Artificial Intelligence Pipeline for Infrared Thermography, Ultrasonic, Visual and other Unconventional Data for Non-Destructive Inspection of Aircraf

Aircraft inspection, particularly of the exterior and through non-destructive methods, presents unique challenges due to the extensive surface areas and the complex structure and composition of aircraft. Factors such as battle damage and flight patterns, can present distinct challenges that require tailored maintenance protocols.  Artificial Intelligence (AI) techniques can significantly enhance the efficiency of these maintenance processes. These inspections involve various data modalities, including visual, ultrasonic, infrared thermography, and other unconventional data types. This diverse data can be processed to train computer vision and other multi-modal models. We propose a unified pipeline capable of processing and encoding this data, training individual models for each data type, and integrating them into an ensemble to detect and classify defects. The data pipeline includes techniques for data formatting and optional augmentation. The model pipeline utilizes a computer vision model based on the YOLO (You Only Look Once) single-stage architecture, incorporating Archetype AI, Inc.’s Newton Physical AI Model. Newton combines sophisticated encoder/decoder architectures for various sensor modalities, and creates a universal embedding space for fused understanding. Finally, the inference pipeline deploys edge-optimized models on commodity hardware, such as laptops, smartphones, and on-premise servers. This proposal covers technology area of interest #3: Emerging Technologies for Autonomous Aviation Maintenance.