AR-Powered Spatial Inspection Platform

The Problem

Large-part quality control is labour-intensive and error-prone. A single wind turbine blade inspection can require hundreds of individual measurements, each recorded manually and later re-entered into an ERP system.

The conventional workflow results in:

• At least five manual measurements per defect (location, dimensions, distances along curved surfaces)
• Handwritten field forms that introduce transcription errors on re-entry
• No direct link between defect records and the composite structure at that point
• Ad-hoc escalation via email when an inspector encounters an ambiguous finding

Key pain points

• Measurement overhead — Each finding requires multiple independent measurements with tape, calipers, and coordinate notation.
• Re-entry errors — Transcribing field notes into ERP or QMS systems introduces inaccuracies at every handoff.
• No layer context — Field records capture location and dimension but not which composite plies are affected — requiring a secondary engineering analysis.
• Escalation friction — Unclear findings travel via email with photos rather than through a structured workflow.

AR Inspection

Inspectors carry an iPhone to the part. They open ROBOSCOPE, scan the work task router QR code to register the session, and begin placing AR markers directly on the surface.

Each marker captures:

• Spatial location — AR anchor coordinates relative to the part (~5 cm accuracy)
• Defect type — selected from a pre-configured classification list
• Dimensions — entered via guided prompts after initial placement
• Photographic evidence — automatic capture from every angle during placement
• Inspector ID and timestamp — bound to the session record

Workflow

1. Open app, authenticate with credentials
2. Scan work task router to register session and load part geometry
3. Walk to each defect location and place an AR marker directly on the surface
4. Complete dimension entry and defect classification for each marker
5. Review session, escalate any flagged items if needed
6. End session — ROBOSCOPE generates the defect record and syncs to ERP

Figure 1 — AR inspection session workflow

AR system performance

Metric	Value
Placement accuracy	~5 cm on large surfaces
Supported surface types	Composite, painted metal, coated structures
Max session markers	Unlimited
Session sync	On Wi-Fi at session close or deferred
Platform	iOS (iPhone)

Layup Registration Table

The Layup Registration Table (LRT) maps every AR-anchored defect to the exact composite layers at that point on the part geometry — giving engineers a full cross-section view without disassembly.

How it works

1. Locate — AR anchor position is resolved against the part’s layup geometry to identify the zone
2. Layer lookup — LRT returns the ordered stack of plies at that coordinate: material, orientation, and thickness per layer
3. Depth assessment — Defect depth is cross-referenced with the layer stack to determine which plies are affected
4. Severity grading — Each affected layer is graded individually; structural plies trigger automatic escalation
5. Export — Full layer-annotated defect record exported to ERP or repair work order with ply-level detail

Figure 2 — Roboscope components and data flow

LRT data per defect

Field	Description
Zone ID	Named layup zone resolved from AR anchor coordinates
Ply stack	Ordered list of plies with material, angle, and nominal thickness
Affected layers	Subset of plies intersecting the defect depth envelope
Repair class	Repair category derived from affected structural ply count

ML Intelligence

ROBOSCOPE includes purpose-trained vision models covering the full defect spectrum for large industrial surfaces and wind turbine blades.

Models run on-device via CoreML, enabling offline operation in environments without reliable connectivity.

Model coverage

• Defect classification — identifies defect type from photographic evidence
• Severity grading — assigns a severity class based on visual characteristics
• Structural anomaly detection — flags features consistent with subsurface damage
• Paintwork issues — detects delamination, cracking, and coating failures
• Circularity deviation — quantifies how much a real curve deviates from its nominal geometry
• Edge case routing — escalates findings that fall outside model confidence thresholds to qualified engineers

Figure 3 — On-device ML model pipeline

Training and updates

Models are trained on domain-specific data collected during active sessions and continually expanded as new failure modes are encountered. Updates are distributed as app updates through Apple Business Manager.

Platform

ROBOSCOPE is an iOS-native application designed for handheld use on the factory floor.

Deployment

Component	Detail
Mobile app	iOS (iPhone) — native, offline-first
Device management	Apple Business Manager
Backend	On-premise Linux VM or Spiral-managed cloud
Integration	REST API — ERP, QMS, PLM, MES
Export formats	JSON, CSV, image packages with metadata

Connectivity model

The application stores all session data locally on-device. Sync to the backend occurs when Wi-Fi is available — at session close or manually deferred. No inspection capability is lost in areas without connectivity.

Industries

Primary

• Wind turbine blade manufacturing — full blade surface inspection, leading edge assessment, structural anomaly detection

Expanding

• Construction — large concrete and steel structure assessment
• Marine and shipbuilding — hull coating and corrosion inspection
• Aerospace composites — large structural panel and fuselage section QC

Benefits

Accurate	AR anchoring eliminates manual coordinate notation. Every defect is positioned at the physical location and recorded automatically.
Complete	A single AR session captures location, type, dimensions, and photographic evidence without tool switching.
Layer-aware	LRT links every surface defect to the exact composite ply stack at that point — without disassembly or secondary engineering analysis.
Direct	Session data flows from the inspector’s phone to the ERP record. No re-entry, no handwritten forms.
Escalation-ready	Edge cases and structural findings are automatically routed through a structured workflow instead of email chains.
Offline-capable	All capture and AI inference runs on-device. Inspections proceed in environments without Wi-Fi or cellular coverage.