Non-Destructive Testing (NDT)¶

Last updated: 2026-05-28

Recent Finds¶

NDT in Canada 2026 — "Taking NDT to New Heights" Conference, Final Day (CINDE, May 26–28, Whistler BC)¶

Status 2026-05-28: Final day (Day 3 of 3) — conference concludes today. (Day 1 opened May 26.) (Day 2: May 27; final day May 28.) The NDT in Canada 2026 (NDTiC 2026) conference opens May 26–28, 2026 at the Fairmont Chateau Whistler, Whistler, British Columbia, organized by CINDE (Canadian Institute for NDE). Theme: "Taking NDT to New Heights" — a comprehensive technical agenda covering emerging trends, practical case studies, and NDT technology exhibits. This is the primary annual gathering of the Canadian NDT community, drawing practitioners, researchers, and technology vendors from across North America. Coming four days after APCNDT 2026 concluded (May 11–15 Honolulu), NDTiC 2026 is effectively the second major regional NDT conference in a two-week window globally. Significance: the Canadian NDT context is particularly relevant for aerospace and oil & gas applications where Canadian regulatory frameworks (TC Canada, OEB, NRCan) govern inspection programs — content from NDTiC will reflect CGSB (Canadian General Standards Board) certification practice and CANDU reactor NDT, which has a distinct set of inspection methodologies not fully covered in ASNT-focused venues. The DÜRR NDT presence at NDTiC 2026 confirms industrial exhibitor engagement — watch for post-conference proceedings and presentations on AI-assisted phased array and digital radiography deployments in Canadian industrial contexts. APCNDT 2026 proceedings (NDT.net, expected June–July 2026) and NDTiC 2026 together will define the NDT research pipeline for H2 2026.

APCNDT 2026 Opens May 11 — "Breaking Barriers" Theme, Open-Access Proceedings (ASNT + NDT.net)¶

The 17th Asia Pacific Conference on Non-Destructive Testing (APCNDT 2026) opens May 11, 2026 at the Hawaii Convention Center, Honolulu (through May 14) under the theme "Breaking Barriers: NDT Solutions for a Changing World — Innovate, Adapt, Transform." Organized by ASNT with NDT.net as the open-access publishing partner (MOU signed prior to conference), all proceedings will be freely published post-conference — continuing the open-access tradition since 2006 and expanding access for APAC researchers. APCNDT is the largest regional NDT gathering in the Asia-Pacific, covering UT, ECT, radiographic, and AI-assisted inspection across aerospace, civil infrastructure, and O&G. Status: conference concluded May 15, 2026 (all 5 days complete, May 11–15). Proceedings being published on NDT.net in open-access format (ASNT/NDT.net MOU); expected June–July 2026. The ASNT/NDT.net open-access MOU ensures all submitted papers will be freely indexed and accessible post-conference — the primary venue for APAC industrial PAUT/ML deployment papers. AI-NDT content from Japanese, Korean, Australian, and Chinese industry labs will be the key research to monitor post-conference — this is where industrial PAUT ML deployment papers from non-English NDT R&D ecosystems will surface first.

AI Reliability Framework for Eddy Current Rail Inspection (NDT & E International, 2026)¶

Peer-reviewed paper in NDT & E International introducing a formal AI algorithm reliability evaluation framework for eddy current (ECT) inspection in safety-critical rail applications. The key regulatory contribution: it moves beyond reporting accuracy metrics (precision, recall) to addressing quantitative reliability requirements before AI systems can be deployed in high-consequence NDT. The framework specifies what statistical evidence packages need to demonstrate for regulatory acceptance — closing the gap between academic benchmark results (which suffice for publications) and industrial certification (which requires demonstrated reliability at specified defect sizes, material conditions, and false-alarm rates). Rail inspection is a particularly demanding domain: the consequences of missed defects are catastrophic (derailment), creating strong regulatory conservatism. This paper is the ECT analogue to the EASA MLEAP framework for aviation — applying structured assurance evidence requirements to a safety-critical infrastructure domain. Directly relevant to the ASNT AI/ML NDT Standard thread: this is the type of reliability methodology ASNT's standard will need to codify.

Deep Learning B-Scan Classification for Automated UT Defect Detection (IEEE, 2026)¶

IEEE conference paper benchmarking CNN-based classification of ultrasonic B-scan images for automated defect detection in industrial NDT. The work directly addresses the manual analyst bottleneck: B-scan image review is currently performed by Level II/III UT technicians, creating throughput limitations for high-volume inspection programs. The CNN architecture outperforms traditional signal processing on accuracy and speed, contributing empirical evidence for replacing manual visual review with AI in PAUT workflows. Combined with the LSTM ECT paper (already in this wiki), the trend is clear: AI classification is now superior to human review on well-defined defect categories in controlled laboratory conditions. The remaining gap for production deployment is domain shift — ensuring models trained on reference specimens remain accurate on field hardware, coupling variability, and surface conditions. This paper is part of the growing body of peer-reviewed evidence that ASNT's AI/ML standard can reference when specifying performance validation requirements.

