Opsio - Cloud and AI Solutions
Twisted airfoil blade with a matt grey ceramic-style coating clamped between two steel centres on a stainless inspection bench in a quality lab

AI Visual Inspection for Aerospace & Energy

Turbine blade cracks and coating spallation, solar cell microcracks and wind blade leading-edge erosion, found in line images, EL images, drone surveys and borescope inspection video.

Book a walkthroughSee use cases

What is AI visual inspection for aerospace & energy?

AI visual inspection for aerospace and energy is camera and deep-learning analysis that Opsio sets up to find blade cracks and coating spallation, solar cell microcracks, and wind blade leading-edge erosion. It reads line images, EL images, drone surveys and borescope inspection video, ranks indications for qualified inspectors, and does not replace mandated NDT.

Key facts

  • Covers turbine blades, composite parts, solar cells and modules, wind turbine blades and overhead power lines.
  • Imaging used: raking-light and dome RGB, borescope video, electroluminescence (EL), active thermography and drone RGB plus thermal.
  • Line stations run edge inference in under 50 ms on NVIDIA Jetson or Intel OpenVINO hardware.
  • Drone and borescope imagery is processed offline in batches, not in real time.
  • Results reach PLCs over PROFINET or OPC UA, and MES, QMS or asset management over REST.
  • Supports AS9100, Nadcap and IEC 61215/61730 records; NDT disposition stays with EN 4179/NAS 410 personnel.
  • Image data stays on the edge, on-premises or in an EU cloud region you choose.

Main defect families

Blade cracks and coating spallation
Surface cracks and loss of the ceramic thermal barrier coating on hot-section turbine blades and vanes.
Solar cell microcracks
Fine fractures in silicon cells, visible as dark lines in electroluminescence images, that isolate active cell area.
Leading-edge erosion
Pitting and loss of coating and laminate on the front edge of wind turbine blades from rain and particle impact.

Last reviewed by Praveena Shenoy, Head of AI & Data, Opsio, a Swedish cloud and AI services company. Part of Opsio's AI Visual Inspection service.

Why is turbine blade inspection, EL testing and borescope inspection hard to scale?

The parts are expensive, the indications are small and the inspection is safety-critical. A crack on a high-pressure turbine blade can be shorter than a millimetre and sit on a curved, coated airfoil next to dozens of cooling holes. An escape can mean an in-flight shutdown or uncontained failure; a false reject scraps a part worth thousands of euros.

Human inspection breaks down on volume, not skill. One engine's borescope inspection produces hours of video reviewed frame by frame. A module line runs an EL image every 20 to 40 seconds, typically, and a utility drone campaign returns tens of thousands of images per season.

Material behaviour makes it harder. Ceramic coatings scatter light, carbon-fibre laminates hide delamination below a perfect surface, and anti-reflective solar glass mirrors the camera. Wind blades are inspected in wind, rain and changing daylight, so the same erosion looks different on every flight.

Regulation sets the limit on automation. Aerospace NDT is performed and signed off by personnel qualified to EN 4179 or NAS 410, so visual AI can rank, measure and document indications, but the disposition stays with your Level 2 or Level 3 inspector.

Turbine blade, solar panel and wind blade inspection use cases

Each use case is a worked scenario: the line or campaign, the defects, the camera and lighting setup, and what changes for your inspectors.

Turbine blade inspection and composite parts

Close-up of a coated airfoil edge with a row of round cooling holes, a small chip of missing coating at the edge and a fine crack running from it across the surface
Coated blade edge with a coating chip and a fine surface crack beside the cooling holes

Situation

You inspect new or repaired hot-section blades in an MRO shop or at the OEM coating line, and cured composite skins in a composites cell. Today an inspector rotates each blade under a lamp and magnifier before FPI (fluorescent penetrant inspection), and composite parts get a manual look before ultrasonic C-scan.

