Skip to content
Aviation AI

Borescope Inspection: How AI Visual Inspection Flags Defects

Borescope inspection with AI visual inspection: how models flag cracks, burns and coating loss in engine images so inspectors review the right frames first.

Borescope inspection is how maintenance teams look inside an engine without taking it apart, using a camera probe to examine blades, vanes and combustor parts for cracks, burns, nicks and coating loss. This guide explains how AI visual inspection supports borescope inspection by flagging likely defects in the video so inspectors review the right frames first, which engine makers already run it, and what image data and labelling a shop needs to build its own models.

A borescope inspection produces a lot of footage. An inspector feeds the probe through an access port, rotates the engine stage by stage and examines every blade, often across several stages and both engines on a twin. The work demands steady concentration, and the findings decide whether an engine stays on wing, gets a repeat inspection at a shorter interval, or comes off for a shop visit.

AI visual inspection helps by doing the first pass. A computer vision model reviews the video, marks the frames and regions that look like damage, and hands the inspector a shortlist to examine against the manual limits. The inspector makes every disposition.

What a Borescope Inspection Covers

Borescope inspections run on schedule as part of an engine maintenance program and on condition after events such as bird strikes, foreign object ingestion or over-temperature exceedances. The inspector compares each finding against the limits in the engine manual for that stage and part, which define what damage is acceptable, what needs a repeat inspection and what requires removal.

Typical findings fall into a handful of families: cracks, nicks and dents from foreign objects, tip curl and erosion, burns and burn-through in hot-section parts, and thermal barrier coating loss. Each family has its own look on camera, and each is measured against its own limits.

Where AI Visual Inspection Fits

AI fits at three points in the borescope workflow:

  • Guided capture. Templates guide the inspector through the stages and blade positions in a set order, so footage is consistent from one inspection to the next.
  • Automatic defect recognition. A model flags frames and regions that likely show damage, labelled by defect type, for the inspector to examine.
  • Records and trending. Findings are tied to engine serial number, stage, blade position and cycles, so the next inspection can compare like with like.

The model never makes the airworthiness call. It sorts the footage so the inspector spends time on the frames that matter, and it keeps a record of what was flagged and what the inspector decided.

Published AI Borescope Programs

Engine makers and inspection equipment suppliers have published results from AI-assisted borescope inspection. The table lists programs with named sources.

ProgramEnginesWhat the AI doesPublished result
GE Aerospace Blade Inspection Tool (October 2024)GEnx, expanding to LEAP and GE9XAI-assisted blade inspectionGEnx inspection time cut from 3 hours to 1.5
GE Aerospace LEAP deployment (February 2025)CFM LEAPBlade Inspection Tool rolled out to MRO shops and customersInspection times cut in half, per GE Aerospace
GE Aerospace and Waygate Technologies (October 2024)GEnx and LEAP high-pressure compressorAI-assisted inspection on the Mentor Visual iQ+ borescopeRecall up 33.6% and precision up 13.5% against the earlier detection model
Waygate and GE Aerospace templates (April 2026)GEnx high-pressure turbine stages 1 and 2Automated Menu Directed Inspection templatesGuided, repeatable capture for HPT blades
Pratt & Whitney with Aiir Innovations (July 2026)V2500, with pilots on GTF and F135AI borescope software from its acquisition of Aiir InnovationsInspections completed in minutes, per RTX

Rolls-Royce has also described an intelligent borescope method built with Waygate Technologies and Rhinestahl, where AI helps assess high-pressure turbine blade findings.

Wondering where AI fits your roadmap? Get a directional read in 30 minutes — no pitch, no commitment.
Book a strategy session →

What Image Data a Borescope AI Model Needs

OEM tools cover the engines they were built for. Shops that maintain other engine types, auxiliary power units or industrial derivatives often build their own models, and the data decides how well those models work.

  • Volume and coverage. Footage across every engine type, stage and blade row the model will see, captured with the probes and lighting your inspectors use.
  • Damage examples. Serious findings are rare, so a training set needs deliberate collection of past findings, including images from shop visits where the damage was confirmed at teardown.
  • Consistent capture. Guided inspection templates produce footage in a set order, which makes frames comparable across inspections and engines.
  • Metadata. Engine serial number, stage, blade position, cycles since last shop visit, probe type and inspector, stored with every clip.

