Towards an Intelligent Aircraft Surface Inspection System: Image-Based Recognition and Classification of Structural Defects

¹ CTI Aeroespecial, Av. Dom Afonso Henriques - Palácio Vila Flor Nr. 701 4810-431 Guimarães, Portugal
² Chair in Aerospace Science and Technology CEIIA, University of Evora, Largo dos Colegiais 2, 7004-516 Évora, Portugal
³ Instrumentation and Control Laboratory, Center for Sci-Tech Research in Earth System and Energy - CREATE, Escola de Ciências e Tecnologia, University of Évora, Largo dos Colegiais 2, 7004-516 Évora, Portugal

Abstract

With the growth and modernisation of the air transport sector and the aviation industry, ensuring the structural integrity of aircraft has become paramount for maintaining aviation safety, minimising unexpected downtime, and adhering to regulatory standards. Surface anomalies, such as corrosion, cracks, and paint degradation, can significantly affect an aircraft's airworthiness. This study proposes a detailed analysis of various surface defects, classified by type, affected area, and material, to develop a consistent Classification Catalogue. This catalogue will serve as the foundation for creating a structured findings database using Roboflow, which will be employed to train models for the automatic detection and classification of aircraft surface defects using advanced image processing and machine learning techniques, with a particular focus on convolutional neural networks (CNNs). The key innovation of this research lies in the development of a validated image dataset structured around a comprehensive and standardised Defect Classification Catalogue. This catalogue has been compiled through a rigorous review of scientific literature and industrial standards, ensuring terminological consistency, improving annotation accuracy, and facilitating effective model training. A methodological pipeline is introduced, which includes: 1) Image acquisition from open-source databases, industrial archives, and controlled field inspections, 2)Data augmentation techniques (e.g., rotation, noise injection, and cropping) to improve model generalisability, 3) Manual annotation guided by the established catalogue, and 4) Expert validation to ensure the precision and completeness of annotations. To support defect traceability and facilitate integration into predictive maintenance workflows, a centralised platform, the Aircraft Inspection Repository (AIR), has been developed. AIR enhances accessibility, supports interactive data visualisation, and provides access to inspection data along with their corresponding reports, thereby promoting dynamic decision-making and long-term structural monitoring.