AIVisualMRO is an advanced AI platform designed to help inspectors complete visual inspections more efficiently and consistently. It combines audio transcription, computer vision (YOLO), and large language models(LLMs) to support every stage of the inspection workflow while ensuring that the human inspector always maintains full control over the final decision.
AIVisualMRO can be taylored into an visual inspection assistant that sits beside the inspector, either final inspection or when the part is first recieved. Or alternatively it can be part of an automated rig that can reduce the burden of inspection through automation.
Example AI training set used for training for defect detection :
This usecase shows footage shot on mobile phone either by RMDTurbine or by other inspections and aircraft mechanics. This type of rotor disks and rotor blisks can be found on all type of turbine turbojet, turbofan or turboprop engines, from either GE, Rolls-Royce, Pratt & Whitney or CFM. In this case it was easier to found datasets and information related to CFM56 rather then the other manufacturers. There was a weath of data on used parts on an unamed auction site, which was easier to train data on.
Detection of dents, nicks and scratches in aerospace components with AiVisualMRO is a cutting-edge approach.
In this example along side detection of nicks , dents, scratches, it also checks for any missing or wrong size
bolts,
The AI is complimenting
the visual inspection, and accurate identification of damage on critical aerospace parts.
We wanted to inspect the blades (top side) and the bolts, but the same methodology to train the AI can be used on other feautures. The Bolts(topside) are easy to train on. It is easy beacause no matter the angle the mobile phone camera is pointed from the bolt is always round. The AI is looking for:
Deployment
It took around 1 day to 2 days to train for one side, with an AI model of confidence of 99% segmentation model.
This type of AI model was then deployed on the mobile devices on which the inspection can be performed and furhter
tested.
Then this was integrated into the inspection app containing further information needed for the visual inspection.
What can it be improved ?
The Dataset !! Due to the dificulty with the blades , dataset had to be manually segmented. Meaning someone had to
handraw
marking / polygones of the blades at different angles.
The variability can be reduced if teh camera is stationary at a set hight as part of a fixture, same with the
lighting if needed.
but even so teh results are very good.
Tracking ! Tracking and ID-ing each blade was challenging as the AI would re-allocate Ids to the same blades as
they would go around and comeback into view.
This was solved through coding and through counting the blades.
What else can the AI be trained on and used on ?
Rootforms inspection !!
AIVisualMRO can also inspect / can be trained on for the inspection of rootforms at the base of the blades. Here looking for dents,
discoloration, oxidation, corrosion, scratches.
It has the same difficulty level as the blades as the 'shape' changing ad different angles. Maybe a bit more difficult rootforms usually tend to be smaller then the blades.
Some aditional resources
Every company performs visual inspection differently. AIVisualMRO begins with a non-intrusive observation phase to understand your current workflow, tools, constraints, and quality criteria. Requirements are gathered collaboratively to ensure the platform fits seamlessly into your existing operations.
