Severity mix D1–D4 · instance level
13 further regions were excluded as non-localized rail-context evidence (D0) — counted, never silently dropped.
Defect classes 181 localized findings
“Pending confirmation” = candidate patterns held for inspector confirmation rather than force-classified.
Maintenance priority auto-assigned
Priority combines severity, rail zone and geometry; Review-only items carry evidence but no localized defect call.
Per-section surface condition (RSQI)
| Section | Findings | RSQI | Class | Manager action |
|---|---|---|---|---|
| sec_200m:1 | 94 | 0.0 | Poor | Immediate ISC section review — index saturated; rank by defect & R/RU counts |
| sec_200m:2 | 64 | 0.0 | Poor | Immediate R/RU review — 6 findings already at repair-state; 42 flagged for UT |
| sec_200m:0 | 23 | 100.0 | Adequate | PLC/corrugation grinding review — high corrugation burden; do not read as healthy |
Scope: image-based surface-condition RSQI estimating the ISC descriptor (GR.IT.VIA.026). It does not include track geometry, ultrasonic results or traffic history. Corrugation is reported as a separate descriptor, consistent with the standard.
Risk matrix 87 images · worst finding per image
Cell = image count at that severity × likelihood. 4 images sit in the worst cell (Critical × Very likely) — these top the review queue below.
Cost of acting now vs. cost of deferring EUR · scenario envelope
Read this correctly: curves show expected future remediation cost as untreated defects progress through condition states (initial mix: 98 early-stage · 20 progressed · 6 at repair-state · 0 urgent). Clean track and new-defect emergence are out of scope. Deterioration rates are literature-anchored for squat-type defects and extrapolated for other classes — treat 60-month values as indicative scenarios.
Highest-priority images ranked by review-priority score
| # | Image | Classes | Risk | Priority score |
|---|
Score blends severity, likelihood, zone criticality and measured geometry — so inspectors open the right image first. Full queue (166 findings) exports to CSV for your maintenance system.
Finding cards — click to open the full evidence chain
Each card is the audit artefact itself: defect class, zone, millimetre geometry, D-level, priority and the standards-referenced advisory. Multi-defect images additionally receive a combined per-image evidence sheet.
⚠ Validation status — read before quoting numbers
Current status: pilot-ready AI-assisted inspection triage. Not certified for autonomous maintenance decisions. This dashboard provides traceable visual evidence, review prioritisation, an image-based surface-condition index and clearly-labelled cost scenarios. Precision, recall and false-negative rates are published only against expert ground truth — a GT protocol and label template ship with every pilot. Before production use: repeated inspections of the same assets, local cost-table calibration, verified GPS/chainage and expert-confirmed labels.