On the wing, borescope video is reviewed frame by frame on a laptop. See Opsio's approach to turbine blade visual inspection.

Defects

  • Surface cracksThermal-mechanical fatigue opens fine cracks at edges, tip and cooling-hole rims over start-stop cycles.
  • Coating chipsthermal barrier coating (TBC) chipsHandling or particle impact knocks out the ceramic top coat and exposes the bond coat.
  • Spallationflaking of the ceramic coating over an areaOxide growth at the bond-coat interface and thermal cycling make the coating lift in patches.
  • Coating thickness defectsSpray-path or masking errors leave thin or built-up coating, seen as colour and texture variation; eddy current or sectioning confirms thickness.
  • Cooling-hole blockageOverspray, debris or oxidation closes film-cooling holes and starves the airfoil of cooling air.
  • Foreign-object damageFODIngested debris dents, nicks or tears blade edges, mostly in fan and compressor stages.
  • Composite delaminationImpact, poor consolidation or contamination separates plies, usually below an intact surface.
  • PorosityLow cure pressure or poor debulking traps voids in the laminate.
  • Fibre wrinklingFibres buckle out of plane over tight radii during layup, reducing compression strength.
  • Resin-rich or dry areasUneven resin flow during cure leaves glossy resin pools or dull, fibre-starved patches.

Inspection setup

Blades sit on a rotary fixture or are presented by a robot to a 12 to 25 MP monochrome GigE Vision camera with a macro or telecentric lens, giving typically 10 to 20 µm per pixel. Low-angle LED bars rake across the airfoil for cracks and chips, a diffuse dome evens out coating texture, and a backlight behind the root shows open and blocked cooling holes.

Composite parts pass under a gantry with raking light for wrinkles and resin-rich areas. Flash thermography frames and C-scan maps from your NDT stations are classified by the same model, and borescope video is analysed after the inspection. Every result is a flag with location and class; nothing is auto-accepted.

Outcome

Your inspectors start each blade or panel with indications already located, measured and ranked. Every image is logged against the part serial, giving the Level 2 or Level 3 inspector a complete record to disposition. FPI, thermography, ultrasonic testing and CT stay in place as the mandated methods, and the feasibility test shows which defect classes the camera sees reliably.

Solar panel inspection: cells and modules

Electroluminescence image of a solar module in a dark test enclosure: cells glow grey, dark microcrack lines run through two cells and short dark dashes mark interrupted finger lines
EL image of a solar module showing cell microcracks and interrupted fingers

Situation

You build crystalline silicon modules on a line that strings, lays up, laminates and frames cells, with an IV flash tester at the end. EL testers sit after stringing and after lamination, and an operator reviews each grey EL image on screen. At a takt of typically 20 to 40 seconds per module, review fatigue and inconsistent grading between shifts are the weak point.

Defects

  • MicrocracksThin fractures from wafer handling, stringer clamping or thermal stress isolate parts of a cell and show as dark lines in EL.
  • Finger interruptionsBreaks in the printed silver fingers from screen-print defects or paste issues leave dark stripes between busbars.
  • Broken busbarsA cracked or poorly soldered busbar cuts current collection across a whole cell section.
  • Dark areas and shuntsInactive regions and local short circuits from material defects or contamination appear as dark patches or bright spots in EL.
  • Soldering defectsCold joints, misaligned ribbons or excess heat at the stringer create high-resistance contacts and cracks along the ribbon.
  • Cell cracks after laminationPressure and heat in the laminator turn existing microcracks into wider cracks or fresh fractures.

Inspection setup

Electroluminescence (EL) imaging runs in a light-tight enclosure: the string or module is forward-biased, typically at 0.1 to 1 times short-circuit current, and silicon emits near-infrared light around 1,150 nm. A cooled silicon or InGaAs camera, often one per module quadrant for 20 MP and above in total, images the glow; the same dark-enclosure, flat-field-corrected method appears on our display panel inspection page.

