AI-assisted rail surface inspection

Every metre of rail, turned into measured evidence.

InfraVision RailSense analyses the inspection imagery you already collect and returns audit-ready findings: defect type, millimetre geometry, position on the railhead, D1–D4 severity and a prioritised review queue — mapped to the quality indices your engineers already use.

Explore the live dashboard → See capabilities

ALIGNED WITH  GR.IT.VIA.026 (ISC descriptor) · EN 13231-1 · UIC 712R · EUR cost anchors: AMT 2024 / IP S.A. 2023

INSPECTION FINDING · EVIDENCE CHAIN PILOT OUTPUT
Rail surface image with detected defect region highlighted
0
defect instances detected & measured in the pilot run
0
inspection images processed end-to-end
0
surface condition classes — 7 defect types plus healthy-rail discrimination
0
findings auto-ranked into a prioritised expert review queue

Figures from a single locked pilot run on operational inspection imagery · every number traceable to an exported case record.

What the platform does

From raw image to maintenance decision — one traceable chain.

RailSense is not a black box that outputs labels. Each finding carries its full evidence: where it is, how big it is, how severe it is, and what that means for the section it sits in.

DETECT

Multi-class defect detection

Locates squats, wheelburns, corrugation, head checks, shelling, spalling and flaking on the railhead — with pixel-level delineation of each defect region, not just a bounding box.

MEASURE

Millimetre geometry

Calibrated measurement of every finding: length, width, area in mm², and distance from the gauge edge — the numbers your grinding and renewal thresholds are written in.

LOCATE

Zone-aware positioning

Every defect is assigned to its rail zone — gauge corner, gauge shoulder or running band — plus GPS/section matching, so risk is read in the context that determines treatment.

GRADE

D1–D4 severity & standards mapping

Visual severity levels map to ISC condition states under GR.IT.VIA.026, with per-class advisories referencing EN 13231-1 and UIC 712R treatment thresholds.

INDEX

Section quality index (RSQI)

Findings aggregate into a per-200m surface-condition index — an automated, image-based estimator of the ISC descriptor that feeds your existing network quality framework.

DECIDE

Budget & deterioration scenarios

Immediate remediation estimates in EUR, plus no-maintenance deterioration scenarios showing how deferral shifts cost over 6–60 months — clearly labelled as scenarios, never sold as predictions.

How it works

Four stages. No new hardware.

RailSense runs on the visual inspection imagery your measurement trains or track teams already produce.

STAGE 01

Ingest & calibrate

Inspection images are cropped to the railhead, oriented, and calibrated to millimetres. GPS or chainage metadata is matched to each frame.

→ calibrated rail frames
STAGE 02

Detect & delineate

The vision pipeline proposes candidate defects, delineates each region at pixel level, and classifies it into one of eight surface defect classes.

→ classified defect regions
STAGE 03

Grade & index

Each finding is measured, zoned, graded D1–D4, and rolled up into section-level RSQI with standards-referenced advisories.

→ severity + section index
STAGE 04

Review & decide

Findings land in a prioritised dashboard: review queue, risk matrix, cost estimates and exportable case records for your asset system.

→ manager dashboard
Real pilot output

See the evidence, not a promise.

These finding cards are unedited exports from the pilot run — the same artefact an inspector or asset manager receives. Click any card to open it full size.

Severity colours: D4 critical · D3 severe · D2 moderate · D1 minor. Each card embeds its advisory and reference standard.

Manager dashboard

One screen for the whole network story.

Severity mix, section quality, review priorities, risk matrix and budget scenarios — populated live with real pilot data so you can judge the product on its actual output.

No signup. No sales call first. Open it and click around.

Open the live dashboard →

Demo is populated from a locked pilot run (87 images). Deterioration & cost figures are labelled no-maintenance scenarios — not validated forecasts.

194
findings with full evidence chain
46
critical (D4) findings flagged for urgent expert review
33.3 /100
RSQI — active defect sections (full inspected set: 93.2)
€56,950
immediate remediation estimate at current severity
Built for auditability

Honest by design.

Infrastructure decisions demand traceability. RailSense is engineered so every claim on screen can be defended in an audit — including its own limits.

Standards-anchored, not standards-adjacent

The section index is an automated image-based estimator of the ISC descriptor within Portugal's IQC framework (GR.IT.VIA.026) — designed to feed the index your network already reports, not replace it. Advisories cite EN 13231-1 and UIC 712R thresholds.

Traceable numbers

Cost figures are anchored to published infrastructure economics (AMT 2024, IP S.A. 2023) in EUR. Every finding, index value and projection exports to CSV with its assumptions attached.

Expert-in-the-loop by default

RailSense is AI-assisted inspection triage — it prioritises and evidences; your experts decide. Safety-critical findings are explicitly routed to ultrasonic / manual verification.

What we don't claim

  • Deterioration curves are labelled no-maintenance scenarios, not validated predictions.
  • Precision/recall figures are published only against expert ground truth — never proxied.
  • The visual RSQI covers surface condition; it does not replace geometry or ultrasonic inspection.