Drone Aircraft Exterior Inspection

Our 'droneInspect' project focused on improving aircraft inspection efficiency through Artificial intelligence and simulated data. The aim was to employ drones equipped with high-definition cameras, to detect damage during inspections, particularly in hard-to-reach areas.

Our project leveraged Unreal Engine 4 (UE4) and Microsoft AirSim to simulate drone flights accurately, considering factors such as cameramovement, reflections, and shadows. A realistic hangar scene with an A320 aircraft was built, introducing variations in lighting conditions to enhance the diversity of training data. To simulate damage, common defect types like scratches, dents, and lightning strike burn marks were virtually applied to the aircraft using a combination of real-world data and procedural generation. The use of Materialize and API in Blender generated realistic-looking defects onthe plane's surface. Using RTX2080Ti GPUs, the team rendered hundreds of images per light/defect position variation, producing a comprehensive dataset for training. Object segmentation images were created simultaneously, eliminating the need for manual labeling.

The simulation setup was not only able to defect >95% of damage typesin the simulated environment but also eliminated the need for time-intensive manual labeling, a common bottleneck in real-world data preparation. The labeled images can then serve as input for a MaskRCNN detection model, trained to identify defects with high accuracy. The experience gained in data simulation from'droneInspect' was instrumental in increasing the speed and accuracy of computer vision applications. As a result, this approach has broader implications for future maintenance and inspection-related projects, showcasing its potential to revolutionize the aircraft inspection process and contribute to the advancement of AI driven solutions.