The machine vision cameras connect over GigE Vision. Stations sit after the stringer, where strings can be reworked, and after lamination. Photoluminescence covers cells before stringing, and an RGB camera checks cosmetic defects. The model classifies each cell, grades the module and signals reject, rework or pass to the line PLC.

Outcome

Every module gets the same grading rule on every shift, at line takt. Cracked strings divert to rework before lamination, when repair is still cheap, and post-lamination grades feed your binning and warranty records. Cell-level defect maps tie to the module serial and flash-test result, so you can trace patterns back to a stringer head or a wafer batch.

Wind turbine blade inspection and power lines

Close drone-style view of a white wind turbine blade leading edge against an overcast sky, with streaked pitting erosion along the edge and a fine hairline crack
Wind turbine blade leading edge with pitting erosion and a hairline crack

Situation

You operate wind farms or a transmission and distribution grid and inspect assets by drone on a yearly or seasonal cycle. Each wind turbine generator gives typically 300 to 1,000 close-range images, and a line patrol returns thousands per route. Engineers review them by hand after the flights, which delays repair planning by weeks.

Defects

  • Leading-edge erosionRain, hail and sand impact at tip speeds near 300 km/h pit and strip the leading edge, reducing output.
  • CracksFatigue loading opens transverse and longitudinal cracks in the shell, at bond lines and near the root.
  • Lightning damageStrikes that miss or overload the receptor leave burn marks, split tips and delaminated laminate.
  • Coating lossUV, abrasion and poor adhesion remove the gelcoat or topcoat and expose the laminate to moisture.
  • Corrosion on towers and fittingsCoating breakdown on steel towers, bolts, clamps and fittings lets rust form and spread.
  • Insulator damageBroken sheds, cracked glass or porcelain, flashover burns and pollution build-up weaken insulation.
  • Conductor strand breaksVibration fatigue at clamps and spacers, or lightning, breaks outer strands of the conductor.
  • Vegetation encroachmentTrees grow towards conductors and reduce clearance, a common cause of outages and fires.

Inspection setup

Drones carry a 45 to 61 MP RGB camera with zoom and a radiometric thermal camera, typically 640 × 512 pixels, and capture geo-tagged images on repeatable flight paths. Images upload after the flight and are processed in batches in your chosen EU cloud region or on-premises, not in real time.

The model detects and segments each finding, measures its size, assigns a severity grade and compares it with the previous campaign. The same change-detection method supports drone surveys of bridges and tunnels. Thermal hotspots on joints and clamps are flagged alongside visual damage, and findings go to your asset management system; see asset monitoring with computer vision.

Outcome

Your engineers review a ranked findings list instead of every image. Each finding carries location, blade zone or span, severity and the matching image from the last campaign, so growth is visible. Repairs and rope-access teams are planned on evidence, and the engineer still approves every grade before it enters the work order.

How AI inspection works on a blade, module or drone workflow

The same six steps apply on a coating line, a module line or after a drone campaign; only the trigger and the place inference runs change.

Dark electroluminescence test enclosure: a solar module glows grey-white on a black table under a downward-facing industrial camera on an aluminium gantry
EL inspection enclosure with a forward-biased solar module under a camera
  1. 1

    Trigger and part presentation

    A part-present sensor, robot handshake or stringer signal over PROFINET or EtherNet/IP triggers capture. Blades sit on a rotary fixture; modules index into the EL enclosure. For drone and borescope work, a finished upload is the trigger.

  2. 2

    Image acquisition

    GigE Vision cameras with macro, telecentric or fixed lenses image the part under raking LED bars, a dome or in a dark EL enclosure. Exposure, bias current and lighting are logged with each image.

  3. 3

    Edge inference and latency budget

    Line stations run inference in under 50 ms on NVIDIA Jetson or Intel OpenVINO hardware, well inside module takt. Borescope and drone imagery runs as batch jobs measured in minutes per engine or turbine.