Labelling Borescope Images

Labels teach the model what each defect looks like, so the quality of the labels sets the ceiling for the model. Four practices make the difference:

  1. Use the manual’s defect terms. Build the label set from the defect categories in your engine manuals, so model output maps straight onto the limits inspectors apply.
  2. Have inspectors label. Experienced borescope inspectors draw boxes or outline regions around each defect. Data annotation teams can scale the work once inspectors set and check the rules.
  3. Measure agreement. Have two inspectors label the same sample and compare. Where they disagree, refine the label definitions before training.
  4. Link labels to outcomes. Where an engine went to the shop, attach the teardown finding to the borescope frames, which gives the model confirmed examples.

How to Evaluate a Borescope Model

For a review-queue model, recall matters most: the share of real defects the model flags. Measure it per defect type and per stage on a held-out set of inspections the model has never seen, with inspector labels as ground truth. Precision, the share of flags that turn out to be real, governs how many extra frames the inspector reviews. Set both targets with your inspection team before testing, review every missed defect by hand, and rerun the evaluation each time the model or the capture process changes.

Running a Borescope AI Pilot

A pilot on one engine type and one or two stages keeps the scope small enough to measure. A practical sequence:

  1. Pick the engine and stages. Choose an engine type with steady inspection volume and stages where findings are frequent enough to learn from.
  2. Collect and label history. Gather past inspection footage with inspector dispositions and any teardown findings, and label a training set and a separate test set.
  3. Train and test offline. Train the first model and score it on the test set by defect type, with your inspection lead reviewing every miss.
  4. Run in shadow mode. Let the model flag frames on live inspections while inspectors work as usual, then compare its flags with their findings.
  5. Move to review-first. Once the shadow results meet the recall target, show flagged frames to inspectors first while they continue to review the full footage.

Each step has a clear exit decision, and the inspection procedure changes only after the team has seen the model work on its own engines.

Hardware and Deployment at the Shop

Borescope models can run on a GPU workstation or server in the shop, sized to your inspection volume. Many shops process footage after capture and return flagged frames to the inspector’s review screen, while equipment-integrated tools run on the borescope itself. Keeping training data, models and results on local hardware also simplifies OEM and customer data terms, because footage stays inside the facility.

Every flagged frame and every inspector disposition becomes a record tied to the engine serial number, stage, blade and cycles. Over time those records show how damage develops between inspections, which feeds repeat-inspection planning and predictive maintenance models that forecast removals. Images and findings often carry OEM and customer confidentiality terms, so many shops keep the models, footage and records on their own servers, as part of a wider AI in aviation program.

Want AI visual inspection for engines your OEM tools do not cover? NeuralChainAI builds custom computer vision models for borescope footage, trained on your images and run on your hardware. Book a strategy session to scope a pilot on one engine type.

Frequently Asked Questions on Borescope Inspection

A borescope inspection uses a camera probe inserted through access ports to examine the inside of an engine without disassembly. Inspectors check blades, vanes and combustor parts for cracks, nicks, dents, burns, erosion and coating loss, then compare each finding with the limits in the engine manual. Inspections run on schedule and after events such as bird strikes.
AI reviews the inspection video and flags frames and regions that likely show damage, labelled by defect type, so the inspector examines those first. GE Aerospace reported that its AI Blade Inspection Tool cut GEnx inspection time from 3 hours to 1.5. Guided capture templates and automatic defect recognition also make inspections more consistent.
The inspector makes every disposition. The model sorts footage and suggests where damage is likely, and the inspector measures each finding against the manual limits and decides whether the engine stays on wing, needs a repeat inspection or comes off. Both the model's flags and the inspector's decisions are recorded.
You need footage for each engine type and stage the model will see, enough confirmed damage examples, and metadata such as engine serial number, stage, blade position and cycles. Inspectors should label defects using the categories in your engine manuals. Teardown findings from shop visits make the strongest training examples.

Leave a Comment

Stop Guessing Whether AI Fits Your Problem.

30 minutes with a senior consultant. Walk away with a one-page scoping summary either way.

Book Your Session
Discuss your Aviation AI project Discuss your project