Semi-automated procedure for damage detection, localization, and visualization on aircraft fuselages

This study presents a semi-automated inspection framework for aircraft fuselage damage detection and localization based on deep learning and 3D reconstruction. The inspection and detection processes are automated by utilizing computer vision algorithms, reducing human error while enhancing productivity and safety. The research provides an overview of the challenges, methodologies, and recent advancements used to identify damage to various aircraft components using autonomous systems. Data collection was performed using a high-quality acquisition system, which gathered images from a commercial, partially full-scale aircraft fuselage section in primer paint. Convolutional neural networks (CNNs) and machine learning models were trained on an extensive dataset of annotated images, enabling them to recognize complex patterns and features associated with different types of damage. Data augmentation techniques were employed to diversify the training data, and fine-tuning, which utilizes pre-trained models on large-scale image datasets, has proven effective in achieving accurate and robust detection results. The data obtained are subsequently used to generate a 3D scene capable of visualizing the defects on the fuselage skin and performing a preliminary assessment of the damages found, as well as obtaining a visual location of the defect and its position relative to the most significant surrounding structural parts.