When a gas turbine bucket cracks mid-shift, the plant loses generation, the outage crew loses the next 96 to 240 hours of restoration work, and the parts department gets a phone call for a single Stage 1 blade that often costs more than the entire inspection system that could have caught the failure. The really painful part is that the crack was almost certainly visible in the last borescope walkthrough, buried in three hours of curved-mirror footage that a fatigued inspector at 2 AM has no realistic chance of parsing frame by frame. Blades sitting in your hot gas path right now carry nicks, coating chips, and leading-edge erosion patterns that a trained vision model would flag inside a minute. AI-enhanced borescope inspection rewrites the arithmetic: every frame is compared against a library of confirmed damage signatures, and every defect is measured, classified, and mapped back to the exact rotor and airfoil position it came from. You can book a demo to see how iFactory analyzes borescope footage at hot-gas-path resolution.
TURBINE BLADE AI VISION · BORESCOPE ANALYTICS · HOT GAS PATH · POWER PLANT
Detect Erosion, FOD, and Coating Damage on Every Blade Without Pulling the Turbine
iFactory's vision engine analyzes borescope video frame by frame, isolates every airfoil in the shot, and measures leading-edge erosion, foreign object damage, TBC spallation, and cracking against a consistent standard — replacing inspector-to-inspector variability with defect-grade telemetry your engineers can trend.
Defects Detected
Leading Edge Erosion
FOD Nicks and Tears
TBC Spallation
Base Metal Cracking
Oxidation Patterns
Cooling Hole Blockage
Tip Curl
Sulfidation
FOUR DEFECT CATEGORIES
The Four Damage Categories That Define Turbine Blade Life
Every borescope finding on a turbine blade falls into one of four dominant categories, each with its own physics, its own progression rate, and its own repair disposition. iFactory's vision models are trained on annotated datasets covering all four so the system can classify what it sees rather than just flag that something is different from baseline.
01
Leading Edge Erosion
Progressive material loss on the airfoil leading edge caused by particulate impingement, moisture droplets in the intake air, and combustion product carryover. Erosion blunts the aerodynamic profile, drops compressor efficiency, and creates stress concentration sites that accelerate downstream cracking.
ProgressionGradual over thousands of fired hours
02
Foreign Object Damage
Discrete impact damage from ingested debris, upstream component fragments, or dislodged internal deposits. Field studies split FOD roughly 42 percent nicks, 38 percent dents, and 20 percent tears, and the sharp corners at damage sites are the initiation points for downstream fatigue cracks.
ProgressionSudden onset, rapid crack propagation risk
03
Coating Damage and Spallation
Loss of thermal barrier coating exposes bond coat and base superalloy to combustion gas temperatures approaching 900 degrees Celsius. TBCs are designed for 48,000 hours but plants routinely realize only 40 percent of that when spallation, chipping, and delamination go undetected between outages.
ProgressionAccelerated by every start-stop thermal cycle
04
Base Metal Cracking
Fatigue and creep-driven crack initiation at leading and trailing edges, weld heat-affected zones, and cooling hole edges. Cracks may start as micron-scale features invisible during a passive borescope walkthrough and propagate to through-thickness failure within a few thousand fired hours if left untracked.
ProgressionSub-critical growth then rapid final failure
SEVERITY ESCALATION
The Escalation Ladder From Cosmetic Nick to Forced Outage
Turbine blade defects do not stay static. Every finding sits on a severity ladder, and the difference between a cheap blend repair during a scheduled outage and a seven-figure emergency parts order is often just two or three cycles of continued operation. AI vision categorizes every defect on this ladder automatically, giving the reliability team a clear disposition instead of a subjective judgment call.
L1
Cosmetic Wear
Sub-millimeter nicks on trailing edges, minor coating scuffing outside high-stress zones, superficial discoloration. Disposition is log and monitor at the next scheduled outage.
No repair cost
L2
Blend Repair Range
Nicks with sharp corners, coating chips 5 to 10 millimeters exposing bond coat, leading-edge erosion at intermediate depth. Blade is dressed in place, corners rounded to eliminate stress raisers.
Repair during outage
L3
Refurbishment Threshold
Deep FOD damage, significant TBC spallation with bare base metal, measurable creep swelling. Blade is removed from service, sent to strip-and-recoat, and reinstalled or replaced from the spare rotation.
Blade rotation cost
L4
Immediate Replacement
Through-thickness cracks, tip curl over the shroud, TBC delamination on more than 50 square millimeters at the leading edge. Blade cannot return to service; new or refurbished replacement installed before next start.