ASNT AI/ML NDT Standard — NDT Week 2026 Accelerates Standardization Push (ASNT, January–April 2026)¶

NDT Week 2026 (January 18–23, Miami — ASNT + ASTM + AWS) established AI standardization as the single highest-priority agenda item for the NDT community in 2026. More than 150 leaders from the three organizations convened with the explicit goal of integrating AI into established NDT standards, with ASNT leadership calling for urgent acceleration: "accelerate standardizing this fast-moving technology to create a solid regulatory framework companies can lean on." The concrete deliverable: ASNT's proposed standard "Use of AI/ML for NDT/E Applications" (public comment period: May 30–July 14, 2025, now closed — under active review by ASNT Standards Council). The standard establishes minimum requirements for developing, validating, deploying, and maintaining AI/ML tools in NDT, covering data management, data preparation, model validation, and performance metrics including POD (probability of detection) and false call rate. If adopted, this would be the first formal U.S. NDT-specific AI standard — filling the gap that EASA's MLEAP framework addresses for aviation but does not cover for non-aviation sectors (O&G, civil infrastructure, power generation). Timeline: no adoption date yet published. A special issue of Materials Evaluation on digital twin technology in NDT/E is planned for July 2026. For AI-NDT vendors: this standard becoming binding would require structured evidence packages for any commercial AI-assisted NDT system sold in the U.S. — harmonizing direction with EASA MLEAP even if the specific requirements differ.

TRAINDE + AI Simulation for NDT Operator Training (EXTENDE, April 2026)¶

EXTENDE announced improvements to its TRAINDE NDT training platform (April 2, 2026), adding AI-driven data compression and realistic signal generation to its ultrasonic inspection simulation environment. TRAINDE is a software tool used to train UT operators by simulating realistic inspection scenarios — defect signals, scanning artifacts, material variability — without requiring physical test specimens. The April 2026 update introduces AI-generated signal synthesis: instead of a fixed library of recorded defect signals, the system uses generative AI to produce unlimited, material-specific, defect-specific synthetic UT signals for operator practice. Practical significance: the operator training bottleneck (labeled data quality) is now addressable via synthetic signal generation — the same gap identified in MDPI's 2025 PAUT ML review (78%→93.3% accuracy improvement bottlenecked by dataset quality). If AI-generated training signals are validated against real field data, this directly addresses the labeled data scarcity problem that ASNT's AI/ML standard also highlights. Direct implication: NDT training programs can now scale instruction hours without proportional growth in physical specimen libraries or field placements.

Hexagon to Acquire Waygate Technologies for $1.45 Billion — Strategic M&A for AI-NDT (Aerospace Testing International, 2026)¶

Hexagon AB (Swedish industrial metrology/geospatial giant) announced a $1.45B acquisition of Waygate Technologies (Baker Hughes's NDT division). The acquisition positions Hexagon as the single largest provider of industrial inspection technology across metrology (CMMs, laser trackers), NDT (PAUT, eddy current, radiographic), and digital twin infrastructure — a vertical integration play. For the AI-NDT landscape: Hexagon's existing portfolio includes Nexus (manufacturing intelligence platform), Leica (3D scanning), and production measurement — combining these with Waygate's PAUT and borescope hardware creates the most complete inspection-to-digital-twin pipeline in the industry. Timing context: this acquisition was announced alongside (or shortly after) the Waygate + GE Aerospace GEnx MDI deployment — Hexagon is acquiring Waygate at the moment its AI-guided inspection products are entering commercial airline deployment. The change of ownership from Baker Hughes (an energy/oilfield services company) to Hexagon (a measurement/software company) also signals a strategic re-emphasis: under Hexagon, Waygate's development roadmap will likely accelerate toward software-first AI inspection platforms rather than hardware-sensor-first oilfield tooling.

