AI-assisted defect recognition (ADR) layered onto video borescope software is a machine-vision workflow that automatically detects, classifies, and documents surface defects during remote visual inspection, then writes the result back to the inspector's report [S1]. The first commercial deployment tied to a major engine OEM was the GE Aerospace / Waygate Technologies joint program announced on October 28, 2024, targeting high-pressure compressor (HPC) inspections on GEnx and CFM LEAP engines [S2].
Quantitative gain on that program: recall improved by 33.6% and precision by 13.5% versus Waygate's prior Gas Power-assist ADR model version 4.1, with false alerts reduced by more than 13% [S2][S6]. The model was trained on a curated dataset of thousands of representative HPC images sourced from GE Aerospace's Services Technology Acceleration Center (STAC) and labelled by GE Aerospace Research subject-matter experts [S2].
What AI-assisted defect recognition actually does inside a video borescope
ADR runs on the borescope's on-board compute or a paired PC, analysing live video frames and stills from the insertion probe to flag candidate defects such as cracks, corrosion, FOD impact damage, and burn marks before the human inspector finishes the pass [S1]. Waygate's Everest Mentor Visual iQ (MViQ) platform has carried ADR since the MViQ OS 3.6 release on February 10, 2022, with delivery via over-the-air (OTA) update or the InspectionWorks Store, followed by a 90-day full-feature trial on connected devices [S4].
Evident's ViSOL RVI software suite, supporting IPLEX NX, IPLEX GX/GT, IPLEX G Lite, and IPLEX GAir videoscopes, exposes the same AI-assisted defect detection function alongside 3DAssist photogrammetric reconstruction, structured workflows, and cloud-based collaboration through Evident Connect [S5]. Both vendors treat ADR as one module inside a broader RVI pipeline that also handles measurement, image management, and automatic report generation, which is the standard scope of any industrial borescope software stack today [S1][S5].
Selection criteria: model quality, deploy footprint, and data ownership
Three engineering criteria decide whether an AI-ADR feature is worth turning on. First, the precision/recall curve on the exact component family the inspector is probing: 33.6% recall and 13.5% precision gains on HPC stages [S2][S6] do not automatically transfer to gas-path combustor liners or industrial turbine HPT shrouds, so each fleet needs its own validation pass. Second, deploy footprint: Waygate runs ADR on-device inside the MViQ video borescope and pushes model updates OTA [S4], while Evident's ViSOL exposes defect detection through the cloud-linked desktop suite [S5]; the former is preferable for shop-floor MRO without reliable Wi-Fi, the latter for enterprise report workflows. Third, data ownership: joint programs where the OEM supplies labelled training video (as GE Aerospace did with STAC [S2]) produce models that reflect real engine wear patterns, not synthetic defects.
A practical comparison matrix for procurement evaluation:
Criterion A: training data origin. GE-supplied engine videos vs vendor-curated image library. A wins where the asset is the OEM's own engine. Criterion B: model update channel. OTA on-device (Waygate MViQ OS 3.6 [S4]) vs cloud-managed software release (Evident ViSOL [S5]). A wins for air-gapped MRO cells, B wins for multi-site fleets. Criterion C: false-positive tolerance. AI-ADR does not eliminate human sign-off; the 13% false-alert reduction reported in 2024 still leaves a non-zero residue that an inspector must triage [S2][S6]. Criterion D: defect coverage. AI ADR is best documented for HPC blade and vane inspection; coverage of casting surfaces, pump volutes, and gearbox bores is earlier-stage and vendor-specific [S1][S3].
Who benefits, and where the technology does not yet fit
Operators running scheduled borescope inspections on high-cycle gas-path hardware, including commercial MRO shops handling GEnx, CFM LEAP, and similar high-bypass turbofans, see the most immediate return because HPC inspections are both the highest-volume and the most time-consuming shop task [S2][S3]. Independent power-plant and aero-derivative engine MROs handling industrial gas turbine hot-section borescope work are a close second, since the same defect taxonomy (leading-edge FOD, thermal barrier coating spallation, creep-driven cracking) carries over [S3].
