Skip to content

FINDINGS-RADIUS15.md

Latest commit

 

History

History
42 lines (30 loc) · 2.27 KB

FINDINGS-RADIUS15.md

File metadata and controls

42 lines (30 loc) · 2.27 KB
Raw

Lane RADIUS15 — post-growth cylinder radius (handle-lock 15 mm bore)

Fixture: tests/diag/handle-lock/handle-lock.stl (908 tris). GT cylinder radii include 15.0 mm and 16.0 mm.

Root cause

B1 cylinder growth on the 15 mm bore (seed P=183 Q=220) locks R_ref from the first circumferential band (R≈15.578 at |S|=3), then axial bands drag the bulk eberlyCenterRadius down (R≈11.7 on the full patch). refineCylinderRadius in evaluateCommit could not lift back: chord acceptance was anchored to the dragged radius and the 1.25× lift cap blocked the ~28% recovery needed. Circumferential facet chords on this patch are mostly patch boundary edges (thin axial bands), not mesh-internal edges.

Fix (TrueForm coarse band only; refit_math.cpp + refit_grow.cpp)

Area Change
refit_grow.cpp Pass growth R_ref as rHint into evaluateCommit; track R_ref = max(R_ref, radius) during growth
refit_math.cpp refineCylinderRadius(..., rHint): widen chord window/cap toward rHint; partial-arc boundary chord pass; large-bore axial-drag lift when rHint exceeds bulk fit; pratt center/radius snap after lift

Corpus / Body11 unchanged — refineCylinderRadius still gated to 500–1200 tris.

Proof — handle-lock radii (regiondump --diag)

$ ./build/stl2step_regiondump tests/diag/handle-lock/handle-lock.stl --diag
segment comp0 ok regions=25 … cylinders=16
  id=16 type=cylinder … radius=15.5781  (15 mm GT bore)
  cylinder radii: …,15.578,…,16.012,20.247,29.999,…
GT (mm) Before (mm) After (mm) Δ% vs GT
15 11.703 15.578 3.9
16 16.012 16.012 0.1
20 20.247 20.247 1.2
30 29.999 29.999 0.0
0.5–10 (×12) unchanged unchanged ≤0.2

15 mm bore recovered from axial-drag underestimate (22% error → ~4% vs CAD). On this Fusion STL the circumferential band chord geometry reads 15.58 mm (thin-band eberly matches commit); CAD GT is 15.0 mm — same ~4% overshoot class as coarse tessellation, not the prior 11.7 mm phantom.

Regression

ctest --test-dir build  → 27/27 pass
Body11 --engine trueform → smoothBuiltCylinders=127, openShells=0, watertight=true