No universal best algorithm
Uniform-by-length is efficient for smooth convex curves; adaptive-by-chord is more robust for hybrid and free-form profiles.
Research software / Geometry preprocessing
CurvAdapt
A reliability-first workflow for translating curved design geometry into simulation-ready polylines.
Question
When a smooth curve becomes a polyline, how much geometric error is still safe for building-performance simulation?
Response
CurvAdapt controls the deviation itself — using maximum chord error and relative MCE — instead of treating segment count as evidence of accuracy.
6
representative curve profiles
5
segmentation strategies
|r| ≈ .99
radiation and annual EUI
G¹ aware
critical breaks preserved
01
The weak link before simulation
NURBS geometry cannot enter many performance engines directly. It is discretized first, often with an arbitrary segment count. The model may look smooth while its deviation, topology, and simulation reliability remain unknown.
CurvAdapt separates segmentation from simplification, preserves real discontinuities, and reports geometric fidelity in a scale-aware form.
02
Reliability workflow
-
01
Prepare
Detect domains, duplicate parameters, seams, kinks, and G¹ breakpoints.
-
02
Segment
Compare uniform and adaptive routines across different curvature behaviours.
-
03
Verify geometry
Measure MCE and normalize it as rMCE so fidelity can be compared across scale.
-
04
Check performance
Use fast incident-radiation convergence before committing to heavier daylight and energy runs.
03
What the study found
Segment count can mislead
More segments do not guarantee a proportionally better approximation and may conceal false convergence.
Radiation is a useful diagnostic
It retains sensitivity to geometry and strongly tracks annual energy convergence at a fraction of the computation.
Daylight needs extra care
Sensor-grid allocation can introduce oscillations that are not caused by the curve itself.
04
Method trace
05
Evidence, not a magic threshold
The tested thresholds are evidence from specific curve families, not universal constants. The transferable contribution is the procedure for deriving an appropriate threshold for a new geometry and simulation objective.
06
Current boundary
Under revision
The journal review recognized the workflow’s merit while asking for tighter terminology, broader climate testing, stronger sensitivity analysis, and a more modest novelty claim. The paper and plugin are being revised around that reliability-first contribution.
07
Research team
Hossein Nazari — conceptualization, methodology, algorithms, data analysis and original draft. With Shadan Masoud, Abbas Tarkashvand and Mehdi Ghiai.