$50K to $200K per blade
L5
In-Service Failure
Missed defect propagates during operation. Blade liberates and causes cascading damage to downstream stages, casings, and diffuser sections. Emergency outage triggered, parts sourced under duress.
$2M to $10M event cost
BLADE ANATOMY MAP
Where Damage Concentrates on a Turbine Airfoil
Damage is not random. Certain zones on the airfoil see specific failure modes over and over because of local aerodynamic loading, cooling geometry, and thermal boundary conditions. iFactory's segmentation model localizes every defect to the correct zone before classifying it, producing a spatial defect map for each blade rather than a bag of unlocated findings.
Blade Tip
Tip curl from thermal deformation, tip-shroud rubs, oxidation crust from hot gas leakage across the tip clearance.
Leading Edge
Particulate erosion, FOD impact nicks, TBC spallation from thermal shock, corrosion pitting under deposit accumulation.
Airfoil and Cooling Holes
Cooling hole blockage from deposits, discoloration indicating cooling circuit issues, base metal oxidation, sulfidation attack.
Trailing Edge
Fatigue crack initiation at cooling slot edges, tear damage from downstream FOD, thin-section overheating and burn-through.
Platform
Hot gas ingestion damage, platform edge cracking, wear from mating blade contact, coating loss at gas path fillets.
Root and Fir Tree
Fretting wear at attachment surfaces, low-cycle fatigue at fir tree serrations, corrosion at disk-blade interface.
STAGE-BY-STAGE PROFILE
Different Turbine Sections Fail Through Different Physics
A single blade damage model cannot serve every stage of the machine. Compressor blades, high-pressure turbine buckets, and low-pressure turbine airfoils each experience distinct dominant loads, and their defect profiles reflect that. iFactory ships dedicated model weights per section, calibrated to the defect signatures and geometries characteristic of each.
Compressor Stages
Ambient to 400 degrees C, high mass flow, particulate-laden intake air, thousands of blades across many stages
Leading edge erosion from intake particulates
FOD from filter breaches and internal debris
Tip clearance rubs and abradable coating loss
Corrosion pitting at low compressor stages
Salt deposition in marine and coastal service
AI Model Focus
Aerodynamic profile change quantification, FOD localization across long blade rows, salt deposit pattern classification.
High-Pressure Turbine
900 degrees C hot gas path, single-crystal superalloy blades, TBC-coated hot section, film-cooling networks
Thermal barrier coating spallation
Bond coat oxidation and TGO growth
Creep-driven bulging on Stage 1 buckets
Cooling hole blockage from deposits
Sulfidation attack in high-sulfur fuel service
AI Model Focus
TBC coverage grading, cooling hole open-area estimation, discoloration classification for cooling circuit health.
Low-Pressure Turbine
Intermediate temperature, larger blades with longer aspect ratio, higher vibratory stress, load-cycling exposure
Trailing edge fatigue crack initiation
Erosion from moisture droplets in last stages
Tip cover and shroud interaction wear
Corrosion in cold-end condensing zone
Vibratory fatigue from bucket resonance
AI Model Focus
Crack initiation detection at sub-pixel scale, moisture erosion pattern mapping, tip cover interaction wear grading.
See Every Defect Category Analyzed on Your Own Borescope Footage
Upload a recent hot gas path inspection video and iFactory returns a graded defect map by rotor position within hours instead of weeks. Book a demo to walk through the workflow on a sample dataset from a comparable unit.
THE INSPECTION LOOP
How Borescope Footage Becomes a Graded Defect Map
The workflow is simple to describe and rigorous to execute. Raw borescope video comes in, structured inspection intelligence comes out. Every stage in the loop has a defined role, and the intermediate outputs are auditable so plant reliability engineers can trace any defect back to the frame it was detected in.
A
Ingest
Borescope video is uploaded directly from the inspection tool. iFactory tags every frame with stage, blade position, access port, and outage identifier so context travels with the imagery from first ingest.
to
B
Localize
A segmentation model isolates each individual blade or vane from the background, separating the airfoil under review from casing structure, lighting artifacts, and reflection patterns that routinely confuse manual video review.
to
C
Classify
Defects on each isolated airfoil are classified into erosion, FOD, coating damage, cracking, oxidation, and cooling hole categories. Every classification carries a confidence score and a spatial mask so severity can be computed geometrically.
to
D
Report
Structured defect report generated per rotor: coating loss area, crack length, pitting count, and erosion depth measured against a consistent standard for every frame, delivered as a graded map with disposition recommendations.
