Aircraft structural repairs involving welding operations — fuselage skin patches, stringer replacements, frame splices, bulkhead reinforcements, and landing gear lug repairs — require weld integrity verification that meets strict airworthiness standards. Manual weld inspection relying on visual-tactile methods and spot-check NDT sampling misses an estimated 15-20% of subsurface defects, with the highest miss rates on complex geometry repairs and multi-pass weld joints. The global welding robotics market reached USD 10.62 billion in 2025 and is projected to hit USD 32.11 billion by 2036 as AI-enabled inspection systems become standard. Robotic weld inspection systems combining structured light scanning, phased array ultrasonics, and AI-based defect classification reduce missed defects by 82% while cutting inspection cycle time by 60% compared to manual methods. This article examines the defect types that robotic systems detect, the sensor technologies deployed, and how iFactory's Structural Repair Tracking module integrates weld inspection data with the full repair lifecycle from work order creation through final certification.
AI Weld Inspection · Phased Array UT · Robotic NDT · Structural Repair Tracking
82% Fewer Missed Weld Defects. 60% Faster Inspection Cycles. 100% Audit-Ready Documentation.
iFactory Structural Repair Tracking connects robotic weld inspection data directly to repair work orders, defect records, and certification documentation — turning every inspection point into an auditable maintenance event.
The Real Cost of Uninspected Weld Defects
Every weld defect that escapes inspection carries a cascading cost: unplanned downtime for emergency repair, accelerated fatigue cracking in adjacent structure, and ultimately the regulatory and liability exposure of an in-service failure. The three categories below represent the most common and financially consequential weld defect types found in aircraft structural repairs.
Category 01 • Structural
Critical Cracks & Fusion Defects
Cracks propagating from the weld zone, lack of fusion between weld passes or base metal, and incomplete penetration at the weld root reduce load-bearing capacity below design limits. These defects are the leading cause of in-service structural weld failures. Robotic phased array ultrasonics detects them with 96% accuracy at depths up to 50 mm.
Category 02 • Volumetric
Porosity & Slag Inclusions
Gas pockets trapped during solidification and non-metallic particles lodged between weld passes create stress concentration points that reduce fatigue life by 30-50%. Manual visual inspection catches less than 40% of subsurface porosity. Robotic eddy current arrays detect volumetric defects down to 1.5 mm in aluminium alloy welds.
Category 03 • Profile
Undercut, Excess Reinforcement & Spatter
Surface profile defects including undercut — a groove melted into the base metal adjacent to the weld toe — and excess reinforcement create stress risers and complicate subsequent NDT interpretation. Robotic structured light scanning maps surface topography at 0.05 mm resolution, detecting profile defects in a single pass at 500 mm per second.
Manual vs Robotic Weld Inspection: A Side-by-Side Comparison
The most effective way to understand the impact of robotic weld inspection is to compare how each method performs across the dimensions that matter most to MRO repair quality and turnaround time.
Detection accuracy
78.3% average classification accuracy across common defect types. Operator fatigue and inconsistent probe positioning create variability of ±12% between inspectors.
94.7% consistent accuracy
AI-based classification delivers 94.7% accuracy regardless of inspector experience, shift timing, or inspection volume. False negative rate: 1.2% vs 8-12% for manual methods.
Coverage per weld joint
10-20% volumetric coverage. Manual spot-check UT at 5-8 locations per 500 mm weld leaves 80-90% of the weld volume uninspected.
100% volumetric coverage
Structured light, phased array UT, and eddy current array scan the entire weld volume. Every point inspected, measured, and recorded with spatial coordinates.
Time per 500 mm weld
20-35 minutes including setup, visual inspection, and spot-check UT at 5 locations. Documentation written manually.
8-12 minutes total
Complete volumetric scan including surface topography, phased array UT, and eddy current. Documentation generated automatically with full traceability.
Data quality & audit trail
Paper forms or basic digital records. No per-point coordinate tracking. No baseline data for trend comparison between inspection cycles.
Full digital audit trail
Every inspection point geo-tagged with timestamp, sensor data, AI classification, confidence score, and severity assignment. Exportable for airworthiness authority review.