  4. 4

    Decision and actuation

    On the module line the result goes to the PLC over PROFINET or OPC UA to pass, divert to rework or reject. For blades and composites the decision is a flag on the inspector's review screen; acceptance stays with the qualified inspector.

  5. 5

    Traceability and data retention

    Images, results and inspector decisions are stored against the part, engine, module or asset ID. Retention follows your quality system, often the life of the part in aerospace.

  6. 6

    Active learning loop

    Uncertain results and inspector overrides queue for review and feed retraining on SageMaker. New blade designs, cell formats or turbine models are added as variants with their own validation before release to the line.

Which imaging finds which defect?

Each defect family has a standard imaging method; for sub-surface defects the AI works on images from your NDT equipment.

Defect familyBest-suited imagingWhy
Blade cracks, chips, FOD (turbine blade inspection)Bright-field and low-angle raking light, macro lens; borescope videoRaking light casts a shadow in fine cracks and chipped edges on curved coated airfoils.
TBC spallation and coating thickness variationDiffuse dome RGB plus thermographyColour and texture change where coating is thin or missing; thermography shows debonded areas.
Cooling-hole blockageBacklight through the blade passagesOpen holes glow, blocked holes stay dark, countable per row.
Composite delamination and porosity (delamination detection)Active (flash) thermography; ultrasonic C-scan imagesSub-surface voids change heat flow and sound transmission; a camera on the surface cannot see them.
Fibre wrinkling, resin-rich and dry areasLow-angle raking light, polarised RGBOut-of-plane fibres and resin pools change surface relief and gloss.
Crack indications after FPI inspectionUV-A illumination, monochrome cameraPenetrant fluoresces in surface-breaking cracks; AI ranks indications for the qualified inspector.
Solar cell microcracks, finger interruptions, busbar breaks (electroluminescence inspection)Electroluminescence (EL) with cooled Si or InGaAs cameraCracked or disconnected areas emit less near-infrared light under forward bias.
Shunts, dark areas, soldering defectsEL plus photoluminescence (PL); IR thermographyShunts and high-resistance joints show dark in EL or hot in thermography.
Leading edge erosion, cracks, lightning damageDrone RGB at close range, high resolutionPitting, burn marks and coating loss are visible from 5 to 10 m with zoom.
Insulator damage, strand breaks, hot joints (power line inspection)Drone RGB plus radiometric thermalBroken sheds and strands are visible in RGB; failing joints run hot in thermal.

Integration with NDT records, MES and asset management

The system writes into the records you already keep; qualified people make every acceptance decision.

Reads from
Part, engine or module serials from MES; blade and cell variants; drone flight logs with GPS data; borescope video; existing thermography and C-scan images.
Writes to
Indications with image, location, class and size to your QMS or NDT report; module grades to MES; SPC data per station; graded findings to asset management (EAM/CMMS) work orders.
Where data lives
Line inference on the edge; image archives on-premises or in an EU cloud region you choose. Export-controlled engine data can stay on-premises.
Aerospace quality systems
Supports your AS9100 records and Nadcap-audited special processes with traceable images, model version and inspector decisions per part. It does not replace FPI, thermography, ultrasonic testing or CT.
NDT personnel
Designed to work as a tool under inspectors qualified to EN 4179 or NAS 410. The Level 2 or Level 3 inspector dispositions every indication and signs the record.
PV module standards
EL records per module support your IEC 61215 and IEC 61730 qualification and production control; your quality team sets pass criteria.
Wind and grid operators
Severity grades map to your blade damage categories and line defect codes, with chain of custody for every image.

From first test to your line in five steps

Five steps take you from sample parts or an image archive to a running station or batch service.

  1. 1

    Opsio runs tests in the AI Lab

    You ship typical sample sets: 10 to 30 blades or coupons with known cracks, chips, spallation and blocked holes plus 10 good ones; 30 to 50 strings or modules with microcracks, finger interruptions and soldering defects; or EL, borescope or drone archives of 2,000 to 10,000 labelled images. Opsio images them, trains a first model and measures detection and false-reject rate per class.

