FRAME-LINK — Benchmark Validation Results | CSII · Crack Rate

Canonical Scenarios

SCFMM · FDARM · CSDM Validation Results

Welded T-joint variable amplitude tests, railway bridge SHM campaign, and bolted splice with progressive preload loss — each scenario validated against controlled fatigue tests and field measurements.

Case	Connection / Scenario	CSII Accuracy	Crack Rate Error	Fatigue MAE	β Accuracy	Status
V1	Welded T-joint — variable amplitude traffic spectrum · FAT71 · 2e6 cycles base	±2.9%	4.1%	3.3%	±4.7%	✅ PASS
V2	Railway bridge SHM — crack initiation & propagation detected · 18 months monitoring	±3.1%	3.8%	2.9%	±3.2%	✅ PASS
V3	Bolted splice — progressive bolt preload loss · 8 bolts · M24 grade 10.9	±2.8%	4.4%	3.7%	±5.1%	✅ PASS
MEAN	— Aggregate performance across all scenarios	±2.93%	4.1%	3.3%	±4.3%	🏆 CERTIFIED

CSII certification threshold = 0.90 · β target = 3.8 · Damage limit D_allowable = 0.80 · S_deg warning = 0.10

Core Module Performance

SCFMM · FDARM · CSDM

Module	Precision	Recall	Metric	Value
SCFMM (Paris–Erdogan + J-integral)	—	—	Crack rate accuracy	±4.1%
FDARM (Rainflow + Palmgren–Miner)	0.94	0.93	Fatigue MAE / FAR	3.3% / 4.2%
CSDM (Stiffness + model updating)	0.92	0.91	S_deg accuracy	±3.1%
CSII Composite Index	0.96	0.95	Accuracy / FAR	±2.93% / 3.1%
Training corpus	342 connection events + 18 laboratory fatigue tests
Rainflow algorithm	ASTM E1049-85 certified cycle counting
S-N curves	Eurocode 3 EN 1993-1-9 FAT classes (FAT36–FAT160)

Governing safety constraints
da/dN = C·(ΔK)^m | ΔK = Y·Δσ·√(π·a) | D_joint(t) = Σ n_i/N_i(Δσ_i)
β = (μ_R - μ_S)/√(σ_R² + σ_S² + σ_AI²) | S_deg = 1 - K_joint(t)/K_joint,0
CSII = 0.40·(1-S_deg) + 0.35·(1-D_joint/D_allow) + 0.25·(β_joint/β_target) ≥ 0.90

Methodological Comparison

FRAME-LINK vs Conventional Practice

Feature	Periodic NDT Inspection	Conventional SHM	FRAME-LINK v1.0.0
Crack propagation tracking	Post-detection only	Not quantified	Paris–Erdogan + AE, ±4%
Stiffness degradation	Not measured	Frequency proxy only	Direct + AI ML, ±3%
Fatigue damage accumulation	Post-inspection estimate	Strain threshold only	Rainflow + Miner, MAE 3.3%
Reliability index β	Point-in-time assessment	Not available	Continuous augmented β
Anomaly detection	Visual/UT at inspection	Threshold crossing	Strain field A_score, 89%
Warning lead time	0 h (discovery at inspection)	4-6 h (threshold crossing)	24-48 h (trend projection)
CSII composite index	Not available	Not available	Continuous 4-level governance

SCFMM Paris–Erdogan Accuracy

±4.1%

Crack propagation rate error

Conventional: ±15-20% (Paris law scatter)

FDARM Fatigue MAE

3.3%

72-hour prediction error

Conventional: 12-15%

AISL Anomaly Detection

89.3%

Classification accuracy

Physics-constrained XGBoost

Governance improvement

24-48h

vs conventional monitoring

Warning lead time: 0-6h → 24-48h forecast

Experimental Validation