Military Aircraft Surface Defect Detection Using Machine Learning

This paper presents a novel approach to automating aircraft surface inspection by leveraging deep learning techniques for the classification of thirteen different types of defects across the entirety of the aircraft exterior. Building upon existing research, which has primarily focused on specific areas of the aircraft, this study extends the scope to encompass comprehensive defect detection across the entire exterior surface. A dataset comprising labeled defect images, encompassing categories such as dents, scratches, radome damage, air intake damage, various types of rivet damage, and surface corrosion, was collected for training purposes. Google Vertex AI, an AutoML platform, was employed to train deep learning models for image classification of the 13 defect categories. Additionally, detailed literature review was conducted, exploring the utilization of deep learning models including YOLO, GAN, and CNN for detecting defects on aircraft surface. The results demonstrate the feasibility of fully automating aircraft inspection using either unmanned aerial vehicles (UAVs) or static cameras, based on the developed deep learning defect classification models. This improvement holds significant promise for enabling more frequent and detailed inspections, thereby enhancing aircraft safety and maintenance practices. The 13 defect categories cover a wide spectrum of potential issues, showcasing the versatility and effectiveness of the trained models.