Waygate Technologies + GE Aerospace Deploy AI-Guided MDI Templates for GEnx Borescope Inspection (GE Aerospace, April 10, 2026)¶

Waygate Technologies (Baker Hughes) and GE Aerospace deployed automated Menu Directed Inspection (MDI) templates for GEnx-1B and GEnx-2B engine borescope inspection, specifically targeting High-Pressure Turbine (HPT) stage 1 and stage 2 blades — the highest-stress, highest-consequence components in commercial turbofan engines. The system runs on Waygate's Mentor Visual iQ+ video borescope and provides AI-driven on-screen overlays, reference imagery, and structured guided workflows that lock inspectors to AMM-aligned viewing angles, reducing operator variability and training time for new technicians. All data — images, video, 3D measurements — flows in real-time into the InspectionWorks Insight cloud platform for fleet-level traceability and trend analysis. This is a commercial deployment, not a trial: built under the JTDA (Joint Technology Development Agreement) signed in 2023, it represents the first AI-guided structured inspection workflow shipped to airlines for a specific turbofan model family. The significance for aviation NDT is that it closes the consistency gap: borescope inspection quality has historically been highly operator-dependent, and MDI templates impose a repeatable procedure that can generate auditable, AMM-traceable inspection records — directly relevant to EASA Part-145 quality assurance requirements.

XAI-Enabled Inspection Agent: 3D CNN + LLM Reporting for Volumetric NDT (Fraunhofer, Procedia CIRP 139, 2026)¶

Peer-reviewed paper presenting a novel agentic AI system for volumetric NDT. Three integrated components: (1) a memory-efficient 3D CNN for defect detection in ultrasonic volumetric data, (2) a custom 3D Grad-CAM engine producing depth-filtered XAI heatmaps visualized in ParaView (addressing the "black box" trust gap for field inspectors), and (3) an LLM-backed reporting module that performs vector-similarity retrieval against ASME BPVC and ISO 9712 standard templates to auto-generate compliance-grade inspection reports. The combination is the novel contribution: XAI explainability + agentic architecture + standards-aligned LLM reporting addresses the full compliance pipeline, not just detection accuracy. Published in Procedia CIRP 2026 — CIRP is the International Academy for Production Engineering, giving this industrial credibility beyond academic ML NDT papers. Most AI-NDT papers stop at classification accuracy; this is the first paper (to this wiki's knowledge) to close the loop from volumetric detection → explainability → auto-generated standards-compliant report as a unified agent. Directly relevant to EASA Level 1/Level 2 AI framework requirements where audit trails and standards traceability are expected in assurance evidence packages.

EASA NPA 2025-07 + 2026 Second NPA: AI Trustworthiness Framework Entering Aviation Binding Rules (EASA, 2026)¶

EASA published NPA 2025-07, its first regulatory proposal on AI trustworthiness in aviation. Consultation closed 10 February 2026. RMT.0742 is now proceeding to its second NPA, planned for 2026, which will translate the generic AI trustworthiness framework into domain-specific regulations — explicitly including aviation maintenance (Part-145/Part-CAO). This second NPA is where AI-assisted inspection and NDT tools will face concrete regulatory requirements. It implements technical requirements aligned with the EU AI Act for high-risk AI systems and defines a structured assurance pathway covering Level 1 (AI-based decision support) and Level 2 (Human-AI teaming). A second NPA in 2026 will extend this generic framework into domain-specific aviation regulations, including maintenance and inspection — directly addressing AI-assisted NDT tools that make or support airworthiness calls. The practical consequence for NDT vendors: EASA's assurance evidence package (not just performance benchmarks) will become a formal certification prerequisite for AI-assisted NDT deployed on EASA-regulated aircraft. Meanwhile, the FAA's parallel effort remains the 2024 Roadmap for AI Safety Assurance — a policy intent document, not prescriptive certification criteria. The 2026 EASA-FAA International Aviation Safety Conference (June 16–18, Chantilly, Virginia) will bring both agencies together, but the regulatory gap for ML-NDT is unlikely to close before 2027. This confirms the dual-certification asymmetry: EASA demands structured evidence packages, FAA allows performance-based compliance — forcing vendors to meet the stricter bar to serve both markets.

Voliro PEC (Pulsed Eddy Current) Drone Payload: CUI Detection Through 100 mm Insulation (Voliro, 2026)¶

Voliro is introducing a new PEC (Pulsed Eddy Current) payload for the Voliro T platform capable of detecting corrosion-under-insulation (CUI) through up to 100 mm of insulation thickness — without removal of insulation. This is a significant capability addition: CUI is one of the most economically damaging and inspection-resistant defect types in petrochemical and power-generation assets, and traditional UT or ECT requires insulation removal (costly, time-consuming, creates personnel hazard). PEC's broadband excitation drives a signal through insulation and cladding to measure remaining wall thickness below. Deploying this on a contact-capable UAV closes the access gap entirely: no scaffolding, no insulation removal, quantitative thickness data at any attitude. Combined with the Voliro T's tiltable-rotor contact stability, this is the most complete drone NDT capability for CUI-prone vertical assets (pressure vessels, piping runs, columns) currently available commercially.

