Digital optics and machine learning algorithms for aircraft maintenance

Abstract. The objective of this study is to present a novel approach for airplane inspection to identify damages on the fuselage. Machine learning algorithms in the aeronautic industry can be used as an instrument for automating the process of inspection and detection, decreasing human error, and increasing productivity and security for operators. An overview of the problems, methods, and recent developments in the field of deep learning algorithms used for general damage detection on aircraft components is provided in this extended abstract. Data were collected using a high-quality acquisition system and the dataset was populated by collecting defect images from 2 typologies of aircraft: a commercial partial full-scale airplane fuselage section in primer paint and a general aviation fuselage white painted, both situated in the laboratory of heavy structures of University of Naples Federico II. The Convolutional Neural Networks (CNNs) and machine learning models were trained on large datasets of annotated images, enabling them to learn complex features associated with different types of damage. Data augmentation techniques are adopted to add diversity to the training data. Transfer learning techniques, which leverage pre-trained models on large-scale image datasets, have also proved to be effective in achieving accurate and robust detection results.