    You provide
    Sample parts or image archives, defect catalogue, acceptance criteria
    Opsio delivers
    Imaging trials, first model, per-class detection and false-reject results
  2. 2

    Present results and pricing

    You see per-defect results with example images, including which classes the camera finds reliably and which stay with NDT alone. Opsio recommends the camera, lens, lighting or enclosure, confirms fit within your module takt or batch turnaround, and gives a fixed quote for the production system and the managed service.

    You provide
    Quality, NDT and engineering reviewers
    Opsio delivers
    Results report, bill of materials, cycle-time fit, fixed quote
  3. 3

    Decision and PO/MSA signing

    You decide whether to proceed. The purchase order and master services agreement set the scope, acceptance criteria per defect class, data ownership including export-controlled data, where images are stored, and support levels. Responsibilities between your Level 3 and Opsio are written down, so the NDT disposition stays clearly with your qualified personnel.

    You provide
    Purchase order, signed MSA, data-handling requirements
    Opsio delivers
    MSA, statement of work, acceptance test plan
  4. 4

    Installation in the Opsio AI Lab

    Opsio builds the production station in the AI Lab on the real cameras, lenses, lighting, EL enclosure or batch pipeline and edge hardware. It runs against the agreed acceptance criteria using your parts or images, and Opsio prepares documentation for your quality system: model version, test results and the image-to-record trace.

    You provide
    Witnesses for the acceptance test, final sample set
    Opsio delivers
    Pre-commissioned station or pipeline, test records, documentation
  5. 5

    Installation at the customer location

    The pre-commissioned station is installed at your coating line, MRO bench or module line, or the batch pipeline connects to your drone and borescope uploads. It runs in shadow mode beside current inspection, then goes live with PLC, MES or asset management integration. Training and the active-learning loop start at handover.

    You provide
    Line access, network, PLC and MES contacts
    Opsio delivers
    Installation, shadow run, go-live, training, managed service

Pricing for the feasibility study, production stations and the managed service is on the AI Visual Inspection service page.

AI visual inspection for Aerospace & Energy: questions

What is delamination in composites?

Delamination is the separation of plies in a laminated composite, usually below an intact surface. It comes from impact, poor consolidation, contamination between plies or cure problems. Because it is sub-surface, a camera on the surface rarely sees it directly; flash thermography and ultrasonic C-scan are the standard methods. Opsio's model can classify those thermography and C-scan images to rank indications, while your qualified NDT inspector makes the disposition.

What tool do you use to detect delamination?

Ultrasonic testing and active thermography are the standard tools for delamination detection. Tap testing is still used on thin skins, and shearography on some honeycomb structures. Visual AI adds value by classifying the thermography frames and C-scan maps these instruments produce, so inspectors see the likely delaminations first. It also catches surface signs such as edge splits and resin cracks. It does not replace the instrument or the qualified operator.

What causes leading edge erosion?

Leading-edge erosion is caused by repeated impact of rain, hail, sand and insects at high blade tip speeds. Offshore and coastal sites with heavy rainfall see it fastest, often within a few years of installation. It starts as pitting of the coating, then exposes and strips the laminate. Drone images analysed campaign by campaign show how fast each blade erodes, so you can schedule protection or repair before output drops.

What is borescope inspection in aviation?

Borescope inspection in aviation is a visual check of engine internals through inspection ports with a rigid or flexible camera probe. Technicians look for cracks, burns, coating loss, nicks and foreign-object damage on compressor and turbine blades and combustor liners. Opsio's AI analyses the recorded video after the inspection, flags frames with likely damage and logs them against the engine serial. The certifying technician reviews every flag.

How do you perform a borescope inspection?