ML-Enhanced PAUT for Functionally Graded Materials + LSTM ECT for Heat Exchanger Tubes (NDT&E International / Springer, 2025–2026)¶

Two papers closing the gap between lab AI-NDT and industrial field deployment. First: a 2025 NDT & E International paper demonstrates ML-enhanced PAUT on functionally graded materials (Ti6Al4V-ZrO2 — the interface-rich composites used in turbine blades, biomedical implants, and aerospace panels) achieving 92.5% overall defect detection accuracy across five defect types, outperforming traditional PAUT by 12.7 percentage points. The gain comes from neural networks distinguishing defect signatures in graded-material interfaces where impedance varies continuously — making conventional amplitude thresholding unreliable. Second: a companion 2026 Springer paper (Journal of Nondestructive Evaluation) reports an LSTM-based automated flaw classifier for eddy current testing (ECT) data from heat exchanger tubes, reducing detectable defect size to 0.3 mm and achieving 99% classification accuracy on real field data. Together these papers represent state-of-art on two distinct AI-NDT challenges: heterogeneous material inspection (PAUT/FGM) and high-rate inline tube inspection (ECT/LSTM). Both use real industrial datasets, not simulated, which is critical for regulatory acceptance arguments.

FAA Part 108: BVLOS Final Rule — Enabling Large-Scale UAV NDT Programs (UAVHQ, 2026)¶

FAA Part 108, with final rule publication targeted March 16, 2026, is the single most consequential regulatory development for UAV-based NDT inspection at scale. It replaces the current Part 107 waiver system (which required 20+ individual site approvals for a nationwide pipeline or powerline inspection program) with an "operational area approval" model: one approval covers all routine flights within a defined zone. Drones up to 1,320 lb with 25-ft wingspan are permitted — large enough to carry PAUT, EMAT, or EMAT sensor payloads required for industrial NDT. Mandatory ADS-B signal reception and integration with Automated Data Service Providers (ADSPs) solves airspace deconfliction. Infrastructure inspection — pipelines, powerlines, bridges, railways — is explicitly recognized as an ideal use case due to predictable linear flight paths and predominantly rural operating environments. 6–12 month compliance window follows publication. Practical consequence for NDT vendors: the business case for Voliro T and Skygauge-class contact UAV NDT platforms (already in this wiki) just became structurally more viable in the U.S. market — the operational overhead of per-site approval was the primary commercial barrier.

European Data Act: A Turning Point for NDT Data Interoperability (Engineer Live, 2026)¶

The European Data Act (in force from September 2025) applies directly to industrial NDT systems by mandating data portability and interoperability for connected industrial devices. For NDT operators and vendors, this has immediate practical consequence: inspection data produced by connected NDT devices (PAUT scanners, eddy current instruments, drone inspection platforms) must be available for export in interoperable formats upon user request, breaking proprietary data lock-in. The convergence with AI-NDT is structural: digital twins, AI analytics, and predictive maintenance workflows all depend on data flows that are currently fragmented across proprietary vendor ecosystems. The EU Data Act forces a partial unbundling: hardware vendors can no longer use data format as a competitive moat. This directly accelerates AI-NDT adoption by lowering the barrier to training cross-vendor ML models on pooled inspection datasets — previously nearly impossible due to proprietary format divergence. Practical implication for NDT labs: audit your inspection data contracts now for portability clauses, particularly if operating within EU jurisdiction.

Digital NDT 4.0: AI-Driven Inspections Feeding Real-Time Digital Twins (OnestopNDT / Industry 4.0, 2026)¶

Synthesis of the 2026 state-of-market for industrial digital NDT: inspections now flow from AI-powered scanners through IoT connectivity into asset digital twins, creating live structural health models that update in near-real-time. The canonical industrial example documented: an oil and gas pipeline manufacturing plant combined PAUT (phased array ultrasonic testing) and digital radiography with AI-driven analysis — resulting in 35% reduction in rework and 45% shorter production and inspection cycles. The enabling architectural shift: replacing batch offline inspection (periodic manual scans) with inline continuous inspection where the NDT instrument is part of the manufacturing process loop, not a separate QC step. The digital twin layer closes the loop: defect data auto-annotates the asset model, triggering maintenance work orders without manual data entry. This is the production-deployment picture for where PAUT + AI ML (CATT-S, MDPI studies) is landing in mature industry segments — not just research but live operational workflows in O&G.

