/patch bugs, face size guards, bounds guards throughout. Deterministic results on Rotor 37.
v0.7.1 feature/bl-transition-topology BL/neutral extrusion constraint + crossing clamp + layer rollback scaffold
v0.7 feature/bl-transition-topology BL/neutral edge detection + BL-side projection constraint + VTK diagnostics
v0.6 feature/deterministic-optimizer Partial deterministic optimizer 8 unsafe OMP loops serialized
v0.5 openfoam13 Pure quality evaluator, thread-safe, tight skewness variance
v0.4 openfoam13 Targeted repair of validity-rejected points
v0.35 openfoam13 VTK diagnostics for validity-rejected points
v0.2 openfoam13 Skewness proxy in quality evaluator
v0.1-beta-stable openfoam13 Pipeline reorder + relative validity thresholds

Running the regression test

An automated validation script is included that runs cartesianMesh on the Rotor 37 case and checks all key quality metrics against known-good baselines.

~/cfMesh/tutorials/rotor37/regression_test.sh ~/rotor37_mesh

Expected output will vary by thread count and active controls, but current Rotor 37 development runs are typically in this range:

cells:            ~1.77M
Mesh non-orthogonality Max: ~89 degrees  average: ~8–13 degrees
Number of severely non-orthogonal (> 70 degrees) faces: case-dependent
Max skewness: ~8.5–11.6

Run this after any code change to catch regressions before committing.

## Troubleshooting

**Build fails with missing header:** Make sure OpenFOAM 13 is sourced before building. Run `echo $WM_PROJECT_VERSION` should return `13`.

**`cartesianMesh` not found after build:** Check `$FOAM_USER_APPBIN` is in your PATH. Run `source /opt/openfoam13/etc/bashrc` again.

**Mesh hangs or stalls:** Several known OMP and topology-repair hangs have been fixed, but this is still an active-development mesher. If a case hangs, rerun with `OMP_NUM_THREADS=1` to separate serial logic from parallel nondeterminism, then open an issue with the STL, `meshDict`, thread count, and relevant log output.

---

## Tested on
- Ubuntu 22.04 WSL2, OpenFOAM Foundation v13
- NASA Rotor 37 two-passage turbomachinery sector with boundary layers

## NASA Rotor 37 boundary-layer mesh preview

<p align="center">
  <img src="NASA_Rotor_37_1.png" alt="NASA Rotor 37 cfMesh OpenFOAM 13 boundary-layer mesh preview 1" width="48%">
  <img src="NASA_Rotor_37_2.png" alt="NASA Rotor 37 cfMesh OpenFOAM 13 boundary-layer mesh preview 2" width="48%">
</p>

<p align="center">
  <img src="NASA_Rotor_37_3.png" alt="NASA Rotor 37 cfMesh OpenFOAM 13 boundary-layer mesh preview 3" width="48%">
  <img src="NASA_Rotor_37_4.png" alt="NASA Rotor 37 cfMesh OpenFOAM 13 boundary-layer mesh preview 4" width="48%">
</p>

<p align="center">
  <em>NASA Rotor 37 two-passage turbomachinery sector generated with the OpenFOAM 13 cfMesh port.</em>
</p>

---

## Original cfMesh
- Source: https://sourceforge.net/projects/cfmesh/
- Authors: Franjo Juretic, Philippose Rajan, Ivor Clifford

---
## Tutorial: meshing NASA Rotor 37

A complete worked example using the included Rotor 37 two-passage turbomachinery sector geometry.

### Prerequisites
- cfMesh built and sourced (see Quick start above)
- ParaView installed (for visualization)

### 1. Set up the case

```bash
mkdir -p ~/rotor37_mesh/constant/triSurface
mkdir -p ~/rotor37_mesh/system
cp /path/to/Rotor37_cfmesh.stl ~/rotor37_mesh/constant/triSurface/

2. Copy the included meshDict

cp /path/to/cfMesh/tutorials/rotor37/system/meshDict ~/rotor37_mesh/system/

The included meshDict is pre-configured for the Rotor 37 geometry with 3-layer blade BL and 5-layer hub/shroud BL.

3. Run the mesher

cd ~/rotor37_mesh
cartesianMesh 2>&1 | tee mesh.log

Expected runtime depends strongly on whether constrainOptimizerBoundaryMotion is enabled. On a Ryzen 7950X3D / WSL2 test setup, unconstrained runs are significantly faster; v0.10 surface-constrained runs are currently slower because the octree projection pre-pass is still being optimized.

Finished renaming boundary patches

4. Check mesh quality

checkMesh -case ~/rotor37_mesh 2>&1 | grep -E "cells:|average non-orth|Failed|aspect|negative"

Expected output: cells: ~1.77M Mesh non-orthogonality Max: ~89 degrees average: ~8.6 degrees *Number of severely non-orthogonal (> 70 degrees) faces: ~600 ***Max skewness = ~8.2-8.4

5. Visualize in ParaView

paraFoam -case ~/rotor37_mesh &

Or create a .foam file and open manually:

touch ~/rotor37_mesh/rotor37.foam
# Open rotor37.foam in ParaView
# Apply → select patches: blade_1-4, hub, shroud
# Color by: p or U after running a solver

To inspect boundary layers: in ParaView, use Filters → Extract Block to isolate wall patches, then apply Surface with Edges representation. The 3–5 layer prism stacks on blade and endwall surfaces should be clearly visible.

License and Attribution

This project is licensed under the GNU General Public License v3.0, consistent with the original cfMesh license. Original cfMesh authorship and copyright remain with the original cfMesh authors, including Franjo Juretić / Creative Fields. This fork includes OpenFOAM 13 compatibility updates, build fixes, and meshing robustness modifications by Mitch Stolk. Modifications in this fork Copyright (C) 2026 Mitch Stolk. If this fork is useful in academic, research, commercial evaluation, or public engineering work, attribution to this repository is appreciated. Mitchell Stolk, cfMesh-OpenFOAM13: OpenFOAM 13 port and robustness-focused fork of cfMesh, 2026.

Disclaimer

Unofficial community port. Not affiliated with Creative Fields or the OpenFOAM Foundation.

About

cfMesh port for OpenFOAM Foundation v13 with boundary layer robustness improvements. Tested on NASA Rotor 37.

Topics

gplv3 meshing mesh-generation cfd openfoam computational-fluid-dynamics hex-mesh cfmesh openfoam13 hex-dominant cartesianmesh cfd-meshing

Resources

Readme
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases 1

cfMesh OpenFOAM 12 and 13 Port Latest
May 9, 2026

Packages

 
 
 

Contributors

Languages