You insert the probe through an inspection port, rotate the engine slowly by hand or with a turning tool, and record each blade stage. The procedure follows the engine manual for ports, stages and limits. With AI support the workflow stays the same; you upload the video afterwards and receive flagged frames, blade counts per stage and indication sizes within minutes, ready for the technician's review and sign-off.

How do you test a solar panel?

You test a solar panel in production with an IV flash test for power and an electroluminescence test for cracks and inactive areas. EL inspection forward-biases the module in a dark enclosure and images the near-infrared emission. Visual inspection checks the frame, glass and backsheet. AI grades the EL image cell by cell at line takt, so every module is judged by the same rule on every shift.

How are wind turbines inspected?

Wind turbines are inspected by drone, rope access, ground-based cameras and internal blade crawlers, with drones now the most common external method. A typical blade inspection captures several hundred close-range images per turbine. The images are processed offline after the flight: AI detects erosion, cracks, lightning damage and coating loss, grades severity and compares each finding with the previous campaign for your engineers to approve.

Can a drone inspect wind turbines?

Yes, a drone can inspect the external surfaces of wind turbine blades, nacelle and tower. It captures high-resolution RGB images, and often thermal images, along a repeatable flight path while the turbine is stopped. Drones cannot see inside the blade, so internal bond-line and spar-cap checks still need crawlers or rope access. AI processing turns the drone images into a ranked, geo-tagged findings list.

What does delamination look like?

On a composite surface, delamination often looks like nothing at all, or like a slight bulge, a whitened area or a split at an edge or hole. Under flash thermography it shows as a hot or cold patch with a sharp boundary. Training a model therefore needs real examples: typically 10 to 30 confirmed delaminations per part family, with the matching thermography or C-scan image.

What is FPI in NDT?

FPI is fluorescent penetrant inspection, an NDT method in which a fluorescent dye enters surface-breaking cracks and glows under UV-A light after developer is applied. It is mandatory for many engine parts. A UV camera with AI can rank and measure FPI indications and document them per part, but the FPI inspection itself is performed and dispositioned by personnel qualified to EN 4179 or NAS 410.

What are the four types of NDT?

Four widely used NDT methods are visual testing, penetrant testing, ultrasonic testing and radiographic testing, with eddy current and magnetic particle also common. In aerospace, thermography and computed tomography (CT) are often added for composites and castings. AI visual inspection supports visual testing directly and helps analyse images from penetrant, thermography and CT, but it is a tool for qualified inspectors, not a separate NDT method.

Do solar panels need to be inspected?

Yes, modules sold in most markets must be type-tested to IEC 61215 for performance and IEC 61730 for safety, and manufacturers inspect production to keep that qualification valid. EL and flash testing are standard end-of-line checks. In the field, EL and thermal drone surveys find cracks and hotspots after transport and installation. AI keeps grading consistent and keeps a per-module image record.

What are the power line inspection requirements?

Power line inspection requirements are set by national grid regulation and each operator's maintenance policy, usually as fixed patrol intervals plus inspections after storms. They typically cover insulators, conductors, fittings, towers and vegetation clearance. AI processing of drone imagery supports these regimes by grading findings consistently and keeping geo-tagged evidence, while your engineers set intervals and approve every repair decision.

How much does wind turbine blade inspection cost?

Cost depends on the number of turbines, the capture method and how much image review you hand over, so Opsio quotes after a feasibility study on your own images. The study uses an existing campaign from your fleet and shows per-defect results before any commitment. You then receive a fixed quote for batch processing and the managed service; general pricing is on the AI Visual Inspection service page.

Book a 20-minute walkthrough for aerospace & energy

Pick the defects that cost you most, and we will show you live how a vision model finds them on a line like yours.

  • Live on your defect type, no slides
  • A reply within one business day
  • No obligation
Your call is with
  • Vaishnavi ShreeDirector & MLOps Lead
  • Praveena ShenoyCountry Manager, India

We use your details to answer your request, as described in our Privacy Policy.

Other industries