FAA AI Roadmap vs. EASA MLEAP: The ML-NDT Regulatory Gap (FAA, 2026)¶

The FAA has published a "Roadmap for Artificial Intelligence Safety Assurance" but has not issued a dedicated ML-NDT certification framework — its NDT rules remain governed by personnel-focused advisory circulars (AC 43-16A, AC 65-31B). EASA's MLEAP (Machine Learning Application Approval) final report is significantly more advanced, with Level 3 AI guidance targeting end-2025. This confirms and deepens the EASA-leads/FAA-lags dynamic noted previously: the FAA roadmap is a policy intent document, not actionable certification criteria for AI-assisted NDT systems. Practical consequence for vendors: ML-NDT systems targeting dual EASA/FAA certification face an asymmetric compliance burden — EASA requires structured assurance evidence packages (MLEAP Level 1/2/3), while the FAA's performance-based approach gives more flexibility but less regulatory clarity. The gap is unlikely to close before 2027 given FAA rulemaking timelines.

Voliro T and Skygauge: Contact-Based UAV NDT Platforms Going Industrial (2026)¶

Two UAV-NDT platforms are gaining industrial traction in 2026. The Voliro T uses a tiltable-rotor design enabling stable contact with structures at arbitrary orientations, carrying EMAT, UT, and dry-film-thickness (DFT) sensors for wall-thickness and corrosion detection. Skygauge similarly focuses on contact-based ultrasonic inspection from a drone body. Both differ fundamentally from visual-only inspection drones by delivering quantitative thickness data directly replacing rope-access inspectors in oil/gas and power-generation asset inspection. Sector projected at 11% CAGR through 2029. The key engineering challenge for contact-based UAV NDT is attitude control during contact — maintaining stable couplant pressure against a curved or inclined surface while the airframe compensates for reaction forces. Voliro's tiltable-rotor approach handles this more elegantly than fixed-rotor designs with mechanical compensation arms.

IAEA AI-Augmented NDT for Disaster Management: International Standardization Pressure (IAEA, 2026)¶

The IAEA launched a coordinated research project to integrate AI-augmented NDT into disaster management and civil infrastructure assessment workflows, targeting real-time defect detection on civil structures post-disaster. Directly parallels a January 2026 ScienceDirect paper on AI-enhanced NDT for civil infrastructure (https://www.sciencedirect.com/science/article/pii/S0926580525000366). The IAEA involvement is significant beyond the technical scope: it signals normative pressure toward standardized AI-NDT protocols at the international level, distinct from the aviation-specific EASA/FAA frameworks. If the IAEA project produces reference guidance, it could accelerate AI-NDT adoption in nuclear power plant inspection — a domain with extreme regulatory conservatism — by providing a multilateral endorsement framework that individual national regulators can point to.

EASA's First Regulatory Proposal on AI for Aviation — Consultation Open (EASA, 2026)¶

EASA has moved from guidance documents into binding rulemaking, with a second NPA (Notice of Proposed Amendment) targeting 2026 extending the AI trustworthiness framework into domain-specific rules for maintenance, flight operations, and ATM. This NPA explicitly addresses ML-based decision support (Level 1 AI) and human-AI teaming (Level 2 AI) — directly bracketing AI-assisted NDT tools where the algorithm recommends an airworthiness call and a human confirms it. The FAA's parallel AI Safety Assurance Roadmap (July 2024) remains performance-based and non-prescriptive. This regulatory asymmetry is the central certification risk: ML-NDT developers targeting both markets must satisfy EASA's structured assurance evidence requirements and the FAA's more open-ended compliance simultaneously. Practical implication: any vendor building AI-assisted NDT for aviation must budget for EASA's formal assurance evidence package, not just performance benchmarks.

Human-Machine Collaborative Automation for PAUT Data Analysis of CFRP (ScienceDirect, 2025)¶

Evaluates a three-tier automation architecture on complex-geometry CFRP components (aircraft fuselage panels, wind turbine blades) using a phased-array roller probe on an industrial manipulator. Layer 1: supervised object detection on amplitude C-scans. Layer 2: unsupervised anomaly detection on B-scans. Layer 3: self-supervised full-volumetric model. Multi-model ensemble improved F1 by up to 17.2% over any single model and reduced analysis time from hours (manual) to under 100 seconds per sample. Critically: the paper explicitly stops short of full autonomy and keeps a human in supervisory role — a design choice that maps directly onto EASA's Level 1/Level 2 AI framework and makes regulatory approval more tractable. This is the paper to cite when designing AI-NDT systems for aerospace certification.