Where AI-ADR is the wrong tool: low-volume, single-event inspections where the model has no training overlap, ultra-high-temperature components where probe insertion windows are seconds rather than minutes, and any regulated sign-off where the inspector cannot demonstrate they personally confirmed each defect call. Foundries and casting houses doing first-article inspection on novel alloys should treat AI-ADR as a screening layer, not a release gate, because the training set is dominated by in-service wear imagery rather than as-cast surfaces [S3][S7].
Operational impact on the inspection workflow
The 2024 GE/Waygate program reduced HPC inspection time per engine by an undisclosed but described as "significant" margin, while improving detection capability at the same accuracy target [S2]. That time saving compounds across MRO shop throughput: a single HPC borescope that previously required a senior inspector for the full visual sweep can be pre-screened by AI, with the human reviewer focused only on flagged frames and ambiguous regions [S1][S6].
Side effects of the workflow shift are real. First, the inspector role moves from full-pass operator to triage-and-confirm reviewer, which changes training requirements: technicians now need to interpret AI confidence scores, reject false positives, and feed corrected labels back into the model when permitted [S1]. Second, report generation becomes largely automatic, with ADR-detected defect metadata, severity tags, and image clips pre-populated into the inspection record, then pushed to InspectionWorks (Waygate) or Evident Connect (Evident) for archival and trend analysis [S4][S5]. Third, model versioning matters: the published 33.6% recall gain is specific to a successor of the Gas Power-assist model 4.1 [S2][S6], so shops must track which ADR build is loaded on each probe.
Failure modes and limits the data already shows
ADR false-positive rate is not zero. The Waygate/GE program reports a 13%-plus reduction versus the prior model, which is an improvement, not an elimination [S2][S6]. Lighting artefacts on polished compressor blades, oil streaks from the borescope insertion path, and probe-tip shadowing can all trigger spurious defect calls that the inspector must dismiss. Model drift is a separate failure mode: as engine fleets accumulate cycles, new wear mechanisms emerge that the original training set did not cover, requiring periodic retraining with freshly labelled field imagery [S1][S2].
Standards alignment is still being shaped. AI-ADR output is advisory; the human inspector's signature remains the regulatory artefact accepted by aviation authorities, and shops should not treat ADR confidence as a substitute for a level-2 or level-3 borescope inspector's judgement. Industrial users running pressure transmitter calibration cycles and flow meter verification on the same rotating equipment will also want ADR results cross-referenced against vibration and process data before any maintenance release decision, because the borescope sees surface damage, not bulk material degradation.
How to roll AI-assisted borescope software into a working MRO cell
A staged deployment works better than a fleet-wide switch. Step one: pilot one AI-ADR-equipped borescope on the highest-volume HPC or HPT inspection, capture both AI flags and inspector findings, and compute your own recall and precision against the ground truth your level-3 inspector sets. Step two: lock the model version, training data scope, and update cadence into the MRO quality manual, so any audit can reconstruct which ADR build produced which report. Step three: integrate the ADR export into your existing inspection data management system, whether that is Waygate's InspectionWorks platform [S4] or a third-party archival tool, and verify that image metadata, defect coordinates, and severity tags survive the round-trip. Step four: train the inspection team on AI confidence interpretation and on feeding corrected labels back, which closes the loop and protects model quality over the next fleet update cycle [S1][S2].
Trackable signals to watch next: whether the joint Waygate/GE program extends beyond GEnx and CFM LEAP to additional engine platforms, whether Evident ships a competitive on-device ADR model for the IPLEX NX line comparable to Waygate's MViQ OS 3.6 footprint, and whether any tier-1 MRO publishes independent field-data recall numbers rather than vendor-supplied figures.
For related coverage, see Squeeze Casting Wrought 6061 Aluminum: Process, Properties, Trade-offs.