DETECTION ACCURACY
Detection Performance by Defect Category
The chart below reflects benchmarked detection accuracy on labeled borescope validation sets across gas and steam turbine deployments. Individual results at your site depend on borescope resolution, illumination quality, and the range of representative defect examples included during the initial calibration phase.
Oxidation and Discoloration
DEPLOYMENT OUTCOMES
What Plants Get in the First Twelve Months of Operation
The figures below reflect aggregated results from iFactory turbine blade vision deployments on gas and combined-cycle units during the first year of production use. Numbers represent typical performance rather than a specific site guarantee.
3 hrs to 20 min
Borescope Review Time Cut
Per-outage engineering time spent reviewing raw borescope footage compressed from a full shift to a supervised acceptance step on the model output.
100 percent
Airfoil Coverage Per Outage
Every accessible blade in every inspected stage is measured against the same standard rather than the subset a fatigued inspector had time to focus on.
$1M plus
Coating Life Extension Value
Value unlocked per blade row when TBC replacement is deferred toward its 48,000-hour design life instead of the 40 percent typically realized.
24 hours
Report Turnaround Target
Documented graded defect findings delivered to the plant engineering team within one working day of the borescope video being uploaded.
FREQUENTLY ASKED QUESTIONS
Questions Plant Reliability and Turbine Owners Ask
Does iFactory work with the borescope hardware we already have, or do we need to buy new inspection equipment?
iFactory analyzes standard video output from any commercial borescope currently used for hot gas path inspections. There is no requirement to replace inspection hardware, retrain field crews on a new tool, or change how the physical inspection is conducted in the plant. The system accepts recorded video files or live streams from the borescope, tags every frame with contextual metadata such as stage and access port, and returns a graded defect report against the incoming footage. Most plants deploy iFactory in parallel with existing inspection workflows for the first two outage cycles and then transition to using the AI report as the primary record.
Book a demo to walk through the ingest process with your borescope model.
How does the system distinguish acceptable operational wear from actual defects that need action?
Every defect is measured against a plant-specific acceptance standard that is loaded during the initial calibration phase. The standard defines what counts as acceptable wear, blend-repair range, refurbishment threshold, and immediate replacement for each stage and blade position. When the vision model detects a defect, it computes measured dimensions and grades severity against these thresholds rather than making a qualitative call. The result is a disposition that maps directly to what the reliability team would decide with full information, applied consistently across every blade in every stage rather than varying by which inspector reviewed the footage.
Contact support to discuss the acceptance criteria for your specific fleet.
Can the models detect very small defects like sub-millimeter nicks or micro-cracks that human reviewers routinely miss?
The models are trained on annotated imagery that includes defects down to the sub-pixel scale where micro-cracks and small nicks live. Sub-pixel edge detection is combined with feature enhancement filters that highlight low-contrast damage against the surrounding airfoil surface. In practice this means the system detects hairline cracks and small FOD nicks that a fatigued inspector would not catch on a first-pass video review, without generating the false-positive rate that would swamp a plant engineering team. Detection sensitivity is tunable during calibration so the plant can decide how aggressive the model should be for each stage.
Book a demo to see sub-pixel defect detection on a sample dataset.
How is the inspection data secured, and does anything leave the plant network?
iFactory offers on-premise deployment where all image processing, model inference, and inspection data storage stay inside the plant network on ruggedized edge hardware. No borescope footage or defect metadata is transmitted to external cloud servers, which satisfies the strict data residency and network isolation policies enforced by NERC-registered generators and industrial operators. Model updates are delivered via secure local network transfer following utility change-management protocols. For plants without strict cloud restrictions, a hosted option is also available.
Contact support to discuss the deployment model for your cybersecurity posture.
What happens the first time we introduce a blade design or coating type the model has not seen before?
The base models are pre-trained on annotated imagery covering the dominant gas and steam turbine platforms in service across the US utility fleet, so the system delivers usable defect detection for common blade geometries and TBC systems from the first inspection. When a genuinely new blade design or coating is introduced, iFactory runs a short calibration phase where a small volume of borescope footage from the new part is used to fine-tune the model against ground-truth annotations from the plant engineering team. This calibration usually takes one to two outage cycles and does not require the plant to stop using the system in the meantime.
Book a demo to discuss calibration for your specific turbine fleet.
Stop Losing Blade Life to Inspection Footage No One Has Time to Read
iFactory turns raw borescope video into a graded defect map you can act on the same day the outage crew comes off shift. Book a demo to see the analysis running on a real hot gas path inspection dataset.