Phased-Array Thermography (PAT): Beam-Steering Applied to Infrared NDT (Scientific Reports, Nov 2025)¶

Southampton aeronautics researchers introduce PAT — applying phased-array beam-steering logic to infrared thermography rather than ultrasound. An array of independently controlled embedded heating elements steers and focuses thermal wavefronts into the structure. This overcomes a core limitation of conventional pulsed thermography: the inability to detect defects perpendicular to the heated surface (inclined cracks, certain delaminations). Validated on aerospace specimens — aluminium plates with flat-bottom holes and CFRP composite panels with impact damage. PAT found flaws that standard pulsed thermography missed entirely. A 3D post-processing step further resolved small, deep defects invisible in 2D analysis. Particularly promising for in-situ structural health monitoring (SHM) of composite airframes because heating elements can be embedded into the structure.

Eddy Current In-Line Inspection for Oil & Gas Pipelines: Systematic Review (MDPI Processes, Jan 2026)¶

Open-access systematic review comparing five ECT methods deployed on pipeline inspection gauges (PIGs) running inside live O&G lines: conventional ECT, multi-frequency ECT, remote-field ECT (RFECT), pulsed ECT (PECT), and array ECT (AECT). Key differentiators: RFECT excels at through-wall corrosion in ferromagnetic pipes; PECT (broadband excitation) gives superior depth-profiling of wall-thickness loss; AECT provides high-speed, high-resolution mapping suitable for corrosion-under-insulation (CUI). The review maps each technique to defect type (corrosion pits, stress corrosion cracking, weld anomalies) and identifies AI-augmented signal interpretation + miniaturized sensor arrays as the next frontier for high-pressure gas transmission infrastructure.

CATT-S: CNN + Transformer for PAUT Defect Detection at 99.4% Accuracy, Real-Time CPU (Sensors MDPI, Oct 2025)¶

Changwon National University benchmarks deep learning architectures on PAUT data from welded and composite specimens. Their novel CATT-S model (Convolutional Attention Temporal Transformer for Sequences) combines CNN layers for local signal morphology with Transformer self-attention to capture cross-beam spatial dependencies — the relational information between adjacent scan lines that pure CNNs miss. Result: 99.4% accuracy, F1 = 0.905 on real experimental data vs. 94.9% for standard CNNs. Critically: the exported ONNX model runs real-time on CPU, making deployment on portable field hardware practical without GPU. Directly applicable to aerospace weld inspection and pipeline girth-weld screening where throughput and operator-independence are both requirements.

Advances of Machine Learning in Phased Array Ultrasonic NDT (MDPI, 2025)¶

Comprehensive review of ML applied to phased array ultrasonic testing (PAUT). Key finding: ML integration improved average detection precision from 78% to 93.3% and recall from 66% to 91.5% across benchmark datasets. The paper surveys CNN-based defect classifiers, U-Net segmentation for B-scan images, and transformer-based anomaly detection. Takeaway: the bottleneck is now labeled data quality, not model architecture — industrial NDT datasets are small and proprietary.

Advanced NDT Integration in Steel Processing (MDM Metals, March 2026)¶

Practical engineering overview of multimodal NDT in steel manufacturing. Combines radiographic testing (RT), eddy current (EC) for surface cracks, and infrared thermography for subsurface delamination. Real-time AI analytics fused across modalities enables comprehensive coverage without the blind spots of any single technique. Highlights the trend toward inline NDT (in the production line) vs. offline batch inspection.

Automated Defect Detection in Multimodal Phased Array Plane Wave Imaging (NDT.net)¶

Research on automating defect detection across multiple PAUT imaging modalities (plane wave, full matrix capture). ML classifiers trained on multimodal representations outperform single-modality models. Particularly relevant for aerospace structural health monitoring where false negatives have catastrophic consequences and false positives drive costly rework.

Hybrid TOFFD + CNN for Aerospace Composite PAUT: 94.7% Detection vs. 78.3% Traditional (CAJMNS, 2026)¶

Published in the Canadian Journal of Modern and Natural Sciences (2026), this paper proposes a hybrid inspection pipeline for carbon-fiber reinforced polymer (CFRP) aerospace composite panels — the dominant structural material in modern airframes — combining Total Focusing Method Full Data (TOFFD) acquisition with a CNN defect classifier. The key innovation: TOFFD imaging reconstructs high-fidelity volumetric data from full-matrix capture acquisitions (every transmitter-receiver pair), giving the CNN significantly richer spatial information than standard phased-array sectorial scanning. Detection accuracy: 94.7% on aerospace-grade CFRP specimens vs. 78.3% for conventional PAUT — a 16.4pp gain on the hardest case (complex-geometry composites with mixed delamination, fiber waviness, and impact damage). Why the improvement: CFRP composites have anisotropic acoustic properties — wave speed and attenuation vary with fiber orientation and layup sequence. TOFFD's post-processing algorithm accounts for anisotropy in the reconstruction, where standard PAUT assumes isotropic media and produces artifacts at ply interfaces. The CNN then classifies in a geometrically clean image space rather than fighting physics artifacts. Directly extends the CATT-S (99.4% on metals, already in this wiki) and ML-Enhanced PAUT on FGMs (92.5%, already in this wiki) to the architecturally critical case of aerospace-grade composites, where the inspection problem is significantly harder than metals or lab specimens.

