---
name: feedback-kds-rebar-bend-radius
description: "Rebar bend rule = KDS 24 14 21 Table 4.5-1 ONLY (KDS 14 20 50 not used) + validated Rhino method for building bent rebar"
metadata: 
  node_type: memory
  type: feedback
  originSessionId: 3f00c5a2-89d3-4ff6-bce0-294eebc0a639
  modified: 2026-09-22T05:31:31.332Z
---

Rebar bend geometry follows **KDS 24 14 21 4.5.2.1 Table 4.5-1** (min inner bend **diameter**, not radius). KDS 14 20 50 (building code) does NOT apply to this project - user ruled it out 2026-10-22.

- Deformed bar: D16 or less 2db / D19 or more 3.5db (inner radius 1db / 1.75db). Plain bar: 1.25db / 2.5db.
- Standard hook (4.5.4.2): 90-150 deg bend, extend >= 5db past the bend; >=150 deg also >= 5db.
- Nominal bar diameters (KS D 3504): H16=15.9mm (commonly 16), H19=19.1, H22=22.2, H25=25.4, H29=28.6.
- Already-built P1 models used R=3db (H25 R=76.2mm) / 4db (H29 R=114.4mm): old values, larger than the KDS 24 14 21 minimum, so not violations - but they are NOT the standard.
- Which R to use for NEW modeling (code minimum vs previous practice) is the user's decision - [미확정], ask before building.

**Why:** user wants only the bridge code (KDS 24 14 21) carried in this project; the earlier 14 20 50 table was a building-code carry-over that does not fit.

**How to apply:** compute inner radius from Table 4.5-1 (or the user-confirmed value), then centerline fillet radius = inner radius + db/2. See [[reference_kds_24_14_21]] and [[project-pyeonggang1-modeling-standard|future pier projects]].

## How to actually build a real KDS-compliant curved bend in Rhino (method validated 2026-09-23, [[project_p1_pier_structure]] C8/C10 rebuild)

**Do not build bent rebar as multiple straight `Brep.CreatePipe` segments joined at sharp/mitered corners** — this was the original (wrong) construction found in C8/C10: many tiny straight sub-segments (some as short as ~18-30mm) meeting at sharp vertices with zero bend radius. A real KDS bend needs an actual circular fillet arc at the corner, and short straight segments physically cannot fit a realistic bend radius (e.g. R=76.2mm for H25 needs the arc's two tangent legs to each have that much straight room on both sides — a corner between two ~20-30mm segments cannot hold it at all). If straight segments were pre-built that short, first **simplify the path to only its真正 structural corners** (e.g. collapse "hook tip → tiny kink → tiny kink → long diagonal" down to just the long diagonal's own two endpoints) before fillet-ing — don't try to fillet every pre-existing micro-corner.

**The actual working method — one clean call, not manual arc/tangent-point math:**
1. Build the full **centerline** path as an ordered list of `Point3d` (the bar's true corner/anchor points — hook tips, bend points, straight-span ends). Convert to a `Rhino.Geometry.Polyline` then `.ToPolylineCurve()`.
2. **Centerline fillet radius = KDS inner radius R + bar radius (db/2)** — e.g. (past P1 case, old value, not the standard) H25: R_inner=76.2mm + db/2=12.7mm → centerline radius = **88.9mm**. (The inner radius is the radius of the bend's *inner* face, not the centerline — always add the bar radius to get the sweep-rail radius.)
3. Call **`Rhino.Geometry.Curve.CreateFilletCornersCurve(polylineCurve, centerlineRadius, tolerance, angleTolerance)`** — this single static method fillets *every* interior corner of the polyline in one call and returns a `PolyCurve` with the true endpoints preserved exactly (start/end untouched, only interior corners rounded). Far simpler and more robust than manually computing per-corner tangent points and building/trimming arcs by hand.
4. Sweep **one single pipe** along the whole resulting smooth `PolyCurve` via `Brep.CreatePipe(filletedCurve, barRadius, False, PipeCapMode.NONE, False, tol, angleTol)` — this gives ONE clean Brep for the entire bent bar (not a fragmented multi-piece assembly), since the curve itself is already one continuous smooth path.
5. **Verify the fillet radius actually landed correctly** by sampling curvature along the filleted curve *before* piping it: loop `t` across `curve.Domain`, call `curve.CurvatureAt(t).Length` (=kappa), and check `1.0/kappa` at its minimum equals the intended centerline radius (88.9 in the H25 example) — this was the actual validation step used and it matched to 5 decimal places. (Checking curvature on the final *pipe's* boundary edges afterward is unreliable — pipe edges mix in the cross-sectional circle's own curvature; always validate on the rail/centerline curve itself, pre-pipe.)
6. If the corner spacing is too tight for `CreateFilletCornersCurve` to succeed (returns `None`), that's the signal the point list still has micro-segments — go back to step 1 and simplify the path further, don't lower the radius to force it to fit.

**Gotcha carried over from the cover-fix work**: when re-deriving centerline points from an already-built (possibly mirrored) Brep, don't assume a vertex read via `.Vertices` is on any particular side (top/bottom) of the pipe — use the **bounding-box center** of the relevant flat/straight piece (`(bbox.Min.Z + bbox.Max.Z) / 2.0`) to get the true centerline Z robustly regardless of which side the vertex happens to sit on for a given (possibly mirror-reflected) bar; this was mixed up once (added the bar radius in the wrong direction for the mirrored C10) and caught only by comparing C8/C10's post-fix bounding boxes and finding them asymmetric when they should have matched.