The AI Classification Pipeline: From Sensor to Severity Score
A robotic weld inspection cell integrates multiple sensor modalities on a single platform. The AI pipeline processes fused sensor data in under 120 milliseconds per frame, transforming raw measurements into actionable defect classifications with severity prioritisation.
01
Multi-Sensor Data Acquisition
Structured light cameras capture surface topography at 0.05 mm resolution. Phased array UT probes scan the full weld volume at 5-15 MHz, detecting subsurface discontinuities. Eddy current arrays map surface and near-surface defects at 100-500 kHz. All three data streams are synchronised to the same spatial coordinate reference.
02
Signal Fusion & Feature Extraction
Raw sensor data is fused into a composite data tensor by spatial coordinate. The AI model extracts spatial features from the structured light point cloud, frequency-domain features from UT A-scans, and impedance-plane features from eddy current signals — creating a multi-dimensional feature vector for every inspection point.
03
CNN Inference & Defect Classification
A trained convolutional neural network processes the feature vector and outputs probability scores across seven defect classes plus a no-defect class. The model is trained on over 50,000 labelled weld inspection images and achieves 94.7% classification accuracy. Inference time is under 120 ms per inspection point.
04
Severity Assignment & Work Order Push
Each classified defect is assigned a severity level based on defect type, size, and location. Critical and moderate findings generate work orders automatically in the iFactory Structural Repair Tracking module. The technician receives the precise location, sensor data, classification, and recommended corrective action for every flagged defect.
Measured Performance Across MRO Operations
94.7%
AI classification accuracy across seven defect types — compared to 78.3% for manual inspection
Source: MRO field study, 14 repair stations
82%
Reduction in missed subsurface defects across all weld geometries and material types
Source: Comparative NDT audit data
60%
Shorter inspection cycle time per weld joint — from 20-35 minutes down to 8-12 minutes
Source: Robotic cell deployment data
Regulatory Standards That Apply to Robotic Weld Inspection
Robotic weld inspection systems must satisfy the same regulatory framework as manual NDT methods. Each relevant standard places specific requirements on inspection procedure qualification, personnel certification, and data documentation.
EASA Part 145
European MRO
Key requirement
Robotic NDT procedure must be approved as part of the repair station's NDT programme. Robot operator must work under a certified NDT Level 2 or 3. Final accept-or-reject decisions remain with the certified inspector.
AI integration requirement
AI classification results are treated as assisted inspection data. Every finding must be reviewed by a certified inspector before maintenance action. Full audit trail of AI confidence scores required.
Key requirement
AC 21-54A provides guidance for advanced NDT technologies. The robotic system is qualified as an NDT method with procedure qualification per ASTM E3169. Inspection results must be reproducible within defined tolerances.
AI integration requirement
AC 21-54A permits AI-assisted inspection with human-in-the-loop validation. Robot positioning repeatability must be verified annually. Thermal and UT calibration records must be maintained for each sensor.
Key requirement
Defines standard practice for qualification of automated NDT systems. Requires demonstrated probability of detection (POD) for each defect type the system is qualified to detect, with statistical confidence intervals.
AI integration requirement
AI model updates that change classification parameters require re-qualification of the affected defect detection capabilities. POD demonstration must include the AI pipeline end-to-end, not just the sensor hardware.
iFactory Structural Repair Tracking
Every Weld Inspected. Every Defect Classified. Every Repair Certified.
iFactory Structural Repair Tracking connects robotic weld inspection systems directly to repair work orders, defect tracking databases, and certification documentation. Every inspection point is geo-tagged to the aircraft structural coordinate system with full sensor data, AI classification results, and severity assignment. Supports single and multi-site MRO operations with consolidated dashboards for repair programme managers, quality assurance teams, and airworthiness authorities.
Pilot in 30 days. Full deployment in one quarter.
Frequently Asked Questions
Every weld on your aircraft carries a structural load. The defects you cannot see are the ones that cost the most. A robotic inspection system can show you every single one.
iFactory Structural Repair Tracking connects your robotic weld inspection cell to your entire repair workflow — from work order creation through AI classification, severity assignment, and airworthiness certification. Book a Demo to see how your structural repair data maps to your inspection obligations.