AIMM Center Opens: 4,000 sq ft Advanced NDT Facility (April 2026)¶

The AIMM Center (Advanced Inspection Methods & Materials), a joint venture between Composite Inspection Consulting/NDT and Omni NDE, opened its 4,000-square-foot facility in April 2026. The facility offers robotic computed tomography (CT), digital X-ray, laser ultrasonic testing, and laser shearography — representing a notable expansion of non-contact, stand-off NDT capabilities for composite and aerospace components. Laser ultrasonics in particular addresses a persistent gap: conventional UT requires couplant and direct probe contact, making it difficult for complex geometries and high-temperature surfaces. Laser UT excites and receives ultrasonic waves optically, enabling inspection of surfaces inaccessible to contact probes. Shearography (interferometric surface displacement mapping under stress) is particularly effective for delamination detection in honeycomb sandwich structures and CFRP panels. The AIMM Center's robotic integration suggests a push toward automated inspection workflows — reducing operator-dependent variability. For the AI-NDT trend: robotic platforms with laser UT/shearography generate high-volume structured datasets well-suited to ML classification pipelines, potentially closing the labeled-data gap that constrains current AI-NDT training.

IAEA Launches 5-Year Coordinated Research Project: AI-Augmented NDT for Disaster Management (2026)¶

The International Atomic Energy Agency has launched a new five-year Coordinated Research Project (CRP) to integrate AI with advanced NDT for disaster management applications. The CRP will explore how AI can be applied to NDT techniques to deliver real-time, data-driven insights — enabling faster damage detection in infrastructure after disasters and accelerating emergency interventions. Key focus: AI/ML translation of measurement signals from multiple NDT modalities for rapid structural integrity assessment. The IAEA framing is significant: it moves AI-NDT from a precision/efficiency story into a safety and resilience story — where real-time automated damage assessment directly affects emergency response timelines. This is a new application domain for AI-NDT beyond the industrial inspection context (aerospace, O&G, power generation) that dominates current literature. The 5-year timeline and multi-country CRP structure suggests standardization potential: CRP outputs often feed into IAEA safety guides and member-state regulatory frameworks.

Hexagon Acquires Waygate Technologies from Baker Hughes — $1.45B Deal (April 13, 2026)¶

Hexagon AB announced a definitive agreement to acquire Waygate Technologies from Baker Hughes for ~$1.45 billion in cash (April 13, 2026). Waygate — headquartered in Germany, 1,500 employees, 25 locations, ~$630M annual revenue, 130+ years of NDT heritage — makes industrial radiographic, CT, ultrasonic, and eddy current inspection systems used across aerospace, automotive, oil & gas, and power generation. The deal closes H2 2026 pending regulatory approval. Strategic rationale: Hexagon (precision measurement software, manufacturing intelligence, Leica Geosystems) extends its digital twin and asset integrity platform into the core NDT instrumentation market. For the NDT industry: this is the most significant ownership change since Baker Hughes acquired Waygate's predecessor (GE Inspection Technologies) in 2017. Hexagon's Nexus platform (cloud-based, connects measurement and design workflows) is the likely destination for Waygate's AI-NDT capabilities — potentially creating a vertically integrated NDT software + hardware + cloud analytics stack. The open question raised in the wiki persists: will Hexagon integrate Waygate into Nexus (favoring software-first customers) or maintain Waygate as an independent industrial brand (favoring field-service customers)? No integration roadmap has been published; closing expected H2 2026.

The Korea Research Institute of Standards and Science (KRISS) developed a waveguide-based ultrasonic sensor that overcomes two fundamental limitations of conventional contact UT: the need for direct probe attachment to the inspection surface, and the inability to inspect in all directions from a single sensor position. The KRISS design applies a waveguide element that remotely excites and receives ultrasonic energy — the transducer does not need to touch the part under inspection and can detect defects in 360° without repositioning. Industrial significance: (1) enables inspection in extreme environments (high temperature, hazardous chemicals, confined spaces) where conventional couplant-based contact probes cannot survive or reach; (2) eliminates inspection blind spots that arise from probe repositioning dead zones in current PAUT setups. Practical consequence for industrial inspection programs: components that were previously either uninspectable or required time-consuming robotic repositioning can now be continuously monitored via a stationary waveguide fixture. This complements the TRAINDE AI signal synthesis (also in this wiki) at the opposite end of the pipeline: TRAINDE improves training data quality; the KRISS sensor improves raw signal accessibility for field measurements. Potential integration path: waveguide-acquired remote signals + CNN B-scan classification (IEEE 2026, also in this wiki) for hands-free automated detection in hazardous environments.

Core Concepts¶

NDT Method Taxonomy¶

Method	Principle	Best For	Limitation
Ultrasonic Testing (UT)	Sound wave reflection/attenuation	Subsurface cracks, thickness	Couplant needed, rough surfaces
Phased Array UT (PAUT)	Electronically steered beam array	Complex geometries, rapid scanning	Cost, data volume
Eddy Current (ECT)	Electromagnetic induction	Surface/near-surface cracks in conductors	Depth limited, lift-off sensitivity
Radiographic (RT/DR)	X-ray / gamma-ray transmission	Porosity, inclusions, welds	Radiation safety, 2D only
Magnetic Particle (MT)	Ferromagnetic field disruption	Surface defects in ferrous materials	Ferrous only
Thermography (IRT)	Thermal anomaly from defects	Delaminations, moisture ingress	Emissivity issues
Visual (VT)	Direct or camera-aided inspection	Surface defects	Surface only

Key Terminology¶

PAUT (Phased Array UT): Array of piezoelectric elements fired with controlled delays to steer and focus the beam electronically. Produces B-scan (cross-section) and C-scan (top-view) images.
FMC/TFM (Full Matrix Capture / Total Focusing Method): All elements fire and receive independently; TFM reconstructs high-resolution images in post-processing. Computationally heavy but gold standard.
POD (Probability of Detection): Statistical measure of an NDT system's ability to detect defects of a given size. Critical for certification (e.g., ASME, EN standards).
TOFD (Time of Flight Diffraction): Uses diffracted waves at crack tips for precise depth sizing.

AI in NDT — Current State (2026)¶

Detection: CNNs and transformers applied to B-scan / C-scan images approach human expert accuracy on known defect types.
Characterization: Harder — requires regression for defect size/shape, not just classification.
Challenges: Domain shift (different materials, geometries, equipment), lack of standardized datasets, regulatory acceptance.
Trend: Physics-informed neural networks (PINNs) incorporating wave propagation equations to reduce data requirements.

Standards & Certification Bodies¶

ASNT (American Society for Nondestructive Testing) — NDT Level I/II/III certification
ISO 9712 — International NDT personnel qualification
ASME Section V — NDT procedures for pressure vessels and piping
EN 4179 / NAS 410 — Aerospace NDT certification

Open Questions¶

What is the minimum labeled dataset size needed to train a reliable PAUT defect classifier for a new material/geometry combination?
How do PINNs compare to purely data-driven models when labeled data is scarce?
EASA's Level 1/2 AI framework is now the de-facto structure for aviation ML-NDT certification — what equivalent is emerging for ASME (pressure vessels) and offshore/O&G regulation?
How does multimodal sensor fusion handle conflicting indications across NDT modalities?
Can PAT (Phased-Array Thermography) be miniaturized and embedded into structural panels for continuous in-flight SHM without adding prohibitive weight?
CATT-S runs real-time on CPU — what's the performance floor? Can it maintain accuracy at full PAUT scan speeds (>100 mm/s mechanical scanning)?
RFECT vs MFL for ferromagnetic pipelines: under what wall-thickness and defect-geometry conditions does RFECT outperform magnetic flux leakage?
EASA NPA 2025-07 consultation closed Feb 10, 2026. Second NPA (domain-specific, maintenance/NDT) planned for 2026 — NPA 2025-12 (continuing airworthiness regular update) comment period closed March 31, 2026; no NDT-specific 2026 NPA published yet.
Hexagon acquires Waygate Technologies ($1.45B): how does change of ownership from Baker Hughes (energy/oilfield) to Hexagon (measurement/software) affect Waygate's AI-NDT roadmap? Will Hexagon integrate Waygate into its Nexus platform?
Voliro PEC payload: what is the quantified accuracy of through-insulation CUI detection vs. conventional removal-and-inspect? Any published POD curves?
FAA-EASA Joint Safety Conference (June 2026): will it produce any joint statement on AI-NDT certification harmonization?